DE102018211575B4 - Method for operating an electrical system of a motor vehicle using artificial intelligence and corresponding overall system - Google Patents

Abstract

Method for operating an electrical on-board network (10) of a motor vehicle (9), wherein the on-board network (10) comprises as components at least one rechargeable electric battery (11) and an electric motor (12) that can be used as a drive motor and generator, and at least one control device (14) assigned to at least one component is used to control the components according to operating parameters describing the energy consumption and energy output of the components, wherein the operating parameters are determined as output data (5) of a determination algorithm (16) that uses artificial intelligence and/or is programmed by artificial intelligence, which determines the output data (5) from input data (1) that describe the current operating state of the motor vehicle (9), wherein reinforcement learning is used as a training method for the artificial intelligence of the determination algorithm (16), wherein input data (1) of the motor vehicle (9) is recorded when the determination algorithm (16) is used during operation of the motor vehicle (9), and at least one minimization of energy consumption is used as a reward criterion for reinforcement learning, wherein output data (5) with associated input data (1) are transmitted from the motor vehicle (9) and other motor vehicles (9) equivalent in terms of the on-board network (10) to a backend device (7) at least when a transmission criterion is fulfilled and are made available to other motor vehicles (9).

Description

The invention relates to a method for operating an electrical system of a motor vehicle, wherein the electrical system comprises as components at least one rechargeable electric battery and an electric motor that can be used as an electric motor and generator, and at least one control device assigned to at least one component is used to control the components according to operating parameters describing the energy consumption and energy output of the components. In addition, the invention relates to an overall system with a backend device and several motor vehicles.

Electric vehicles, as known in the prior art, have an on-board electrical system that contains both the electric motor and an electric battery that feeds the electric motor (or other on-board electrical systems). The electric motor can also be used as a generator for recuperation. In this case, a configuration is usually provided in which a high-voltage network and a low-voltage network with a lower voltage are provided in the electric vehicle, which can be connected, for example, via a DC/DC converter, whereby the low-voltage network can be fed from its own low-voltage battery. The low-voltage network can be part of the on-board electrical system as a low-voltage component, but will usually be at least connected to it. Both the high-voltage component of the on-board electrical system and the low-voltage network often contain other consumers as components, for example components of an air conditioning system, which require significant amounts of energy from the battery of the on-board electrical system.

A key quality feature of an electric vehicle is the range that results from the battery charge levels. The range of the electric vehicle can decrease when a lot of electrical energy is required in other devices in the electric vehicle. However, this also applies if an operating strategy is chosen for the electric motor whose energy efficiency is not optimal, for example if there is frequent, strong acceleration or something similar. This means that the interaction of the energy requirements of all components of the on-board network determines the final range of the electric vehicle.

Approaches have already been proposed to determine operating strategies, for example for upcoming journeys, by arranging certain operating states of the electric motor in such a way that the greatest possible range is achieved. However, such considerations are usually based on predetermined route choices and only consider the electric motor as a component, without considering the overall efficiency of the on-board network.

The invention is therefore based on the object of specifying an operating method which enables an optimization of the consumption of electrical energy and thus an increase in the range, particularly for electric motor vehicles.

EN 10 2011 082 336 A1 discloses a method for controlling energy distribution processes in a motor vehicle. First, a current environment of the vehicle is recorded, based on which a future driving state of the vehicle is determined. Furthermore, an expected energy requirement of at least one consumer of the vehicle is determined, with charging or discharging of an energy storage device being controlled depending on the expected energy requirement. Within the scope of the method, the expected energy requirement can be determined using neural networks.

The Wikipedia article “ Reinforcement learning”, edited on 26 June 2018, 19:11, URL: https://de.wikipedia.org/w/index.php?title=Best%C3%A4 rkendes Lernen&oldid=178653865 , describes the machine learning technique also known as “reinforcement learning”.

EN 10 2015 108 270 A1 relates to hybrid electric vehicles and an associated energy management control strategy. It is provided that a control of the vehicle is programmed to operate a drive train according to a predicted vehicle speed profile. The control device can have a neural network, wherein the vehicle speed profile is determined, for example, based on route planning. The prediction of the vehicle speed can serve, for example, to support vehicle energy planning.

