Method for providing a predictive model for determining the amount of electrical energy consumed by a vehicle, computer-implemented method for determining a vehicle configuration recommendation, and method for setting a configuration parameter for a portal

Abstract

The invention relates to a method for providing an artificial intelligence-based prediction model for determining the amount of electrical energy consumed during a journey with a battery-electric vehicle. The method according to the invention is characterized by the following method steps: a) collecting journey data (101, 102) by the vehicles of a vehicle fleet; b) feeding the driving data (101) to a combustion engine machine learning model (106); c) Conversion of combustion engine energy consumption parameters into initial electrical energy consumption parameters based on a defined set of rules; d) feeding the travel data (102) to an electrical machine learning model (107); and e) feeding the trip data (101, 102) and the first and second electrical energy consumption parameters to an energy consumption machine learning model (108), wherein the trip data (101, 102) are used as ground truth, so that the energy consumption machine learning model (108) learns to estimate the electrical energy consumption (109) of electric motor-driven vehicles (BATT) from trip data (101, 102) while neutralizing an overfitting imposed by the combustion engine machine learning model (106) and the electrical machine learning model (107).

Description

[0001] The invention relates to a method for providing a prediction model based on artificial intelligence according to the type defined in more detail in the preamble of claim 1, a computer-implemented method for determining a vehicle configuration recommendation and a method for specifying a configuration parameter for a portal for configuring a vehicle to be manufactured or already manufactured.

[0002] The share of purely battery-electric vehicles in traffic is constantly increasing. The traction battery is a particularly relevant component of electrically powered vehicles. The traction battery is one of the most expensive components of a vehicle, and the vehicle's range depends on the capacity of the traction battery. Since charging electrical drive energy takes a comparatively long time compared to refilling liquid fuel, the available range of an electric vehicle is particularly important for users. Therefore, a compromise must be made between a large and expensive traction battery, which is associated with a long range, and a small, but cheaper traction battery, which, however, results in a reduced range.It is desirable to provide methods and means to identify suitable traction battery capacity for the vehicles offered in a particular market.

[0003] DE 10 2011 014 413 A1 discloses a system and method for predicting vehicle energy consumption. The vehicle energy consumption is based on an extrapolation of statistical information about the driving history. For this purpose, the driving behavior of a specific person is determined and a user-specific vehicle energy consumption is calculated. It can be assumed that the user is using a battery-electric vehicle. Depending on the calculated vehicle energy consumption, a detailed vehicle recommendation can be issued to the user.

[0004] A system and method for simulating the performance of a virtual vehicle is also known from US 2013 / 0253782 A1. This document describes the calculation of performance parameters of differently configured vehicles. The performance parameters are compared to suggest a cost-optimal vehicle for a user. Fuel consumption for each vehicle can also be estimated and compared.

[0005] The present invention is based on the object of providing methods and means with which it is possible to give an improved vehicle configuration recommendation.

[0006] According to the invention, this object is achieved by a method for providing a prediction model based on artificial intelligence having the features of claim 1 and a computer-implemented method for determining a vehicle configuration recommendation having the features of claim 4. Advantageous embodiments and further developments as well as a method for specifying a configuration parameter for a portal emerge from the dependent claims.

[0007] A generic method for providing a prediction model based on artificial intelligence for determining the amount of electrical energy consumed during a journey with a battery-electric vehicle is further developed according to the invention by the following method steps: a) collecting journey data by the vehicles of a vehicle fleet for a training period and aggregating the journey data on a central computing device, wherein the vehicle fleet comprises at least one internal combustion engine-powered vehicle and at least one battery-electric vehicle, and wherein the journey data for each journey describe the respective environmental conditions, journey conditions and vehicle performance parameters; b) feeding the driving data of combustion engine-powered vehicles to a combustion engine machine learning model as ground truth, so that the combustion engine machine learning model learns to predict combustion engine energy consumption parameters; c) Conversion of the combustion engine energy consumption parameters into initial electrical energy consumption parameters based on a defined standard rate; d) feeding the driving data of electric motor-driven vehicles to an electric machine learning model as ground truth, so that the electric machine learning model learns to predict second electric energy consumption parameters; and e) feeding the trip data of combustion engine-powered vehicles and electric motor-powered vehicles as well as the first and second electric energy consumption parameters to an energy consumption machine learning model, wherein the trip data are used as ground truth, so that the energy consumption machine learning model learns to estimate the electric energy consumption of electric motor-powered vehicles from trip data while neutralizing an overfitting imposed by the combustion engine machine learning model and the electric machine learning model.

