Energy flow management system
By customizing energy system simulation models and artificial intelligence algorithms to train the target value controller, and optimizing energy flow management, the problem of traditional systems failing to consider user and system behavior is solved, and the efficiency and cost reduction of energy system are improved.
Patent Information
- Application Number
- CN202380036929.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-04-29
- Filing Date
- 2023-04-27
- Publication Date
- 2024-12-10
AI Technical Summary
Traditional energy flow management systems fail to effectively consider user or system behavior, resulting in inefficiency of the energy system.
By customizing the simulation model of the energy system based on configuration data, the target value controller is trained using an artificial intelligence algorithm to calculate and control the target value of the controllable device of the energy system to optimize energy flow management.
Improve the efficiency of the energy system and reduce energy costs. By dynamically adjusting the controller in consideration of user and system behavior, more accurate energy management is achieved.
Smart Images
Figure CN119111026A_ABST
Abstract
Description
[0001] The invention relates to an energy flow management system adapted to perform a computer-implemented method for optimizing energy flows of controllable devices within an energy system.
[0002] US2019 / 288508A1 describes a method for operating an electrical energy supply network, which is connected to a connecting station with a lower-level distribution network, wherein the operation of the energy supply network is controlled by a network control system.
[0003] Renewable energy sources lead to increased fluctuations in energy production. Therefore, energy management systems are becoming increasingly important. An energy management system is provided to distribute the generated renewable energy as efficiently as possible and to support the grid stability of the power supply grid. An energy system comprising a photovoltaic array and a battery storage device can be implemented in a building. An energy management system can be provided to optimize the use of the generated photovoltaic power. In addition to locally available data (such as the generated photovoltaic power or the local current energy consumption or the current state of charge of the battery storage unit), conventional energy management systems can also use forecast data (such as weather or power consumption forecasts) or adaptive electricity prices to optimize the efficiency of the local energy system.
[0004] However, conventional energy flow management systems do not take into account the local behavior of the corresponding energy management system and / or the usage behavior of users of the energy system.
[0005] It is therefore an object of the present invention to provide a method and a system for improving the efficiency of an energy system taking into account user or system behavior.
[0006] The above objects are achieved by a computer-implemented method for optimizing energy flows of controllable devices within an energy system comprising the features of claim 1 .
[0007] According to a first aspect, the present invention provides a computer-implemented method for optimizing energy flow of controllable devices within an energy system, the method comprising the following steps:
[0008] Customize simulation models of energy systems based on configuration data,
[0009] Using a customized simulation model of the energy system, training individual target value controllers associated with controllable devices of the energy system by means of an artificial intelligence algorithm based on measured data and / or based on generic default data,
[0010] calculating target values for associated controllable devices of the energy system by the trained target value controller, and
[0011] Energy flow related functions of the controllable devices of the energy system are controlled in response to the calculated target values of the corresponding controllable devices.
[0012] In a possible embodiment of the computer-implemented method according to the first aspect of the invention, a simulation model of an energy system is selected from a set of predetermined simulation models stored in a data storage device and loaded into a processing device, which can be used to customize the loaded simulation model based on configuration data input by a user via a user interface and / or based on configuration data received from the corresponding energy system.
[0013] In a possible implementation of the computer-implemented method according to the first aspect of the invention, computing resources are used to execute an artificial intelligence algorithm to train respective target value controllers associated with controllable devices of the respective energy system. These computing resources may include cloud-based computing resources.
[0014] In a possible implementation of the computer-implemented method according to the first aspect of the invention, the trained target value controller is deployed on a control entity of the respective system.
[0015] These control entities may include a cloud-based control unit, a system control unit of the corresponding energy system or a distributed control unit of the controllable devices of the corresponding energy system.
[0016] In another possible implementation of the computer-implemented method according to the first aspect of the invention, the deployed trained target value controller is executed by a respective control entity to calculate target values of associated controllable devices of the energy system.
[0017] In another possible embodiment of the computer-implemented method according to the first aspect of the invention, the calculated target values of the controllable devices are encapsulated in a common schedule for the energy system and / or in separate device-specific schedules for each controllable device of the energy system.
[0018] In another possible embodiment of the computer-implemented method according to the first aspect of the present invention, a common schedule of the energy system is transmitted to a central energy management entity of the energy system, wherein target values of the common schedule are provided by the central energy management entity of the energy system to associated controllable devices of the energy system to control energy flow-related functions of the individual controllable devices of the energy system.
[0019] In another possible implementation of the computer-implemented method according to the first aspect of the invention, the device-specific schedule is transmitted to a distributed energy management entity of the controllable devices of the energy system,
[0020] The target value of the device-specific schedule received by the distributed energy management entity of each controllable device of the energy system is used to control the energy flow related functions of each controllable device of the energy system.
