A cloud-edge collaborative power distribution network scheduling method based on digital twins

Abstract

This invention relates to the field of distribution network technology, specifically providing a cloud-edge collaborative distribution network scheduling method, equipment, medium, and program product based on digital twins. The method includes: injecting random scheduling actions and disturbances into the digital twin distribution network to generate a training dataset; training a proxy model using the training dataset, wherein the proxy model is a deep neural network model; using the proxy model as an environment, training a scheduling strategy model of the physical distribution network corresponding to the digital twin distribution network using deep reinforcement learning methods; and inputting real-time status data of the physical distribution network into the scheduling strategy model to determine the scheduling strategy of the physical distribution network. This invention, through a collaborative architecture of "cloud training - edge execution," transforms complex online optimization calculations into forward inference of a lightweight strategy model on the edge side, reducing decision response time to milliseconds and achieving extreme real-time scheduling decisions and high operational security.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a cloud-edge collaborative power distribution network scheduling method, equipment, medium, and program product based on digital twins. Background Technology

[0002] With a high proportion of distributed renewable energy (such as photovoltaic and wind power) being connected to modern power distribution networks, their operating states exhibit strong uncertainty and rapid fluctuations, placing high demands on the real-time performance, adaptability, and security of power grid dispatch control. Traditional dispatch methods relying on manual experience, offline calculations, or centralized optimization models are no longer sufficient to handle power fluctuations and voltage disturbances on a minute or even second level. Existing technical solutions mainly fall into two categories: one is online model predictive control (MPC), which solves the optimization model online based on the current state and predicted disturbances in each dispatch cycle, achieving control through rolling optimization and feedback correction. However, when dealing with refined models containing a large number of discrete variables and nonlinear constraints, the solution time often reaches tens of seconds to several minutes, making it difficult to meet the real-time requirement of second-level response, and there is an inherent contradiction between computation time and model accuracy. The other is offline policy tables or rule bases, which pre-define control strategies for a limited number of typical scenarios and execute them online through scenario matching. However, their coverage is limited, making it difficult to exhaust all possible combinations of operating states. When facing new scenarios composed of random fluctuations, control failure or performance degradation is prone to occur, lacking sufficient generalization ability and adaptability. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a cloud-edge collaborative distribution network scheduling method based on digital twins, which ensures the real-time performance and accuracy of distribution network scheduling decisions.

[0004] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0005] In a first aspect, embodiments of the present invention provide a cloud-edge collaborative power distribution network scheduling method based on digital twins, applied to a cloud server, comprising:

[0006] Inject random scheduling actions and disturbances into a digital twin distribution network to generate a training dataset;

[0007] Using the training dataset, a proxy model is trained, wherein the proxy model is a deep neural network model;

[0008] Using the aforementioned proxy model as the environment, a scheduling strategy model for the physical distribution network corresponding to the digital twin distribution network is trained using deep reinforcement learning methods.

[0009] Send the scheduling strategy model to the edge device;

[0010] Based on the response data of the physical distribution network uploaded by the edge device, the digital twin distribution network, the proxy model, and the scheduling strategy model are adjusted.

[0011] In some embodiments of the present invention, before injecting random scheduling actions and disturbances into the digital twin distribution network to generate a training dataset, the method further includes: constructing the digital twin distribution network;

[0012] Building a digital twin distribution network includes:

[0013] The topology, equipment parameters, and historical power data of the physical power distribution network are obtained; wherein, the historical power data includes: load data, engine data, line data, and transformer data;

[0014] Construct a digital twin distribution network that maps 1:1 to the topology and equipment parameters of the physical distribution network;

[0015] The parameters of the digital twin distribution network are calibrated using the historical power data.

[0016] In some embodiments of the present invention, random scheduling actions and disturbances are injected into a digital twin distribution network to generate a training dataset, including:

[0017] Collect the status data of the digital twin distribution network;

[0018] Random scheduling actions and disturbances are injected into the digital twin distribution network to obtain the response status data of the digital twin distribution network; wherein, the random scheduling objects are one or more of distributed power sources, energy storage systems, grid topology, voltage reactive power devices, and loads; the disturbances are used to simulate random fluctuations in photovoltaic and wind power output;

[0019] The state data, scheduling actions, disturbance and response state data are combined into a single training dataset.

[0020] In some embodiments of the present invention, a proxy model is trained using the training dataset, including:

[0021] For the training data, the state data, scheduling actions, and disturbance actions are used as inputs to the proxy model, and the response state data is used as the output of the proxy model to train the proxy model.