[0008] The method according to the invention serves to develop the combustion engine machine learning model, the electric machine learning model, and the energy consumption machine learning model or to make them available for later use. The respective machine learning models can, in particular, be artificial neural networks. They are therefore, in particular, so-called deep learning models. The combustion engine machine learning model and / or the electric machine learning model preferably comprise three hidden and fully connected layers. Training is preferably carried out using MSE loss and the so-called Adam Optimizer.

[0009] The method according to the invention therefore also considers the usage behavior of combustion engine-powered vehicles when estimating the energy consumption of battery-electric vehicles. It should be noted that not only the amount of electrical drive energy, but also the total electrical energy consumption and the amount of electrical energy required to cover this energy consumption are taken into account. For this purpose, the energy consumption of said combustion engine-powered vehicles is converted into a corresponding equivalent for electric motor-powered vehicles.

[0010] This means that an even broader database is available for training the machine learning models, so that even more realistic results can be obtained.

[0011] To generate the database, the aforementioned travel data is collected by the vehicles in a fleet and transmitted to a central computing device for further processing. The central computing device may be, for example, a server or server network, also referred to in this context as a cloud server or backend. For communication with the central computing device, the vehicles in the fleet may include suitable communication tools, such as a telecommunications unit. Such a telecommunications unit enables data to be sent via a mobile internet connection.

[0012] The trip data describes the environmental conditions, driving conditions, and vehicle performance parameters present during a particular trip. The environmental conditions are those conditions that exist in the environment when a particular trip is carried out, such as the ambient temperature, weather, traffic density, and the like. The driving conditions describe all information that can be used to define a trip, such as the starting point and destination of the trip, the start time, the arrival time, the route traveled, the distance traveled, and the like. The transition between the individual pieces of information is fluid. For example, the gradient of the route traveled can be assigned to both the environmental conditions and the driving conditions.The vehicle performance parameters describe all vehicle parameters present during a particular journey, particularly those that are time-dependent. These include, for example, vehicle speed, longitudinal acceleration, engine torque, drive energy consumption, the power consumption of additional electrical consumers, and the like.

[0013] The training period can vary in length. The training period is at least long enough to collect sufficient driving data to ensure a desired minimum prediction quality for the respective machine learning models. The larger the vehicle fleet, the shorter the training period can be. For example, the training period lasts a few months or years. It can also be a rolling period, so that the corresponding machine learning models are continuously trained. Due to the ever-increasing amount of data used for training, the prediction quality of the machine learning models can also be gradually improved.

[0014] The combustion engine machine learning model is trained using the driving data from the combustion engine-powered vehicles, and the electric machine learning model is trained using the driving data provided by the electric motor-powered vehicles. Before the respective driving data is fed into the machine learning models, it can be grouped. Proven cluster ring algorithms, such as the k-means algorithm, can be used for this purpose. For example, 3 can be used as the value for k. The optimal group dispersion value, such as the silhouette coefficient, can be used to determine k.

