CN-122001096-A - AI-based micro-grid distributed cooperative control method

CN122001096ACN 122001096 ACN122001096 ACN 122001096ACN-122001096-A

Abstract

The invention discloses an AI-based micro-grid distributed cooperative control method, which relates to the technical field of micro-grid intelligent control and comprises the steps of collecting distributed power supply operation data and constructing a micro-grid dynamic topological graph, wherein edges between nodes are endowed with comprehensive weights for fusing electric distances and communication delays. The characteristics of each node and the neighbors thereof are input into a pre-trained graph neural network, the network aggregates multi-hop neighbor information through multi-layer message transmission, the node state is updated, and a cooperative control strategy is generated for each distributed power supply according to the node state, and the cooperative control strategy is issued to a local controller for execution. The system collects new data in a fixed period to realize closed loop optimization. The method realizes accurate modeling of the micro-grid information physical coupling characteristic, enhances the feasibility and robustness of a control strategy, enables a local decision to have global coordination capability through multi-hop information aggregation, and optimizes the dynamic response of the system.

Inventors

ZHOU XIAOWEI
WANG FENG
DING WUJUN
LI JIANG

Assignees

杭州鸿途智慧能源技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260212

Claims (10)

1. The AI-based micro-grid distributed cooperative control method is characterized by comprising the following steps: acquiring an operation characteristic data set of all distributed power supply nodes in the micro-grid; Modeling the graph structure of the operation characteristic data set, modeling each distributed power supply node as a vertex in the graph structure, modeling the electrical connection and communication link between the nodes as edges, and endowing each edge with comprehensive weight reflecting the electrical distance and communication time delay to form a dynamic topological graph of the micro-grid; extracting local neighborhood characteristics of each vertex from the dynamic topological graph of the micro-grid, wherein the local neighborhood characteristics comprise vertex self operation characteristic data and operation characteristic data of neighbor vertices connected through edges to form vertex characteristic vectors; inputting vertex feature vectors of all vertexes into a pre-trained graphic neural network, wherein the graphic neural network aggregates information of each vertex and multi-hop neighbors thereof through multi-layer message transmission and updates hidden state vectors of all vertexes; calculating a cooperative control strategy for each vertex based on the updated hidden state vectors of all the vertices; the cooperative control strategy of each vertex is fed back to the corresponding distributed power supply local controller to be executed, and the output characteristic of the distributed power supply is changed; and after a preset time interval, the operation characteristic data of each node is collected again and combined with the execution result at the previous moment to generate a new operation characteristic data set for starting the next control cycle.
2. The AI-based microgrid distributed cooperative control method of claim 1, wherein graph structure modeling the operational characteristic dataset comprises: The operation characteristic data set comprises output power of a node, node voltage, local load and energy storage state; determining the connection relation of distributed power supply nodes according to the physical wiring diagram of the micro-grid and the communication network structure; Calculating impedance modulus values between any two nodes which are electrically connected as electrical distance weights; Measuring or estimating the round-trip time of a data packet between any two nodes with communication connection as a communication delay weight; The electrical distance weight and the communication time delay weight are subjected to linear superposition after normalization to obtain comprehensive weights of edges connecting two nodes; And constructing an adjacency matrix and a feature matrix of the dynamic topological graph of the micro-grid based on the comprehensive weights of all the nodes, the connection relations and the corresponding edges.
3. The AI-based microgrid distributed cooperative control method of claim 2, wherein extracting local neighborhood features of each vertex from the microgrid dynamic topology map comprises: identifying all neighbor vertexes directly connected with each vertex through the edge aiming at each vertex in the dynamic topological graph of the micro-grid; Reading operation characteristic data of the vertex; Reading the operation characteristic data of all the neighbor vertexes, and weighting the operation characteristic data of the neighbor vertexes according to the comprehensive weight of the connecting edge; and splicing the operation characteristic data of the vertex with the weighted neighbor vertex operation characteristic data to form a vertex characteristic vector of the vertex.
