CN-122026515-A - Cloud-edge collaborative power distribution network scheduling method based on digital twin

CN122026515ACN 122026515 ACN122026515 ACN 122026515ACN-122026515-A

Abstract

The invention relates to the technical field of power distribution networks, and particularly provides a cloud edge collaborative power distribution network scheduling method, equipment, medium and program product based on digital twin, wherein the method comprises the steps of injecting random scheduling actions and disturbance into a digital twin power distribution network to generate a training data set; training to obtain a proxy model through a training data set, wherein the proxy model is a deep neural network model, training to obtain a scheduling strategy model of a physical distribution network corresponding to the digital twin distribution network through a deep reinforcement learning method by taking the proxy model as an environment, and inputting real-time state data of the physical distribution network into the scheduling strategy model to determine a scheduling strategy of the physical distribution network. According to the method, the complex online optimization calculation is converted into forward reasoning of the edge side lightweight strategy model through the cooperative framework of cloud training and edge execution, so that the decision response time is shortened to the millisecond level, and the extremely real-time performance and the high operation safety of scheduling decisions are realized.

Inventors

DING LONG
LIU CHANGYI

Assignees

国网山西省电力有限公司经济技术研究院

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. A cloud edge collaborative distribution network scheduling method based on digital twinning is characterized by being applied to a cloud server and comprising the following steps: Injecting random scheduling actions and disturbance into the digital twin power distribution network to generate a training data set; Training to obtain a proxy model through the training data set, wherein the proxy model is a deep neural network model; training to obtain a scheduling strategy model of a physical power distribution network corresponding to the digital twin power distribution network by taking the agent model as an environment through a deep reinforcement learning method; Transmitting the scheduling policy model to an edge device; And adjusting the digital twin power distribution network, the proxy model and the scheduling strategy model based on the response data of the physical power distribution network uploaded by the edge equipment.
2. The cloud-edge collaborative distribution network scheduling method based on digital twinning according to claim 1, wherein prior to injecting random scheduling actions and disturbances into the digital twinning distribution network and generating a training data set, further comprising constructing the digital twinning distribution network; constructing a digital twin power distribution network, comprising: The method comprises the steps of obtaining a topological structure, equipment parameters and historical power data of a physical power distribution network, wherein the historical power data comprise load data, engine data, line data and transformer data; Constructing a digital twin power distribution network which is 1:1 mapped with the topological structure and equipment parameters of a physical power distribution network; And calibrating parameters of the digital twin power distribution network through the historical power data.
3. The digital twin based cloud edge collaborative distribution network scheduling method according to claim 1, wherein injecting random scheduling actions and perturbations into a digital twin distribution network to generate a training data set includes: collecting state data of the digital twin power distribution network; Injecting random scheduling actions and disturbance into the digital twin power distribution network to obtain response state data of the digital twin power distribution network, wherein the random scheduling objects are one or more of distributed power sources, energy storage systems, power grid topological structures, voltage reactive power equipment and loads; the state data, scheduling actions, perturbations, and response state data are combined into one training data.
4. The cloud-edge collaborative distribution network scheduling method based on digital twinning according to claim 3, wherein training through the training data set to obtain a proxy model comprises: For the training data, the state data, scheduling actions and perturbations are taken as inputs to a proxy model, and the response state data is taken as an output of the proxy model, so as to train the proxy model.
5. The cloud edge collaborative distribution network scheduling method based on digital twin according to claim 1, wherein the scheduling policy model of the physical distribution network corresponding to the digital twin distribution network is obtained through training by taking the agent model as an environment and using a deep reinforcement learning method, and the method comprises the following steps: at virtual time steps The intelligent agent for deep reinforcement learning obtains the state data of the digital twin power distribution network through observation And to compare the status data Inputting the scheduling policy model to obtain an execution action ; The state data is processed And performing an action Input to the proxy model to obtain response state data Wherein the status response data As the next virtual time step Status data of (2); based on the response status data Generating the action to be executed by a predefined reward function Is awarded of (a) ; The state data is processed Executing actions Responsive status data And rewards And merging and forming empirical data, and feeding back to the intelligent agent to update parameters of the scheduling strategy model.
6. The digital twinning-based cloud-edge collaborative distribution network scheduling method according to claim 5, wherein the reward function is comprised of at least one negative penalty term in combination with at least one positive reward term; The negative penalty term comprises at least one of a penalty for voltage out-of-limit severity, a penalty for network loss, and a penalty for device operation cost; the forward prize term includes at least a prize for maintaining a voltage stability margin.
7. The cloud edge collaborative distribution network scheduling method based on digital twinning is characterized by being applied to edge equipment and comprising the following steps: Receiving a scheduling policy model sent by a cloud server; inputting real-time state data of a physical power distribution network into the scheduling policy model, and determining a scheduling policy of the physical power distribution network; issuing the scheduling strategy to a terminal executing mechanism of the physical power distribution network; and sending the response data of the physical distribution network to a cloud server.
8. An electronic device comprising a processor and a memory storing a program, characterized in that the program comprises instructions that when executed by the processor cause the processor to perform the digital twinning-based cloud-edge collaborative distribution network scheduling method according to any one of claims 1-7.
9. A non-transitory machine readable medium storing computer instructions for causing the computer to perform the digital twinning-based cloud-edge collaborative distribution network scheduling method of any one of claims 1-7.
10. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the digital twin based cloud edge collaborative distribution network scheduling method of any one of claims 1 to 7.

