Edge computing with ai at mains supply network interconnection points

Abstract

The invention relates to a network interconnection module (10) for a network station (100), in particular in the form of a local network station or a substation, wherein the network station (100) connects different network nodes (Ni), preferably different consumers and / or different energy generators, via electrical lines (L), with the network interconnection module (10) comprising: a communications unit (11) for obtaining operating parameters (BP) from the different network nodes (Ni); a memory unit (12) in which a, in particular offline, trained artificial neural network (KNN), preferably a graph neural network (GNN) is stored; a computer unit (13) which is designed to process the operating parameters (BP) using the artificial neural network (KNN), in order to provide optimal control parameters (SP) and / or control variables (SG) for the different network nodes (Ni).

Description

[0001] Title: Edge Computing with Kl at Power Grid Transfer Points

[0002] Description

[0003] The invention relates to a grid transfer module for a grid station, in particular in the form of a local grid station or a transformer station, wherein the grid station connects various grid nodes, preferably different consumers and / or different energy generators, via electrical lines. Furthermore, the invention relates to the use of a corresponding grid transfer module for planning and / or operating energy distribution networks, comprising at least one grid station, in particular in the form of a local grid station or a transformer station, and various network nodes. Furthermore, the invention relates to a corresponding grid station, in particular in the form of a local grid station or a transformer station, with a corresponding grid transfer module. Furthermore, the invention relates to a training method for training an artificial neural network, preferably a graph-based artificial neural network, for a corresponding grid transfer module.Furthermore, the invention relates to an operating method for operating a network station, in particular in the form of a local network station or a transformer substation, with a corresponding network transfer module.

[0004] As part of the energy transition, significant changes are occurring in electricity demand and generation. Renewable energies, in particular, are becoming increasingly important on the generation side. On the consumer side, a significant increase in the penetration of electric vehicles and heat pumps is imminent. These decentralized developments are leading to dynamic energy generation, flexible loads, and thus to more complex, often highly dynamic, supply requirements for the distribution grids. Overall, these decentralized developments are leading to increasing overall capacity utilization of the distribution grids. Modern distribution grids are therefore increasingly being expanded and equipped with smart devices that monitor grid areas via sensors and can perform certain grid management tasks. Particularly in the medium and low voltage ranges, grid stations, such asSubstations or distribution substations face challenges because they must coordinate a multitude of variable measured variables and a multitude of dynamic control variables. Complex calculation methods are often used at a central grid station to optimize load flow between different energy producers and various consumers. These calculations usually require considerable computing power and a relatively long computing time, making them of very limited use in practical operation, especially for small grid areas.

[0005] The object of the invention is therefore to provide a grid transfer module for a (local or regional) grid station, in particular in the form of a local grid station (in the low-voltage range) or a transformer substation (in the medium-voltage range). In particular, the object of the invention is to provide an intelligent grid transfer module for a grid station that enables improved control of optimized load flows through the grid station that exist between different energy producers and different consumers. Preferably, the object of the invention is to provide an intelligent grid transfer module for a grid station that enables improved control of optimized load flows with little computational effort, quickly and reliably.Preferably, the object of the invention is to create a modular, individually manageable module that can be flexibly used at different grid stations in the low-voltage and / or medium-voltage range in order to enable intelligent control of optimized load flows through the grid station. Furthermore, the object of the invention is to enable improved use of a corresponding grid transfer module, preferably for planning and / or operating (in particular in real time) energy distribution networks. Furthermore, the object of the invention is to provide an improved grid station, in particular in the form of a local grid station or a transformer substation, with a corresponding grid transfer module.Furthermore, it is an object of the invention to provide a training method for training an artificial neural network, preferably a graph-based artificial neural network, for a corresponding grid transfer module. Furthermore, it is an object of the invention to provide an improved operating method for operating a grid station, in particular in the form of a local grid station or a transformer substation, with a corresponding grid transfer module.

[0006] The problem is solved by: a network transfer module for a network station with the features of the independent device claim. Furthermore, the problem is solved by: a use of a corresponding network transfer module for planning and / or operating energy distribution networks with the features of the independent use claim. Furthermore, the problem is solved by: a corresponding network station, in particular in the form of a local network station or a transformer substation, with a corresponding network transfer module with the features of the independent device claim. Furthermore, the problem is solved by: a corresponding training method for training an artificial neural network, preferably a graph-based artificial neural network, for a corresponding network transfer module with the features of the independent method claim.Furthermore, the problem is solved by: a corresponding operating method for operating a network station, in particular in the form of a local network station or a transformer substation, with a corresponding network transfer module having the features of the independent method claim.