[0022] In some embodiments of the present invention, using the proxy model as the environment, a scheduling strategy model of the physical distribution network corresponding to the digital twin distribution network is trained using a deep reinforcement learning method, including:

[0023] In virtual time step The deep reinforcement learning agent obtains the state data of the digital twin distribution network through observation. and the status data Input the scheduling strategy model to obtain the execution action. ;

[0024] The status data and perform actions The input is given to the proxy model to obtain response status data. The state response data As the next virtual time step Status data;

[0025] Based on the response status data The execution action is generated through a predefined reward function. Rewards ;

[0026] The status data Execution of actions Response status data and rewards The data is merged to form empirical data, which is then fed back to the intelligent agent to update the parameters of the scheduling strategy model.

[0027] In some embodiments of the present invention, the reward function is composed of at least one negative penalty term and at least one positive reward term;

[0028] The negative penalty items include at least one of the following: penalties for the severity of voltage exceedance, penalties for network losses, and penalties for equipment operating costs;

[0029] The positive reward items include at least a reward for maintaining voltage stability margin.

[0030] Secondly, embodiments of the present invention provide a cloud-edge collaborative power distribution network scheduling method based on digital twins, applied to edge devices, including:

[0031] Receive scheduling strategy model sent by the cloud server;

[0032] The real-time status data of the physical distribution network is input into the scheduling strategy model to determine the scheduling strategy of the physical distribution network;

[0033] The scheduling strategy is then distributed to the terminal execution mechanism of the physical distribution network;

[0034] The response data of the physical power distribution network is sent to the cloud server.

[0035] Thirdly, the present invention also provides an electronic device, including: a processor, and a memory storing a program, the program including instructions, which, when executed by the processor, cause the processor to perform the above-described cloud-edge collaborative power distribution network scheduling method based on digital twins.

[0036] Fourthly, the present invention also provides a non-transitory machine-readable medium storing computer instructions for causing the computer to execute the above-described cloud-edge collaborative power distribution network scheduling method based on digital twins.

[0037] Fifthly, the present invention also provides a computer program product, including a computer program / instruction, characterized in that, when the computer program / instruction is executed by a processor, it implements the above-mentioned cloud-edge collaborative power distribution network scheduling method based on digital twin.

[0038] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0039] First, the method of this invention achieves extreme real-time performance and high operational security in scheduling decisions. Through a collaborative architecture of "cloud training-edge execution," complex online optimization computations are transformed into forward inference of a lightweight strategy model on the edge, reducing decision response time to milliseconds and effectively suppressing rapid power fluctuations. Simultaneously, the exploration and training of all new strategies are conducted in a high-fidelity digital twin environment in the cloud, realizing "virtual trial and error, physical application," completely avoiding the security risks associated with directly applying new strategies to the physical power grid.

[0040] Secondly, the method of this invention utilizes a digital twin calibrated with historical data to generate a large amount of training data, and uses this training data to drive a deep neural network as a surrogate model. The surrogate model inherits the accuracy of the high-fidelity physical model while possessing the efficiency of neural network forward computation, thus providing an ideal environment that combines accuracy and speed for subsequent reinforcement learning training and rapid decision-making at the edge.

[0041] Furthermore, by feeding back real-world response data uploaded from edge devices to the cloud, the digital twin distribution network, agent model, and scheduling strategy model can be dynamically adjusted, enabling the system to continuously adapt to slow changes in the power grid. In addition, within the reinforcement learning framework, the agent interacts with the agent model to directly learn the end-to-end mapping from system state to optimal scheduling actions, inherently unifying situation prediction and decision optimization, breaking down the barriers of traditional modular methods. Attached Figure Description

[0042] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other embodiments based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating a cloud-edge collaborative power distribution network scheduling method based on digital twins, provided by an embodiment of the present invention.

[0044] Figure 2 This is a flowchart illustrating a cloud-edge collaborative power distribution network scheduling method based on digital twins for edge devices, provided by an embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0046] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0047] like Figure 1 As shown, this embodiment of the invention provides a cloud-edge collaborative power distribution network scheduling method based on digital twins applied to cloud servers. Figure 1 This is a flowchart illustrating the cloud-edge collaborative distribution network scheduling method based on digital twins. This flowchart only shows the logical sequence of the method described in this embodiment. In other possible embodiments of the invention, different sequences may be used, provided they do not conflict. Figure 1 Complete the steps shown or described in the order indicated.

