DE102024101836A1 - METHOD FOR OPERATING A DATA MANAGEMENT DEVICE OF A MOTOR VEHICLE - Google Patents

Abstract

The invention relates to a method for operating a data management device of a motor vehicle, wherein
the data management device has at least the following components:
- a computer;
- at least one recording unit; and
- a vehicle-side communication interface which is in data-transmitting connection with a backend-side communication interface of a backend,
the method comprising at least the following steps:
a. by means of the at least one recording unit of the motor vehicle, recording raw data;
b. by means of the vehicle-side communication interface, receiving an order sent from the backend communication interface, which order requests at least one set of order data and includes at least one noise factor;
c. by means of the computer, noise-generating a set of raw data selected on the basis of the order data requested in the order on the basis of the at least one noise factor in a noise function to form a set of noise data; and
d. using the vehicle-side communication interface, sending the set of order data requested in step b. on the basis of the data obtained in step c.
noisy noise data to the backend communication interface.
The method proposed here for operating a data management facility ensures compliance with privacy costs.

Description

The invention relates to a method for operating a data management device of a motor vehicle, as well as a method with such a method for operating a model data platform for motor vehicles.

Generative machine learning models in motor vehicles are used, for example, for autonomous vehicle control, image recognition, speech recognition and other functions.

Large data sets are required to develop and train such machine learning models. Providing sufficient real-world data for such models requires extensive data collection and processing. This raises data privacy concerns, as, for example, driving behavior and location data of drivers are collected, processed, or shared between different market participants.

Based on this, the present invention is based on the object of at least partially overcoming the disadvantages known from the prior art. The features of the invention arise from the independent claims, for which advantageous embodiments are presented in the dependent claims. The features of the claims can be combined in any technically reasonable manner, whereby the explanations from the following description and features from the figures, which comprise additional embodiments of the invention, can also be consulted for this purpose.

• - einen Computer;
• - zumindest eine Erfassungseinheit; und
• - eine kraftfahrzeugseitige Kommunikationsschnittstelle, welche in datenübertragender Verbindung mit einer backendseitigen Kommunikationsschnittstelle eines Backends steht,

wobei das Verfahren zumindest die folgenden Schritte umfasst:

1. a. mittels der zumindest einen Erfassungseinheit des Kraftfahrzeugs, Erfassen von Rohdaten;
2. b. mittels der kraftfahrzeugseitigen Kommunikationsschnittstelle, Empfangen eines von der backendseitigen Kommunikationsschnittstelle versandten Auftrags, welcher zumindest einen Satz an Auftragsdaten anfordert und zumindest einen Rauschfaktor umfasst;
3. c. mittels des Computers, Verrauschen eines auf Basis der in dem Auftrag angeforderten Auftragsdaten ausgewählten Satzes an Rohdaten auf Basis des zumindest einen Rauschfaktors in einer Rauschfunktion zu einem Satz an Rauschdaten; und
4. d. mittels der kraftfahrzeugseitigen Kommunikationsschnittstelle, Senden des in Schritt b. angeforderten Satzes an Auftragsdaten auf Basis der in Schritt c.

verrauschten Rauschdaten an die backendseitige Kommunikationsschnittstelle.The invention relates to a method for operating a data management device of a motor vehicle, wherein
the data management device has at least the following components:

• - a computer;
• - at least one recording unit; and
• - a vehicle-side communication interface which is in data-transmitting connection with a backend-side communication interface of a backend,

the method comprising at least the following steps:

1. a. by means of the at least one recording unit of the motor vehicle, recording raw data;
2. b. by means of the vehicle-side communication interface, receiving an order sent from the backend communication interface, which order requests at least one set of order data and includes at least one noise factor;
3. c. by means of the computer, noise-generating a set of raw data selected on the basis of the order data requested in the order on the basis of the at least one noise factor in a noise function to form a set of noise data; and
4. d. using the vehicle-side communication interface, sending the set of order data requested in step b. on the basis of the data obtained in step c.

noisy noise data to the backend communication interface.

