DE102023002489A1 - Weather data-based predictive management of an electric vehicle charging station - Google Patents

Abstract

The invention relates to a charging system comprising at least one charging station (1) connected to a power grid (5) for supplying electric vehicles (3) with electrical energy, an energy storage device (7) connected to the power grid (5) and the charging station (1), and a control unit (9), wherein the control unit (9) is designed to determine i) a current and future time-dependent power grid share and a complementary energy storage share of the electrical current of the charging station (1) for a respective electric vehicle (3) and ii) a current and future time-dependent energy state increase of the energy storage device (7) by drawing energy from the power grid (5) by executing a machine learning model adapted to the individual charging system with current and future weather data from a weather forecast and with a determined current and future time-dependent requirement for charging energy quantity and charging power for vehicles (3) as the respective input variable of the machine learning model.

Description

The invention relates to a charging system comprising at least one charging station connected to a power grid for supplying electric vehicles with electrical energy.

The widespread expansion of electromobility and the corresponding charging infrastructure is giving rise to new problems, e.g. when setting up a new charging park or several neighboring charging stations, high costs are necessary due to the corresponding power requirements, for example to lay new power lines over several kilometers.

Charging stations or charging parks are known which have an integrated power storage unit or power storage units in the vicinity of which charging stations are arranged, for example to cover peak power when many vehicles are charging in parallel at high power. The disadvantage of the previously known combination of charging stations and power storage units arranged nearby is that they only have simple controls, so that situations can still arise where the required peak power cannot be covered or, on the other hand, the arrangement is greatly over-dimensioned (in terms of the installed storage capacity) or an increase in the connected load may be necessary. Both are particularly disadvantageous because either the charging power is capped when peak power is required, or a greatly over-dimensioned installed capacity or connected load must be provided. Up to now, charging parks or charging stations have typically only been set up at locations where a correspondingly high connected load is available or at least nearby. Often, the places with a high available connection capacity do not coincide with places where there is a corresponding volume of traffic and a corresponding need for charging infrastructure. In addition, more and more vehicle manufacturers are setting up their own charging network, especially in places (e.g. dealers, workshops) etc. that are already owned by the car manufacturers or in a particularly lucrative area where customers of the relevant manufacturers have a particular advantage because they prefer to stay there anyway. Here, too, the demand and the available connection capacity often do not match.

The EN 10 2019 207 128 A1 relates to an energy supply station with a connection to an electrical supply network with a plurality of charging parking spaces for electrically powered vehicles, wherein the energy supply station comprises a plurality of electrical energy storage devices for buffering electrical energy, wherein each charging parking space comprises a charging connection, wherein charging parking spaces located directly adjacent to one another have charging connections which are electrically firmly connected to different electrical energy storage devices.

The object of the invention is to make the operation of a charging system for electrically charging electric vehicles more flexible and adapted to the demand and supply of electrical power.

The invention results from the features of the independent claims. Advantageous further developments and embodiments are the subject of the dependent claims.

A first aspect of the invention relates to a charging system having at least one charging station connected to a power grid for supplying electric vehicles with electrical energy, an energy storage device connected to the power grid and the charging station, and a control unit, characterized in that the control unit is designed to determine i) a current and future time-dependent power grid share and a complementary energy storage share of the electrical current of the charging station for a respective electric vehicle and ii) a current and future time-dependent energy state increase of the energy storage device by drawing energy from the power grid by executing a machine learning model adapted to the individual charging system with current weather data and future weather data from a weather forecast as an input variable of the machine learning model and from a determined current and future time-dependent requirement for charging energy quantity and charging power for vehicles as the respective input variable of the machine learning model.