EN 10 2017 219 489 A1 discloses a method for controlling a transmission of a hybrid vehicle, in particular to reduce unnecessary gear changes and improve fuel efficiency. A neural network is provided here, by means of which it is determined whether or not a gear change is to be carried out by predicting a torque required in the near future. This prediction is made based on a status of an accelerator pedal operation or a brake operation status, a detected vehicle speed, information about the behavior or distance of a vehicle driving ahead, GPS-based navigation information and/or a recorded driving style of the respective driver. An evaluation unit can be mounted on the vehicle or a modified model can be received from outside, whereby for example a cloud server can be provided in which the machine learning process takes place based on the corresponding input values, whereby the resulting model is transmitted to the vehicle.

US 2018 / 0 170 356 A1 relates to a system in which gear shifts are controlled by means of artificial intelligence depending in particular on a driver's recorded driving style or an acceleration prediction.

In the subsequently published publication US 2018 / 0 334 170 A1 A method for controlling or adjusting the temperature, in particular of an energy storage device in an electric vehicle, is described. This control is carried out, for example, using a deep learning method.

To achieve this object, the features of claim 1 are provided according to the invention in a method of the type mentioned at the outset.

The invention therefore proposes to optimize the energy consumption and preferably also other positive effects in the on-board network of the motor vehicle by using artificial intelligence to recognize patterns that lead to optimal operation of the motor vehicle, which is designed in particular as an electric vehicle, and to provide corresponding operating parameters for the control units.

The use of the method according to the invention brings with it a number of advantages. Firstly, the energy consumption in motor vehicles, in particular electric vehicles, is optimized so that the range of the electric vehicles is increased and the longevity of the batteries is improved. The approach according to the invention can be applied both to motor vehicles driven by human drivers and to fully automatic motor vehicles. Furthermore, the method described here can be applied to both traditional electric vehicles and electric vehicles with solar cells, in particular also to motor vehicles conceivable in the future that are powered solely by renewable energy sources (e.g. solar and wind energy). Another advantage is that the use of optimal operating parameters also increases the longevity of components and control units, so that maintenance costs can be reduced overall.

When using artificial intelligence in the investigation algorithm, it has proven particularly useful if the investigation algorithm includes at least one neural network. Neural networks are particularly suitable for the learning techniques presented below.

In this case, a particularly preferred embodiment provides that the determination algorithm, from input data describing the current operating state of the motor vehicle, initially predicts, in a first step, pending task data describing tasks relating to electrical energy flow between the components for a prediction period within the on-board network, for which the output data is determined in a second step. This creates a basis for a particularly advantageous use of artificial intelligence after the operating state is ultimately abstracted and can be better related to similar facts. In this embodiment, a prediction of pending tasks relating to energy flows in the on-board network is provided for a prediction period which can be between 2 and 30 minutes long, in particular 15 to 25 minutes. In advantageous embodiments, it is also conceivable to adjust the length of the prediction period to the informative value of the available input data, which describe the current operating state of the motor vehicle, so that, for example, the prediction of the upcoming tasks can be carried out until a certain reliability threshold is undershot during their prediction. Tasks include, on the one hand, the drive of the motor vehicle by the electric motor itself or the recuperation of electrical energy by the electric motor, but also other components, for example heating and/or cooling processes within the motor vehicle. It should be noted that the on-board network will in most cases have other components. In concrete terms, it can generally be provided that the on-board network has at least one consumer as a further component, in particular an air conditioning component, and/or at least one low-voltage component and/or power electronics for connection to a low-voltage network.

Essential input data of the determination algorithm can include a current position of the motor vehicle and a current destination of the motor vehicle. The destination does not necessarily have to represent a final destination of a current operating phase of the motor vehicle, but can of course also be an intermediate destination. For example, if it is not known or deducible within the motor vehicle what the final destination of the journey is, but it is known that the motor vehicle is on a section of motorway, it can be assumed that at least until the next Exit this section of the motorway or the like. In advantageous embodiments, however, input data that goes beyond the examples of input data mentioned above will also be used, for example information on the environmental conditions of the motor vehicle, for example including weather and/or lighting conditions, since this can result in heating and/or cooling requirements as well as lighting requirements and the like, which also lead to corresponding tasks in the on-board network. A particularly useful source of input data is the motor vehicle diagnostic data that is already used in many bus systems of motor vehicles, which describe the current and possibly also future state of the motor vehicle and are therefore excellently suited as a basis for predicting the activities of the motor vehicle in the near future, i.e. the forecast period. Input data can also include the current state of the on-board network and its components. For example, the state of the battery can be described by information such as battery voltage, temperature, state of charge, aging state, battery power, battery current, input impedance, number of charging cycles and/or other data. Useful input data also include the current temperature of other components besides the battery and their electrical power output and/or power consumption. The input data can also be provided by the control units, which can influence not only the other operation of the components assigned to them but also their energy consumption and/or output.

