Neural network-based power grid service microservice reconstruction and cross-isolation collaboration method

Abstract

The present application relates to the technical field of power system informatization and computer software engineering, in particular to a power grid business microservice reconstruction and cross-isolation collaboration method based on neural network, a multi-channel program graph with security domain attributes and data frequency attributes is constructed, a graph neural network is constructed by using GraphSAGE or GCN, a heterogeneous security potential energy loss function is introduced, so that different security zone nodes form a high potential energy barrier in the embedding space, and K-Means clustering is performed based on the fused node embedding vector to generate a microservice candidate set conforming to the security partition, then a cross-zone shadow proxy mechanism is constructed through a frequency adaptive shunting strategy, one-way transaction collaboration across physical isolation devices is realized, finally, the scheme quality is quantified by an evaluation system composed of security boundary penetration entropy and real-time closed-loop time delay tolerance, and closed-loop optimization is formed; the problems of power grid microservice splitting ignoring physical isolation, high-frequency data transmission bottleneck and cross-zone transaction collaboration difficulty are solved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of power system information technology and computer software engineering, specifically a method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks. Background Technology

[0002] With the development of smart grids, power business systems are facing the need to transform from monolithic architectures to microservice architectures. However, the power grid business environment has significant unique characteristics, namely strict protection requirements of "security partitioning, dedicated networks, horizontal isolation, and vertical authentication". The core control area (security zone I / II) and the management information area (security zone III / IV) are separated by forward and reverse physical isolation devices (network gateways). This physical unidirectional transmission characteristic renders the general Internet microservice decomposition method ineffective.

[0003] Currently, existing technologies have the following drawbacks:

[0004] 1. Splitting algorithms ignore physical isolation constraints: Existing splitting methods based on graph neural networks (GNNs) or clustering typically divide based solely on code structure or call frequency. Aggregating classes belonging to the control area and classes belonging to the management area into the same service violates power grid safety regulations; even with later filtering using hard rules, service fragmentation is likely to occur, compromising the integrity of business logic.

[0005] 2. Lack of transaction support for unidirectional networks: General microservice architectures rely on bidirectional communication (such as RESTful APIs and RPCs) to implement distributed transactions (Saga / TCC). However, in a power grid environment, services in low-density zones (management zones) cannot directly call back to services in high-density zones (control zones), rendering standard transaction modes unavailable.

[0006] 3. High-frequency time-series data triggers network storms: The power grid contains massive amounts of millisecond-level time-series data (such as AGC commands and PMU phasor data). If the general microservice model is adopted to distribute all data through message queues (Kafka / MQ), it is very easy to cause serialization delay accumulation, which will endanger the real-time performance of the power grid closed-loop control.

[0007] 4. The evaluation indicators lack industry specificity: Existing cohesion and coupling indicators cannot measure the impact of the splitting scheme on the "number of times the security boundary is penetrated" or the "control loop delay".

[0008] Therefore, there is an urgent need for a method for automated decomposition, physical isolation adaptation, and quality assessment of microservices for individual power grid applications across security zones (control zones / management zones). Summary of the Invention

[0009] This invention aims to provide a neural network-based method for reconstructing microservices for power grid operations and coordinating across isolation devices. By improving the loss function of the graph neural network, physical isolation constraints are embedded in the feature learning stage. Through frequency adaptive splitting and shadow proxy mechanisms, the invention solves the problems of high-frequency data processing and transaction coordination across unidirectional isolation devices unique to the power grid.

[0010] This invention solves the above-mentioned technical problems through the following technical solution: a method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks, comprising the following steps:

[0011] S1. Construct a multi-channel program diagram with safety potential energy and frequency attributes;

[0012] S2. Graph representation learning based on heterogeneous security potential loss function: Graph neural network is used to embed nodes into vectors. Heterogeneous security potential loss function is designed. Through this loss function, the model forces nodes of different security partitions to move further apart in the vector space during backpropagation, forming a high potential barrier.

[0013] S3, Embedding Fusion and Microservice Clustering: The learned node embedding vectors H are fused and a clustering algorithm is executed to generate a set of microservice candidates that conform to the security partition.

[0014] S4. Data splitting and reconstruction based on adaptive sampling frequency: In the data layer reconstruction stage, data frequency attributes of nodes are considered. Implement a hot / cold message tiering strategy to replace the single message queue model;

[0015] S5, Cross-Physical Isolation Device “Shadow Proxy” Transaction Collaboration: For business transactions that cross forward and reverse physical isolation devices, a cross-regional shadow proxy mechanism is constructed.

[0016] S6. Quality assessment of power grid characteristics, establishing safety boundary penetration entropy. Real-time closed-loop delay tolerance An evaluation index system is used to quantitatively score the refactoring solution. If the solution meets the criteria, a microservice architecture and deployment configuration are generated; otherwise, the parameters are adjusted. Return to S2 and retrain.