[0021] In another possible embodiment of the computer-implemented method according to the first aspect of the invention, retraining of individual target value controllers associated with controllable devices of the energy system by means of an artificial intelligence algorithm is triggered by a triggering event, or on the fly in parallel with execution on the device.
[0022] The triggering event may include a passing retraining period, a change in energy system settings or a change in energy system configuration and / or a detected change in the surroundings of the energy system and / or a received retraining command.
[0023] In a further possible implementation of the computer-implemented method according to the first aspect of the invention, the calculated target value comprises a parameter target value of an equipment parameter,
[0024] Parameter limits of a device parameter and / or control status of a device parameter.
[0025] In a further possible implementation of the computer-implemented method according to the first aspect of the invention, the calculated target values are encapsulated in a time grid of a timetable and / or are individually time stamped.
[0026] In another possible implementation of the computer-implemented method according to the first aspect of the invention, the controllable devices of the energy system include power consuming devices, power storing devices, power generating devices and / or power converting devices of the energy system.
[0027] In another possible embodiment of the computer-implemented method according to the first aspect of the invention, the simulation model used for customization comprises an electrical simulation model and / or a thermal simulation model of an energy system of the real-world building, in particular a MATLAB simulation model.
[0028] In another possible implementation of the computer-implemented method according to the first aspect of the invention, the artificial intelligence algorithm for training the target value controller comprises a self-learning algorithm.
[0029] The self-learning algorithm may include a genetic programming algorithm with symbolic regression.
[0030] In one possible implementation, the self-learning algorithm includes a non-dominated sorting genetic algorithm II or a offspring selection genetic algorithm.
[0031] According to another aspect, the invention also provides an energy flow management system comprising the features of claim 15 .
[0032] According to a second aspect, the invention provides an energy flow management system having a processing device adapted to perform the computer-implemented method according to the first aspect of the invention.
[0033] In the following, possible implementations of different aspects of the present invention are described in more detail with reference to the accompanying drawings.
[0034] Figure 1 A flow chart showing a possible exemplary implementation of a computer-implemented method for optimizing energy flows of controllable devices within an energy system according to the first aspect of the invention;
[0035] Figure 2 shows another flow chart for illustrating a possible exemplary embodiment of a computer-implemented method for optimizing energy flows of controllable devices within an energy system according to the first aspect of the present invention;
[0036] Figure 3 Another flow chart showing another exemplary embodiment of a computer-implemented method for optimizing energy flow of controllable devices within an energy system according to the first aspect of the present invention;
[0037] Figure 4 Shown for illustration Figures 1 to 3 A block diagram of the operation of a computer-implemented method for optimizing energy flow to controllable devices within an exemplary energy system in accordance with the present invention is shown;
[0038] Figure 5 Shown for illustration Figures 1 to 3 A schematic diagram of a possible exemplary embodiment of a computer-implemented method for optimizing energy flows of controllable devices within an energy system according to the first aspect of the present invention is shown;
[0039] Figure 6 Shown for illustration Figures 1 to 3 Another schematic diagram of another exemplary embodiment of a computer-implemented method for optimizing energy flow of controllable devices within an energy system is shown;
[0040] Figure 7 shows another schematic flow chart for illustrating sub-steps of a training process within the computer-implemented method according to the first aspect of the present invention;
[0041] Figure 8 The invention shows a possible exemplary implementation of the computer-implemented method according to the first aspect of the invention using a genetic programming algorithm. Figure 7 The mutation sub-step within the training process is shown.
[0042] like Figure 1 As shown in the flowchart of Figure 4 The computer-implemented method of controlling energy flows of controllable devices within the energy system 10) shown may include several main steps in a possible implementation.
[0043] In a first step S1, a simulation model of an energy system may be customized based on available configuration data. In a possible implementation, the configuration data may be input by a user of the energy system via a user interface. Furthermore, the configuration data may also be received from the corresponding energy system, for example from a configuration data memory of the energy system via a communication data link. In step S1, the simulation model of the energy system is customized to match the corresponding energy system, i.e. to match the system behavior of the energy system, but also to match the user behavior of at least one user of the corresponding energy system (e.g. the behavior of a person living in a building comprising the energy system) to heat a room in the building or to heat water in a boiler for hot water supply or to store electricity consumed by electrical devices in the building.
[0044] In a further step S2, the individual target value controllers associated with the controllable devices of the respective energy system are trained by means of an artificial intelligence algorithm using the measured data and / or generic default data and using the simulation model of the energy system customized in step S1. Figure 7 A possible implementation of the training process TP performed in step S2 is shown in FIG. The training process TP comprises Figure 7 The cycle shown.
[0045] In a further step S3, the target values TV of the associated controllable devices of the corresponding energy system are calculated by the trained target value controller in time intervals ranging from milliseconds to several hours. The calculated target values TV can be stored in a time table TT for transmission to the control entity of the energy system. The trained target value controller can also be deployed on the control entity of the energy system (such as a PLC).