[0015] Preferably, the combustion engine machine learning model and the electric machine learning model are so-called "deep learning regression models." The external boundary conditions, i.e., the environmental boundary conditions and the driving boundary conditions, are used as independent variables. At least some of the vehicle performance parameters, at least the vehicle's energy consumption during a particular journey, serve as dependent variables. The total energy consumption is composed of various partial energy consumptions, such as drive energy consumption, the energy consumption of additional consumers, for example, an air conditioning system or a device for playing media content, the operation of the vehicle's headlights, the operation of a seat heater, and the like.

[0016] Several trips can be transferred to a respective machine learning model in a group, for example, grouped together over a period of one or more weeks.

[0017] The combustion engine machine learning model and the electric machine learning model allow us to identify the influence of different vehicle components on propulsion energy consumption patterns. This allows us to establish a correlation between the corresponding components and their respective influence on propulsion energy consumption.

[0018] Since the target variable is the (total) energy consumption of a battery-electric vehicle, particularly in the form of an amount of energy and not just a time-dependent power output, corresponding consumption parameters of combustion engine-powered vehicles must be converted, which is done in step c). The efficiency of the vehicle components of combustion engine-powered vehicles and battery-electric vehicles, as well as the respective energy flow within the vehicle, are fundamentally different, so simply converting liquid fuel consumption into a quantity of electrical energy consumed is not expedient. The conversion of the combustion engine energy consumption parameters into initial electrical energy consumption parameters is based on a defined set of rules. This set of rules can, for example, be defined by a developer, in particular the vehicle manufacturer.Driving data can also be analyzed to find suitable rules. The respective energy consumption depends in particular on the selected driving speed, the ambient temperature, the driving duration, the driving distance, and the like. The efficiency of the drive unit, the energy requirements of additional consumers, the propulsion energy consumption, the electrical energy recovered through recuperation, and the consumption of liquid fuel have been identified as influencing factors for determining energy consumption or energy requirements.

[0019] By applying the set of rules, it is possible to estimate how high the corresponding energy consumption of electrical energy would be for a vehicle powered by an internal combustion engine if the vehicle were designed as an electric motor-powered vehicle.

[0020] The first and second electrical energy consumption parameters are then fed into the energy consumption machine learning model. This is also an AI model, which, together with the combustion engine machine learning model and the electric machine learning model, forms a hierarchical model for determining trip-specific energy consumption. The training of the energy consumption machine learning model is based on trips recorded with both combustion engine-powered vehicles and electric motor-powered vehicles. The task of the energy consumption machine learning model is to improve the accuracy of the prediction by removing the aforementioned overfitting. The final result is the expected electrical energy consumption for a given trip.

[0021] An advantageous development of the method according to the invention provides that the driving data is cleaned and normalized before being fed to at least one of the combustion engine machine learning model or the electric machine learning model. This facilitates corresponding processing or interpretation by the respective machine learning model. Irrelevant data can be discarded from a corresponding data set. Every vehicle has a wide variety of control units that monitor a wide variety of tasks and provide a wide variety of functions. This makes it possible to operate a wide variety of subsystems in the vehicle. Due to the limited computing capacity available in the vehicle, the data processed in the vehicle is typically aggregated and summarized. This may require the corresponding driving data to be prepared before being read into said machine learning models.For example, individual parameters can be integrated over a relevant period of time or subjected to other mathematical operations to convert the data to a specific target format or calculate a dependent variable. Corresponding data can also be normalized, which allows for easier comparison of the data.

[0022] According to a further advantageous embodiment of the method according to the invention, in step c), the amount of electrical drive energy consumed by an electric motor-driven vehicle is calculated from the amount of mechanical propulsion energy required by a combustion engine-driven vehicle to generate propulsion. As already mentioned, directly converting liquid fuel consumption into consumed electrical drive energy is not expedient. Therefore, the conversion is preferably based on the amount of mechanical propulsion energy required by the combustion engine-driven vehicle to generate propulsion. This allows the calculation of an electrical energy quantity that better corresponds to the actual electrical energy consumption. The amount of mechanical propulsion energy is the energy used to overcome driving resistance during the entire journey.These include rolling resistance, air resistance, inclination resistance, acceleration resistance and the like.