Unless explicitly stated otherwise, ordinal numbers used in the preceding and following descriptions serve only to clearly distinguish them and do not reflect the order or ranking of the designated components. An ordinal number greater than one does not necessarily imply that another such component must be present.

A method for operating a data management device of a motor vehicle is proposed. The data management device is configured to provide a model, for example, a generative machine learning model, in a motor vehicle. For example, the model is used for vehicle control, image recognition, and/or speech recognition.

The data management device is further configured to communicate with a backend in order to transmit data resulting from the operation of the motor vehicle to the backend in order, for example, to further improve the models.

For this purpose, the data management device comprises a computer, at least one recording unit and a motor vehicle-side communication interface.

Preferably, the computer on board the motor vehicle is capable of independently executing the method, for example, without the use of additional resources (other than those explicitly mentioned). Alternatively or additionally, the computer or parts of the computer are arranged outside the motor vehicle in a data room of the motor vehicle. The data room of the motor vehicle is In this case, this is to be understood as meaning that, for example, the data of the motor vehicle or a driver are stored separately and protected from access by third parties in the data room.

A computer comprises, for example, one or more general-purpose processors (CPU) or microprocessors, RISC processors, GPU and/or DSP. The computer has, for example, additional elements such as memory interfaces or communication interfaces. Optionally or additionally, the terms refer to such a device that is capable of executing a provided or embedded program, preferably using a standardized programming language (for example, C++, JavaScript or Python) and/or controlling and/or accessing data storage devices and/or other devices such as input interfaces and output interfaces. The term computer also refers to a plurality of processors or a plurality of (sub-)computers that are connected to one another via physical lines and/or are otherwise communicatively connected and possibly share one or more other resources, such as data storage. A (data) storage device is, for example, a hard disk drive (HDD) or a (non-volatile) solid-state memory, such as ROM memory or flash memory [Flash EEPROM]. The memory often comprises a number of individual physical units or is distributed across a large number of separate devices, so that it is accessed via data communication, such as package data services. The latter is a decentralized solution, where the memory and processors of a large number of separate computers are used instead of a (single) central server or in addition to a central server.

• - Positionserfassungseinheit (beispielsweise Global Positioning System [GPS]);
• - Temperatursensor;
• - Geschwindigkeitsmesseinheit;
• - Beschleunigungsmesseinheit;
• - Mikrofon;
• - Kamera;
• - Lidar-Sensor;
• - Radar-Sensor; und
• - Ultraschallsensor.

The at least one detection unit comprises, for example, one or more sensors or a measuring system. For example, a detection unit comprises at least one unit from the following list:

• - Position detection unit (e.g. Global Positioning System [GPS]);
• - Temperature sensor;
• - speed measuring unit;
• - Accelerometer unit;
• - Microphone;
• - Camera;
• - Lidar sensor;
• - radar sensor; and
• - Ultrasonic sensor.

The vehicle-side communication interface is configured to enable data transmission between the motor vehicle and a backend, or a backend-side communication interface of the backend. The communication interfaces are preferably transceivers of a telematics system, for example, a mobile network (e.g., 3G, 4G, or 5G). Alternatively or additionally, a stationary data exchange, for example, via WLAN, Bluetooth, and/or Ethernet, takes place to transfer data in specified environments, such as workshops.

The procedure comprises at least the following four steps a., b., c. and d.

In step a, raw data is collected using the vehicle's data acquisition unit. The collected raw data is buffered, i.e., temporarily stored, in the data management system or further processed in real time.

In step b, an order is received via the vehicle-side communication interface. The order is previously sent via a backend-side communication interface of the backend. The order requests at least one set of order data, which the vehicle is to transmit to the backend. Furthermore, the order includes at least one noise factor, which is to be used to add noise to the data underlying the requested order data.