[0046] In a final step S4, the energy flow related functions of the controllable devices of the energy system are controlled by the control entity of the controllable devices in response to the (pre) calculated target values TV received by the control entity within the time table TT transmitted to the control entity. The target values TV may also be calculated in real time on the fly by trained target value controllers deployed on the control entities of the respective controllable devices of the energy system. The target values TV applied by the control entity to the controllable devices may comprise set values.
[0047] In a possible embodiment of the computer-implemented method, the simulation model customized in step S1 may be selected from a set of predetermined simulation models stored in a data storage device and loaded into a processing device. The processing device is configured to customize the loaded simulation model in step S1 based on configuration data input by a user via a user interface and / or based on configuration data received from a configuration data storage of the energy system itself. The processing device may include computing resources for executing an artificial intelligence algorithm to automatically train respective target value controllers associated with controllable devices of the respective energy system.
[0048] The target value controller trained by the artificial intelligence algorithm in step S2 can be used to calculate the optimized target value TV embedded in the timetable TT, which can be transmitted to different kinds of control entities via a communication data link. These control entities may include a cloud-based control unit for remotely controlling the controllable devices of the energy system, a local system control unit of the corresponding energy system, and also a distributed control unit of different controllable devices within the corresponding energy system. The controllable devices may include different kinds of devices, in particular power consuming devices, power storage devices, power generating devices and / or power conversion devices within the energy system.
[0049] The corresponding energy system may be represented by a simulation model. In a possible embodiment, the simulation model may include an electrical simulation model and / or a thermal simulation model of an energy system of a real-world building (such as a user's residential building).
[0050] exist Figure 1 In step S3 of the computer-implemented method illustrated in the flow chart of , a trained target value controller is executed to calculate optimized target values TV for associated controllable devices within the energy system represented by the customized simulation model.
[0051] In a possible embodiment, the target values TV of the associated controllable devices of the energy system calculated in step S3 may be encapsulated in a common timetable TT-SYS for the energy system. In an alternative embodiment, the calculated target values may also be encapsulated in separate device-specific timetables TT-DEV for each controllable device of the energy system. A mixture of using a common timetable TT-SYS for a group of controllable devices and using dedicated device-specific timetables TT-DEV for other controllable devices is also possible.
[0052] If the calculated target values are encapsulated in a common timetable TT-SYS for the energy system, the common timetable can be communicated to a central energy management entity of the respective energy system in a possible embodiment. The target values TV of the common timetable TT-SYS can be provided or forwarded by the central energy management entity of the energy system to the associated control entities of the controllable devices of the energy system to control energy flow related functions of the respective controllable devices of the energy system in step S4, i.e. functions having an impact on the energy or power flow between components of the respective energy system.
[0053] On the other hand, if the calculated target values TV are encapsulated in individual device-specific schedules TT-DEV of the controllable devices of the energy system, the device-specific schedules can be transmitted to the distributed energy management entities of the controllable devices of the energy system. The target values TV of the device-specific schedules TT-DEV received by the distributed energy management entities of the individual controllable devices of the energy system are used to control energy flow related functions of the controllable devices within the energy system.
[0054] In a possible implementation of a computer-implemented method according to the first aspect of the present invention, retraining of each target value controller associated with a controllable device of an energy system may be performed by an artificial intelligence (AI) algorithm. The retraining may be triggered by a triggering event. The triggering event for performing the retraining may include a retraining period that has passed, a change in an energy system setting, or a change in an energy system configuration. The triggering event may also include a detected or predicted change in the surroundings or environment of the corresponding energy system. In addition, the triggering event may include a received retraining command. The retraining command may be issued by an operator of a manufacturing plant of a component of the energy system (such as an inverter or a photovoltaic array of an energy system that monitors the energy flow of the energy system), or by a local user of the energy system or a person living in a building where the energy system is installed. Retraining may also be triggered by a third party such as an operator of the power supply grid to which the energy system is connected to exchange electrical energy on demand.
[0055] The target values TV calculated in step S3 may include parameter target values of device parameters, parameter limits of device parameters and / or control states of corresponding device parameters. The calculated target values TV may be encapsulated in a time grid of a timetable TT and / or may be individually time stamped. The timetable TT may include columns for different target values TV and rows for different time points or time slots within the time grid lasting, for example, a predetermined number of milliseconds, seconds, minutes or hours depending on the use case.
[0056] In a possible embodiment, the simulation model of the customized energy system in step S1 of the computer-implemented method may include a MATLAB simulation model. In addition, the artificial intelligence algorithm used to train the target value controller in step S2 may include a self-learning algorithm, in particular a genetic programming algorithm that performs symbolic regression. In a preferred embodiment, the genetic programming algorithm may include an NSGA-II algorithm. In an alternative embodiment, the genetic programming algorithm may also include an OSGA algorithm.