[0023] By taking into account other energy consumption parameters, such as the aforementioned active additional consumers, electrical energy recovered from recuperation and the like, the electrical energy consumption of the electric motor-driven vehicle can then be determined particularly reliably.

[0024] According to the invention, a computer-implemented method for determining a vehicle configuration recommendation comprises the following method steps: - feeding journey data collected during a relevance period and a specified number of vehicles, comprising a number of combustion engine-powered vehicles and a number of electric motor-powered vehicles, to a generative adversarial network to generate synthetic journey data for a future period; - Reading the synthetic driving data for combustion engine-powered vehicles into a combustion engine machine learning model trained according to a method described above; - Reading the synthetic driving data for electric motor-driven vehicles into an electric machine learning model trained according to a method described above; - feeding the results of the combustion engine machine learning model and the electric machine learning model to an energy consumption machine learning model trained according to a method described above; and - Determining a traction battery capacity recommendation depending on the trip-specific electrical energy consumption determined by the energy consumption machine learning model.

[0025] With the aid of the computer-implemented method according to the invention for determining the vehicle configuration recommendation, it is thus possible to make a recommendation for the capacity of a traction battery to be provided in an electric motor-driven vehicle based on the trip data collected during the relevant period and the use of said trained machine learning models. For this purpose, the respective electrical energy consumption is examined for each trip. A traction battery capacity is then recommended that allows comfortable use of the respective vehicle for the respective trip. This can mean, for example, that for a specified number of trips, no charging stop at a charging station to recharge electrical drive energy should be made.A tolerated number of charging stops, particularly taking into account a desired maximum charging time, could also be specified as a boundary condition for determining the appropriate traction battery capacity.

[0026] Preferably, not just a single traction battery capacity recommendation is provided, but rather a range of possible traction battery capacities. This allows the specification that, for a specific vehicle model, the traction battery to be installed in the vehicle should have a capacity between a lower and upper value determined in this way.

[0027] A wide variety of relevance periods can be considered. The relevance period can coincide with the training period or deviate from it partially or entirely. For example, a specific calendar year, such as 2017, can be selected as the relevance period. The relevance period can also be a sliding period, such as the most recent year, the most recent calendar year, or even fractions or multiples thereof.

[0028] The specified number of vehicles can particularly preferably be a forecast from a vehicle manufacturer for sales figures for a specific vehicle model or models. For example, it could be the number of vehicles the vehicle manufacturer is expected to sell in a specific region next year. Based on the previous behavior of the users of the vehicles in the vehicle fleet and the forecast sales figures, an estimate can then be made of how many and which journeys are expected for the future period. The future period can also have any fixed or sliding duration. The future period is preferably the next calendar year.

[0029] A generative adversarial network is an unsupervised learning algorithm. A generative adversarial network comprises two artificial neural networks. The first network is called the generator, and the second network is called the discriminator. Typically, the generator maps a vector of latent variables to a desired result space. The goal of the generator is to learn to generate results according to a specific pattern. The task of the discriminator is to recognize whether the data generated by the generator corresponds to real input data or was artificially generated by the generator. This is used to adapt the generator so that it produces results that can no longer be distinguished by the discriminator. In the context of the invention, the generator serves to artificially generate synthetic driving data.To generate the synthetic driving data, the generator can read in a vector comprising random noise as an additional input variable.

[0030] According to a further advantageous embodiment of the method according to the invention, individual traction battery capacity recommendations are determined for different geographical regions. Different geographical regions, for example countries, continents, or the like, differ from one another in many ways. Due to different mentalities, the driving behavior of people assigned to a particular geographical region can differ. External conditions also differ, such as climatic conditions, the average applicable speed limit, topography, and the like. In sparsely populated areas, journeys with long journey distances and journey times are more frequently made than in densely populated areas, such as a metropolitan area or a large city.Thus, the expected electrical energy consumption for specific journeys between different geographical regions will also vary. Therefore, it is preferable to calculate different traction battery capacity recommendations for different geographical regions. For example, a different traction battery capacity recommendation can be determined for each calendar year for Asia, Europe, North America, and Australia.