For example, the noise factor is an input variable for a noise function. The noise function is designed, for example, to apply classic noise consisting of a random number multiplied by the signal data. In this case, the noise factor describes the range of a random value, for example, a floating-point number between 0 and 1. Alternatively or additionally, the noise function includes an encryption technique. For example, the position of a vehicle is obscured by cutting up and removing the start and end points of a local route, so that the start and destination locations cannot be traced. For timestamps, for example, the noise function is a factor that is consistently applied to a series of points in time.

For example, the noise factor is a numerical value, for example, between 1 (one) and 10 (ten), which is used as an input to a noise function to reduce the noise in the data. For example, the corresponding noise function is provided on the vehicle side.

In step c, a selected set of raw data acquired in step a is noise-generated by the computer. The set of raw data is selected based on the order data requested in the order received in step b. The noise factor and the selected set of order data form the input variables of the noise function, which delivers a corresponding set of noise data as the output variable.

In step d., the set of order data requested in step b. is sent to the backend via the vehicle-side communication interface. For example, the set of requested order data corresponds to the set of noise data generated in step c. Preferably, however, further intermediate steps take place in which the set of noise data is further processed into the set of order data.

The collection, processing, and sharing of data necessary, for example, to create generative machine learning models may conflict with data protection interests. For example, vehicle-related data and/or personal data are subject to guidelines and/or laws, and the relevant individuals, such as drivers or vehicle owners, have an overriding interest in such data protection.

The extent to which the collection, processing, and sharing of data compromises a person's data protection interests can be quantified, for example, through privacy costs. Training generative machine learning models still requires large amounts of data from the field—that is, data actually collected during the operation of a motor vehicle. Here, it is proposed to reduce privacy costs by making the data noisy. By making the data noisy as described above, it is possible to prevent, for example, model data from being used to extrapolate back to the raw data, making the raw data—such as times, position data, weather data, audio data, driving behavior, and/or driver preferences—no longer accessible. The more noisy the data is—i.e., the higher the noise factor—the lower the privacy costs associated with the resulting data.

It is further proposed in an advantageous embodiment of the method that in a sub-step c.1, model data are generated by means of the computer on the basis of the noise data,
wherein preferably, by means of the vehicle-side communication interface
at least part of the model data is sent as order data in step d. to the backend communication interface.

According to the embodiment of the method proposed here, in a substep c.1, a set of model data is generated from the set of noisy data by means of the computer. The model data are preferably model parameters of a generative machine learning model. For example, the generative machine learning model is trained using the noisy raw data, thus adapting the model data.

For example, the requested order data includes at least a portion of the model data determined in substep c.1. Thus, at least this portion of the model data is sent to the backend communication interface in step d. via the vehicle-side communication interface.

It is further proposed in an advantageous embodiment of the method that step c.1 comprises training a generative machine learning model, and
where the model data includes model parameters of the generative machine learning model.

Here, it is proposed that in substep c.1, a machine learning model be trained based on the raw data. For example, such a machine learning model could be a model for recognizing traffic signs, other road users, objects, speech, or similar.

For example, the machine learning model includes a neural network, a decision tree, or a support vector machine.

For example, the machine learning model is validated continuously or repeatedly and the model data is adjusted.

The model data are preferably model parameters, i.e. values that define, for example, boundary conditions, functions or algorithms of the model.

• - Temperatur;
• - Ort;
• - Geschwindigkeit;
• - Abstand zu einem Objekt;
• - Lenkwinkel;
• - Beschleunigung;
• - Wetter;
• - Bremskraft;
• - Bremsbefehl; und
• - Zeit.

It is further proposed in an advantageous embodiment of the method that the raw data comprise at least one data type from the following list:

• - temperature;
• - Location;
• - Speed;
• - Distance to an object;
• - steering angle;
• - acceleration;
• - Weather;
• - braking force;
• - brake command; and
• - Time.

For example, a data type includes a plurality of measured values. For example, a data type includes the measured values of a single sensor and/or the measured values of multiple sensors. For example, a data type includes the measured values of multiple temperature sensors, preferably multiple temperature sensors of a component.