[0048] See Figure 1 The method of this invention specifically includes the following steps:

[0049] Step S101: Inject random scheduling actions and disturbances into the digital twin distribution network to generate a training dataset.

[0050] Digital twin technology is a cutting-edge technology that simulates, analyzes, predicts, and optimizes the entire lifecycle of a physical entity or system by creating a dynamic digital model that is fully synchronized with the physical entity. Its core lies in building a virtual model that is synchronized with the physical entity in real time to support simulation analysis and decision optimization.

[0051] The methods for constructing a digital twin distribution network according to embodiments of the present invention include:

[0052] Acquire the topology, equipment parameters, and historical power data of the physical distribution network; the historical power data includes load data, engine data, line data, and transformer data, etc.; construct a digital twin distribution network that maps the topology and equipment parameters of the physical distribution network in a 1:1 manner; and calibrate the parameters of the digital twin distribution network using the historical power data.

[0053] A high-fidelity simulation model is established in a cloud server cluster, maintaining a 1:1 mapping relationship with the physical distribution network in terms of topology, equipment parameters, and physical laws. This model is constructed based on physical laws such as distribution network power flow equations and equipment dynamic characteristic equations. Using SCADA (Supervisory Control and Data Acquisition) and PMU (Phasor Measurement Unit) data obtained from the historical database of the physical distribution network, key parameters in the model, such as line parameters, transformer ratios, and load characteristics, are inverted and calibrated to ensure that the simulation output of this digital twin distribution network during steady-state and typical transient processes is consistent with the actual operating data of the physical distribution network within a set error range.

[0054] The digital twin power distribution network serves as the foundational environment for all subsequent virtual training and testing.

[0055] The core of step S101 is to build a "lightweight simulator" that can quickly and accurately simulate the dynamics of the distribution network. Its goal is to learn the state transition law of the distribution network, that is, "given the current state and actions, what will the distribution network become in the next moment".

[0056] With a virtual time step For example, this explains how random scheduling actions and disturbances are injected into a digital twin distribution network to generate training data. The specific steps include:

[0057] Data collection of digital twin distribution networks in virtual time steps Status data Injecting random scheduling actions into digital twin distribution networks and disturbance Obtain response status data of the digital twin distribution network. ; transfer status data Scheduling actions disturbance and response status data Merge into one training data , }

[0058] Status data and response status data It can be data such as voltage amplitude and phase angle of each node in the digital twin distribution network, active and reactive power of branches, and status of key switches.

[0059] Random scheduling actions It needs to cover various operating conditions and conform to physical laws. Essentially, it involves performing randomized operational simulations of all controllable objects in the physical distribution network within a high-fidelity virtual environment, adhering to safety constraints. This exhausts all possible combinations of operating states and control strategies, thereby training an intelligent dispatching strategy capable of handling all operating conditions. Specifically, the dispatching objects and actions should cover the entire "source-grid-load-storage" chain of the physical distribution network, mainly including the following five categories:

[0060] The first category is the scheduling of distributed power sources, primarily targeting grid-connected inverters for renewable energy sources such as photovoltaics and wind power. Random scheduling actions include randomly setting active power output values ​​within their rated capacity range to simulate scenarios such as power curtailment and solar power abandonment; simultaneously, randomly setting reactive power output values ​​within their four-quadrant operating capabilities to enable their flexibility in voltage regulation. The second category is the scheduling of energy storage systems, with battery energy storage converters as the core target. Random scheduling actions include randomly setting charge and discharge power values ​​within the maximum charge and discharge power range, and potentially randomly switching their operating modes (charge / discharge / standby) to simulate their role in supporting power balance and frequency. The third category is the scheduling of network topology, with tie switches and sectionalizing switches as key targets. Random scheduling actions simulate the "on" or "off" state changes of these switches, thereby randomly reconstructing the radial or interconnected operating structure of the power grid. This is crucial for training the agent to master load transfer and fault isolation capabilities. The fourth category is the scheduling of voltage and reactive power equipment, involving on-load tap-changing transformers and parallel capacitors / reactors. Random dispatching actions include randomly adjusting the transformer tap position within tap limits and randomly setting the number of capacitor banks in operation to simulate various reactive power compensation and voltage regulation methods. The fifth category is the dispatching of flexible loads, mainly targeting interruptible and adjustable loads. Random dispatching actions include issuing interruption commands or randomly adjusting their operating power and time periods within the agreed-upon limits to simulate demand-side response.