The energy storage device is a local part of the charging system and is located in close proximity to the charging station; preferably both are located within a radius of 500 m, particularly preferably within a radius of 50 m from each other. The power grid, on the other hand, is a geographically widely branched network. The determined current and future time-dependent requirement for charging energy quantity and charging power for vehicles can, on the one hand, be determined by the machine learning model itself on the basis of given boundary conditions and known information in the sense of a prediction in order to then be used for the further determination of i) and ii) (and if used as explained below, also iii)), but can alternatively also be determined in a separate module and used as an input variable of the machine learning model to determine i) and ii) (and if used as explained below, also iii)).

The control unit is designed in particular to determine i) and ii) (and if used as explained below, also iii)) and to bring about, i.e. to carry out. For this purpose, the control unit then has corresponding means to regulate and limit the flows of energy, in particular of electric current, in each case according to their specific temporal progressions. This is advantageously done with corresponding power electronics of the control unit.

One or more charging stations can be provided for the charging system. An energy storage device is arranged on the charging station(s), in the respective charging station(s) or in the immediate vicinity. A single energy storage device can be connected to all charging stations, each charging station can have individual energy storage modules, or something similar - the topology plays a subordinate role here. However, it is advantageous to keep the distance between a respective charging station and an energy storage device as small as possible in order to keep transfer losses of electrical energy as low as possible.

The charging and discharging strategy, i.e. the time course for increasing or decreasing the storage state of the energy storage, as well as the time course of the withdrawal of electrical energy from the general power grid is determined by the machine learning model. The machine learning model is preferably a predefined algorithm or a pre-trained artificial neural network, which receives at least the future weather data as an input variable. This future weather data represents the minimum proportion of an input variable, in the sense of a system input, for the machine learning model. The output variable of the machine learning model are the time-dependent variables i) and ii) (and if used as explained below, also iii)).

This future weather data and, if necessary, other input variables are taken into account by the machine learning model in order to specify a corresponding action (charging or discharging, or with what power and at what times) of the energy storage system. One goal can be to charge electrical energy into the energy storage system when it is very cheap (and available in large quantities) and then to make it available from the energy storage system to charge the battery vehicles when electrical power from the power grid is very expensive or the supply of electricity is scarce, for example when very little energy from wind and solar power can be fed into the power grid at that moment.

The fact that the machine learning model is adapted to the individual charging system means in particular that individual parameters of the machine learning model are used for each individual charging system. Depending on the machine learning model used, such parameters can be interpreted intuitively, such as the number of charging stations or the maximum capacity of the energy storage system, but they can also not be interpreted intuitively, such as the trained weights of a deep artificial neural network.

The individual adaptation of the machine learning model reflects the average energy requirement based on, for example, the transport infrastructure or number of vehicles, the respective location of the charging stations, as well as the availability of electrical infrastructure with the corresponding power. The weather data also takes into account the time-dependent solar radiation for the use of photovoltaics and wind conditions for the use of wind power. The energy storage can thus be used as a degree of freedom in the optimization of energy management through the machine learning model.

Because the energy storage system is not only regulated, but also the charging system's own and time-dependent energetic boundary conditions are taken into account in the energy management of the energy storage system through the use of artificial intelligence of the machine learning model, high individual peak powers can be avoided and ideally smoothed out so that the available storage capacity can be used optimally and a possibly necessary increase in the connected load of a charging park can be dispensed with or this increase can at least be mitigated.

The energy storage system is preferably an electric battery storage system, but can alternatively or additionally have a flywheel storage system, a redox flow battery, an electrolysis system with re-conversion to electricity via fuel cells or similar. Depending on the storage system, however, the efficiency in energy management and the maximum current flow capacity must be taken into account.

By taking into account a temporal progression from current weather data to predicted future weather data in the machine learning model, as well as a requirement for total energy, which is dependent on current and future boundary conditions and which will have to be transferred to electric vehicles in a given period of time, as well as the requirement for charging power in the given period of time, and also even in the simplest embodiment of only one energy storage device, connected to the charging station and to the power grid, a large number of degrees of freedom are available (times and power of the energy output of the energy energy storage to the charging station or back into the power grid, times and performance of energy extraction from the power grid into the energy storage or directly to the charging station), it can be expected that a simple control depending on limit values regarding the costs from the power grid or a charge state of the energy storage cannot yet achieve a global optimum of cost optimization and/or optimization of other target variables.