4. The AI-based microgrid distributed cooperative control method of claim 3, wherein said inputting vertex feature vectors of all vertices to a pre-trained graph neural network comprises: Inputting an adjacency matrix of the dynamic topological graph of the micro-grid and vertex feature vector matrices of all vertexes into a graph neural network; At each layer of the graph neural network, each vertex receives messages which are transmitted by the neighbor vertexes through edges and are processed by the hiding states of the neighbor vertexes and the edge weights; Each vertex aggregates all received neighbor messages, fuses the neighbor messages with the last round of hiding state of the vertex through nonlinear transformation, and updates the hiding state vector of the vertex; and after the message passing and the state updating of the preset layer number, outputting the final hidden state vector after updating of all the vertexes.
5. The AI-based microgrid distributed cooperative control method of claim 4, wherein said calculating a cooperative control strategy for each vertex based on updated hidden state vectors of all vertices comprises: The cooperative control strategy comprises a power reference value adjustment instruction and a voltage compensation instruction; inputting the final hidden state vector of each vertex into a fully connected strategy network; The full-connection strategy network outputs continuous values of two dimensions, and the continuous values respectively correspond to the power reference value adjustment quantity and the voltage compensation quantity; Adding the power reference value adjustment quantity to the output power of the current node to obtain a power reference value adjustment instruction; Adding the voltage compensation quantity to the current node voltage to obtain a voltage compensation instruction; The power reference value adjustment command and the voltage compensation command together form a cooperative control strategy of the vertex.
6. The AI-based microgrid distributed cooperative control method of claim 5, wherein said feeding back the cooperative control strategy of each vertex to its corresponding distributed power source local controller is performed, comprising: transmitting a power reference value adjusting instruction to a power controller corresponding to the distributed power supply, and adjusting the reference value of a power control loop of the power controller; the voltage compensation instruction is issued to a voltage regulator corresponding to the distributed power supply and used as an additional compensation signal for voltage droop control or reactive voltage control; The distributed power supply local controller drives the converter through pulse width modulation according to the new reference value and the compensation signal, and changes the output active power and reactive power.
7. The AI-based microgrid distributed cooperative control method according to claim 1, wherein the re-collecting the operation characteristic data of each node after a preset time interval and combining the operation characteristic data with the execution result at the previous time to generate a new operation characteristic data set comprises: When a preset control period arrives, synchronously acquiring the output power, node voltage, local load and energy storage state of all distributed power nodes in the micro-grid; Acquiring the actual completion degree of a power reference value adjustment instruction and a voltage compensation instruction executed in the last control period from each distributed power supply local controller; And correlating and combining the newly acquired operation characteristic data with the actual completion degree data of the previous period instruction according to the nodes to form a new operation characteristic data set containing historical action feedback.
8. The AI-based microgrid distributed cooperative control method according to claim 7, wherein after associating and merging the newly collected operation characteristic data with the actual completion data of the previous period instruction by node, further comprising: Checking consistency of the new operation characteristic data set, and checking whether the time stamp of each node data is synchronous and whether an abnormal outlier exists; Performing standardized processing on the operation characteristic data passing the verification to eliminate the influence of different physical dimensions; and taking the normalized data as the input of the graph structure modeling in the next round of control loop.
9. The AI-based microgrid distributed cooperative control method of claim 1, wherein the training method of the pre-trained graph neural network comprises: Constructing a historical data set containing typical operation scenes and fault scenes of various micro-grids; the historical data set is utilized, the global running cost of the micro-grid is the lowest, the voltage deviation is the smallest, and the parameters of the graph neural network and the full-connection strategy network are trained offline by adopting a deep reinforcement learning algorithm; after training, parameters of the graph neural network and the full-connection strategy network are fixed and used for online distributed cooperative control.
10. The AI-based microgrid distributed cooperative control method of claim 1, further comprising an exception handling mechanism: in each round of control loop, the graph neural network outputs uncertainty evaluation values of each vertex hiding state vector at the same time; If the uncertainty evaluation value of a certain vertex exceeds a preset confidence threshold, judging that the vertex data or communication is abnormal; And switching to a conservation control mode locally preset by the abnormal vertex, and simultaneously notifying the abnormal state of the abnormal vertex to a neighbor vertex through a message transmission mechanism of a graph neural network, and guiding the neighbor vertex to adapt to the abnormal condition of the abnormal vertex in the next round of calculation.

