Procedure for adjusting an air conditioner

Abstract

The invention relates to a method (1) for user-adapted adjustment of an air conditioning system (2) of a vehicle (3) for a user (13) using a control architecture (4). In this process, a model-relevant data package (5) is collected and subsequently a comfort model (9) is trained in an AI unit (8). Furthermore, a setting-relevant data package (11) is collected and transmitted to the trained comfort model (9). Based on this, user-adapted settings (12) are predicted using the comfort model (9), and the air conditioning system (2) is adjusted accordingly. The invention also relates to the control architecture (4) for carrying out the method (1).

Description

[0001] The invention relates to a method for user-adapted adjustment of an air conditioning system of a vehicle for a user by means of a control architecture according to the preamble of claim 1. The invention also relates to the control architecture for carrying out the method.

[0002] An air conditioning user can adjust the temperature according to their thermal sensitivity by changing several parameters. Map-based climate control allows the user to adjust the air conditioning system, simplifying its operation. However, each user's perception of thermal comfort varies greatly. Conventional map-based climate control systems are disadvantageously designed for the "average" person and cannot adapt to each individual user. If the user feels uncomfortable with map-based climate control, they must laboriously readjust the air conditioning system manually. Another solution is offered by AI-supported comfort models (AI: Artificial Intelligence), which can adapt to the individual preferences of the user. It is desired to make the existing comfort models available for driving in other vehicles as well.

[0003] Solutions for user-adapted adjustment of the air conditioning system are known, for example, from DE 10 2020 107178 A1, DE 10 2015 221416 A1, and DE 10 2018 113074 A1. Furthermore, AI-assisted control of any device is known from DE 10 2018 129417 A1.

[0004] The object of the invention is therefore to provide an improved or at least alternative embodiment for a method of the generic type, in which the described disadvantages are overcome. The object of the invention is also to provide a corresponding control architecture for implementing the method.

[0005] This object is achieved according to the invention by the subject matter of the independent claims. Advantageous embodiments are the subject matter of the dependent claims.

[0006] The present invention is based on the general idea of ​​training an AI unit and creating an individual comfort model for a user.

[0007] The method according to the invention is intended for user-adapted adjustment of a vehicle's air conditioning system for a user and is carried out using a control architecture. A modeling loop and a setting loop are repeatedly executed in the method. In the modeling loop, at least one model-relevant data packet from the user is first collected. Then, a comfort model is trained in an AI unit (AI: Artificial Intelligence) with data from at least one of the model-relevant data packets from the user. In the setting loop, a setting-relevant data packet from the vehicle is collected and transmitted to the trained comfort model. User-adapted settings for user-adapted adjustment of the air conditioning system are then predicted using the comfort model based on data from the setting-relevant data packet. The air conditioning system is subsequently adjusted according to the user-adapted settings.

[0008] The modeling loop is repeatedly executed to train the comfort model using data from the model-relevant data packet. The respective modeling loop can be ended as soon as the comfort model has been trained once. A new modeling loop can then be started. Multiple model-relevant data packets can be transmitted within the respective modeling loop. The model-relevant data packet can be generated repeatedly when the air conditioning system is used and used to further train the comfort model. This allows the comfort model to be continually adapted and individualized to the user as the air conditioning system is used. The settings loop is repeatedly executed to predict user-adapted settings based on data from the settings-relevant data packet and to adjust the air conditioning system.The user-adapted settings are necessary for using the air conditioning system and are predicted using the trained comfort model. In other words, the comfort model is increasingly customized as the air conditioning system is used, and the air conditioning system is adjusted to be more customized or better adapted to the user using the customized comfort model. The comfort model is trained using data from the user's model-relevant data package, thus reducing the computing power in the vehicle. Furthermore, the trained comfort model can be ported to other vehicles belonging to the same user. Using this method, a customized comfort model can be created and trained centrally. It is understood that the described method can also be used to train multiple comfort models for multiple users.

[0009] In the context of the present invention, user-adjusted air conditioning settings are understood to mean settings that are user-adjusted or that adjust the air conditioning system according to the user's desired experience. The user-adjusted air conditioning settings can include target settings for the vehicle's air conditioning system. The target settings for the air conditioning system can be user-adjusted. In other words, the trained comfort model can map the entire set of settings to the target settings for the air conditioning system. The target settings for the air conditioning system correspond to the optimal settings for the air conditioning system for the specific user under given ambient conditions.