[0061] The “random injection” of these scheduling actions is not completely disordered, but follows principles such as combined injection (changing parameters of multiple objects at the same time to simulate collaborative control), timing association (simulating logic with an operational sequence), and safety boundary constraints (all values ​​must be within the physical limits of the equipment and the operating procedures).

[0062] Injecting disturbances into a digital twin distribution network serves to introduce real-world uncertainties into the virtual training environment, simulating random fluctuations in photovoltaic and wind power output, thereby ensuring that the surrogate model trained on simulated data can maintain robust adaptability and reliability in the complex and ever-changing real power grid.

[0063] Step S102: Train the surrogate model using the training dataset, wherein the surrogate model is a deep neural network model.

[0064] During the training phase of the surrogate model, the training data obtained in step S101... , the state data Scheduling actions and disturbance The response state data will be used as input to the proxy model. The output of the proxy model is used to train the proxy model.

[0065] The training process is a standard supervised learning process. The loss function is defined as the mean squared error between the predicted state and the true state. The gradient descent algorithm is used to iterate repeatedly on the training dataset, continuously adjusting the surrogate model parameters to minimize the prediction error.

[0066] The trained surrogate model can approximate the original time-consuming physical simulation with high accuracy at a speed of microseconds. However, it can only predict state changes and also has the ability to judge the quality of actions.

[0067] Step S103: Using the agent model as the environment, train the scheduling strategy model of the physical distribution network corresponding to the digital twin distribution network through deep reinforcement learning.

[0068] Deep Reinforcement Learning (DRL) is an artificial intelligence approach that combines the perception capabilities of deep learning with the decision-making capabilities of reinforcement learning. It learns complex decision-making paradigms through trial-and-error interactions between the agent and the environment.

[0069] Step S103 uses the agent model obtained in step S102 as the environment to train a scheduling strategy model that can output the optimal scheduling action through reinforcement learning. Its core is to solve the decision problem of "what action should be taken in a given state".

[0070] With a virtual time step For example, the specific steps of step S103 are explained, including:

[0071] In virtual time step Deep reinforcement learning agents obtain state data of the digital twin distribution network through observation. and the status data Input the scheduling strategy model to obtain the execution action. ; transfer status data and perform actions Input is given to the proxy model to obtain response status data. Among them, status response data As the next virtual time step Status data; based on response status data By using a predefined reward function, a reward for the action is generated. Rewards Reward value To perform actions In state Real-time quantitative evaluation of the effects produced; state data Execution of actions Response status data and rewards Merging to form empirical data The data is fed back to the agent to update the parameters of the scheduling policy model. Then, the process proceeds to the next virtual time step. This process is repeated, and will not be elaborated further.

[0072] The reward function is used to guide the agent to choose the correct path. It encodes the complex optimization objectives (safety, economy, stability) of distribution network scheduling into a computable scalar signal.

[0073] The reward function typically consists of at least one negative penalty term and at least one positive reward term. The negative penalty term and positive reward term data are generated directly or indirectly from the response state data obtained at each virtual time step. In this embodiment of the invention, the negative penalty term includes penalties for the severity of voltage exceedance, penalties for network losses, and penalties for device operating costs; the positive reward term includes a reward for maintaining voltage stability margin. Furthermore, a weighting coefficient is designed for each penalty term to measure its importance.

[0074] The above process is repeated tens of millions or even hundreds of millions of times at extremely high speeds (milliseconds per interaction) in a cloud-based virtual environment to conduct risk-free trial and error. Through massive amounts of trial and error, the intelligent agent explores different actions in different states and continuously adjusts its strategy based on the reward signals returned by the environment.

[0075] After each interaction, the environment calculates a reward based on a reward function. If an action causes voltage to exceed limits (unsafe), the reward will be a large negative value (penalty). Through algorithmic learning, the agent will reduce the probability of choosing that action in similar future situations. If an action stabilizes the voltage while reducing network loss and the action magnitude is small, the reward will be a positive value or a small negative value (reward or slight penalty), and the agent will reinforce that behavior. Through continuous trial and error and reward feedback, the agent gradually learns to select action sequences that maximize long-term cumulative rewards.