For this purpose, the machine learning model is used, which takes into account the complex relationships between current and future weather data, the demand and the scope for meeting the demand, taking into account local boundary conditions such as the maximum charging power of a respective charging station, the capacity of the energy storage system, the geographical distance from solar parks and wind farms, and optionally other boundary conditions. In particular, a linear model can be used to predict the influence of the weather data on the surplus or shortage of electrical energy in the power grid, while the machine learning model takes the linear model into account, but also implements other relationships that take the above-mentioned boundary conditions into account. The machine learning model can be advantageously adapted over the operating time of the charging system, so that changed boundary conditions and changed demand, such as those caused by the evolution of electric vehicles, can also be taken into account.

If, as described below, it is taken into account that battery-electric vehicles are increasingly equipped with the ability to charge bidirectionally, i.e. that they can feed energy stored in their battery back to the power grid via a charging station, this results in an additional degree of freedom for the machine learning model, which increases the complexity of the problem exponentially.

The values of the degrees of freedom i) and ii) (and optionally, if the ability of bidirectional charging of battery-electric vehicles is taken into account, also iii)) are specified by the control unit for a current point in time and future points in time. The specification refers in particular to a certain, predefined period of time, which can, however, also be selected variably, for example depending on the probabilities of the weather data actually occurring. For example, a degree of cloud cover can be calculated with a certain probability; if this is very close to 50%, the weather data is to be considered unreliable in its forecast and the period of time is to be shortened. In any case, however, the values of the degrees of freedom of at least i) and ii) (and possibly iii)) are also determined in advance on the basis of current and future data by applying the machine learning model in order to be able to carry out a global optimization over the specified period of time. This is in contrast to a local optimization, which in particular takes into account a single point in time, in particular the present. However, with the latter, only a local optimum can be found, while global optimization over the entire period optimizes a target variable (see below) in such a way that no better solution can be found on the basis of the predicted variables (there is no other local minimum that is lower than the minimum found), unless the case of Pareto optimality occurs, in which values of the target variables can be changed mutually without a change in the target variable. However, this is also seen as a global minimum.

According to an advantageous embodiment, the control unit is designed to specifically use weather data as input variables of the machine learning model that apply to areas with wind turbines and/or photovoltaic systems in a predetermined geographical proximity to the charging system.

The closer the location of electricity generation is to the place of consumption, the lower the transmission losses between the generator and the receiver are in principle. Appropriately developed power grids can be used to specifically switch energy supply sources to demand sinks. This takes into account the ability of wind power and photovoltaic systems to generate electricity depending on the location and weather. The energy storage system of the charging system serves in particular to provide a buffer storage for nearby wind farms or photovoltaic systems and to better compensate for the lack of base load capacity of such power generators. By specifically taking weather data relating to a wind turbine and/or a photovoltaic system near the charging system into account in the machine learning model, the energy storage system in the charging system can be used even more optimisedly in order to better avoid power peaks that reach the technical limits of the charging system.

According to a further advantageous embodiment, the control unit is designed to use weighted weather data as input variables of the machine learning model, wherein weather data for areas with wind turbines and/or photovoltaic systems in a predetermined geographical proximity to the charging system are weighted higher than weather data for other areas.

This weighting advantageously allows the property of local power grids to be taken into account, namely to supply consumers from nearby sources wherever possible. However, as explained above, this is not always possible, so this circumstance is taken into account in a weighting, as it can be assumed that, particularly in future power grids, sources closer to a consumer will supply a higher proportion of electricity than sources further away.