It is further proposed in an advantageous embodiment of the method that each data type of raw data, raw data which is required to determine the order data, is assigned a noise factor which is received in the order from the motor vehicle-side communication interface.

According to the embodiment of the method proposed here, a noise factor is assigned to each data type of raw data required to determine the order data. Accordingly, a plurality of noise factors, corresponding to one for each data type required to determine the set of order data, is preferably sent from the backend to the motor vehicle with the order, or received from the motor vehicle.

Different data types may have different effects on the overall privacy costs of a dataset. For example, location data generates higher privacy costs than temperature data from thermal management. For example, the accuracy of different data types is of varying importance for the respective use case, such as the generative machine learning model. For example, if the accuracy of a data type, such as the driver's driving behavior, is particularly relevant for the use case, low noise factors for the driving behavior data are transmitted in the job. If the location data is of little relevance to the use case, the noise factors transmitted in the job for the location data are high. For example, high noise factors for data types that are less relevant to the corresponding machine learning model can be compensated for by low noise factors for data types that are more relevant to the machine learning model, so that the overall privacy costs of the set of job data can be balanced out.

According to a further aspect, a method for operating a model data platform for motor vehicles is proposed, wherein
the model data platform, the data management device of at least one motor vehicle and a backend with a backend-side
communication interface,
wherein the method for operating the model data platform comprises the method for operating the data management device of a motor vehicle according to an embodiment as described above, and wherein the order received in step b. by means of the motor vehicle-side communication interface is sent by means of a backend-side communication interface of a backend.

A method for operating a model data platform for motor vehicles is proposed here. The method comprises the method described above for operating a data management device of a motor vehicle.

The model data platform comprises at least one data management device of a motor vehicle according to the above description and a backend with a backend-side communication interface. The model data platform preferably comprises a plurality of data management devices of a plurality of motor vehicles, for example, a motor vehicle fleet.

The backend comprises, for example, a physical and/or virtual server, which is set up to communicate with the motor vehicle(s) via the backend communication interface, and a memory.

The backend communication interface is configured to communicate with the vehicle-side communication interface(s).

The backend is configured to send the order, which is received via the vehicle-side communication interface of one of the motor vehicles, via the backend-side communication interface in step b. Furthermore, the backend is configured to receive the set of order data sent via the vehicle-side communication interface in step d. via the backend-side communication interface.

It is further proposed in an advantageous embodiment of the method that in the backend a relationship between noise factors and privacy costs and/or
a connection between order data and privacy costs is maintained.

According to this embodiment, a relationship between privacy costs and noise factors is maintained. Furthermore, alternatively or additionally, a relationship between order data and privacy costs is maintained. For example, the relationship between privacy costs and order data can be used to determine how high the privacy costs of a set of order data are. For example, the relationship between privacy costs and noise factors can be used to determine how much the privacy costs can be reduced using a specific noise factor. For example, one or more noise factors can be selected such that the privacy costs of the order data are below a target cost.

For example, the relationship between privacy costs and noise factors is stored in a cost catalog in the backend.

For example, the relationship between order data and privacy costs is stored in a data catalog in the backend. For example, the data catalog includes a mapping of the raw data required to determine the respective order data.

Preferably, the backend is designed to determine the noise factors by means of the explained relationships between privacy costs and noise factors or order data in such a way that the target cost value is met by the set of order data requested in the order.

For example, the target cost is predetermined based on legal regulations, agreements with the vehicle owner or driver, and/or the requirements for the requested order data. For example, at the request of the vehicle owner, location data that could provide information about the location, such as frequent start and end points of routes, is highly noisy.

It is further proposed in an advantageous embodiment of the method that in a step e. the order data of several of the motor vehicles, which were each sent to the backend in step d., are combined into an overall data set.

According to this embodiment, for example, an overall model is formed on the basis of several sets of model parameters of several motor vehicles.