Today's motor vehicles usually have a large number of control units, engine control units, battery management control units, infotainment control units, control units of driver assistance systems, steering control units and the like. For example, motor vehicles are already known in which several hundred control units are present. The method according to the invention can either be carried out by one of these control units, but preferably a special, separate, global energy control unit is provided which carries out the determination algorithm and passes the operating parameters on to the corresponding control units if the determination algorithm, as will be discussed in more detail below, is to be implemented within the motor vehicle.

In the first step, the next activities of the motor vehicle are ultimately predicted so that the energy intake or consumption is also predicted, including, for example, energy consumption and/or generation based on regenerative braking, traffic, downtime, weather conditions, speed, torque and the like. For example, a predictive analytics model can be used to predict the task data, such as a model with principal component analysis and/or logistic regression analysis. This means that the prediction in the first step does not necessarily have to use (or be based on) artificial intelligence itself, although the use of artificial intelligence can also be useful in this context. For example, a neural network can be used for the first step. Preferably, artificial intelligence is used in the determination algorithm itself at least in the second step, as will be explained in more detail.

As already explained, artificial intelligence can be used for both the first and second steps. In this context, it is preferred if different and, in particular, separately trained sub-algorithms of artificial intelligence are used for the first and second steps. This means that the logical separation of prediction and the determination of the operating parameters from the task data, which is demonstrated by the mention of a first and a second step, can also be transferred to a separation of the associated sub-algorithms, since the complexity of the relationships is thus also logically separated. This lowers the demands on the two sub-algorithms, so that overall the computing power required, in particular for both training and final execution, if both do not happen in one go, can be reduced.

In principle, classic approaches can be used to train sub-algorithms of artificial intelligence that are included in the determination algorithm. However, such more classic approaches require training data that already contain a basic truth. Such training data can be determined, for example, when operating data recorded in the operation of a large number of motor vehicles with the same equipment regarding the on-board network and the energy balance are collected and evaluated in terms of energy consumption (and, if necessary, other criteria). However, it is also possible to use training data that is determined "theoretically" or results from simulations. In the case of motor vehicles that regularly cover the same route, for example, or otherwise regularly have the same task constellation in the on-board network, training data can also be determined automatically through variation and tests within this motor vehicle and subsequent evaluation. Classic cal training mechanisms are particularly suitable for the first step, since it is easy to determine afterwards whether the prediction was correct based on the task data. However, a different variant is preferred for the second step.

Accordingly, according to the present invention, reinforcement learning, in particular deep reinforcement learning, is used as a training method for the artificial intelligence of the determination algorithm, in particular in the second step, whereby input data of the motor vehicle are recorded when the determination algorithm is used during operation of the motor vehicle and at least a minimization of energy consumption is used as a reward criterion for reinforcement learning. The technique of so-called reinforcement learning is particularly suitable as a type of "learning by doing" for artificial intelligence, since the present invention aims at certain optimization goals, represented by the reward criteria, specifically at least the reduction of energy consumption, so that the use of reinforcement learning during operation, and thus the use of the determination algorithm, enables a continuous improvement of the determination algorithm. Reinforcement learning, for which the term "reinforcing learning" is also known in German, is a type of machine learning in which the determination algorithm independently learns a strategy in order to maximize the rewards received. It is not necessary to specify a ground truth, that is, to know in advance which set of operating parameters is the optimal one, but rather a utility function is approximated based on the rewards determined using the reward criteria, which describes the value of a certain set of operating parameters.

It is particularly advantageous to combine reinforcement learning with deep learning methods, especially for neural networks. The approach is based on letting the detection algorithm construct its knowledge directly from constellations of input data, similar to a human, without any special specifications and/or heuristics.