[0007] The invention provides: a grid transfer module for a (local or regional) grid station, in particular in the form of a local grid station (in the low-voltage range, e.g., 400 / 230 V) or a transformer substation (in the medium-voltage range, e.g., 10 kV to 36 kV), wherein the grid station connects various grid nodes, preferably different consumers and / or different energy generators, via electrical lines or busbars. The grid transfer module comprises the following elements: a communication unit for receiving operating parameters or

[0008] Power parameters (such as an electrical power, in particular an active power and / or a reactive power, a voltage and / or a phase angle) from the various network nodes or from corresponding electrical lines or busbars, a storage unit in which an artificial neural network, preferably a graph-based artificial neural network, trained in particular offline, is stored, and a computing unit which is designed to process the operating parameters with the aid of the artificial neural network in order to provide optimal control parameters (such as a power, in particular an apparent power, a voltage and / or a phase angle on corresponding lines) and / or manipulated variables (such as switch positions) for the various network nodes or for the corresponding electrical lines or busbars.

[0009] The present idea involves the application of specific methods implemented using machine learning (load flow calculations, or LF for short, and optimal power flow calculations, or OPF for short) for the coordination, regulation, and control of grid nodes and substations in medium- and / or low-voltage grid areas. These methods are substituted by the use of learning methods, particularly unsupervised ones, such as graph-based ones.

[0010] The invention recognizes that artificial neural networks, in particular graph-based artificial neural networks, can advantageously enable different network areas to be mapped, for example, using corresponding network nodes, connected using associated load flow calculations, and learned using an optimal load flow calculation to determine optimal control parameters for the existing network nodes. Artificial neural networks, in particular graph-based artificial neural networks, can advantageously also enable rapid adaptation to changing supply tasks. The learning process can advantageously take place "offline," so that on-site execution can be carried out very quickly, for example, in the millisecond range, using the already trained artificial neural network.This runtime-efficient determination of control parameters and manipulated variables enables integration into decentralized computing units (“edge computing devices”) for the implementation of real-time control systems as well as simultaneous calculation of large network areas in SCADA systems.

[0011] The trained artificial neural network can preferably be implemented in an intelligent device, a so-called grid transfer module, which can also be referred to as a "grid cube." A ready-made grid transfer module can be deployed in existing (local or regional) grid stations, particularly in the form of local grid stations (in the low-voltage range) or transformer substations (in the medium-voltage range), e.g., using standardized fastening elements and electrical connection elements, and can perform communication and control tasks there. The grid transfer module can advantageously receive the corresponding operating parameters from the individual grid nodes, particularly including the grid station itself, have these operating parameters processed by the trained artificial neural network, and deliver ready-made control parameters for different lines.

[0012] For offline training, no historical data from the real operation of the grid station is required. Instead, synthetic training data can be used, allowing for a wide range of load situations, including rare ones. Diverse load situations can include various scenarios: providing control power, operating according to specified schedules, responding to fluctuations in energy generation, responding to changing demand, and even congestion management.

[0013] A further advantage of the proposed network handover module is the autonomous detection of invalid states by the (self-learning) artificial neural network. This offers significant advantages, particularly over supervised learning methods, because it is very fast on-site execution, requires little computing power, does not require real-time communication with a central processing unit, does not require outsourcing of computations to the cloud, and does not require verification by trained personnel.

[0014] In this way, a grid transfer module can be provided for a (local or regional) grid station, in particular in the form of a local grid station (in the low-voltage range) or a transformer substation (in the medium-voltage range). In particular, an intelligent grid transfer module can be provided for a grid station in this way, which enables improved control of optimized load flows or power flows through the grid station that exist between different energy producers and different consumers. Preferably, an intelligent grid transfer module can be provided for a grid station in this way, which enables improved control of optimized load flows with little computational effort, in real time during operation of the grid station, and reliably.Preferably, in this way, a modular, individually manageable module can be provided which can be used flexibly at different network stations in the low voltage and / or medium voltage range in order to enable intelligent control of optimized load flows through the network station.

[0015] Furthermore, it can be provided that the artificial neural network was trained offline from the grid station, e.g. somewhere at a suitable learning station. This makes it possible for all calculations (load flow calculations or LF for short and an optimal load flow calculation or power flow calculation, so-called “optimal power flow” or OPF for short) to be carried out during training on the learning station. Subsequently, the trained artificial neural network no longer needs to carry out any calculations on-site at the grid station. The obtained operating parameters or performance parameters (such as electrical power, in particular active power and / or reactive power, voltage and / or phase angle) are only processed by the fully trained artificial neural network in order to determine optimal control parameters (such asA power (in particular, apparent power, voltage, and / or phase angle on corresponding lines) and / or control variables (such as switch positions) are provided to the various network nodes in a fully automated, very fast, and efficient manner. A learning station can be designed, for example, as a data center or a cloud infrastructure.