[0057] In a possible embodiment, the computer-implemented method according to the present invention is designed to keep the energy cost of the energy system as low as possible. This is achieved by taking into account the individual user behavior and / or local system behavior of the energy system. Once the behavior of the real-world energy system changes, the controller is adjusted. The behavior of the real-world energy system may change, for example, due to additional electrical appliances or due to changes in the individual user behavior of the user operating the corresponding energy system. The computer-implemented method according to the present invention can be used for various different types of energy systems or energy system types. The energy system can be implemented in a small building, an apartment or even a large building (such as a production site). The simulation model for the accurate simulation of the energy system implemented in the corresponding building can include all components and electrical appliances of the energy system, and user behavior can also be modeled. The self-learning artificial intelligence algorithm can use a customized simulation model and historical data (especially measurement data) to train the optimal energy flow controller. The trained target value controller can then be forwarded or deployed on the control entity or computing unit. The control entity may include a cloud-based control unit, but may also include a local computer, in particular a system control unit of the corresponding energy system or even a distributed control unit of different controllable devices implemented in the corresponding local energy system. The trained and deployed target value controller is then executed within the energy flow management system in order to create a timetable TT with optimal control values, ie target values TV for different controllable appliances or devices within the corresponding energy system.
[0058] In a possible embodiment, the simulation model of the energy system representing the building can be constructed by building blocks of single electrical and / or thermal components of the real-world building. These components can include controllable devices of the energy system, in particular controllable power consumption devices or loads of the energy system, controllable storage devices (such as battery storage devices), controllable power generation devices (such as photovoltaic arrays of energy systems) and / or power conversion devices. In order to model user behavior and system behavior as accurately as possible, several different user inputs collected from the website can be used. These user inputs can be used to provide configuration data for customizing the corresponding simulation model of the local energy system. The configuration data input by the user via the user interface can, for example, include the number of people living in the corresponding building implementing the energy system or how many of them are employed. Another example of the configuration data input by the user can include the living area in square meters of the building implementing the energy system. The building blocks of the simulation model can be implemented in a general way, and then customized for each single real-world energy system in step S1 by configuring them with accurate real-world parameters and measurement data. In addition, the model blocks of the simulation model can be connected to each other in different ways, so that a small number of single building blocks of the simulation model can be used to accurately model various different and complex energy systems.
[0059] In a possible implementation of the computer-implemented method according to the first aspect of the present invention, the artificial intelligence algorithm may use the simulation model customized in step S1 and the measured historical system data. If the measured data is not available, the artificial intelligence algorithm may use universal default data to train the various target value controllers associated with the controllable devices of the corresponding energy system. These default data may include, for example, a standard load or a warm water tipping profile. The default data may be used to train a heuristic controller, which is suitable for minimizing / reducing the energy cost of the energy system, i.e., maximizing / improving the efficiency of the corresponding energy system. Due to the self-learning nature of the artificial intelligence algorithm, once a triggering event occurs, retraining of the controller may be automatically performed. The re-triggering event may include a predetermined retraining period. The triggering event may also include a change in the energy system settings, such as if the system changes significantly. In addition, the triggering event may include a detected change in the surrounding environment of the energy system, such as when the season changes from winter to summer.
[0060] Once the training of the heuristic energy flow controller is completed, the trained target value controller can be deployed in the hardware control entity of the energy flow management system. The trained target value controller can replace the previously existing operating target value controller of the energy system. Using the currently received system data and other data (such as calculated prediction data), the trained and deployed target value controller can calculate the optimal target values or control target values TV for different controllable devices of the corresponding energy system. These target values TV can be encapsulated in a timetable TT and sent to the corresponding controllable devices.
[0061] Figure 2 A possible exemplary implementation of the computer-implemented method according to the invention is shown.
[0062] In the illustrated embodiment, in step S0, a system configuration may be loaded and / or the user may enter details about his house or family via a user interface. These details may, for example, include the number of people living in the household or whether they are employed. These details or information may be used as configuration data for customizing a standard initial simulation model of a corresponding energy system.
[0063] In step S1 the information input in step S0 is used to customize the simulation model of the energy system to match the corresponding user energy system, including the behavior of controllable devices within the set energy system and the individual user behavior of the user(s) of the energy system.
[0064] In step S2, an artificial intelligence algorithm can be used to train individual target value controllers associated with controllable devices of the energy system. The energy consumption or energy generation of the device over time can be controlled. This can be performed in step S2A based on available measurement data MDATA, or in step S2B based on universal default data. In step S2A, measurement data MDATA received from sensors implemented in the energy system of the user's building can be aggregated and used to train the individual target value controllers of the corresponding energy system. In the event that no measurement data MDATA is available from the user's energy system, other available universal default data can be used to train the target value controller of the energy system in step S2B.