According to a further advantageous embodiment, the control unit is designed to take into account at least one of the following in the weather data: temporal progression of the solar radiation intensity, temperature, temporal progression of the wind speed, variance of the wind speed.

The choice of weather data is advantageously adapted to the type of power generator. Since wind power and photovoltaics in particular are extremely dependent on the weather, weather data on the wind and solar radiation are provided in particular, but the current temperature can also be taken into account in order to take into account the fact that photovoltaic systems with higher temperatures enable a significantly lower power output due to an increase in electrical resistance.

This improves the energy management of the energy storage system in that electrical energy is purchased particularly cheaply (especially to charge the energy storage system) when a lot of relatively inexpensive energy is available, since wind turbines, for example, provide a relatively large amount of electrical power and the energy stored in the energy storage system is then used to charge an electric vehicle if the energy in the energy storage system was obtained at a cheaper price than it could be obtained from the power grid at the current moment of need. This enables a charging park operator to make a higher profit or pass on the low electricity price to customers who drive battery-powered vehicles. It also prevents the charging station operator from having to buy electricity at a higher price than it has to sell at the charging station.

According to a further advantageous embodiment, the control unit is designed to determine, by executing the machine learning model i) and ii), such that a predictive target variable based on monetary costs and/or a lack of available electrical energy for the vehicles to be charged and/or an emission of a predetermined substance during the production of the electrical power is minimized in global optimization.

This involves applying multi-objective optimization, whereby individual objectives are expressed in a common cost function in order to minimize the functional value of the cost function. Of course, the cost function can also be formulated in a quality function, the value of which is maximized.

According to a further advantageous embodiment, the control unit is designed to adapt the machine learning model over the operating times of the charging system by means of reinforcement learning.

Reinforcement learning, also called reinforcement learning (RL for short), is based in particular on the concept of reward and punishment functions in order to be able to adapt parameters of the machine learning model. For example, if an artificial neural network is used which has an extremely high number of internal and uninterpretable parameters, the adaptation of these parameters leads to an optimized determination of i) and ii) based on the available data, in particular weather data.

According to a further advantageous embodiment, the control unit is designed to determine an updated value of an actual target variable in the further course of operation of the charging system and to compare it with the previously determined value of the predictive target variable, and to activate a reward function for reinforcement learning when the actual target variable is so close to the previously determined predictive target variable that a predetermined comparison condition is met.

According to a further advantageous embodiment, the control unit is designed to determine, in the execution of the machine learning model with the current weather data and future weather data from a weather forecast as well as with the determined time-dependent requirement for charging energy quantity and charging power as the respective input variable of the machine learning model iii), a time-dependent feed-back of electrical energy from the battery of an electric vehicle into the power grid and/or into the energy storage device.

This is particularly, but not exclusively, relevant for companies (e.g. logistics companies) with a large fleet of vehicles, which have their own charging facility with a large number of charging stations and a relatively large vehicle fleet. Often these vehicles (especially bat battery-electric vehicles) are not used at weekends. These vehicles connected to the charging system can then, if they have the bidirectional charging function, be used as additional energy storage (at least at weekends or Sundays), i.e. so that standard control of the charging system with an energy storage device can be carried out via normal energy management control mechanisms, since, for example, the connection of the vehicles at the weekend (which can vary) must be taken into account, which would lead to complex management of the energy management of the charging system's energy storage device. This results in a further degree of freedom in the management of the charging system: Both the power grid and the energy storage device as well as a battery-electric vehicle connected to the charging system with the option of bidirectional charging are each in principle able to emit energy and receive energy at a different time. This means that there is another energy flow that can be used by the control unit for global optimization.