[0016] Furthermore, it can be provided that the artificial neural network is a self-learning artificial neural network, in particular a graph-based artificial neural network, and / or that the artificial neural network was trained using a graph-based method. The advantage of such a self-learning neural network is that no monitoring, e.g. by trained personnel, is required, neither during training nor at the site of use. For example, no real training data is required, neither from operation nor from an associated control system. At the site of use, such a self-learning artificial neural network can also directly evaluate the delivered result, e.g. by providing it with a quality metric. A verification function can preferably be optimal power flow calculation (OPC).(Optimal Power Flow or OPF for short) that enables the calculation of optimal control parameters and / or manipulated variables while adhering to permissible power flows or load flows. The English term "Optimal Power Flow" (OPF) is generally more common. Furthermore, it can be provided that the operating parameters include a power, in particular an active power and / or a reactive power, a voltage and / or a phase angle on electrical lines connecting the various grid nodes. Furthermore, it can be provided that the control parameters include a power, in particular an apparent power, a voltage and / or a phase angle on corresponding lines, and / or that the manipulated variables include switch positions and / or power levels. In this way, a load flow calculation between the grid nodes can be substituted by the artificial neural network.Electrical power, such as reactive power and active power at the grid node, is sent to the artificial neural network, which has been trained to consider different cost functions, such as permissible operating parameter limits and / or economic efficiency functions. As a result, optimal control parameters and / or manipulated variables are provided for the various grid nodes, such as voltage level and phase angle, or active and reactive power on each line. As a result, it can be determined, for example, that certain energy generators are switched on or off and / or that certain consumers are switched on or off, whereby the switching on can occur at different power levels.

[0017] Furthermore, it can be provided that the artificial neural network was trained using at least one model of a network section comprising a network station and various network nodes, wherein the network station connects the various network nodes via electrical lines. It is conceivable that the neurons of the artificial neural network can map the various network nodes. Furthermore, it is conceivable that the edges between the neurons of the artificial neural network map power flow calculations for different lines. In this way, the artificial neural network can advantageously map the network section with its network nodes and corresponding load flow calculations. Thus, the trained artificial neural network can replace the performance of computationally intensive load flow calculations at the site of use.

[0018] Advantageously, load flow calculations between the various network nodes can be used as connection functions between corresponding neurons of the artificial neural network when training the artificial neural network. Thus, for example, power flow equations can be used as objective functions at the neurons and edges of the artificial neural network when training the artificial network. Thus, the trained artificial neural network can advantageously replace the performance of computationally intensive load flow calculations at the site of use. In this way, the artificial neural network can learn a network section to provide optimal control parameters and / or manipulated variables for the various network nodes.

[0019] According to a further advantage, it can be provided that, when training the artificial neural network, an optimal power flow calculation at the grid station, comprising at least one cost function, in particular having permissible operating parameter limits for the various grid nodes and / or an economic function for energy generation and / or energy distribution, is used as an evaluation function and / or an optimization function. This ensures that the artificial neural network independently learns the associated grid section and, for example, optimally designs grid operation according to all of the aforementioned aspects. It can also be provided that the results of the artificial neural network are automatically evaluated on-site during operation of the grid station. Invalid grid states can thus be quickly identified and avoided.

[0020] Furthermore, it can be provided that the artificial neural network was trained using a model of a network section, wherein the model represents an arbitrary and / or virtual network section with a network station and various network nodes. In this way, a decoupling between a real network section and a virtual network section can be created for training.

[0021] Furthermore, it can be provided that the artificial neural network was trained using a model of a network section, wherein the model maps a real network section at a deployment location of the network transfer module with a network station and various network nodes. In this way, a specially adapted artificial neural network for a specific real network section can be enabled. Furthermore, it can be provided that the artificial neural network was trained using different models of different network sections, which may have a network station and various network nodes. The artificial neural network can thus learn different possible / virtual deployment situations for a network station. At the deployment location at a specific real network station, such a trained artificial neural network can respond in an improved manner to changing supply requirements.

[0022] Furthermore, it can be provided that the artificial neural network has been trained using different operating scenarios on at least one model of a network section, which may have different operating parameters for the various network nodes. The artificial neural network can thus learn different possible / virtual scenarios for a network station. When deployed at a specific real network station, such a trained artificial neural network can respond to changing supply requirements in an improved manner.