For example, the machine learning models can be optimized across multiple vehicles, for example for an entire vehicle fleet.

It is further proposed in an advantageous embodiment of the method that, by means of the backend, the at least one noise factor is determined for each data type contained in the set of order data of the order as a function of an influence of the data type on a target variable.

According to the embodiment proposed here, the backend determines at least one noise factor for each data type depending on the influence of the data type on a target variable, for example, the requested order data or model parameters. For example, location data for a thermal management model is susceptible to significant noise, whereas, for example, the interior temperature and the ambient temperature are less susceptible to noise.

It is further proposed in an advantageous embodiment of the method that the relationship between privacy costs and order data and/or the relationship between noise factors and privacy costs is generated and/or adjusted by means of the backend by summing the privacy costs on the basis of the set of order data which are sent in the order.

According to this embodiment, the relationship between privacy costs and order data and/or the relationship between noise factors and privacy costs is generated and/or adjusted by summing the privacy costs based on the set of order data. In other words, after the order data or model parameters have been received on the backend, the actual privacy costs incurred are determined or summed. This allows the aforementioned relationships or catalogs to be trained. Furthermore, models or algorithms for determining the optimal noise factors can be adapted or generated, for example using artificial intelligence.

• - eine Mehrzahl an Kraftfahrzeugen; und
• - ein Backend;

wobei jedes der Kraftfahrzeuge, eine Mehrzahl an Erfassungseinheiten und einen Computer aufweist und in kommunizierender Verbindung mit dem Backend steht.According to a further aspect, a model data platform for executing a method is proposed, which has at least the following components:

• - a plurality of motor vehicles; and
• - a backend;

wherein each of the motor vehicles has a plurality of detection units and a computer and is in communicative connection with the backend.

It is further proposed in an advantageous embodiment that the plurality of motor vehicles are each designed, by means of the computer and the detection unit, to carry out step a., step b. and step c., and the backend is designed to carry out step b.

• 1: ein Prozessablauf eines Verfahrens zum Betreiben einer Modelldatenplattform; und
• 2: eine Modelldatenplattform für eine Kraftfahrzeugflotte in deiner schematischen Darstellung.

The invention described above is explained in detail below against the relevant technical background with reference to the accompanying drawings, which show preferred embodiments. The invention is in no way limited by the purely schematic drawings, whereby it should be noted that the drawings are not to scale and are not suitable for defining proportions. It is shown in

• 1 : a process flow of a method for operating a model data platform; and
• 2 : a model data platform for a vehicle fleet in your schematic representation.

In 1 A process flow of a method for operating a model data platform 13 is shown. As shown at the top, a data management device 1 of a motor vehicle 2 is shown, and as shown at the bottom, a backend 7. The backend 7 and the data management device 1 of the motor vehicle 2 are connected by a data transmission connection 15.

For the sake of clarity, the method is presented in a block structure. The block structure illustrates, for example, the software architecture or the processes for executing the method. The individual blocks are referred to below as units (order management unit 17, order generation unit 18, communication unit 21, data reception unit 22, data generator unit 23, aggregation unit 24, dispatch unit 25). A unit does not necessarily represent a separate hardware component. Rather, several units are, for example, in a single processor 14 and/or memory 16 (cf. 2 ) can be implemented.

In the backend 7, an order 9 is generated by means of an order management unit 17. The order 9 comprises a request for a set of order data 10 and noise factors. The requested order data 10 are preferably model parameters for a machine learning model. The noise factors are, for example, numerical values, each of which is assigned to a data type of order data 10 in the order 9. For example, the order management unit 17 requests the order 9 and/or the information necessary to generate the order 9 from an order generation unit 18. The noise factors are thus specified in the order generation unit 18 for each data type requested as order data 10 in the order 9. For this purpose, a cost catalog 26 is maintained, for example, using the memory 16. The cost catalog 26 provides a relationship between noise factors for a data type and the privacy costs resulting from the noise factor. For example, a data catalog 19 is also maintained, which maintains a connection between the order data 10 and the raw data 8 necessary to determine the order data 10. Alternatively or additionally, the data catalog 19 includes, for example, privacy costs for the order data 10 or the corresponding raw data 8, for example, in the event that they are not noisy. Using the cost catalog 26 and the data catalog 19, noise factors can thus be determined, for example, that are sufficient to comply with limit values for privacy costs.