[0010] In the context of the present invention, data of the model-relevant data packet is understood to mean data or values ​​that are relevant for training the comfort model or can be taken into account when training the comfort model. In the context of the present invention, data of the setting-related data packet is understood to mean data or values ​​that are relevant for setting the air conditioning system and / or can be taken into account in the comfort model when predicting the user-adapted settings. The model-relevant data packet and the setting-relevant data packet can comprise partially identical data or values. The model-relevant data packet and the setting-relevant data packet can comprise partially different data or values. In particular, the model-relevant data packet can contain the setting-relevant data packet and further information.

[0011] The model-relevant data package and / or the setting-relevant data package can include context data acquired by sensors of the vehicle. The context data can include at least one context variable of the vehicle. The at least one context variable can preferably be interior temperature and / or exterior temperature and / or air humidity and / or position of the sun and / or light intensity and / or air pressure. It is understood that the above list is not exhaustive. The context data can thus be taken into account when training the comfort model and / or when adjusting the air conditioning system, so that the comfort model can be customized and / or the air conditioning system can be adjusted to the user's individual needs.

[0012] The model-relevant data packet and / or the setting-relevant data packet can include user data. The user data can include at least one user parameter of the user. The at least one user parameter can preferably be physiological user information such as seating position and / or user identity and / or skin temperature and / or pulse and / or respiratory rate and / or age and / or gender and / or clothing condition and / or emotional state and / or biometric inputs and / or physiological inputs. It is understood that the above list is not exhaustive. The user data can, for example, be recorded by sensors of the vehicle. The vehicle sensors can be imaged by a camera installed in the vehicle or an infrared camera installed in the vehicle. Furthermore, the user data can, for example, be provided by a mobile device of the user.User data can be determined, for example, from information stored on the mobile device. For example, calendar data can be used to determine whether the user has just completed a planned sporting activity.

[0013] The creation of the model-relevant data package can be initiated by a manual user input. The user input can be intended to change the current user-adjusted settings of the air conditioning system. The user input can then be transmitted in the model-relevant data package, if necessary, along with other data and thus taken into account when training the comfort model. The creation of the model-relevant data package can be initiated, in particular, with each user input. The user input is entered manually by the user and is intended to change the current settings of the air conditioning system. The user input can be entered by the user, for example, if the desired sensation is not achieved by the predicted user-adjusted settings and thus by the current comfort model.

[0014] The user input can then initiate the creation of the model-relevant data package and data from the current model-relevant data package can be taken into account in the comfort model. The current model-relevant data package includes the user input currently entered by the user and, if necessary, additional data so that the comfort model can be adapted to the individual user and their individual perception. In addition to the user input, the model-relevant data package can include the context data and / or the user data described above so that the comfort model can be trained based on the user input in the context of the currently available context data and / or the currently available user data. Since the user input is entered manually by the user, the model-relevant data package can be created accordingly without a predefinable frequency and / or without a predefinable frequency.

[0015] The user input can comprise setting data entered by the user. The setting data can then comprise at least one setting variable. The at least one setting variable can preferably be the desired temperature and / or the fan speed to be set and / or the seat heating to be set and / or the heating area to be set and / or the heating intensity to be set. It is understood that the above list is not exhaustive. Alternatively or additionally, the user input can comprise optimization data entered by the user. The optimization data can then comprise at least one optimization variable. The at least one optimization variable can preferably comprise inputs relating to user-specific sensations. For example, the optimization data can take into account information such as “too cold on the head”, “too warm on the right leg”, “the draft is unpleasant”, etc.

[0016] Alternatively or additionally, the creation of the model-relevant data package can be initiated after a predefined time interval if no user input is made within the predefined time interval. As already explained above, the user input is made manually by the user and causes a change to the current user-adjusted settings of the air conditioning system. If the user does not make any changes to the user-adjusted settings of the air conditioning system, the desired sensation has already been achieved using the current comfort model or the user-specific settings are sufficiently predicted under the current ambient conditions. This can be used as positive feedback from the user. The predefined time interval can, for example, be between 1 minute and 30 minutes, in particular between 5 and 20 minutes.