[0023] Advantageously, fictitious operating parameters from the various network nodes can be used when training the artificial neural network. This allows the training of the artificial neural network to be decoupled from the actual operation of the network station. At the same time, the artificial neural network can be improved in this way to prepare for a variety of load situations, including rare ones, such as the provision of control power, running on specified schedules, responding to fluctuations in energy generation, responding to changing demand, and even congestion management.

[0024] Furthermore, it may be advantageous for the computing unit to be designed to provide the control parameters and / or control variables for the various network nodes, in particular only depending on the output of the artificial neural network, preferably offline from the network station. This allows for rapid on-site execution using the already trained artificial neural network, for example, in the millisecond range. Preferably, no significant computing power needs to be provided, because no large calculations need to be performed.Furthermore, it may be advantageous for the communication unit, when providing optimal control parameters and / or control variables for the various network nodes, to be used only to receive operating parameters from the various network nodes and to transmit control parameters for the various network nodes. In particular, no traditional optimization algorithms for load flow calculations and / or optimal power flow calculations are performed online during operation of the network station. This enables automatic operation of a network station, which can take place independently of central data centers.

[0025] Furthermore, it can be advantageous for the computing unit to be designed to evaluate the output of the artificial neural network, comprising the control parameters for the various network nodes. Preferably, the computing unit can be designed to evaluate the output of the artificial neural network, comprising the control parameters for the various network nodes, using a quality metric that, in particular, maps an optimal power flow calculation at the grid station, comprising at least one cost function, in particular having permissible operating parameter limits for the various grid nodes and / or an economic function for energy generation and / or energy distribution. In this way, the result can be evaluated on-site. Invalid control parameters and / or control variables can thus be quickly identified and reliably avoided during operation of the grid station.

[0026] The invention can offer further advantages if the grid transfer module, which can be designed as described above, is used for planning and / or operating (particularly in real time) energy distribution networks. In this way, expansions of energy supply networks can be planned in an improved manner.

[0027] Furthermore, the object is achieved by a grid station, in particular in the form of a local grid station or a transformer substation, comprising a grid transfer module, which can be designed as described above. Standardized fastening elements and / or electrical connecting elements for mechanically and / or electrically connecting the grid transfer module can preferably be provided on the grid station. This achieves the same advantages as those described above in connection with the grid transfer module. These advantages are fully incorporated herein by reference.Furthermore, the object is achieved by: a training method for training an artificial neural network, preferably a graph-based artificial neural network for a network transfer module, which can be designed as described above, wherein the artificial neural network is trained using at least one model of a network section which has a network station and various network nodes, wherein the network station connects the various network nodes via electrical lines, wherein the neurons of the artificial neural network map the various network nodes, and wherein the edges between the neurons of the artificial neural network map power flow calculations for different lines.

[0028] This achieves the same advantages described above in connection with the network transfer module. These advantages are fully referenced here.

[0029] Advantageously, when training the artificial neural network, load flow calculations between the various network nodes can be used as connection functions between corresponding neurons of the artificial neural network, whereby, in particular, the neurons of the artificial neural network can map the various network nodes of the network section. During training, the load flow calculations can thus be used, for example, as target functions at the neurons and edges of the artificial neural network.

[0030] Furthermore, it may be advantageous if, when training the artificial neural network, an optimal power flow calculation at the grid station, comprising at least one cost function, in particular having permissible operating parameter limits for the various grid nodes and / or economic efficiency function in energy generation and / or energy distribution, is used as an evaluation function and / or an optimization function.

[0031] One possible training method is to train the artificial neural network using a model of a network section, where the model represents an arbitrary and / or virtual network section with a network station and various network nodes. This allows the training method to be decoupled from the real situation at the deployment location of the network transfer module.

[0032] A training method can also involve training the artificial neural network using a model of a network section, with the model representing a real network section at a deployment location of the network transfer module, including a network station and various network nodes. This allows the training method to be adapted to the real situation at the deployment location of the network transfer module.

[0033] A training method can further provide for the artificial neural network to be trained using different models of different network sections, which may include a network station and various network nodes. The artificial neural network can thus learn different possible / virtual deployment situations for a network station in order to be able to respond more effectively to changing supply requirements on-site.

[0034] A training method can further provide for the artificial neural network to be trained using different operating scenarios on at least one model of a network section, which may have different operating parameters from the various network nodes. The artificial neural network can thus learn different possible / virtual scenarios at a network station in order to be able to respond more effectively to changing supply requirements on site.