• - Eine geplante Route eines jeweiligen mit der Ladeanlage aktuell oder künftig verbundenen elektrischen Fahrzeugs;
• - ein aktueller Ladezustand eines jeweiligen mit der Ladeanlage aktuell oder künftig verbundenen elektrischen Fahrzeugs;
• - eine Ladekurve einer Batterie eines jeweiligen mit der Ladeanlage aktuell oder künftig verbundenen elektrischen Fahrzeugs;
• - eine Temperatur einer Batterie eines jeweiligen mit der Ladeanlage aktuell oder künftig verbundenen elektrischen Fahrzeugs;
• - eine installierte Kapazität der Batterie eines jeweiligen mit der Ladeanlage aktuell oder künftig verbundenen elektrischen Fahrzeugs;
• - aktuelle Uhrzeit und/oder aktueller Wochentag und/oder aktueller Monat und/oder aktuelle Jahreszeit und/oder Ferienzeiten und/oder Feiertage;

According to a further advantageous embodiment, the control unit is designed to take into account at least one of the following in addition to the weather data when determining i), ii) and iii):

• - A planned route of each electric vehicle currently or in the future connected to the charging station;
• - a current charge status of each electric vehicle currently or in the future connected to the charging system;
• - a charging curve of a battery of an electric vehicle currently or in the future connected to the charging system;
• - a temperature of a battery of an electric vehicle currently or in the future connected to the charging system;
• - an installed capacity of the battery of each electric vehicle currently or in the future connected to the charging station;
• - current time and/or current day of the week and/or current month and/or current season and/or holiday periods and/or public holidays;

This also allows the daily, weekly, monthly and annual requirements for charging power and capacity to be taken into account - the requirements can vary greatly at different times of the day. For example, a distinction is made between working days and Saturdays and Sundays, holiday periods in federal states or abroad, etc.; this information can also be used for a more precise forecast of the charging energy requirements.

According to a further advantageous embodiment, the machine learning model comprises one of the following: artificial neural network, A2C algorithm, Q-learning algorithm, DQN algorithm.

Further advantages, features and details emerge from the following description, in which - if necessary with reference to the drawing - at least one embodiment is described in detail. Identical, similar and/or functionally identical parts are provided with the same reference numerals.

• 1: Eine Ladeanlage gemäß einem Ausführungsbeispiel der Erfindung.

It shows:

• 1 : A charging system according to an embodiment of the invention.