[0017] The adjustment loop can be performed at a predefined frequency. The predefined frequency can preferably be between 0.1 and 10 Hertz, particularly preferably 1 Hertz. The adjustment loop can be initiated independently of user input. This allows the air conditioning system to be adjusted depending on changing ambient conditions without any manual action by the user. Accordingly, the modeling loop and the adjustment loop can be performed independently of one another and / or at different times. As already explained above, the modeling loop can be initiated without a predefined frequency and / or without a predefined frequency. Accordingly, the modeling loop and the adjustment loop can be performed at different frequencies or frequencies.

[0018] Offset settings can be created from data from the model-relevant data package. The offset settings can then be saved in the vehicle and additionally used to adjust the air conditioning system. However, as soon as the offset settings have been taken into account in the trained comfort model, the offset settings can be deleted. When adjusting the air conditioning system, data from the model-relevant data package that is not yet included in the current comfort model can then also be taken into account. The offset settings can, for example, take into account a user input that is not yet considered in the comfort model. The offset settings can, in particular, include offsets requested by the user to the current settings or to the currently set settings of the air conditioning system.For example, the offset settings can include an offset of the interior temperature set by the air conditioning system if the user wants to increase or decrease the interior temperature set by the air conditioning system. If multiple model-relevant data packages arise that cannot yet be considered in the comfort model, all of these data packages can be used sequentially to create the offset settings.

[0019] In this method, the comfort model can be trained periodically at a predefined time—for example, hourly, daily, or weekly. Between two consecutive points in time, all model-relevant data packets from the user can be saved and used to train the comfort model. In other words, multiple model-relevant data packets from the user can be saved—for example, in the vehicle and / or in a cloud—and the comfort model can be trained using data from multiple data packets at once. In other words, creating the model-relevant data packet does not automatically mean training the comfort model. The model-relevant data packets can be saved—for example, in the vehicle and / or in a cloud—until the comfort model is trained and used as soon as the comfort model is trained.It is conceivable that not all model-relevant data packets are used to train the comfort model. In other words, the comfort model can also be trained using some model-relevant data packets. The comfort model can be trained periodically at a predefined time—for example, hourly, daily, or weekly. It is understood that the comfort model will not be trained if no model-relevant data packet is available.

[0020] The model-relevant data packets can be weighted when training the comfort model. For example, the time interval between two consecutive model-relevant data packets can be used for weighting. If the time interval is small, data from the first data packet apparently did not lead to the desired user experience and is of little relevance when training the comfort model. If the time interval is large, data from the first data packet apparently led to the desired user experience and is highly relevant when training the comfort model. The terms "small" and "large" should be considered relative to each other. The "small" time interval can be, for example, 1-5 minutes, and the "large" time interval can be, for example, 10-30 minutes.

[0021] In a possible first embodiment of the method, the modeling loop can be performed in the vehicle. The comfort model can then be trained in the vehicle's AI unit. The method can be performed entirely locally in the vehicle, without the cloud.

[0022] In the first embodiment of the method, the comfort model can be stored in the modeling loop after training in the vehicle and executed in the settings loop in the vehicle. The user-adapted settings can be predicted locally in the vehicle. The comfort model trained in the vehicle can, of course, be used for this purpose.

[0023] In a possible second embodiment of the method, the modeling loop can be performed in a cloud. The user's model-relevant data packet can then be collected and transmitted to the cloud, and the comfort model can be trained in the cloud's AI unit. The cloud is a network that provides computing resources such as servers and / or data storage and / or applications.

[0024] In the second embodiment of the method, if transmission of the user's model-relevant data packet to the cloud is not possible, the model-relevant data packet can be stored in a vehicle memory. The model-relevant data packet can then be transmitted to the cloud as soon as transmission of the user's model-relevant data packet to the cloud is possible. This means that data from the model-relevant data packet that cannot be transmitted directly can also be taken into account when training the comfort model. If multiple model-relevant data packets arise that cannot be transmitted directly to the cloud, all of these data packets can be stored in the memory and then transmitted together to the cloud. Accordingly, all of these data packets can be taken into account when training the comfort model without any data loss.