[0035] A training procedure can also be designed to use fictitious operating parameters of the various network nodes when training the artificial neural network. This can improve training and increase the adaptability of the artificial neural network to a wide variety of on-site scenarios.

[0036] A training method can further provide for the artificial neural network to be trained again when a network section changes, in particular to provide an update to the artificial neural network. Furthermore, the object is achieved by: an operating method for operating a network station, in particular in the form of a local network station or a transformer substation, wherein the network station connects various network nodes, preferably different consumers and / or different energy producers, via electrical lines, wherein a network transfer module is used to operate the network station, which module can be designed as described above to provide optimal control parameters for the various network nodes using a trained artificial neural network, preferably a graph-based neural network.

[0037] This achieves the same advantages described above in connection with the network transfer module. These advantages are fully referenced here.

[0038] Advantageously, the grid transfer module can provide the control parameters and / or control variables for the various grid nodes during operation of the grid station, in particular only depending on the output of the artificial neural network, preferably offline from the grid station. When providing optimal control parameters and / or control variables for the various grid nodes, the communication unit can preferably be used only to receive operating parameters from the various grid nodes and to transmit control parameters for the various grid nodes, wherein, in particular, no load flow calculations and / or optimal power flow calculations are performed online during operation of the grid station.

[0039] Furthermore, it can be advantageous for the network transfer module to evaluate the output of the artificial neural network, comprising the control parameters and / or control variables for the various network nodes. Advantageously, the network transfer module can evaluate the output of the artificial neural network, comprising the control parameters and / or control variables for the various network nodes, using a quality metric that, in particular, maps an optimal power flow calculation at the network station, comprising at least one cost function, in particular having permissible operating parameter limits for the various network nodes and / or an economic function for energy generation and / or energy distribution. In an operating method, it can further be provided that, when a network section changes, the artificial neural network is sent again to a learning station for training, in particular to receive an update from the artificial neural network.

[0040] The invention and its further developments, as well as their advantages, are explained in more detail below with reference to the drawing. It shows schematically:

[0041] Fig. 1 is an exemplary representation of a network transfer module in the sense of the present disclosure.

[0042] Fig. 1 shows a grid transfer module 10 for a (local or regional) grid station 100, in particular in the form of a local grid station (in the low voltage range) or a transformer substation (in the medium voltage range), wherein the grid station 100 connects various grid nodes Ni, preferably different consumers and / or different energy producers, via electrical lines L or busbars.

[0043] The grid transfer module 10 has the following elements: a communication unit 11 for receiving operating parameters BP or power parameters (such as an electrical power, in particular an active power and / or a reactive power, a voltage and / or a phase angle) from the various grid nodes (Ni) and / or from corresponding electrical lines L or busbars, a storage unit 12 in which an artificial neural network KNN, preferably a graph-based artificial neural network GNN, trained in particular offline, is stored, and a computing unit 13 which is designed to process the operating parameters BP with the aid of the artificial neural network KNN in order to determine optimal control parameters SP (such as a power, in particular an apparent power, a voltage and / or a phase angle on corresponding lines) and / or manipulated variables SG such asTo provide switch positions for the various network nodes Ni or for the corresponding electrical lines L.

[0044] The present idea involves the application of specific methods implemented using machine learning (load flow calculations, or LF for short, and optimal power flow calculations, or OPF for short) for the coordination, regulation, and control of grid nodes and substations in medium- and / or low-voltage grid areas. These methods are substituted by the use of learning methods, particularly unsupervised ones, such as graph-based ones.

[0045] The invention recognizes that artificial neural networks (ANNs), in particular graph-based artificial neural networks, can advantageously enable different network areas to be mapped, for example, using corresponding network nodes, connected using associated load flow calculations, and learned optimal control parameters for the existing network nodes using an optimal load flow calculation. Advantageously, artificial neural networks (ANNs), in particular graph-based artificial neural networks, can also enable rapid adaptation to changing supply tasks. The learning process can advantageously take place "offline," so that on-site execution can be carried out very quickly, for example, in the millisecond range, using the already trained artificial neural network.This runtime-efficient determination of control parameters and manipulated variables enables integration into decentralized computing units, “edge computing devices,” for the implementation of real-time control systems, as well as simultaneous calculation of large network areas in SCADA systems.