1 shows a local charging system. The charging system has a control unit 9 and an energy storage device 7. A charging station 1 is connected to both the energy storage device 7 and the power grid 5, with the energy storage device 7 also being connected to the power grid 5. The energy storage device 7 itself is arranged in the immediate vicinity of the rest of the charging park in order to be able to implement the shortest possible cable lengths and thus keep infrastructure costs as low as possible. The charging station 1 is used to supply electric vehicles 3 with electrical energy and has corresponding cable routing and connections for electric vehicles 3. The power grid 5 is an extensive, country-wide power grid 5, with a wind farm 11 and a solar park 13 connected to the power grid 5 in a geographical proximity to the charging system. However, these systems primarily serve to supply the charging park with energy. The wind farm 11 and the solar park 13 therefore typically supply the largest share of electrical energy compared to other sources for electrical power generation from the power grid 5. However, since demand and supply rarely match, the energy storage unit 7 is installed with a correspondingly high capacity in order to at least partially compensate for this circumstance. Although the energy storage unit 7 is still connected to the public power grid 5, this connection also has a power limitation, i.e. any peak demand for battery vehicles 3 to be charged (in the case of a large number of simultaneous charges) can no longer be covered by the installed connection power, i.e. a costly increase in power or a reduction in the charging power of the battery vehicles 3 by the charging stations 1 would have to be carried out - both of which are undesirable. However, in order not to use too large a storage capacity, which would be correspondingly costly, on the energy storage unit 7, optimal energy management is to be sought. This is done by the control unit 9 with its implemented artificial intelligence and the corresponding neural networks and/or algorithms. Ideally, the energy storage unit 7 is charged or fully charged when electricity is available cheaply in the power grid 5 and/or has a high proportion from wind turbines 11 or solar systems 13 and/or has caused low carbon dioxide emissions. At least some of the electric vehicles 3 are also equipped with the ability to charge bidirectionally. Connected to a charging station 1, which in turn is connected to the energy storage unit 7, such an electric vehicle 3 with its own battery, which can be charged by the charging station 1, but can also supply energy to the energy storage unit 7 via the charging station 1, effectively represents additional capacity for the energy storage unit 7. This is particularly relevant for the operation of a vehicle fleet, e.g. of a logistics company: For example, at weekends, when the vehicles 3 are not needed for logistics operations, these vehicles 3 can be available as an extended power storage unit 7. For example, it can make sense to make part of the energy stored in the fleet vehicles 3 available on a power exchange when the price of electricity is particularly high. This allows the logistics company to make a profit even though the vehicles 3 are actually just standing still and serving as part of the energy storage unit 7. Even if a car rental company provides such a charging park for its customers of electric vehicles 3, it is possible that the vehicles 3, which are, for example, near the charging park and which transmit their state of charge value, and possibly also the route planning if this leads past the charging park, also use this data and determine the probability of the energy requirements of these vehicles 3 when they arrive at the charging park. Furthermore, the information can also be evaluated, for example, whether the charging park has been stored in the navigation system as a stopover for charging. Based on this information, it can then be determined or predicted what energy requirements there are for the arriving vehicles 3 and since the manufacturer of its vehicles 3 also knows the charging curves, the battery temperature and the installed capacity of the battery of the respective models (this information is also transmitted), a very precise determination of the required energy can take place here. The demand can thus be forecast by the control unit 9, but the supply is primarily determined by the control unit 9 using current and future weather data. All this information, symbolized by blocks in the left column of the 1 are therefore to be taken into account by the control unit 9 in order to achieve a global optimization of the current decision on the control of energy flows between electric vehicles 3, battery storage 7 and power grid 5 over a predetermined period of time. With the help of the current and future weather data, the control unit 9 can determine what output power the wind farm 11, the photovoltaic park 13 and the electricity prices in the general power grid 5 will have, in particular over a predetermined period of time that covers the next few hours and/or the next few days. Accordingly, weather data that is valid for the geographical area of the wind farm 11 and the photovoltaic park 13 are taken into account in particular. The resulting highly complex optimization problem for achieving the specified goal with a cost function that takes into account electricity prices, demand and availability from the power grid 5 and optionally freely defined environmental balance terms is solved by the control unit 9 by applying a machine learning model comprising a deep neural network. This deep neural network is only to be used for the one charging system under consideration, since the multitude of boundary conditions such as the number of charging stations 1, the maximum electrical power in the connections between the charging station 1, the energy storage unit 7, the connection to the power grid 5, as well as the local conditions of power generation such as the named wind farm 11 and possibly the behavior of the drivers of the electric vehicles 3 must be taken into account. However, this artificial neural network is further trained during the course of the operation of the charging system in order to be able to determine and direct the electrical currents more effectively over time and to be able to adapt to any changed boundary conditions. The decisive factor for the application of the machine learning model by the control unit 9 is to make a prediction based on heuristic data, current and estimated future input variables as to which electrical currents will be optimal at which times and at what level in order to be able to achieve the complex multi-objective optimization.

Although the invention has been illustrated and explained in detail by preferred embodiments, the invention is not limited by the disclosed examples and other variations can be derived therefrom by the person skilled in the art without departing from the scope of the invention. It is therefore clear that a multitude of possible variations exist. It is also clear that embodiments mentioned as examples really only represent examples that are not to be understood in any way as a limitation of the scope of protection, the possible applications or the configuration of the invention. Rather The foregoing description and the description of the figures enable the person skilled in the art to implement the exemplary embodiments in concrete terms, whereby the person skilled in the art, with knowledge of the disclosed inventive concept, can make a variety of changes, for example with regard to the function or arrangement of individual elements mentioned in an exemplary embodiment, without departing from the scope of protection defined by the claims and their legal equivalents, such as further explanations in the description.