[0025] As already described above, the comfort model can be trained periodically at a predefined time. Between two consecutive points in time, all model-relevant data packets from the user can be stored in the vehicle or in the cloud and used to train the comfort model. If all model-relevant data packets from the user are stored in the vehicle, they can be periodically transmitted from the vehicle to the cloud at a predefined time. In other words, several model-relevant data packets from the user can be transmitted to the cloud at once. The comfort model can then be trained with all transmitted model-relevant data packets at once.

[0026] In the second embodiment of the method, the comfort model can be stored in a model pool in the cloud. To retrain the comfort model, the once-created comfort model can then be retrieved from the model pool and trained. The comfort model stored in the model pool can then be replaced or updated with the new, current, or trained comfort model. In addition, the comfort model can be transmitted from the model pool to the vehicle and stored in the vehicle. The comfort model can be transmitted from the model pool to the vehicle periodically at a predefined time. For example, the comfort model can be transmitted to the vehicle hourly, daily, or weekly. In particular, the comfort model can be transmitted from the cloud model pool to the vehicle after each completed training session.

[0027] In its second embodiment, the method can be carried out in a different manner. Three different alternatives are described in more detail below. It is understood that an extension of the described alternatives with the features described above is conceivable. Furthermore, it is understood that further alternatives can also be implemented in the method.

[0028] In the first alternative, the comfort model in the modeling loop can be saved in a cloud model pool after training. Furthermore, the comfort model can be transmitted from the model pool to the vehicle and additionally stored in the vehicle. In the settings loop, the settings-relevant data packet can then be transmitted to the comfort model stored in the vehicle. Accordingly, the comfort model can be executed in the vehicle, and the user-adapted settings can be predicted locally in the vehicle. In the first alternative, the cloud is therefore only used in the modeling loop. As described above, the modeling loop and the settings loop can be executed at different frequencies. In particular, the frequency of the modeling loop can be several times lower than the frequency of the settings loop.This can significantly reduce the amount of data transmitted to and received from the cloud.

[0029] In the second alternative, the comfort model in the modeling loop can be saved in a cloud model pool after training. In the adjustment loop, the setting-relevant data packet can then be transmitted from the vehicle to the comfort model stored in the model pool. The comfort model can then be executed in the cloud, and the user-adapted settings can be predicted in the cloud and subsequently transmitted to the vehicle. In the second alternative, the comfort model is not saved in the vehicle, and the modeling loop and the adjustment loop are executed via the cloud.

[0030] In the third alternative, the comfort model can be saved in a cloud model pool after training in the modeling loop. Furthermore, the comfort model can be transmitted from the model pool to the vehicle and additionally saved in the vehicle. In the settings loop, the settings-relevant data packet is transmitted to the comfort model stored in the vehicle if transmission of the vehicle's settings-relevant data packet to the cloud is not possible. The comfort model is then executed in the vehicle, and the user-adapted settings are predicted locally in the vehicle. However, if transmission of the vehicle's settings-relevant data packet to the cloud is possible, the settings-relevant data packet is transmitted from the vehicle to the comfort model stored in the model pool.The comfort model is then executed in the cloud, and the user-adapted settings are predicted in the cloud and subsequently transmitted to the vehicle. The third alternative thus represents a hybrid solution combining the first and second alternatives of the method.

[0031] The invention also relates to a control architecture for executing the method described above. The control architecture is designed to execute the method. Specifically, the control architecture is designed to repeatedly collect a model-relevant data packet from the user; to repeatedly train a comfort model with data from the model-relevant data packet from the user in an AI unit; to repeatedly collect a setting-relevant data packet from the vehicle and to repeatedly transmit the setting-relevant data packet to the comfort model; to repeatedly predict user-adapted settings for user-adapted adjustment of the air conditioning system based on data from the setting-relevant data packet using the comfort model; and to repeatedly adjust the air conditioning system according to the user-adapted settings.It is understood that the control architecture includes the necessary hardware and / or software to execute the method. The necessary hardware and / or software can be part of the vehicle and / or part of the cloud. To avoid repetition, reference is made to the above explanations.