[0017] Preferably, step S1 specifically includes: performing static and dynamic analysis on the target monolithic application and constructing a multi-channel program dependency graph. For each program node Extract and label security domain attributes and data frequency attributes Let E be an edge, specifically:

[0018] S11. Static structure analysis: Use static code analysis tools to scan the source code of a monolithic application;

[0019] S12. Industry Attribute Feature Extraction and Labeling: Unlike general software analysis, this involves extracting and labeling features for each node. Add specific feature vectors Including security domain attributes and data frequency attributes Two power grid attribute fields.

[0020] Preferably, in S12:

[0021] Security Domain Attributes To identify the physical security zone to which the node belongs, specifically Zone I - Real-time Control Zone, Zone II - Non-Control Production Zone, and Zone III - Management Information Zone;

[0022] For data frequency attribute If a node is involved in data processing, mark the average sampling or update frequency of its data.

[0023] Preferably, the graph neural network in S2 uses GraphSAGE or GCN as its basic skeleton, and the network input is the initial features of the nodes. After K layers of neighbor aggregation and nonlinear transformation, the embedded representation of the output node is obtained. .

[0024] Preferably, the total loss function of the heterogeneous security potential energy loss function in S2 is defined as: ,in For graph structure reconstruction loss, For the safety potential energy loss term, For hyperparameter weights;

[0025] The specific formula for the safety potential energy loss term is:

[0026] ;

[0027] in, For nodes The embedding vector; d is the preset safety boundary distance threshold; For the safety potential energy penalty coefficient: if The coefficient is 0; if If there is no predefined compliance gateway interface between the two, then the coefficient is set to the maximum value. .

[0028] Preferably, S3 specifically includes:

[0029] S31. Multi-channel embedding fusion: This involves fusing the node's embedding data obtained from the call relationship graph. Embedded with data dependency graph Weighted concatenation is performed to obtain the comprehensive embedding vector. ;

[0030] S32. Unsupervised clustering generates microservices: for The K-Means clustering algorithm will be executed, and the clustering results will automatically form the control area service set. Management Area Service Collection Based on this, the algorithm further optimizes the control area service set according to the density of its internal functions. Management Area Service Collection Subdivide into smaller parts.

[0031] Preferably, the data layer reconstruction stage in S4 includes:

[0032] Set frequency threshold ;

[0033] In high-frequency real-time streaming channels, for For microservices, deploy a Sidecar proxy to directly transfer data between services on the same physical machine or within a Pod using shared memory or zero-copy technology, bypassing the network protocol stack and ensuring millisecond-level closed loop.

[0034] In low-frequency service flow channels: For The microservices use standard message middleware for publish / subscribe decoupling.

[0035] Preferably, the cross-region shadow proxy mechanism in S5 includes:

[0036] Source write: After the microservice in the high-density area completes the local transaction, it writes the changed data to the local ferry buffer file / table;

[0037] Physical transfer: Using a hardware unidirectional transmission protocol of a forward isolation device to transfer data to a low-density area;

[0038] Shadow mirroring: Data falls into the shadow database on the low-density side, which is a read-only mirror of the data in the high-density area;

[0039] Event triggering: The low-density microservice listens for changes in the shadow database, triggering subsequent business logic to achieve eventual consistency coordination based on the physical unidirectional link.

[0040] Preferably, the security boundary penetration entropy in S6 The formula used to measure the dispersion of business processes across physical security partitions is as follows:

[0041] ;

[0042] in Entropy represents the probability that the i-th microservice involves cross-isolation device calls. The lower the entropy value, the more concentrated the cross-zone calls are in a few gateway services, and the higher the security.

[0043] Preferably, the real-time closed-loop delay tolerance in S6 It is used to identify strong real-time closed-loop paths in the system and calculate the additional network hops introduced by splitting the path. If the high-frequency closed-loop path is split to different physical nodes, a high penalty score is given to evaluate the impact of splitting on power grid stability.

[0044] The positive and progressive effects of this invention are as follows:

[0045] 1. By using a heterogeneous safety potential energy loss function, the AI ​​model can understand the physical partitioning of the power grid, and the generated partitioning scheme naturally complies with safety specifications, solving the problem of traditional algorithms requiring a lot of post-processing.

[0046] 2. The proposed shadow proxy mechanism achieves automated collaboration of cross-regional businesses without violating the red line of prohibiting access from low-density areas to high-density areas;

[0047] 3. The frequency adaptive load shedding strategy avoids the performance bottleneck of general microservice architecture when processing high-frequency power grid data, and ensures low latency of control closed loop. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the overall method flow provided by the present invention.

[0049] Figure 2 This is a diagram of the data splitting and cross-isolation device shadow proxy architecture based on adaptive sampling frequency provided by the present invention.