[0031] As already mentioned, a duration of one year is preferred for the future period. Shorter or longer time periods could also be considered.

[0032] A method for defining a configuration parameter for a portal for configuring a vehicle to be manufactured or manufactured provides according to the invention that a traction battery capacity recommendation determined according to a method described above is selected as a minimum or maximum traction battery capacity of a traction battery available for the vehicle that can be selected in the portal.

[0033] The portal could, for example, be an online portal for configuring a vehicle available for purchase from the vehicle manufacturer, also known as a "configurator." The vehicle manufacturer may offer purely battery-electric vehicles. Various traction batteries with varying capacities can be selected via the portal. A larger capacity also entails increased costs. Using a procedure described above, traction battery capacity recommendations can be determined for specific regions and time periods. The vehicle manufacturer can use these recommendations to determine the model-specific, configurable or buildable vehicle configurations.For example, it can be determined that traction batteries that allow a range of between 600 and 900 kilometers can be installed in a specific vehicle model in Germany by 2025. This ensures that the vehicle manufacturer's customers select the optimal traction battery for their needs. This prevents a customer from choosing an oversized and therefore too expensive traction battery, or an undersized traction battery that would allow for too limited a range, in their desired vehicle.

[0034] Preferably, the maximum usable traction battery capacity in the vehicle can be set via the portal for an order period to any value between zero and the actual maximum traction battery capacity of the traction battery installed in the vehicle. In other words, the usable traction battery capacity in the vehicle can be changed according to a subscription model. For this purpose, an actually oversized traction battery is installed in a vehicle, whereby the respective vehicle user can adjust the actual usable traction battery capacity in the vehicle depending on their current needs. If the user uses their vehicle frequently, they can book a high traction battery capacity for the corresponding period, while they can book a lower traction battery capacity during a period when the vehicle is used less frequently.The traction battery capacity provided by the vehicle is electronically locked or released by a corresponding battery management system. This allows the customer to save even more, as the full price of a particularly large traction battery does not have to be paid all at once when purchasing the vehicle. This gives the customer a great deal of flexibility. If long journeys are planned, for example during vacation time, the full traction battery capacity can be booked for that period as needed. If, however, the vehicle user only commutes to work, a reduced maximum usable traction battery capacity can be set. The respective order period can vary in length. Predefined order periods, for example, weekly or monthly, can be specified via the portal.The order period can also be freely specified by the user, for example using a digital calendar, by entering a specific start date and end date.

[0035] For example, a specific vehicle model may be equipped with a traction battery capable of providing a technically feasible range of 1,000 kilometers. Based on the traction battery capacity recommendation made above, it is determined that for Germany, the maximum usable traction battery capacity for 2026 at the time of the vehicle model's sales launch will be limited to 600 kilometers. The user can then change the corresponding usable traction battery capacity at any time via the portal. For example, the user can book a range of 400 kilometers from January to March, 500 kilometers from April to June, 800 kilometers from July to August, and 500 kilometers from September to December.

[0036] To provide the respective method steps comprised by the respective methods, the invention also encompasses respective computer program products, comprising computer-interpretable instructions that enable a processor of a computing system, for example a computer, to execute the respective method steps. A computer-readable storage medium storing such a computer program product also falls within the scope of the invention.

[0037] Further advantageous embodiments of the method according to the invention for providing a prediction model based on artificial intelligence for determining the amount of electrical energy consumed during a journey with a battery-electric vehicle and of the computer-implemented method for determining a vehicle configuration recommendation also emerge from the exemplary embodiments which are described in more detail below with reference to the figures.