[0046] The trained artificial neural network (ANN) can preferably be implemented in an intelligent device, a so-called grid transfer module 10, which can also be referred to as a "grid cube." A ready-made grid transfer module 10 can be deployed in existing (local or regional) grid stations, in particular in the form of local grid stations (in the low-voltage range) or transformer substations (in the medium-voltage range), e.g., using standardized fastening elements and electrical connection elements, and can perform communication and control tasks there. The grid transfer module can advantageously receive the corresponding operating parameters (BP) from the individual grid nodes, have these operating parameters (BP) processed by the trained artificial neural network (ANN), and deliver ready-made control parameters (SP) and / or control variables (SG) for different lines.

[0047] For offline training, no historical data from the real operation of the grid station is required. Instead, synthetic training data can be used, allowing for a wide range of load situations, including rare ones. Diverse load situations can include various scenarios: providing control power, operating according to specified schedules, responding to fluctuations in energy generation, responding to changing demand, and even congestion management.

[0048] A further advantage of the proposed network handover module is the autonomous detection of invalid states by the self-learning artificial neural network (ANN). This offers significant advantages, particularly over supervised learning methods, because it is very fast on-site, requires little computing power, does not require real-time communication with a central processing unit, does not require outsourcing of computations to the cloud, and does not require verification by trained personnel.

[0049] The invention also recognizes that load flow calculations can play a beneficial role in grid planning and / or grid operation processes. Load flow calculations can be used to avoid grid bottlenecks. Violations of operating parameter limits and overloading of grid resources can also be avoided. Furthermore, load flow calculations can be used to consider various economic functions in energy generation and / or energy distribution. Based on load flow calculations, operating parameter limits, and economic functions, optimal power flow (OPF) calculations can enable optimal operation of a grid station. Generating plants can be switched on and / or off, or operated to a specific extent.Consumer devices can still be supplied with a controllable voltage.

[0050] In connection with load flow calculations, operating parameter limits, and economic functions, or with an optimal load flow calculation, the invention proposes the use of artificial intelligence. Advantageously, the invention proposes the use of graph-based artificial neural networks (GNNs) for load flow calculations, which can also substitute for the cost functions in an optimal load flow calculation. Furthermore, artificial intelligence methods (such as reinforcement learning) can support and be used in planning and operational issues and optimization problems such as the identification of switching operations in grid operation or the determination of cost-minimized grid expansion plans, which are usually solved using meta-heuristics, e.g., genetic methods.

[0051] To replace load flow calculations, graph-based artificial neural networks (GNNs) can advantageously be used to simulate load flows between different network nodes Ni, e.g., different energy producers and / or energy consumers, in a network section A. Graph-based artificial neural networks (GNNs) preferably allow the processing of weighted graphs. The network model on which the load flow calculation is performed comprises a structure with different nodes and edges. Matrices can be used to describe the structure of a graph. Such a matrix can, for example, represent an adjacency matrix, which in the application of graph-based artificial neural networks (GNNs) is referred to as a graph shift operator (GSO).In this way, the load flow result can be interpreted as a graph signal, allowing the load flow problem to be simulated by graph-based artificial neural networks.

[0052] In this way, a grid transfer module 10 can be provided for a (local or regional) grid station 100, in particular in the form of a local grid station (in the low-voltage range) or a transformer substation (in the medium-voltage range). In particular, an intelligent grid transfer module 10 can be provided for a grid station 100, which enables improved control of optimized load flows or power flows through the grid station 100 that exist between different energy producers and different consumers. Preferably, an intelligent grid transfer module 10 can be provided for a grid station 100, which enables improved control of optimized load flows with little computational effort, in real time during operation of the grid station 100, and reliably.In this way, a modular, individually manageable grid transfer module 10 can preferably be provided, which can be flexibly used at different grid stations 100 in the low-voltage and / or medium-voltage range to enable intelligent control of optimized load flows through the grid station 100. Furthermore, the object is achieved by using a corresponding grid transfer module for planning and / or operating (in particular in real time) energy distribution networks.

[0053] Furthermore, the problem is solved by: a corresponding network station 100, in particular in the form of a local network station or a transformer substation, with a corresponding network transfer module.

[0054] Furthermore, the task is solved by: a corresponding training method for training an artificial neural network, preferably a graph-based artificial neural network for a corresponding network transfer module.

[0055] Furthermore, the problem is solved by: a corresponding operating method for operating a network station, in particular in the form of a local network station or a transformer substation, with a corresponding network transfer module.

[0056] The above explanation of the embodiments describes the present invention exclusively by way of examples. Of course, individual features of the embodiments can be freely combined with one another, provided they are technically feasible, without departing from the scope of the present invention.