[0065] The measurement data MDATA can be provided to a processing device executing an artificial intelligence training algorithm via a wired or wireless communication link. In the event of an interruption of the communication link, available universal default data can be used instead to train the target value controller in step S2B. In a possible embodiment, the universal default data can be loaded from a common database, in particular a cloud-based database.
[0066] The trained target value controller may then be deployed and used to calculate the target values TV for the associated controllable devices of the energy system in step S3. A single trained controller may be deployed and used to calculate, for example, the power profile of each controllable device within the energy system over the next milliseconds or hours, depending on the respective implementation and the availability of prediction and measurement data.
[0067] exist Figure 2 In a possible implementation manner shown in the flowchart of , a single trained controller can output a schedule TT with corresponding power targets in step S4A, and the schedule TT can be sent to the control entity of the controllable device of the energy system. In step S4B, the controllable device of the energy system follows the power target value indicated in the received schedule TT until the control entity of the controllable device receives a new schedule.
[0068] Figure 3 Another schematic diagram for illustrating the operation of the computer-implemented method according to the present invention is shown.
[0069] Similar to Figure 2 In the embodiment shown, Figure 3 In the embodiment shown, in step S0 a system configuration may be loaded and / or the user may enter details about his house or family via the user interface.
[0070] from Figure 3It can be seen that in step S1, the simulation model of the energy system is customized using the user input and system configuration and model database from step S0. Figure 3 In a possible embodiment, the simulation model of the energy system is selected from a set of predetermined simulation models stored in a data storage device and loaded into a processing device, which is used to perform customization of the loaded simulation model in step S1. Figure 3 As shown, customization is performed based on system configuration and / or based on user input.
[0071] In a further step S2, the individual target value controllers associated with the different controllable devices of the energy system are trained by an artificial intelligence algorithm based on the measurement data MDATA using a customized simulation model. The measurement data MDATA used for training may include Figure 3 The historical system data shown can be obtained from Figure 4 If no such historical measurement data are available, then the measurement database 7 shown can be read. Figure 4 The other data (in particular the common default data) read from the database 8 shown are used for the training process TP. Figure 7 A possible implementation of the training process TP is shown in the flow chart of FIG.
[0072] In step S3, a target value TV for a controllable device is calculated by the trained target value controller and transmitted to an associated control entity of the energy system. In sub-step S3A, the trained target value controller is sent to an execution environment. The execution environment may include a control entity including a cloud-based control unit, a system control unit of the corresponding energy system, or a distributed control unit of different controllable devices implemented in the corresponding energy system.
[0073] like Figure 3 As shown, in another sub-step S3B, it can be checked whether a single target value controller has been executed or is running in a corresponding execution environment. If this is the case, Figure 3 In the sub-step S3C shown in FIG. 1 , the previously executed target value controller is exchanged with a trained or retrained new target value controller. If there is no target value controller already running on the corresponding execution environment, then the target value controller can be replaced in the following manner: Figure 3 In the illustrated sub-step S3D, the newly trained single target value controller is simply stored in the corresponding control entity. In sub-step S3E, the trained target value controller is run to optimize the energy flow in the energy system.
[0074] In step S4, the exchanged or implemented trained and deployed target value controller is executed to control the energy flow related functions of the controllable devices of the energy system. The executed controller may use the available current system data of the energy system read from the database, such as Figure 3 The calculated target values TV can be encapsulated in a common timetable TT-SYS for the energy system in sub-step S4A and / or in separate device-specific timetables TT-DEV for each controllable device of the energy system. The created timetable TT can then be sent to the control entity of the controllable device of the energy system in sub-step S4A. The control entity of the device can then Figure 3 The received control values TV from the received timetable TT are executed in the sub-step S4B shown. The common timetable TT-SYS for the entire energy system can be transmitted to the central energy management entity of the corresponding energy system in sub-step S4A. The target values TV of the common timetable TT-SYS can be provided or distributed by the central energy management entity of the energy system to the energy management entities of the associated controllable devices of the energy system to control the energy flow related functions of the controllable devices of the energy system.
[0075] As an alternative, the device-specific timetable TT-DEV may be transmitted to the distributed energy management entity of the controllable devices of the energy system in sub-step S4A. The target value TV of the device-specific timetable TT-DEV received by the distributed energy management entity of the controllable devices may be used to control energy flow related functions of the respective controllable devices within the energy system of the user.
[0076] Figure 7 Shown in Figure 1 , Figure 2 , Figure 3 A possible implementation of the training process TP performed in step S2 of an embodiment of the computer-implemented method according to the invention is shown in the flow chart of FIG. Figure 7 Possible sub-steps of a training procedure TP performed in step S2 of the computer-implemented method are shown.