The noise factors depend on the requested order data 10 or the respective raw data 8, which are necessary to determine the respective order data 10 in the data management facility 1. Furthermore, the noise factors depend, for example, on legislative requirements, voluntary (e.g., regional) requirements, or individual agreements or contracts with individual persons or legal entities.

For example, the backend communication interface 6 is controlled by the order management unit 17, thus sending the order 9. The order 9 is received via the motor vehicle communication interface 5. For this purpose, the data management device 1 comprises a communication unit 21. The receipt of the order 9 is also referred to as step b in this disclosure.

The communication unit 21 instructs a training unit 20 to generate the order data 10. For this purpose, a recording unit 4 (see 2 ) raw data 8 is recorded. The recording unit 4 is, for example, a position recording unit (for example, Global Positioning System [GPS]), a temperature sensor, a speed measuring unit, an acceleration measuring unit, a microphone, a camera, a lidar sensor, a radar sensor and/or an ultrasonic sensor. The raw data 8 are, for example, temporarily stored. For example, the raw data 8 has already been acquired prior to the receipt of the order 9. Such raw data 8 includes, for example, temperature, location, speed, distance to an object, steering angle, acceleration, weather, braking force, braking command, and/or time. The acquisition of the raw data 8 is also referred to as step a. in this disclosure.

The raw data 8 is noise-added according to task 9, also referred to as step c. For this purpose, the raw data 8 and the noise factors are used as input variables for a noise function. As shown, the noise data 11 thus obtained is collected, i.e., temporarily stored, referred to here as step c.0.

Based on the noise data 11, a machine learning model is trained in a step c.1. For example, model parameters of the machine learning model are determined or adjusted based on the noise data 11, thus generating the requested order data 10.

The order data 10 are sent to the backend 7 via the vehicle-side communication interface 5, step d. There, the order data 10 are received via the backend-side communication interface 6, for example via a data receiving unit 22.

For example, the order data 10 of several data management devices 1, i.e. several motor vehicles 2, are aggregated by means of an aggregation unit 24.

By aggregating model parameters of the machine learning models of multiple motor vehicles 2, the privacy costs are further reduced, since, for example, it becomes less accurate and more complex to calculate back to the underlying raw data 8 of a motor vehicle 2. For example, the number of motor vehicles 2 whose model parameters are aggregated is estimated or calculated based on the privacy costs and the target cost. Preferably, the number of motor vehicles 2 is set at least high enough that a specified limit for the privacy costs is not exceeded.

In a further step, the aggregated model parameters are checked, for example, to determine whether they actually comply with the limit value, which may arise, for example, from legal regulations. For example, the aggregated model parameters are processed using a data generator unit 23 to generate a data packet for a data customer or another application. Such a data packet is then forwarded, for example, using a dispatch unit 25.

For example, privacy costs determined for a set of order data 10 and/or aggregated model parameters are used to train the cost catalog 26. For example, a more precise noise factor for the desired privacy costs can be determined during a subsequent run of the method.

In 2 A model data platform 13 for a motor vehicle fleet is shown in a schematic representation. The model data platform 13 is designed by means of a method according to 1 to be operated and to execute the method. The model data platform 13 comprises a plurality of data management devices 1 and a backend 7.

Each motor vehicle 2 has a data management device 1, which, as shown, is arranged entirely within the motor vehicle 2. Each data management device 1 comprises a motor vehicle-side communication interface 5 and a computer 3 with a processor 14 and a memory 16.