[0057] Bez uqszeic hen list

[0058] 10 Network transfer module

[0059] 11 Communication unit

[0060] 12 storage units

[0061] 13 Computing unit

[0062] 100 network stations

[0063] A network section

[0064] Ni network nodes

[0065] L lines

[0066] M1 Model

[0067] M2 Model

[0068] BP operating parameters

[0069] SG control variables

[0070] SP control parameters

[0071] GNN Network

[0072] KNN Network

[0073] LF power flow calculations

[0074] OPF optimal power flow calculation

Claims

Patent claims 1. A network transfer module (10) for a network station (100), in particular in the form of a local network station or a transformer substation, wherein the network station (100) connects various network nodes (Ni), preferably different consumers and / or different energy producers, via electrical lines (L), the network transfer module (10) comprising: a communication unit (11) for receiving operating parameters (BP) from the various network nodes (Ni), a memory unit (12) in which an artificial neural network (KNN), preferably a graph-based neural network (GNN), trained in particular offline, is stored, a computing unit (13) which is designed to process the operating parameters (BP) with the aid of the artificial neural network (KNN) in order to provide optimal control parameters (SP) and / or manipulated variables (SG) for the various network nodes (Ni).

2. Network handover module (10) according to claim 1, wherein the artificial neural network (KNN) was trained offline of the network station (100), and / or wherein the artificial neural network (KNN) is a self-learning artificial neural network, in particular a graph-based artificial neural network (GNN), and / or wherein the artificial neural network (KNN) was trained using a graph-based method.

3. Network transfer module (10) according to one of the preceding claims, wherein the operating parameters (BP) comprise a power, in particular an active power and / or a reactive power, a voltage and / or a phase angle on electrical lines (L) connecting the various network nodes (Ni).

4. Network transfer module (10) according to one of the preceding claims, wherein the control parameters (SP) comprise a power, in particular an apparent power, a voltage and / or a phase angle on corresponding lines (L), and / or wherein the manipulated variables (SG) comprise switch positions and / or power levels.

5. Network handover module (10) according to one of the preceding claims, wherein the artificial neural network (KNN) was trained using at least one model (M1, M2) of a network section (A) which has a network station (100) and various network nodes (Ni), wherein the network station (100) connects the various network nodes (Ni) via electrical lines (L), which in particular the neurons of the artificial neural network (KNN) map the various network nodes (Ni), wherein preferably the edges between the neurons of the artificial neural network (KNN) map power flow calculations (LF) for different lines (L).

6. Network transfer module (10) according to one of the preceding claims, wherein, during training of the artificial neural network (KNN), load flow calculations (LF) between the various network nodes (Ni) were used as connection functions between corresponding neurons of the artificial neural network (KNN), and / or wherein, during training of the artificial neural network (KNN), an optimal power flow calculation (OPF) at the network station (100), comprising at least one cost function, in particular having permissible operating parameter limits for the various network nodes (Ni) and / or economic efficiency function in energy generation and / or energy distribution, was used as an evaluation function and / or an optimization function.

7. Network handover module (10) according to one of the preceding claims, wherein the artificial neural network (KNN) was trained using a model (M1) of a network section (A), wherein the model (M1) maps any desired and / or virtual network section (A) with a network station (100) and various network nodes (Ni), or wherein the artificial neural network (KNN) was trained using a model (M2) of a network section (A), wherein the model (M2) maps a real network section (A) at a location of the network handover module (10) with a network station (100) and various network nodes (Ni).

8. Network handover module (10) according to one of the preceding claims, wherein the artificial neural network (KNN) was trained using different models (M1, M2) of different network sections (A), which may have a network station (100) and different network nodes (Ni).

9. Network handover module (10) according to one of the preceding claims, wherein the artificial neural network (KNN) was trained using different operating cases on at least one model (M1, M2) of a network section (A), which may have different operating parameters (BP) of the various network nodes (Ni).

10. Network transfer module (10) according to one of the preceding claims, wherein fictitious operating parameters (BP) of the various network nodes (Ni) were used in training the artificial neural network (KNN). 1 1. Network transfer module (10) according to one of the preceding claims, wherein the computing unit (13) is designed to provide the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), in particular only as a function of the output of the artificial neural network (KNN), preferably offline of the network station (100), and / or wherein the communication unit (11) is used, when providing optimal control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), only for receiving operating parameters (BP) from the various network nodes (Ni) and for transmitting control parameters (SP) for the various network nodes (Ni), wherein in particular no load flow calculations (LF) and / or optimal power flow calculations (OPF) are carried out online during operation of the network station (100). 1 . Network transfer module (10) according to one of the preceding claims, wherein the computing unit (13) is designed to evaluate the output of the artificial neural network (KNN), comprising the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), and / or wherein the computing unit (13) is designed to evaluate the output of the artificial neural network (KNN), comprising the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), with a quality metric which, in particular, maps an optimal power flow calculation (OPF) at the network station (100), comprising at least one cost function, in particular having permissible operating parameter limits for the various network nodes (Ni) and / or economic efficiency function in energy generation and / or energy distribution.