[0077] In the first step S2 - 1 of the training process TP, possible solution candidates may be randomly initialized. Figure 8 An example of two (per column) possible parent solution candidates is shown, namely, parent solution candidate 1 and parent solution candidate 2. Figure 8 As can be seen in , each parent solution candidate may include a subtree provided for each heuristic controller within the modeled energy system 10 .
[0078] In a further step S2-2, a fitness assessment of the parent solution candidates may be performed. The quality (eg, energy cost reduction or other metrics) of different parent solution candidates may be assessed using the DLL and historical system data created once before the training process TP begins.
[0079] DLL generation works, for example, as follows: An optimization framework such as HeuristicLab can be used to initiate C code generation from a MATLAB Simulink model, forming a simulation model of the energy system 10. The code can be extended with other functions. The optimization framework can then compile the code into a dynamic link library (DLL). This DLL can then be used in the actual training process TP, where the framework can use a genetic algorithm to solve a symbolic regression optimization problem. The framework can create Figure 8 An initial population of solution candidates is shown, which is randomly initialized in step S2-1.
[0080] Two candidates may be selected from the parent solution candidates in the parent selection step S2-3. Figure 8 As shown, parent selection is done column by column. This selection is used in the pairing process in the mutation step S2-4, in which crossover and mutation are applied so that new child solution candidates are generated.
[0081] Step S2-4 is an example of having multiple (three) controllers. Figure 8 . One controller is responsible for one column. Each crossover is done between a selection from "parent solution candidate one" and a selection from "parent solution candidate two", represented by an "X" symbol. The selection is done column by column in this case. The pairing process is done in such a way that there is an exchange of information between two solution candidates in order to create a new child solution candidate - represented by an "arrow" symbol.
[0082] To exchange information, a branch is cut from each tree of solution candidates. The cut point can be chosen using different strategies (e.g. randomly). The cut branches will be exchanged. The result is two new child solution candidates.
[0083] After such crossover, there is a mutation, for example, of the sub-solution candidates. This is done, for example, in one part of the tree (branch). The mutation can be a change in a mathematical operator or a change in a constant value.
[0084] Then, in a further fitness evaluation step S2-5, the quality of these sub-solution candidates is evaluated by injecting them into the DLL, simulating the model and extracting the resulting energy cost from the DLL again. In a possible implementation, the energy cost can then be used as a quality metric. Figure 7 The best solution candidate is selected for the next generation based on this quality metric in the survivor selection step S2-6 shown. During the course of these generations, the quality metric (e.g., the energy cost of the system) is minimized. As a quality metric, one can also use, for example, the degree of autarky.
[0085] The training process TP may be performed until a predetermined termination criterion is reached, and the training process TP ends in step S2-7.
[0086] from Figure 7 As can be seen from the figure, sub-steps S2-3, S2-4, S2-5, and S2-6 are iterated or repeated in a loop until the termination criterion is met or reached.
[0087] A solution candidate in this context consists of an entity that holds one or more controller formulas represented by one or more subtrees, such as Figure 8 As shown, each subtree represents a target value controller 6 within the energy system. The controller formula indicates the functional correlation of the target control value TV represented by an operator ( / ; *; +; -) or a functional relationship such as log, exp, etc. with other entities. Symbolic regression forms a subtree or formula and optimizes specific parameters or values. In a preferred embodiment, the artificial intelligence algorithm used includes a genetic programming algorithm with symbolic regression, in particular a non-dominated sorting genetic algorithm II (NSGA-II). The algorithm not only attempts to minimize the energy cost, but also attempts to minimize the complexity of the trained target value controller 6 used in the simulation model, that is, the symbolic regression tree contained in the solution candidates. In an alternative embodiment, the genetic programming algorithm may also include an offspring selection genetic algorithm (OSGA). This forms a single-objective optimization algorithm designed to minimize the energy cost of a simulated energy system of a building.
[0088] Figure 4 A schematic diagram showing the architecture of an energy flow management system 1 according to the present invention is shown. In the exemplary embodiment shown, Figure 1 The energy flow management system 1 shown can be implemented in a data cloud. The energy flow management system 1 can include one or more user interfaces 2 and a simulation model database 3 in the cloud 4, such as Figure 4As shown. The simulation model database 3 may include a plurality of different predetermined simulation models, for example for different kinds of buildings. A specific simulation model may be selected from a set of predetermined simulation models and loaded into a processing unit 5 forming a cloud-based computing resource, which is used to customize the selected simulation model in step S1 and to train at least one individual target value controller 6 in step S2. Training may be performed in a training loop in step S2 based on the measurement data MDATA of the user's home or energy system loaded from the database 7, as shown. Figure 4 If no measurement data MDATA is available, the artificial intelligence algorithm can also use generic default data for the user's home to train a single target value controller 6 using a customized simulation model of the user's energy system. The generic default data can be provided to the processing unit 5 from the database 8, such as Figure 4 A single target value controller may be executed to calculate the target value encapsulated in step S3. Figure 4 The target value TV in the time table (TT) 9 shown is given by Figures 1 to 3 The target value TV calculated by the trained target value controller 6 in step S3 of the flowchart shown in FIG. Figures 1 to 3 In step S4 of the flowchart shown in FIG. Figure 4 The energy flow related functions of the controllable devices 12-i shown. The target values TV can be encapsulated in a time grid, for example, of a timetable TT. In an alternative implementation, the target values TV can also be individually time stamped with a system time stamp TS. The calculated target values TV of the timetable TT can include parameter target values of device parameters of the controllable devices. The calculated target values TV can also include parameter limits of device parameters of the controllable devices within the corresponding energy system and / or control states of device parameters.