In addition to a reward criterion relating to the minimization of energy consumption, other reward criteria can of course also be used. Other reward criteria, particularly those used as secondary reward criteria, can preferably relate to the heat generation within the vehicle electrical system and/or the load on components of the vehicle electrical system, in particular the battery. For example, another reward criterion can aim to maximize the lifespan of the battery. It should be noted that in order to evaluate the reward criteria and thus to determine the reward, in particular in addition to input data, other operating data of the motor vehicle, in particular result data, can be recorded, which are then evaluated by the reward criteria. These can include constant tracking of the power flows within the vehicle electrical system, in particular the energy consumption and/or energy absorption of components, from which, for example, energy consumption results. Temperatures, power cycles and the like can be tracked in a similar way. In other words, this means that the determination algorithm receives feedback for each of its actions, which in turn is used to continuously improve it. The resulting reward of reinforcement learning can also depend on previous, comparable investigations, so that, for example, if input data or task data occur repeatedly in a motor vehicle, e.g. due to driving on a certain route, constant optimization is strived for and the currently achieved, best performance is considered the benchmark.

According to the invention, output data with associated input data and/or task data are transmitted from the motor vehicle and other motor vehicles that are equivalent in terms of the on-board network to a backend server and made available to other motor vehicles at least when a transmission criterion is met. The transmission criterion can particularly advantageously indicate that the determination algorithm has been sufficiently trained with regard to these and, if appropriate, also within an equivalence criterion of equivalent input data and/or task data, so that optimization can be assumed. While it is conceivable, for example, to run counters for certain equivalence classes of input data and/or task data in order to determine this, it can also be provided additionally or alternatively that the determination algorithm itself breaks down according to these equivalence classes or is trained for these equivalence classes. The output data is stored on the backend device, assigned to the corresponding input data or tasks, so that vehicles with a similar task, i.e. with input data and/or task data that at least meet the equivalence criterion, have access to this optimization carried out elsewhere. It is also particularly advantageous if, when the transmission criterion is met, the input data and/or task data and the output data, result data describing their effect are also transmitted to the backend device so that if, in particular within an equivalence class, matching input data and/or task data with associated output data from several motor vehicles are available, it can be checked which set is the better one.

Specifically, it can be provided that if there is input data and/or task data within the motor vehicle for which an ignorance criterion is met, which indicates that the motor vehicle's determination algorithm has not yet been sufficiently trained with these, the input data and/or task data are transmitted to the backend server, which checks whether output data from another motor vehicle is provided for input data or task data that are equivalent at least within the scope of an equivalence criterion, which are transmitted back to the motor vehicle and used there. This will be explained in more detail using an example.

If, for example, one of the motor vehicles regularly travels the same route, for example every working day, it is extremely likely that a particularly optimal solution for the input data or the resulting set of tasks described by task data will be found within this motor vehicle over time. If these findings are also made known to the backend device, a driver of another motor vehicle who wants to travel this route for the first time can also benefit from these findings.

In summary, it can be said that optimal operating strategies for certain input data and thus task data are regularly sent to the backend device, which accumulates these over time and makes them available to other vehicles. In order to be able to find operating parameters and result data that relate to the same or similar facts, in particular task data, the backend device can generally use methods of classification and clustering, in particular within the framework of equivalence criteria, in order to create appropriate comparison and optimization bases.

Artificial intelligence can also be used in the context of such an experience exchange system, for example to be able to assign the correct data sets to a requester. For example, so-called recommender systems can be used within the backend device and/or similar energy profiles can be found by using statistical similarity methods, such as cosine similarity, Jaccard similarity and/or Hamming similarity. The advanced techniques already mentioned, such as clustering and classification, can be improved using deep learning methods and other "big data" approaches, such as dimensionality reduction. Overall, a system can be implemented that links comparable energy profiles, in particular described by task data, for new routes and routes used frequently by other road users based on statistical similarity.

It should be noted that a corresponding exchange approach would also be conceivable in principle, independent of the use of reinforcement learning, if other optimization methods are available within individual motor vehicles in order to create increasingly improved output data, i.e. operating parameter sets, for certain sets of input data and/or task data.

As already explained, in a first embodiment it is possible for the determination algorithm to be carried out by a particularly dedicated control unit of the motor vehicle. Such a dedicated control unit can be, for example, an energy control unit. In this way, the operating parameters are determined directly within the motor vehicle, with the control unit providing the corresponding computing power to carry out the determination algorithm. In this way, the motor vehicle is therefore independent of a connection to a backend device as far as the optimization achieved by the determination algorithm is concerned. However, the optimization can then only benefit indirectly from the wealth of experience of other motor vehicles, for example via the exchange described or a higher-level use of machine learning on the backend based on the results of the individual reinforcement learning determination algorithms.