Description

AI-based micro-grid distributed cooperative control method Technical Field The invention belongs to the technical field of intelligent control of micro-grids, and particularly relates to a distributed cooperative control method of a micro-grid based on an AI. Background The stable and efficient operation of a micro-grid depends on the coordinated control of numerous distributed power sources within it. The existing cooperative control method is mainly divided into two types, namely centralized type and distributed type. The centralized control relies on the central controller to collect global information and perform optimization calculation, but the central controller has single-point fault risk, and has heavy communication and calculation burden, so that the central controller is difficult to adapt to the scenes of plug and play and topology change of the distributed power supply. The distributed control, such as a method based on a consistency protocol, has higher reliability and expandability because each node only communicates with the neighbor, and becomes the current main stream research direction. Existing distributed control schemes are typically based on a simplified network model. Most methods only design control laws according to the topology of the communication network, neglect the difference of actual electric coupling strength between nodes, and other methods consider electric connection, but do not incorporate the quality and time delay of the communication link into a unified analysis framework. This way of modeling, which splits or simply equates the information layer with the physical layer, results in control strategies that may not perform well under practical electrical constraints or that may degrade when communications are limited. In addition, the traditional distributed algorithm often only depends on one-hop information of a direct neighbor to make a decision, lacks the perceptibility of a wider area network state, and is difficult to quickly coordinate the whole network resources to realize globally optimal dynamic response when coping with complex power fluctuation or local faults. Disclosure of Invention The present invention aims to solve at least one of the technical problems existing in the prior art; for this reason, the invention provides an AI-based micro-grid distributed cooperative control method, which comprises the following steps: acquiring an operation characteristic data set of all distributed power supply nodes in the micro-grid; Modeling the graph structure of the operation characteristic data set, modeling each distributed power supply node as a vertex in the graph structure, modeling the electrical connection and communication link between the nodes as edges, and endowing each edge with comprehensive weight reflecting the electrical distance and communication time delay to form a dynamic topological graph of the micro-grid; extracting local neighborhood characteristics of each vertex from the dynamic topological graph of the micro-grid, wherein the local neighborhood characteristics comprise vertex self operation characteristic data and operation characteristic data of neighbor vertices connected through edges to form vertex characteristic vectors; inputting vertex feature vectors of all vertexes into a pre-trained graphic neural network, wherein the graphic neural network aggregates information of each vertex and multi-hop neighbors thereof through multi-layer message transmission and updates hidden state vectors of all vertexes; calculating a cooperative control strategy for each vertex based on the updated hidden state vectors of all the vertices; the cooperative control strategy of each vertex is fed back to the corresponding distributed power supply local controller to be executed, and the output characteristic of the distributed power supply is changed; and after a preset time interval, the operation characteristic data of each node is collected again and combined with the execution result at the previous moment to generate a new operation characteristic data set for starting the next control cycle. Further, performing graph structure modeling on the operation characteristic data set, including: The operation characteristic data set comprises output power of a node, node voltage, local load and energy storage state; determining the connection relation of distributed power supply nodes according to the physical wiring diagram of the micro-grid and the communication network structure; Calculating impedance modulus values between any two nodes which are electrically connected as electrical distance weights; Measuring or estimating the round-trip time of a data packet between any two nodes with communication connection as a communication delay weight; The electrical distance weight and the communication time delay weight are subjected to linear superposition after normalization to obtain comprehensive weights of edges connecting two nodes; And constructing a