[0089] exist Figure 4 In the illustrated embodiment, the time schedule (TT) 9 may be provided to a central energy management entity of the energy system 10 and / or transmitted to distributed energy management entities of the controllable devices 12 within the energy system 10. Figure 4 In the illustrated embodiment, the energy system 10 includes an inverter 11 for different kinds of controllable devices 12-1 to 12-5. These controllable devices 12-i may include power consuming devices such as household loads, power storage devices such as batteries, power generating devices such as photovoltaic arrays connected to the inverter 11, or power conversion devices such as power-to-heat conversion devices, wallboxes, or heat pumps. The controllable devices 12-i may or may not be connected to the inverter 11. Figure 4 The dashed lines in show communication and data transfer connections via a data bus system, while the straight, uninterrupted lines represent energy flows between the entities shown.
[0090] Figure 5 , Figure 6 Different possible implementations of the computer-implemented method according to the invention are shown. Figure 5 A first possible embodiment is shown, in which a single target value controller 6 is trained and deployed on a control entity of an inverter 11 of an energy system 10. In the example shown, the inverter 11 is connected to an electricity storage device formed by a battery 12-4. The inverter 11 is further connected to a power supply grid 13 in the embodiment shown, and to one or more household load devices 12-1 forming power consumption devices. The inverter 11 of the energy system 10 may include a control entity, which forms a local system control unit of the corresponding energy system 10 of the user. The control entity implemented in the inverter 11 can be used to collect measurement data MDATA of the energy system 10. As Figure 5 As shown, the measurement data MDATA may include different kinds of data, such as the current energy consumption tariff, the current energy feed-in tariff, the current photovoltaic production, the current household load or the current state of charge SoC of the battery 12-4. Additional measurement data MDATA may exist, including, for example, the power received from or provided to the power supply grid 13. The target value controller 6 may be trained in the cloud 4, for example in Figure 4 training on the processing unit 5 shown, or on Figure 5 The target value controller 6 is trained locally on the control entity of the inverter 11 shown. In step S2, the training of the target value controller 6 is performed by an artificial intelligence algorithm based on the measurement data MDATA that can be collected by the control entity of the inverter 11. The target value TVS provided by the trained target value controller 6 can, for example, include a battery charging or discharging power for the battery 12-4 of the energy system 10. The target value TV can be received by the control entity of the inverter 11 from an external cloud-based control entity, or calculated in real time by the control entity of the inverter 11.
[0091] Figure 6 Another possible implementation of the computer-implemented method according to the present invention is shown. Figure 6 As shown, each or at least a portion of the controllable devices 12-i of the energy system 10 may include an associated single target value controller 6. In the illustrated embodiment, the inverter 11 of the energy system 10 includes a control entity implementing an associated trained target value controller 6.
[0092] The controllable devices 12-i of the energy system 10 may include, for example, an electricity-to-heat conversion device 12-2 or a heat pump 12-5, such as Figure 4The controllable device 12-i may also include other types of controllable devices, such as a charging unit for charging an electric vehicle. For each controllable device 12-i, a target value TV may be provided by an associated trained target value controller 6 deployed on a control entity of the corresponding controllable device 12-i, such as Figure 6 As shown. The energy flow related functions of the controllable device 12-i can be controlled in response to the target value TV of the corresponding controllable device 12-i calculated by the associated trained target value controller 6. Figure 6 In the example shown, the measurement data MDATA provided by the measuring entities or sensors of the energy system 10 for training the target value controller may include measurement data MDATA related to different kinds of controllable devices 12-i, such as the previous power or the current boiler temperature of the electricity-to-heat conversion device 12-2.
[0093] In a preferred embodiment, the simulation model of the energy system 10 represents not only the electrical components of the energy system 10, but also the building itself. In this embodiment, the simulation model comprises an electrical simulation model and a thermal simulation model of the energy system 10 within the user's real-world building. In this embodiment, the simulation model comprises the electrical part and the thermal part of the connections. The computer-implemented method according to the present invention can be used for any kind of energy system, in particular for energy systems implemented in real-world buildings, such as residential buildings of users.
[0094] The configuration data is used to create and customize a simulation model of the energy system 10. The target value controller is trained by an artificial intelligence algorithm using a heuristic optimization framework.