[0038] Showing: Fig. 1 shows a schematic model architecture of a system for estimating the electrical energy consumption of an electric motor-driven vehicle; Fig. 2 a schematic model architecture of a system for determining a traction battery capacity recommendation; Fig. 3 a schematic representation of the process flow for training a combustion engine machine learning model and an electric machine learning model; and Fig. 4 a schematic representation of an energy flow diagram for a vehicle powered by an internal combustion engine and an electric motor.

[0039] Using a method according to the invention, a recommendation for the capacity of a traction battery that can be installed in a vehicle is to be determined, in particular a minimum and a maximum selectable capacity. Various machine learning models are used for this purpose. Fig. 1 shows a system setup of the system used to train the machine learning models. For example, the Fig. The system structure shown in Figure 1 can be implemented in a central computing facility such as a cloud server.

[0040] First, trip data 101, 102 are provided, which were collected from the vehicles of a vehicle fleet within a training period. The trip data 101 originate from combustion engine-powered vehicles VERB (see Fig. 4) and the trip data 102 come from battery-electric vehicles BATT (see also Fig. 4). Respective trip data 101, 102 each comprise environmental boundary conditions 103, trip boundary conditions 104 and vehicle performance parameters 105. Since the environmental boundary conditions 103 and the trip boundary conditions 104 are vehicle-extrinsic values ​​in the extended sense, they are shown as adjacent boxes in Fig. 1, wherein the vehicle performance parameters 105 represent vehicle-intrinsic parameters and are therefore shown separately.

[0041] The trip data 101, 102 also contain the actual energy consumption of the respective vehicles during the journey. This can be defined as time-dependent power. Through time integration, the energy consumption can be converted into energy quantities. In addition to pure drive energy consumption, other consumption or energy conversions are included, such as energy recovered through recuperation, the power requirements of additional consumers, and the like. Energy consumption or an energy quantity can be specified in the form of joules, liters of fuel, liters of fuel per hour, liters of fuel per 100 kilometers, watts, kilowatt hours, kilowatt hours per hour, kilowatt hour per 100 kilometers, and the like.

[0042] The respective trip data 101, 102 can be cleaned and normalized by an optional processing module 110. In particular, corresponding information contained in the trip data 101, 102 can be subjected to mathematical operations such as time integration or distance integration.

[0043] The respective, possibly processed, driving data 101 are fed to a combustion engine machine learning model 106. The combustion engine machine learning model 106 is thereby trained to predict combustion engine energy consumption parameters, in particular liquid fuel consumption, depending on the environmental boundary conditions 103, driving boundary conditions 104, and the vehicle performance parameters 105.

[0044] By means of a conversion module 111, the combustion engine energy consumption parameters are converted into first electrical energy consumption parameters based on a defined set of rules.

[0045] The optionally processed trip data 102 are fed to an electric machine learning model 107. This trains the electric machine learning model 107 to predict second electric energy consumption parameters, in particular an electric drive energy consumption, depending on the environmental conditions 103, trip conditions 104, and vehicle performance parameters 105. The respective output variables supplied by the combustion engine machine learning model 106 and the electric machine learning model 107 are then fed, together with the trip data 101 and 102, to an energy consumption machine learning model 108. This model is thereby trained to estimate the electric energy consumption 109 occurring during a respective trip for corresponding battery-electric vehicles BATT.The results provided by the combustion engine machine learning model 106 and the electric machine learning model 107 may exhibit overfitting. The overfitting can be at least partially neutralized by using the energy consumption machine learning model 108, which improves the accuracy of the estimation of the electrical energy consumption 109.

[0046] The data collected by means of the Fig. The machine learning models 106, 107, 108 trained in the system setup shown in Figure 1 can advantageously be implemented in a Fig. 2 can be used to determine a vehicle configuration recommendation.