[0076] In reinforcement learning, the optimal policy is defined as the policy that maximizes the expected cumulative reward. Through trial and error and learning, the agent causes the parameters of the scheduling policy model to converge to a set of values ​​that maximize the total reward obtained in the long run when acting according to this policy in the virtual environment. Since the reward function fully reflects the optimization objective of power grid scheduling, this policy that maximizes the cumulative reward is considered the optimal or near-optimal scheduling policy sought in this invention. It inherently balances multiple constraints and objectives, including safety, economy, and equipment wear.

[0077] Step S104: Send the scheduling strategy model to the edge device.

[0078] The scheduling strategy model trained in step S103 is transformed into a lightweight version with extremely low computational and storage overhead through model compression techniques such as model pruning, weight quantization, and knowledge distillation. The lightweight scheduling strategy model is then deployed to edge devices in the physical distribution network, such as substation monitoring hosts and feeder automation terminals.

[0079] Step S105: Based on the response data of the physical distribution network uploaded by the edge device, adjust the digital twin distribution network, agent model, and scheduling strategy model.

[0080] The real response data of the physical distribution network is encrypted and synchronously transmitted back to the cloud server. This response data is used for two purposes: first, dynamic calibration of the digital twin distribution network, using new data to fine-tune key parameters of the high-fidelity model in the cloud online, enabling the digital twin to track drifts in the physical distribution network caused by equipment aging, minor changes in network topology, etc., and maintain its high fidelity; second, incremental learning of the agent model and scheduling strategy model, adding new data as new samples to the training dataset, and periodically incrementally training or fine-tuning the agent model and scheduling strategy model. This gives the entire system continuous learning and adaptability, enabling it to continuously optimize strategies to cope with the long-term evolution of the distribution network and new operating modes.

[0081] like Figure 2 As shown in the figure, this embodiment of the invention also provides a cloud-edge collaborative power distribution network scheduling method based on digital twins for edge devices. Figure 2 This is a flowchart illustrating the cloud-edge collaborative distribution network scheduling method based on digital twins. This flowchart only shows the logical sequence of the method described in this embodiment. In other possible embodiments of the invention, different sequences may be used, provided they do not conflict. Figure 2 Complete the steps shown or described in the order indicated.

[0082] See Figure 2 The method of this invention specifically includes the following steps:

[0083] Step S106: Receive the scheduling strategy model sent by the cloud server.

[0084] Step S107: Input the real-time status data of the physical distribution network into the scheduling strategy model to determine the scheduling strategy of the physical distribution network.

[0085] Step S108: Distribute the scheduling strategy to the terminal execution mechanism of the physical distribution network.

[0086] Step S109: Send the response data of the physical power distribution network to the cloud server.

[0087] During online operation, edge devices collect real-time status data of the physical distribution network via the local communication network and input it into a lightweight scheduling strategy model. The model completes forward inference within 10-50 milliseconds and outputs the optimal scheduling action. Based on the optimal scheduling action, corresponding scheduling instructions are generated and sent to the corresponding terminal actuators, such as photovoltaic inverters, energy storage converters, and on-load tap changers, via standard industrial protocols (such as IEC 61850 and Modbus). Afterward, real-time response data from the physical distribution network is collected, encrypted, and sent to the cloud server.

[0088] An embodiment of the present invention also provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the cloud-edge collaborative distribution network scheduling method based on digital twins, an embodiment of the present invention.

[0089] The present invention also provides a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the cloud-edge collaborative distribution network scheduling method based on digital twins, an embodiment of the present invention.

[0090] An embodiment of the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the cloud-edge collaborative power distribution network scheduling method based on digital twins, as described in the embodiments of the present invention.

[0091] refer to Figure 3 The present invention will now describe a structural block diagram of an electronic device that can serve as an embodiment of the present invention, serving as an example of a hardware device applicable to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0092] like Figure 3 As shown, the electronic device includes a computing unit 101, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 102 or a computer program loaded from a storage unit 108 into a random access memory (RAM) 103. The RAM 103 may also store various programs and data required for the operation of the electronic device. The computing unit 101, ROM 102, and RAM 103 are interconnected via a bus 104. An input / output (I / O) interface 105 is also connected to the bus 104.