An alternative embodiment of the present invention therefore provides that the determination algorithm is executed on a vehicle-external backend device that receives the input data from the motor vehicle and that the operating parameters are transmitted to the motor vehicle. In this way, the determination algorithm can use the usually greater computing power of a backend device that communicates with the motor vehicles, for example via a mobile network, in particular for a large number of motor vehicles. The motor vehicle therefore sends the necessary input data for the determination algorithm to the backend device, where the The determination algorithm uses this information to determine operating parameters as output data, which are then transmitted back to the vehicle. There they are used accordingly by the control units.

In particular, if the determination algorithm that is carried out on the backend device is based on reinforcement learning, in particular deep reinforcement learning, it is also expedient if the motor vehicle transmits result data describing the effect of the operating parameters back to the backend device. Such result data, which can describe, for example, energy consumption and/or other effects caused by the operating parameters, can then be used, for example, to be evaluated using reward criteria and to promote the continuous improvement of the determination algorithm. If the determination algorithm on the backend device then uses the input data and result data from a large number of motor vehicles that are equivalent in terms of the on-board network, an excellent basis is provided for excellent optimization of the operation of the on-board network in each of these motor vehicles. Nevertheless, feedback from the motor vehicle is of course also useful with regard to other types of determination algorithm.

In addition to the method, the invention also relates to an overall system with the features of claim 11. The overall system comprises several motor vehicles, each with an on-board network, the on-board network comprising as components at least one rechargeable electric battery and an electric motor that can be used as a drive motor and generator, and at least one control unit assigned to at least one component for controlling the components according to operating parameters describing the energy consumption and energy output of the components, which overall system is characterized in that a central control unit and/or one of the at least one control unit assigned to at least one component is designed to determine the operating parameters as output data of a determination algorithm that uses artificial intelligence and/or is programmed by artificial intelligence, the determination algorithm determining the output data from input data that describe the current operating state of one of the motor vehicles, reinforcement learning being used as a training method for the artificial intelligence of the determination algorithm, input data of the respective motor vehicle being recorded when the determination algorithm is used during operation of the respective motor vehicle and at least a minimization of the Energy consumption is used, the backend device being set up so that output data with associated input data can be transmitted from the respective motor vehicle and other motor vehicles equivalent in terms of the on-board network to the backend device and can be made available to other motor vehicles at least when a transmission criterion is met. The motor vehicles are preferably electric vehicles. All statements relating to the method according to the invention can be transferred analogously to the overall system according to the invention, so that the advantages already mentioned can also be obtained with this. The determination algorithm can preferably comprise a neural network and/or be trained by reinforcement learning, in particular deep reinforcement learning, in particular with regard to the second step. Furthermore, each motor vehicle can preferably have a communication device for communicating with a backend device in order to receive, for example, optimal operating parameter sets from there and/or to transmit its own operating data to there, for example for the global optimization and improvement of energy management for a large number of motor vehicles.

• 1 einen prinzipiellen Ablaufplan des erfindungsgemäßen Verfahrens,
• 2 ein erstes Gesamtsystem zur Ausführung des Verfahrens, und
• 3 ein zweites Gesamtsystem zur Ausführung des Verfahrens.

Further advantages and details of the present invention emerge from the exemplary embodiments described below and from the drawings.

• 1 a basic flow chart of the method according to the invention,
• 2 a first overall system for carrying out the method, and
• 3 a second overall system for carrying out the procedure.

1 shows a basic flow chart of a method according to the invention for operating an on-board network of a motor vehicle, here an electric vehicle. In addition to a rechargeable electric battery, the on-board network comprises an electric motor for driving the motor vehicle, which can also be used as a generator, as well as other components, which in particular can also be located in a low-voltage part of the on-board network or can be a connecting link to a low-voltage network of the motor vehicle. The operation of the individual components of the on-board network is controlled via control units, each of which is assigned to at least one component and uses operating parameters to control the energy intake and/or energy output from or into the on-board network. To determine these operating parameters, a determination algorithm is used which uses artificial intelligence in the exemplary embodiments shown here.

Input data 1 of this algorithm describes the current operating state of the motor vehicle, but in particular at least its current position and a current destination, which can be known, for example, from data from a navigation system of the motor vehicle or can be estimated at least in the short term based on the current journey. Further input data can relate to the state of other vehicle systems of the motor vehicle, in particular also components of the on-board network, and/or environmental conditions of the motor vehicle. It is essential for the input data that they represent a basis for inferring future activities of the motor vehicle and thus tasks in the on-board network.