The backend 7 also includes a backend-side communication interface 6 and a computer 3 with a processor 14 and a memory 16. The backend 7 and the data management devices 1 are connected for data transmission via the communication interfaces 5, 6. For example, the data transmission connection 15 is a satellite-based connection, for example, based on a mobile radio standard.

The method proposed here for operating a data management facility ensures compliance with privacy costs.

List of reference symbols

1

Data management facility

2

motor vehicle

3

computer

4

Registration unit

5

vehicle-side communication interface

6

backend communication interface

7

backend

8

Raw data

9

Order

10

Set of order data

11

Noise data

12

Model data

13

Model data platform

14

processor

15

data transmission connection

16

memory

17

Order management unit

18

Order generation unit

19

Data catalog

20

training session

21

Communication unit

22

Data receiving unit

23

Data generator unit

24

Aggregation unit

25

Shipping unit

26

Cost catalog

Claims (10)

A method for operating a data management device (1) of a motor vehicle (2), wherein the data management device (1) has at least the following components: - a computer (3); - at least one acquisition unit (4); and - a motor vehicle-side communication interface (5) that is in data-transmitting communication with a backend-side communication interface (6) of a backend (7), wherein the method comprises at least the following steps: a. by means of the at least one acquisition unit (4) of the motor vehicle (2), acquiring raw data (8); b. by means of the motor vehicle-side communication interface (5), receiving a job (9) sent from the backend-side communication interface (6), which job requests at least one set of job data (10) and includes at least one noise factor; c. by means of the computer (3), noise-deflecting a set of raw data (8) selected on the basis of the order data (10) requested in the order (9) based on the at least one noise factor in a noise function to form a set of noise data (11); and d. by means of the motor vehicle-side communication interface (5), sending the set of order data (10) requested in step b. based on the noise data (11) deflecting in step c. to the backend-side communication interface (6). Procedure according to Claim 1 , wherein in a sub-step c.1, model data (12) are generated by means of the computer (3) on the basis of the noise data (11), wherein preferably, by means of the motor vehicle-side communication interface (5) at least part of the model data (12) is sent as order data (10) to the backend-side communication interface (6) in step d. Procedure according to Claim 2 , wherein step c.1 comprises training a generative machine learning model, and wherein the model data (12) comprise model parameters of the generative machine learning model. Method according to one of the preceding claims, wherein the raw data (8) comprises at least one data type from the following list: - temperature; - location; - speed; - distance to an object; - steering angle; - acceleration; - weather; - braking force; - braking command; and - time. Method according to one of the preceding claims, wherein each data type of raw data (8), raw data (8) which is required to determine the order data (10), is assigned a noise factor which is received in the order (9) from the motor vehicle-side communication interface (5). Method for operating a model data platform (13) for motor vehicles (2), wherein the model data platform (13) has the data management device (1) of at least one motor vehicle (2) and a backend (7) with a backend-side communication interface (6), wherein the method for operating the model data platform (13) comprises the method for operating the data management device (1) of a motor vehicle (2) according to one of Claim 1 until Claim 5 and wherein the order (9) received in step b. by means of the motor vehicle-side communication interface (5) is sent by means of a backend-side communication interface (6) of a backend (7). Procedure according to Claim 6 , wherein in the backend (7) a connection between noise factors and privacy costs and/or a connection between order data (10) and privacy costs is maintained. Procedure according to Claim 6 or Claim 7 , wherein in a step e. the order data (10) of several of the motor vehicles (2), which were each sent to the backend (7) in step d., are combined to form an overall data set. Procedure according to one of Claim 6 until Claim 8 , wherein by means of the backend (7), the at least one noise factor for each data type contained in the set of order data (10) of the order (9) is determined as a function of an influence of the data type on a target variable. Procedure according to one of Claim 7 until Claim 9 , wherein by means of the backend (7), the relationship between privacy costs and order data (10) and/or the relationship between noise factors and privacy costs is generated and/or adjusted by summing the privacy costs based on the set of order data (10) sent in the order (9).