[0032] Further important features and advantages of the invention emerge from the subclaims, from the drawings and from the associated description of the figures based on the drawings.

[0033] It is understood that the features mentioned above and those to be explained below can be used not only in the combination specified in each case, but also in other combinations or on their own, without departing from the scope of the present invention.

[0034] Preferred embodiments of the invention are illustrated in the drawings and are explained in more detail in the following description, wherein the same reference numerals refer to the same or similar or functionally identical components.

[0035] They show, schematically Fig. 1 shows a schematic flow of a method according to the invention in a first alternative with a cloud; Fig. 2 shows a schematic flow of the method according to the invention in a second alternative with a cloud; Fig. 3 a schematic flow of the method according to the invention in a third alternative without cloud.

[0036] Fig. Figure 1 shows a schematic flow of a method 1 according to the invention in a first alternative. Method 1 is intended for user-adapted adjustment of an air conditioning system 2 of a vehicle 3 and is implemented using a control architecture 4. In method 1, a modeling loop MS and an adjustment loop are repeatedly executed.

[0037] In the modeling loop MS, a model-relevant data packet 5 is collected and transmitted to a cloud 6. In the cloud 6, the model-relevant data packet 5 is stored in a memory 7. Subsequently, a comfort model 9 is trained with data from the model-relevant data packet 5 in an AI unit 8 of the cloud 6. The trained comfort model 9 is then stored in a model pool 10. In the first alternative of method 1, the comfort model 9 is additionally transmitted to the motor vehicle 3 and stored in the motor vehicle 3. In the settings loop ES, a settings-relevant data packet 11 of the vehicle 3 is collected and transmitted to the comfort model 9 stored in the vehicle 3. Based on data from the settings-relevant data packet 11, user-adapted settings 12 are predicted using the comfort model 9.The user-adjusted settings 12 are then used to adjust the air conditioning system 2 to the user's preferences.

[0038] The setting-relevant data package 11 can include context data of the vehicle 3—for example, interior temperature and / or exterior temperature and / or humidity and / or position of the sun and / or light intensity and / or air pressure. Alternatively or additionally, the setting-relevant data package 11 can include user data—for example, physiological user information such as seating position and / or user identity and / or skin temperature and / or pulse and / or respiratory rate and / or age and / or gender and / or clothing condition and / or emotional state and / or biometric inputs and / or physiological inputs. Based on data from the setting-relevant data package 11, the settings 12 for the air conditioning system 2 can be predicted using the comfort model 9, and the air conditioning system 2 can be adjusted. As a result, the air conditioning system 2 is customized or individually adjusted to the needs of a user 13.

[0039] In method 1, the modeling loop MS is initiated by a user input 14. The user input 14 is entered manually by the user 13 of the vehicle 3 if their desired thermal sensation is not achieved with the current user-adjusted settings 12 of the air conditioning system 2. The user input 14 can be setting data entered by the user 13—for example, the desired temperature and / or fan speed to be set and / or seat heating to be set and / or heating surface to be set and / or heating intensity to be set. Alternatively or additionally, the user input 14 can include optimization data entered by the user 13—for example, "too cold on the head," "too warm on the right leg," "the draft is unpleasant," etc.

[0040] The user input 14 is then combined with the setting-relevant data package 11, which contains context data and user data, to form the model-relevant data package 5 and transmitted to the cloud 6. The model-relevant data package 5 is stored in the cloud 6 and used to further train the comfort model 9. For this purpose, the currently applicable comfort model 9 is retrieved from the model pool 10 and further trained. After training, the comfort model 9 is updated in the model pool 10, and the updated comfort model 9 is stored in the vehicle 3. Accordingly, the user input 14 combined with data from the setting-relevant data package 11 is taken into account in the updated comfort model 9.

[0041] If transmission of the model-relevant data packet 5 is not possible, the model-relevant data packet 5 is stored in the vehicle 3. In method 1 according to the first alternative, offset settings 15 are also created from data of the non-transmitted model-relevant data packet 5. The offset settings 15 are stored in the vehicle 3, and the user-adapted settings 12 are corrected with the offset settings 15. The offset settings 15 can, for example, include offsets of the setting variables to be adjusted for the air conditioning system 2. As soon as transmission is possible, the model-relevant data packet 5 is transmitted to the cloud 6, and the offset settings 15 are deleted.