In a first step 2 of the determination algorithm, these input data 1 are used to predict task data 3 that describe these future tasks, for example acceleration periods, speeds, stationary phases and recuperation periods with regard to the electric motor and/or operating times and requirements for other components, for example components of an air conditioning system for heating and/or cooling in the motor vehicle. The tasks can relate to more or less narrowly defined periods within the prediction period, which can be 20 minutes, for example, or can also preferably be selected dynamically based on the reliability of the prediction. This means that certain tasks can, if necessary - which will ultimately be described by the operating parameters - be postponed to certain periods of time.

In the first step 2, a sub-algorithm of the determination algorithm is used, which can also use artificial intelligence, but does not necessarily have to. It is also conceivable to derive the task data 3 without using artificial intelligence, for example by using corresponding prediction models or the like. If the sub-algorithm of the first step 2 uses artificial intelligence, it is advisable to use classic machine learning methods, preferably sub-algorithms that include neural networks. This means that after the ground truth for predicting the task data will be known in the future, classic training data can be generated using target results.

In a second step 4, which is also carried out by a sub-algorithm of the determination algorithm, the task data 3 are then used to determine output data 5, which are or describe the operating parameters, by means of artificial intelligence, here again using a neural network. In the present exemplary embodiment, so-called deep reinforcement learning is used for the sub-algorithm of the second step 4, with the rewards used therein resulting from reward criteria that aim at minimal energy consumption and, secondarily, at the lowest possible heating of at least one of the components and the longest possible battery life. This means that the use of the operating parameters obtained as output data 5 ultimately produces effects that can be recorded in the motor vehicle using result data and can be evaluated by the reward criteria in order to evaluate the performance achieved and to learn from it in an adaptive manner, so that the sub-algorithm of the second step 4 is constantly improving with the determinations to be made.

2 shows an overall system in which a backend device 7, which can be a backend server and/or a backend cloud, for example, communicates via wireless communication connections 8, for example mobile phone connections, with communication devices of motor vehicles 9, which are not shown in detail here for the sake of clarity. The motor vehicles 9 are equivalent with regard to the on-board network 10 contained in each case, which is only indicated for one motor vehicle 9 for the sake of clarity, which means in particular that they are identical with regard to the equipment, at least for the on-board network 10. For example, the motor vehicles 9 can all be the same model and/or of the same type.

As already explained, the on-board network 10 comprises the battery 11, the electric motor 12 and other components 13. The control units 14 controlling the operation of the components 11, 12, 13 are also indicated. A central energy control unit 15 is provided in each of the motor vehicles 9 at a higher level than these control units 14, in which the determination algorithm 16 is carried out, as indicated.

Due to deep reinforcement learning, a confirmed improvement and optimization in terms of energy consumption and secondary criteria is now taking place in the motor vehicles 9 , in particular due to particularly frequent task constellations, which can be tracked accordingly within the motor vehicle. In the design according to 2 However, it is now planned that the input data 1, the output data 5 and the result data within the motor vehicles 9 are logged and at least those data sets for which a transmission criterion is met, which describes sufficient training on these task constellations, are sent to the Backend device 7, where, based on the input data 1 or task data 3 and a suitable classification/clustering, for example a summary of equivalence classes, comparable data sets are summarized and in turn evaluated in order to determine the optimal output data 5 for each equivalence class.

If sufficient reinforcement learning has taken place for an equivalence class of input data 1 or preferably task data 3, for example if a route has been driven repeatedly, it can be assumed that a certain optimization has taken place and the locally optimized results can be made available to other motor vehicles 9. If a new task constellation described by input data 1 and/or task data 3 occurs in the vehicle's own motor vehicle 9 or other motor vehicles 9 and has not yet been sufficiently trained (which can be checked using a lack of knowledge criterion), a request can be sent to the backend device 7, which uses an equivalence criterion to check whether output data 5 from other motor vehicles 9, which were sent based on the transmission criterion, are already available for comparable task constellations. And the backend device 7 transmits the corresponding operating parameter set found back to the requesting motor vehicle 9 so that this output data 5 can be used there.

In this way, all motor vehicles 9 benefit from the learning processes of the detection algorithms in the other motor vehicles 9.

3 shows a further constellation in which the determination algorithm 16 for all motor vehicles 9 is carried out centrally in the backend device 7, thus benefiting there anyway from the operating states and feedback/result data of all motor vehicles 9.