Description

Cloud-edge collaborative power distribution network scheduling method based on digital twin Technical Field The invention relates to the technical field of power distribution networks, in particular to a cloud edge collaborative power distribution network scheduling method, equipment, medium and program product based on digital twin. Background With the high-proportion distributed renewable energy sources (such as photovoltaic and wind power) connected into a modern power distribution network, the running state of the power distribution network presents the characteristics of strong uncertainty and rapid fluctuation, and high requirements are put on the real-time performance, the self-adaptability and the safety of power grid dispatching control. Traditional scheduling methods relying on manual experience, off-line calculation or centralized optimization models have difficulty in coping with power fluctuations and voltage disturbances on the order of minutes or even seconds. The prior art scheme mainly comprises two types, namely online Model Predictive Control (MPC), wherein the MPC solves an optimization model on line based on the current state and predictive disturbance in each scheduling period, realizes control through rolling optimization and feedback correction, but when a refined model containing a large number of discrete variables and nonlinear constraints is processed, the solving time is often tens of seconds to minutes, the real-time requirement of second-level response is difficult to meet, the inherent contradiction between calculation time and model precision exists, and an offline strategy table or rule base prepares a control strategy in advance for a limited typical scene, and is executed through scene matching in online, however, the coverage capacity is limited, and the combination of all possible running states is difficult to be exhausted, and control failure or performance decline easily occurs when the new scene formed by random fluctuation is faced, and the sufficient generalization capacity and the self-adaption are lacked. Disclosure of Invention The invention aims to overcome the defects in the prior art, and provides a cloud-edge collaborative power distribution network scheduling method based on digital twinning, which ensures the instantaneity and accuracy of power distribution network scheduling decisions. In order to achieve the above purpose, the invention is realized by adopting the following technical scheme: In a first aspect, an embodiment of the present invention provides a cloud edge collaborative distribution network scheduling method based on digital twinning, which is applied to a cloud server, and includes: Injecting random scheduling actions and disturbance into the digital twin power distribution network to generate a training data set; Training to obtain a proxy model through the training data set, wherein the proxy model is a deep neural network model; training to obtain a scheduling strategy model of a physical power distribution network corresponding to the digital twin power distribution network by taking the agent model as an environment through a deep reinforcement learning method; Transmitting the scheduling policy model to an edge device; And adjusting the digital twin power distribution network, the proxy model and the scheduling strategy model based on the response data of the physical power distribution network uploaded by the edge equipment. In some embodiments of the invention, the method further comprises the steps of constructing the digital twin power distribution network before injecting random scheduling actions and disturbances into the digital twin power distribution network and generating the training data set; constructing a digital twin power distribution network, comprising: The method comprises the steps of obtaining a topological structure, equipment parameters and historical power data of a physical power distribution network, wherein the historical power data comprise load data, engine data, line data and transformer data; Constructing a digital twin power distribution network which is 1:1 mapped with the topological structure and equipment parameters of a physical power distribution network; And calibrating parameters of the digital twin power distribution network through the historical power data. In some embodiments of the present invention, injecting random scheduling actions and perturbations into a digital twin power distribution network generates a training data set comprising: collecting state data of the digital twin power distribution network; Injecting random scheduling actions and disturbance into the digital twin power distribution network to obtain response state data of the digital twin power distribution network, wherein the random scheduling objects are one or more of distributed power sources, energy storage systems, power grid topological structures, voltage reactive power equipment and loads; the state data, sche