[0042] Fig. 2 shows a schematic flow of the method 1 according to the invention in a second alternative. The method 1 in the second alternative differs from the method 1 in the first alternative in that the comfort model 9 is not transmitted to the vehicle 3 and is executed in the cloud 6. Accordingly, the setting-relevant data packet 11 is transmitted to the cloud 6, and the comfort model 9 is executed in the cloud 6. The user-adapted settings 12 are therefore predicted in the cloud 6 and subsequently transmitted to the vehicle 3, in which the air conditioning system 2 is now adjusted. Here, too, the offset settings 15 can comprise offsets of the setting variables of the air conditioning system 2 to be adjusted.

[0043] Referring to Fig. 1 and Fig. 2, method 1 can also be carried out according to a further alternative. In the further alternative, the trained comfort model 9 is stored in the model pool 10 and in the vehicle 3. If a data transmission connection to the cloud 6 exists, the comfort model 9 in the model pool 10 is used and the adjustment loop ES is carried out in method 1 according to Fig. 2. If there is no data transmission connection with the cloud 6, the comfort model 9 in the vehicle 3 is used and the setting loop ES is carried out in the method 1 according to Fig. 1.

[0044] Fig. Figure 3 shows a schematic flow of method 1 according to the invention in a third alternative without a cloud. Here, the modeling loop MS and the adjustment loop ES are performed locally in the vehicle 3. Otherwise, method 1 in the third alternative corresponds to method 1 in the second alternative. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2020 107178 A1

[0003] DE 10 2015 221416 A1

[0003] DE 10 2018 113074 A1

[0003] DE 10 2018 129417 A1

[0003]