13. Use of a network transfer module (10) according to one of the preceding claims for planning and / or operating energy distribution networks, comprising at least one network station (100), in particular in the form of a local network station or a transformer substation, and various network nodes (Ni).

14. Network station (100), in particular in the form of a local network station or a transformer substation, comprising a network transfer module (10) according to one of the preceding claims.

15. Training method for training an artificial neural network (ANN), preferably a graph-based artificial neural network (GNN) for a network transfer module (10) according to one of the preceding claims, wherein the artificial neural network (ANN) is trained using at least one model (M1, M2) of a network section (A) which has a network station (100) and various network nodes (Ni), wherein the network station (100) connects the various network nodes (Ni) via electrical lines (L), wherein the neurons of the artificial neural network (ANN) map the various network nodes (Ni), and wherein the edges between the neurons of the artificial neural network (ANN) map power flow calculations (LF) for different lines (L).

16. Training method according to the preceding claim, wherein, when training the artificial neural network (ANN), load flow calculations (LF) between the various network nodes (Ni) are used as connection functions between corresponding neurons of the artificial neural network (ANN), and / or wherein, when training the artificial neural network (KNN), an optimal power flow calculation (OPF) at the network station (100), comprising at least one cost function, in particular having permissible operating parameter limits for the various network nodes (Ni) and / or an economic efficiency function in energy generation and / or energy distribution, is used as an evaluation function and / or an optimization function.

17. Training method according to one of the preceding claims, wherein the artificial neural network (KNN) is trained using a model (M1) of a network section (A), wherein the model (M1) maps any and / or virtual network section (A) with a network station (100) and various network nodes (Ni), or wherein the artificial neural network (KNN) is trained using a model (M2) of a network section (A), wherein the model (M2) maps a real network section (A) at a location of the network transfer module (10) with a network station (100) and various network nodes (Ni).

18. Training method according to one of the preceding claims, wherein the artificial neural network (KNN) is trained using different models (M1, M2) of different network sections (A), which may comprise a network station (100) and different network nodes (Ni).

19. Training method according to one of the preceding claims, wherein the artificial neural network (KNN) is trained using different operating cases on at least one model (M1, M2) of a network section (A), which may have different operating parameters (BP) of the various network nodes (Ni).

20. Training method according to one of the preceding claims, wherein fictitious operating parameters (BP) of the various network nodes (Ni) are used in training the artificial neural network (ANN).

21. Training method according to one of the preceding claims, When a network section (A) changes, the artificial neural network (ANN) is retrained to receive an update.

22. Operating method for operating a network station (100), in particular in the form of a local network station or a transformer substation, wherein the network station (100) connects various network nodes (Ni), preferably different consumers and / or different energy generators, via electrical lines (L), wherein a network transfer module (10) according to one of the preceding claims is used to operate the network station (100) in order to provide optimal control parameters (SP) for the various network nodes (Ni) with the aid of a trained artificial neural network (ANN), preferably a graph-based neural network (GNN).

23. Operating method according to the preceding claim, wherein the network transfer module (10) provides the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), in particular only as a function of the output of the artificial neural network (KNN), preferably offline of the network station (100), and / or wherein the communication unit (11) is used, when providing optimal control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), only for receiving operating parameters (BP) from the various network nodes (Ni) and for transmitting control parameters (SP) for the various network nodes (Ni), wherein in particular no load flow calculations (LF) and / or optimal power flow calculation (OPF) are carried out online during operation of the network station (100).

24. Operating method according to one of the preceding claims, wherein the network transfer module (10) evaluates the output of the artificial neural network (KNN), comprising the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), and / or wherein the network transfer module (10) evaluates the output of the artificial neural network (KNN), comprising the control parameters (SP) and / or control variables (SG) for the various network nodes (Ni), with a quality metric, which in particular comprises a optimal power flow calculation (OPF) at the network station (100), comprising at least one cost function, in particular having permissible operating parameter limits for the various network nodes (Ni) and / or economic efficiency function in energy generation and / or energy distribution.

25. Operating method according to one of the preceding claims, wherein, upon a change in a network section (A), the artificial neural network (KNN) is sent again to a learning station for training in order to receive an update.

Images