[0093] Multiple components in the electronic device are connected to I / O interface 105, including: input unit 106, output unit 107, storage unit 108, and communication unit 109. Input unit 106 can be any type of device capable of inputting information into the electronic device. Input unit 106 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 107 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 108 may include, but is not limited to, disks and optical discs. Communication unit 109 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0094] The computing unit 101 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 101 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 101 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as computer programs tangibly contained in a machine-readable medium, such as storage unit 108. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 102 and / or communication unit 109. In some embodiments, the computing unit 101 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0095] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0096] In the context of embodiments of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0097] It should be noted that the term "comprising" and its variations used in the embodiments of this invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "a plurality" mentioned in the embodiments of this invention are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0098] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0099] The steps described in the method embodiments provided by the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0100] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0101] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A cloud-edge collaborative distribution network scheduling method based on digital twins, characterized in that, Applications to cloud servers include: Inject random scheduling actions and disturbances into a digital twin distribution network to generate a training dataset; Using the training dataset, a proxy model is trained, wherein the proxy model is a deep neural network model; Using the aforementioned proxy model as the environment, a scheduling strategy model for the physical distribution network corresponding to the digital twin distribution network is trained using deep reinforcement learning methods. Send the scheduling strategy model to the edge device; Based on the response data of the physical distribution network uploaded by the edge device, the digital twin distribution network, the proxy model, and the scheduling strategy model are adjusted. Using the aforementioned proxy model as the environment, a scheduling strategy model for the physical distribution network corresponding to the digital twin distribution network is trained through deep reinforcement learning, including: In virtual time step The deep reinforcement learning agent obtains the state data of the digital twin distribution network through observation. and the status data Input the scheduling strategy model to obtain the execution action. ; The status data and perform actions The input is given to the proxy model to obtain response status data. The response status data As the next virtual time step Status data; Based on the response status data The execution action is generated through a predefined reward function. Rewards ; The status data Execution of actions Response status data and rewards The data is merged to form empirical data, which is then fed back to the intelligent agent to update the parameters of the scheduling strategy model.

2. The cloud-edge collaborative distribution network scheduling method based on digital twins according to claim 1, characterized in that, Before injecting random scheduling actions and disturbances into the digital twin distribution network to generate a training dataset, the process also includes: constructing the digital twin distribution network; Building a digital twin distribution network includes: The topology, equipment parameters, and historical power data of the physical power distribution network are obtained; wherein, the historical power data includes: load data, engine data, line data, and transformer data; Construct a digital twin distribution network that maps 1:1 to the topology and equipment parameters of the physical distribution network; The parameters of the digital twin distribution network are calibrated using the historical power data.

3. The cloud-edge collaborative distribution network scheduling method based on digital twins according to claim 1, characterized in that, Injecting random scheduling actions and disturbances into a digital twin distribution network generates a training dataset, including: Collect the status data of the digital twin distribution network; Random scheduling actions and disturbances are injected into the digital twin distribution network to obtain the response status data of the digital twin distribution network; wherein, the random scheduling objects are one or more of distributed power sources, energy storage systems, grid topology, voltage reactive power devices, and loads; the disturbances are used to simulate random fluctuations in photovoltaic and wind power output; The state data, scheduling actions, disturbance and response state data are combined into a single training dataset.

4. The cloud-edge collaborative distribution network scheduling method based on digital twins according to claim 3, characterized in that, Using the training dataset, a proxy model is trained, including: For the training data, the state data, scheduling actions, and disturbance actions are used as inputs to the proxy model, and the response state data is used as the output of the proxy model to train the proxy model.

5. The cloud-edge collaborative distribution network scheduling method based on digital twins according to claim 1, characterized in that, The reward function is composed of at least one negative penalty term and at least one positive reward term; The negative penalty items include at least one of the following: penalties for the severity of voltage exceedance, penalties for network losses, and penalties for equipment operating costs; The positive reward items include at least a reward for maintaining voltage stability margin.

6. The cloud-edge collaborative distribution network scheduling method based on digital twins according to claim 1, characterized in that, Applied to edge devices, including: Receive scheduling strategy model sent by the cloud server; The real-time status data of the physical distribution network is input into the scheduling strategy model to determine the scheduling strategy of the physical distribution network; The scheduling strategy is then distributed to the terminal execution mechanism of the physical distribution network; The response data of the physical power distribution network is sent to the cloud server.

7. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the cloud-edge collaborative power distribution network scheduling method based on digital twins according to any one of claims 1 to 6.

8. A non-transitory machine-readable medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the cloud-edge collaborative power distribution network scheduling method based on digital twins according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the cloud-edge collaborative power distribution network scheduling method based on digital twins as described in any one of claims 1 to 6.

Images