[0095] The trained target value controllers 6 can be evaluated using a customized simulation model during simulation to classify the trained target value controllers according to their performance. The trained target value controllers 6 found to perform best for energy cost reduction can be deployed or produced for physical implementation in the corresponding buildings. After deployment in the energy system 10, these high-performance target value controllers 6 then optimize the energy flow of the controllable devices 12-i within the energy system 10 of the corresponding building. The target value controllers 6 can be trained using a genetic artificial intelligence algorithm to minimize the energy cost within the building.
Claims
1. A computer-implemented method for optimizing energy flows of controllable devices within an energy system (10), the method comprising the following steps: - based on the configuration data, customizing (S1) a simulation model of the energy system (10); - using a customized simulation model of the energy system (10), training (S2) individual target value controllers (6) associated with the controllable devices (12) of the energy system (10) by means of an artificial intelligence algorithm based on measured data and / or based on generic default data; - calculating (S3) a target value (TV) of an associated controllable device (12) of the energy system (10) by a trained target value controller (6); as well as - controlling (S4) energy flow related functions of the controllable devices (12) of the energy system (10) in response to the target values (TV) of the corresponding controllable devices (12) calculated by the trained target value controller (6).
2. A computer-implemented method according to claim 1, wherein the simulation model of the energy system (10) is selected from a set of predetermined simulation models stored in a database (3) and loaded into a processing device (5), and the processing device (5) customizes (S1) the loaded simulation model based on configuration data input by a user via a user interface (2) and / or based on configuration data received from the energy system (10).
3. A computer-implemented method according to claim 2, wherein computing resources are used to execute the artificial intelligence algorithm to train (S2) individual target value controllers (6) associated with controllable devices (12) of the corresponding energy system (10).
4. The computer-implemented method of any one of the preceding claims, wherein the trained target value controller is deployed on a control entity, the control entity comprising: Cloud-based control unit, a local system control unit of the corresponding energy system (10), or A distributed control unit for controllable devices (12) of a corresponding energy system (10).
5. A computer-implemented method according to any one of the preceding claims 1 to 4, wherein the deployed trained target value controller (6) is executed by a corresponding control entity to calculate (S3) a target value (TV) of an associated controllable device (12) of the energy system (10).
6. A computer-implemented method according to any one of the preceding claims 1 to 3, wherein the target value (TV) of the controllable device (12) calculated by the trained target value controller (6) is encapsulated in a common schedule of the energy system (10), and / or is encapsulated in separate device-specific schedules for each controllable device (12) of the energy system (10).
7. A computer-implemented method according to claim 6, wherein the common schedule of the energy system (10) is transmitted to a central energy management entity of the energy system (10), wherein a target value (TV) of the common schedule is provided by the central energy management entity of the energy system (10) to the associated controllable devices (12) of the energy system (10) to control (S4) energy flow-related functions of the individual controllable devices (12) of the energy system (10).
8. A computer-implemented method according to claim 6, wherein the device-specific schedule is transmitted to the distributed energy management entity of the controllable device (12) of the energy system (10), wherein the target value (TV) of the device-specific schedule received by the distributed energy management entity of each controllable device (12) of the energy system (10) is used to control the energy flow related functions of each controllable device (12) of the energy system (10).
9. The computer-implemented method according to any one of the preceding claims, wherein retraining of the respective target value controllers (6) associated with the controllable devices (12) of the energy system (10) by the artificial intelligence algorithm is triggered by a triggering event, the triggering event comprising: a retraining period elapsed, a change in the energy system settings or a change in the energy system configuration and / or a detected change in the surroundings of the energy system (10), and / or a received retraining command.
10. The computer-implemented method according to any one of the preceding claims 1 to 9, wherein the calculated target value (TV) comprises: Parameter target values of device parameters of controllable devices, parameter limits of device parameters and / or control states of said device parameters.
11. A computer-implemented method according to any one of the preceding claims, wherein the target values (TV) calculated by a trained target value controller are encapsulated in a time grid of the timetable and / or are individually time-stamped.
12. The computer-implemented method according to any one of the preceding claims, wherein the controllable device (12) of the energy system (10) comprises: Power consuming equipment, Power storage devices, Power generation equipment, and / or The energy system comprises a power conversion device.
13. The computer-implemented method of any one of the preceding claims, wherein the simulation model for customization comprises: An electrical simulation model and / or a thermal simulation model of an energy system (10) of a real-world building, in particular a MATLAB simulation model.
14. A computer-implemented method according to any of the preceding claims, wherein the artificial intelligence algorithm used to train (S2) the target value controller (6) comprises a self-learning algorithm, in particular a genetic programming algorithm performing symbolic regression.
15. An energy flow management system adapted to perform the computer-implemented method according to any one of the preceding claims 1 to 14.