Claims (11)

Method for operating an electrical on-board network (10) of a motor vehicle (9), wherein the on-board network (10) comprises as components at least one rechargeable electric battery (11) and an electric motor (12) that can be used as a drive motor and generator, and at least one control device (14) assigned to at least one component is used to control the components according to operating parameters describing the energy consumption and energy output of the components, wherein the operating parameters are determined as output data (5) of a determination algorithm (16) that uses artificial intelligence and/or is programmed by artificial intelligence, which determines the output data (5) from input data (1) that describe the current operating state of the motor vehicle (9), wherein reinforcement learning is used as a training method for the artificial intelligence of the determination algorithm (16), wherein input data (1) of the motor vehicle (9) is recorded when the determination algorithm (16) is used during operation of the motor vehicle (9), and at least one minimization of energy consumption is used as a reward criterion for reinforcement learning, wherein output data (5) with associated input data (1) are transmitted from the motor vehicle (9) and other motor vehicles (9) equivalent in terms of the on-board network (10) to a backend device (7) at least when a transmission criterion is fulfilled and are made available to other motor vehicles (9). Procedure according to Claim 1 , characterized in that the determination algorithm (16) of the artificial intelligence comprises a neural network. Procedure according to Claim 1 or 2 , characterized in that initially in a first step (2) from the input data (1) for a prediction period within the vehicle electrical system (10) pending task data (3) describing tasks relating to electrical energy flow between the components are predicted, for which the output data (5) are determined in a second step (4). Procedure according to Claim 3 , characterized in that different sub-algorithms of artificial intelligence are used for the first and the second step (2, 4). Procedure according to Claim 3 or 4 , characterized in that output data (5) with associated input data (1) and task data (3) are transmitted from the motor vehicle (9) and other motor vehicles (9) equivalent with regard to the on-board network (10) to the backend device (7) and are made available to other motor vehicles (9) at least when the transmission criterion is met. Method according to one of the preceding claims, characterized in that if input data (1) and/or task data (3) are present within the motor vehicle for which an ignorance criterion is met, which indicates that the determination algorithm (16) of the motor vehicle (9) has not yet been sufficiently trained with these, the input data (1) and/or task data (3) are transmitted to the backend device (7). which checks whether output data (5) of another motor vehicle (9) are provided for input data (1) or task data (3) that are equivalent at least within the scope of an equivalence criterion, which output data are transmitted back to the motor vehicle (9) and used there. Method according to one of the preceding claims, characterized in that the determination algorithm (16) is executed on a vehicle-external backend device (7) receiving the input data (1) from the motor vehicle (9) and the operating parameters are transmitted to the motor vehicle (9). Procedure according to Claim 7 , characterized in that the motor vehicle (9) transmits result data (6) describing the effect of the operating parameters back to the backend device (7). Method according to one of the preceding claims, characterized in that the on-board network (10) has as a further component (13) at least one consumer and/or at least one low-voltage component and/or power electronics for connection to a low-voltage network. Method according to one of the preceding claims, characterized in that the input data (1) comprise at least a current position of the motor vehicle (9) and a current destination of the motor vehicle (9). Overall system comprising a backend device (7) and several motor vehicles (9), each with an on-board network (10), the on-board network (10) comprising as components at least one rechargeable electric battery (11) and an electric motor (12) that can be used as a drive motor and generator, and at least one control unit (14) assigned to at least one component for controlling the components according to operating parameters describing the energy consumption and energy output of the components, a central control unit (15) and/or one of the at least one control unit (14) assigned to at least one component is designed to determine the operating parameters as output data (5) of a determination algorithm (16) that uses artificial intelligence and/or is programmed by artificial intelligence, the determination algorithm (16) determining the output data (5) from input data (1) that describe the current operating state of one of the motor vehicles (9), reinforcement learning being used as a training method for the artificial intelligence of the determination algorithm (16), input data (1) of the respective motor vehicle (9) being used when the Determination algorithm (16) is recorded during operation of the respective motor vehicle (9) and at least a minimization of energy consumption is used as a reward criterion of reinforcement learning, wherein the backend device (7) is set up so that output data (5) with associated input data (1) can be transmitted from the respective motor vehicle (9) and other motor vehicles (9) equivalent in terms of the on-board network (10) to the backend device (7) at least when a transmission criterion is met and can be made available to other motor vehicles (9).

Images