[0047] In this case, trip data 201, 202 and a specified number of vehicles 203 collected for a relevant period are fed to a generative adversarial network 204. The generative adversarial network 204 is capable of forecasting synthetic trip data for a future period, for example, the next calendar year, from the received input data. These synthetic trip data are in turn fed to the aforementioned trained combustion engine machine learning model 106 and electric machine learning model 107. Here, too, combustion engine energy consumption parameters can optionally be converted into initial electric energy consumption parameters using a corresponding conversion module 111.

[0048] The results output by the combustion engine machine learning model 106 and the electric machine learning model 107 are then fed to the trained energy consumption machine learning model 108, which determines corresponding electrical energy consumption 109 for the synthetic trip data. Based on the trip-specific electrical energy consumption 109, respective traction battery capacity recommendations 205 are then determined. For example, a statistical mean energy consumption for an average trip is 110 kWh. This allows a traction battery capacity recommendation 205 of, for example, 120 kWh to be determined (mean plus safety factor).To determine the traction battery capacity recommendation 205, various variables can be taken into account, such as consumption resulting from engine power and efficiency, for example 34.9 kWh / 100 km, an average distance from the travel data, for example in the range of 49 km to 110 km, and the like. Particularly preferably, the traction battery capacity recommendation 205 is a traction capacity range whose lower and upper limits indicate how small the minimum traction battery capacity to be provided in a vehicle should be, and how large the maximum traction battery capacity to be provided should be. A vehicle manufacturer can use this information to determine which capacities it should offer its customers for the traction batteries of a specific vehicle model. Such vehicles are built accordingly by the vehicle manufacturer.These vehicles can be used reliably by customers by providing the requested range. Since the costs of such vehicles are limited by limiting the maximum traction battery capacity, customer satisfaction increases. Following a subscription model, the usable traction battery capacity in the vehicle can be adjusted as needed.

[0049] Fig. Figure 3 shows in detail the steps performed during training of the combustion engine machine learning model 106 and the electric machine learning model 107. In a step 301, trip data 101, 102 relevant for training are obtained. In step 302, the corresponding information is grouped into energy-efficient trips 303, energy-optimal trips 304, and energy-intensive trips 305. This information is then fed into a deep learning regression model 306. The goal of the deep learning regression model 306 is to estimate the respective electrical energy consumption or liquid fuel consumption of a trip.

[0050] Fig. 4 shows the energy flow diagram used by the conversion module 111 for a respective combustion engine-driven vehicle VERB and electric motor-driven vehicle BATT.

[0051] An electrical energy quantity 401 indicates the amount of electrical energy present within the vehicle's traction battery. An energy quantity 402 describes the amount of energy consumed by an electric drive motor of the vehicle. Taking into account the efficiency of the electric drive motor, an energy quantity 403 is provided that corresponds to the amount of propulsion energy. An energy quantity 404 can be provided through recuperation. Taking into account a corresponding recuperation system efficiency, an energy quantity 405 can be made available as usable recuperation energy. Finally, an energy quantity 406 is consumed by additional consumers.

[0052] The internal combustion engine-powered vehicle VERB includes a tank containing liquid fuel. This stores chemical energy 407 depending on the fill level and fuel type. Taking into account the corresponding combustion processes of the drive unit, an amount of energy 408 can be generated that corresponds to the amount of mechanical energy converted during a journey. Part of this energy can be fed to an alternator to generate an amount of electrical energy 409. This is partly used to operate corresponding additional consumers. These consume an amount of energy 410 analogous to the amount of energy 406. Part of the amount of energy 408 is used as propulsion energy 411. The amounts of energy 403 and 411 can be converted into one another. In a plug-hybrid vehicle, the recovery of kinetic energy into electrical energy through recuperation is also possible.Taking into account the corresponding resistances, a corresponding amount of recuperation energy 412 can be determined.