Claims

[1] Method (1) for user-adapted adjustment of an air conditioning system (2) of a vehicle (3) for a user (13) by means of a control architecture (4), wherein in a repeated modeling loop (MS): - at least one model-relevant data packet (5) of the user (13) is collected; - a comfort model (9) is trained in an AI unit (8) with data from at least one of the model-relevant data packets (5) of the user (13); wherein in a repeated setting loop (ES): - a setting-relevant data package (11) of the vehicle (3) is collected and transmitted to the trained comfort model (9); - based on data from the setting-relevant data package (11), user-adapted settings (12) for user-adapted adjustment of the air conditioning system (2) are predicted by means of the comfort model (9); - the air conditioning system (2) is adjusted according to the user-adjusted settings (12). [2] Method according to claim 1, characterized by , - that the creation of the model-relevant data package (5) is initiated by a user input (14) manually made by the user (13) to change the current user-adapted settings (12) of the air conditioning system (2), and / or - that the creation of the model-relevant data package (5) is initiated after a predefined time interval if no user input (14) made manually by the user (13) to change the current user-adapted settings (12) of the air conditioning system (2) is made within the predefined time interval. [3] Method according to claim 1 or 2, characterized by that the adjustment loop (ES) is carried out at a predefined frequency, preferably between 0.5 and 10 Hertz, particularly preferably 1 Hertz. [4] Method according to one of the preceding claims, characterized by , - that offset settings (15) are created from data of the model-relevant data package (5), stored in the vehicle (5) and additionally used to adjust the air conditioning system (2), and - that as soon as the offset settings (15) are taken into account in the trained comfort model (9), the offset settings (15) are deleted. [5] Method according to one of the preceding claims, characterized by , - that the comfort model (9) is trained periodically at a predefined time, whereby between two successive times all model-relevant data packets (5) of the user (13) are stored and used to train the comfort model (9), and / or - that data of the model-relevant data package (5) are weighted when training the comfort model (9). [6] Method according to one of the preceding claims, characterized by , - that the modelling loop (MS) is carried out in the vehicle (3), and - that the comfort model (KM) is trained in the AI ​​unit (8) of the vehicle (3). [7] Method according to claim 6, characterized by that in the modeling loop (MS): - the comfort model (9) is stored in the vehicle (3) after training; that in the setting loop (ES): - the comfort model (9) is executed in the vehicle (3) and the user-adapted settings (12) are predicted locally in the vehicle (3). [8] Method according to one of the preceding claims, characterized by , - that the modelling loop (MS) is carried out in a cloud (6), - that the model-relevant data package (5) of the user (13) is transmitted to the cloud (6) after collection, and - that the comfort model (KM) is trained in the AI ​​unit (8) of the cloud (6). [9] Method according to claim 8, characterized by , - that, if transmission of the model-relevant data packet (5) of the user (13) to the cloud (6) is not possible, the model-relevant data packet (5) is stored in the vehicle (3), and - that as soon as a transmission of the model-relevant data packet (5) of the user (13) to the cloud (6) is possible, the model-relevant data packet (5) is transmitted to the cloud (6) and deleted in the vehicle (3). [10] Method according to claim 8 or 9, characterized by , - that all model-relevant data packets (5) of the user (13) are stored in the vehicle (3) and periodically transmitted from the vehicle (3) to the cloud (6) at a predefined time, and / or - that the comfort model (9) is stored in a model pool (10) of the cloud (6); and / or - that the comfort model (9) is stored in a model pool (10) of the cloud (6) and is additionally transmitted from the model pool (10) to the vehicle (3) and stored in the vehicle (3), and / or - that the comfort model (9) is stored in a model pool (10) of the cloud (6) and the current comfort model (9) is periodically transmitted from the model pool (10) to the vehicle (3) at a predefined time, and / or - that the comfort model (9) is stored in a model pool (10) of the cloud (6) and is transmitted from the model pool (10) of the cloud (6) to the vehicle (3) after each completed training session. [11] Method according to one of claims 8 to 10, characterized by that in the modeling loop (MS): - the comfort model (9) is stored in a model pool (10) of the cloud (6) after training; - the comfort model (9) is transmitted from the model pool (10) to the vehicle (3) and additionally stored in the vehicle (3); that in the setting loop (ES): - the setting-relevant data packet (11) is transmitted to the comfort model (9) stored in the vehicle (3); - the comfort model (9) is executed in the vehicle (3) and the user-adapted settings (12) are predicted locally in the vehicle (3). [12] Method according to one of claims 8 to 10, characterized by that in the modeling loop (MS): - the comfort model (9) is stored in a model pool (10) of the cloud (6) after training; that in the setting loop (ES): - the setting-relevant data packet (11) is transmitted from the vehicle (3) to the comfort model (9) stored in the model pool (10); - the comfort model (9) is executed in the cloud (6) and the user-adapted settings (12) are predicted in the cloud (6) and then transmitted to the vehicle (3). [13] Method according to one of claims 8 to 10, characterized by that in the modeling loop (MS): - the comfort model (9) is stored in a model pool (10) of the cloud (6) after training; - the comfort model (9) is transmitted from the model pool (10) to the vehicle (3) and additionally stored in the vehicle (3); that in the setting loop (ES): - if transmission of the setting-relevant data packet (11) of the vehicle (3) to the cloud (6) is not possible, the setting-relevant data packet (11) is transmitted to the comfort model (9) stored in the vehicle (3); the comfort model (9) is executed in the vehicle (3) and the user-adapted settings (12) are predicted locally in the vehicle (3), and - if transmission of the setting-relevant data packet (11) of the vehicle (3) to the cloud (6) is possible, the setting-relevant data packet (11) is transmitted from the vehicle (3) to the comfort model (9) stored in the model pool (10); the comfort model (9) is executed in the cloud (6); the user-adapted settings (12) are predicted in the cloud (6) and subsequently transmitted to the vehicle (3). [14] Control architecture (4) for carrying out the method (1) according to one of the preceding claims, wherein the control architecture (4): - for repeatedly collecting a model-relevant data packet (5) from the user (13); - for repeatedly training a comfort model (9) with data from the model-relevant data package (5) of the user (13) in an AI unit (8); - for repeatedly collecting a setting-relevant data packet (11) of the vehicle (3) and for repeatedly transmitting the setting-relevant data packet (11) to the comfort model (9); - for repeatedly predicting user-adapted settings (12) for user-adapted adjustment of an air conditioning system (2) of the vehicle (3) based on data of the setting-relevant data packet (11) by means of the comfort model (9); - is designed for repeated adjustment of the air conditioning system (2) according to the user-adjusted settings (12).

Images