List of reference symbols

1

Charging station

3

Vehicles

5

Power grid

7

Energy storage

9

Control unit

11 13

Wind turbine Photovoltaic system

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions.

Cited patent literature

• DE 102019207128 A1 [0004]

Claims (10)

Charging system comprising at least one charging station (1) connected to a power grid (5) for supplying electric vehicles (3) with electrical energy, an energy storage device (7) connected to the power grid (5) and the charging station (1), and a control unit (9), characterized in that the control unit (9) is designed to determine i) a current and future time-dependent power grid share and a complementary energy storage share of the electrical current of the charging station (1) for the electric vehicles (3) and ii) a current and future time-dependent energy state increase or energy state decrease of the energy storage device (7) by exchanging energy with the power grid (5) by executing a machine learning model adapted to the individual charging system with current weather data and future weather data from a weather forecast as respective input variables of the machine learning model and from a determined current and an estimated future time-dependent requirement for charging energy quantity and charging power for electric vehicles (3). Charging system according to Claim 1 , wherein the control unit (9) is designed to specifically use such weather data as input variables of the machine learning model which apply to areas with wind turbines (11) and/or photovoltaic systems (13) in a predetermined geographical proximity to the charging system. Charging system according to Claim 2 , wherein the control unit (9) is designed to use weighted weather data as input variables of the machine learning model, wherein weather data for areas with wind turbines (11) and/or photovoltaic systems (13) in a predetermined geographical proximity to the charging system are weighted higher than weather data for other areas. Charging system according to one of the preceding claims, wherein the control unit (9) is designed to take into account at least one of the following in the weather data: temporal progression of the solar radiation intensity, temperature, temporal progression of the wind speed, variance of the wind speed. Charging system according to one of the preceding claims, wherein the control unit (9) is designed to determine by executing the machine learning model i) and ii) such that a predictive target variable based on monetary costs and/or a lack of available electrical energy for the vehicles (3) to be charged and/or an emission of a predetermined substance during the production of the electrical current is minimized in global optimization. Charging system according to Claim 5 , wherein the control unit (9) is designed to adapt the machine learning model over the operating times of the charging system by reinforcement learning. Charging system according to Claim 6 , wherein the control unit (9) is designed to determine an updated value of an actual target variable during the further course of operation of the charging system and to compare it with the previously determined value of the predictive target variable, and to activate a reward function for reinforcement learning when the actual target variable is so close to the previously determined predictive target variable that a predetermined comparison condition is met. Charging system according to one of the preceding claims, wherein the control unit (9) is designed, in the execution of the machine learning model with the current weather data and future weather data from the weather forecast as respective input variables of the machine learning model and with the current and the further estimated time-dependent requirement for charging energy quantity and charging power iii) to determine a time-dependent feed-back of electrical energy from the battery of an electric vehicle (3) into the power grid (5) and/or into the energy storage device (7). Charging system according to Claim 8 , wherein the control unit (9) is designed to take into account at least one of the following in addition to the weather data when determining i), ii) and iii): - a planned route of a respective electric vehicle (3) currently or in the future connected to the charging system; - a current state of charge of a respective electric vehicle (3) currently or in the future connected to the charging system; - a charging curve of a battery of a respective electric vehicle (3) currently or in the future connected to the charging system; - a temperature of a battery of a respective electric vehicle (3) currently or in the future connected to the charging system; - an installed capacity of the battery of a respective electric vehicle (3) currently or in the future connected to the charging system; - current time of day and/or current day of the week and/or current month and/or current season and/or holiday periods and/or public holidays; Charging system according to one of the preceding claims, wherein the machine learning model of a which includes: Artificial Neural Network, A2C Algorithm, Q-Learning Algorithm, DQN Algorithm.

Images