[0053] Instead of energy quantities, instantaneous power can also be considered. These quantities can be converted into one another at will, since the respective operating map of the vehicles is described by a specific known journey. 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 2011 014 413 A1

[0003] US 2013 / 0253782 A1

[0004]

Claims

[1] Method for providing a prediction model based on artificial intelligence for determining the amount of electrical energy consumed during a journey with a battery-electric vehicle, characterized by the following procedural steps: a) collecting trip data (101, 102) by the vehicles of a vehicle fleet for a training period and aggregating the trip data (101, 102) on a central computing device, wherein the vehicle fleet comprises at least one internal combustion engine-powered vehicle (VERB) and at least one battery-electric vehicle (BATT), and wherein the trip data (101, 102) describe the respective environmental boundary conditions (103), trip boundary conditions (104) and vehicle performance parameters (105) for each trip; b) feeding the travel data (101) of combustion engine-powered vehicles (VERB) to a combustion engine machine learning model (106) as ground truth, so that the combustion engine machine learning model (106) learns to predict combustion engine energy consumption parameters; c) Conversion of the combustion engine energy consumption parameters into initial electrical energy consumption parameters based on a defined standard rate; d) feeding the trip data (102) of electric motor-driven vehicles (BATT) to an electric machine learning model (107) as ground truth, so that the electric machine learning model (107) learns to predict second electric energy consumption parameters; and e) feeding the trip data (101, 102) of internal combustion engine-driven vehicles (VERB) and electric motor-driven vehicles (BATT) as well as the first and second electrical energy consumption parameters to an energy consumption machine learning model (108), wherein the trip data (101, 102) are used as ground truth, so that the energy consumption machine learning model (108) learns to estimate the electrical energy consumption (109) of electric motor-driven vehicles (BATT) from trip data (101, 102) while neutralizing an overfitting imposed by the combustion engine machine learning model (106) and the electric machine learning model (107). [2] Method according to claim 1, characterized by that the travel data (101, 102) are cleaned and normalized before being fed to at least one of the combustion engine machine learning model (106) or electric machine learning model (107). [3] Method according to claim 1 or 2, characterized bythat in step c) the amount of electrical drive energy (403) consumed by an electric motor-driven vehicle (BATT) is calculated from the amount of mechanical propulsion energy (411) required by an internal combustion engine-driven vehicle (VERB) to generate propulsion. [4] Computer-implemented method for determining a vehicle configuration recommendation, characterized by the following procedural steps: - feeding trip data (201, 202) collected during a relevance period and a specified number of vehicles (203), comprising a number of internal combustion engine-powered vehicles (VERB) and a number of electric motor-powered vehicles (BATT), to a generative adversarial network (204) for generating synthetic trip data for a future period; - reading the synthetic driving data for vehicles powered by internal combustion engines (VERB) into an internal combustion engine machine learning model (106) trained according to a method according to one of claims 1 to 3; - reading the synthetic driving data for electric motor-driven vehicles (BATT) into an electric machine learning model (107) trained according to a method according to one of claims 1 to 3; - feeding the results of the combustion engine machine learning model (106) and the electric machine learning model (107) to an energy consumption machine learning model (108) trained according to a method according to one of claims 1 to 3; and - Determining a traction battery capacity recommendation (205) as a function of the trip-specific electrical energy consumption (109) determined by the energy consumption machine learning model (108). [5] Method according to claim 4, characterized bythat individual traction battery capacity recommendations (205) are determined for different geographical regions. [6] Method according to claim 4 or 5, characterized by that a duration of one year is used for the future period. [7] Method for setting a configuration parameter for a portal for configuring a vehicle to be manufactured or manufactured, characterized by that a traction battery capacity recommendation (205) determined according to a method according to one of claims 4 to 6 is selected as a minimum or maximum traction battery capacity of a traction battery available for the vehicle that can be selected in the portal. [8] Method according to claim 7, characterized bythat the maximum usable traction battery capacity in the vehicle can be set via the portal for an order period to any value between zero and the actual maximum traction battery capacity of the traction battery installed in the vehicle.

Images