[0050] Figure 3 This invention provides a quality assessment system diagram that includes security boundary penetration entropy and closed-loop delay tolerance. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] This embodiment takes the reconstruction of a monolithic application of a provincial power grid dispatch automation system to a microservice architecture as an example. The monolithic application mainly includes functional modules such as data acquisition (SCADA), automatic generation control (AGC), state estimation, dispatcher power flow (DPF), report generation and user management, involving Security Zone I (real-time control zone) and Security Zone III (management information zone).

[0053] See Figures 1 to 3 A neural network-based method for refactoring and cross-isolation collaboration of power grid business microservices includes the following steps:

[0054] S1. Construct a multi-channel program graph with "safety potential energy" and "spatiotemporal frequency" attributes.

[0055] This step aims to transform unstructured monolithic application code into graph-structured data that incorporates characteristics of the power grid industry, including:

[0056] 1. Static Structure Analysis

[0057] Static code analysis tools (such as abstract syntax tree (AST) parsers) are used to scan the source code of monolithic applications and establish call relationship channels. Define each class or function as a graph node V. If class A calls a method of class B, then a directed edge is established between nodes A and B. Simultaneously establish data dependency channels Scan the SQL statements and ORM mapping files in the code. If class A reads and writes data table T, and class B also reads and writes data table T, then establish an undirected edge between A and B. .

[0058] 2. Industry Attribute Feature Extraction and Labeling

[0059] Unlike general software analysis, this embodiment analyzes each node. Add specific feature vectors This includes two key power grid attribute fields:

[0060] Security Domain Attributes : Reads the system's deployment configuration file or package namespace. For example, it includes the package path. com.grid.control.* The node is marked as (Security Zone I) will contain com.grid.manage.* The node is marked as (Security Zone III).

[0061] Data frequency attribute Analyze the system's operation logs or measurement point configuration tables. For nodes processing SCADA telemetry data, calculate their average data arrival interval. If the interval is 2 seconds, mark it. =0.5Hz; For AGC instruction processing nodes, marked =0.25Hz (4 seconds); For daily report nodes, mark In addition, each node v is labeled with an extra "business cycle or event triggering attribute C(v)" to identify whether the node logic is driven by fixed-cycle scheduling (such as state estimation every 5 seconds), random events (such as switch changes), or manual requests. This attribute will be used as a weighting factor in the S3 clustering stage to ensure that nodes that work closely together within the same business cycle tend to belong to the same microservice when partitioning, thereby reducing cross-service coordination overhead.

[0062] S2. Graphical representation learning based on heterogeneous security potential loss function

[0063] This step uses a graph neural network (GNN) to map discrete program nodes into continuous low-dimensional vectors, and forces physical isolation constraints to be injected during the mapping process.

[0064] 1. Construct a graph neural network model

[0065] GraphSAGE or GCN is selected as the basic skeleton, and the network input is the initial features of the nodes. After K layers (K=2 in this embodiment) of neighbor aggregation and nonlinear transformation, the embedded representation of the output node is obtained. .

[0066] 2. Define the heterogeneous security potential energy loss function.

[0067] To address the problem of traditional clustering algorithms misclustering nodes with similar functions but belonging to different security zones, this embodiment introduces the concept of "security potential" during the model training phase, and the total loss function... Defined as:

[0068] ;

[0069] in, This is the unsupervised graph reconstruction loss, used to preserve the similarity of the code's call structure; The hyperparameter weight is set to 0.8 in this embodiment; The specific calculation formula for the safety potential energy term is as follows:

[0070] ;

[0071] in, It is the preset Euclidean distance threshold for the safety boundary; It is a potential energy penalty matrix: when hour, ;when hour, (Maximum value).

[0072] 3. Model Training

[0073] The model is trained using the stochastic gradient descent (SGD) algorithm. During forward propagation, the current embedding vector distribution is calculated; during back propagation, if two nodes with a calling relationship are found... Belonging to different security zones (i.e., cross-zone calls), due to great, This will generate a huge gradient signal, forcing the weight matrix W to update, making... and After rapidly moving away from each other in the vector space and training convergence, the resulting embedding vector H naturally forms clusters separated by safe zones.

[0074] S3, Embedded Fusion and Microservice Clustering

[0075] 1. Multi-channel embedding fusion: This involves fusing the nodes' embeddings obtained from the relationship graph. Embedded with data dependency graph Weighted concatenation is performed to obtain the comprehensive embedding vector. .

[0076] 2. Unsupervised clustering for microservice generation: [This is related to...] When performing the K-Means clustering algorithm, due to the potential energy constraint in S2, the nodes in different safe zones already have significant distances in space, and the clustering results will automatically form the control zone service set. Management Area Service Collection ;

[0077] Based on this, the algorithm further optimizes the internal functionality according to its tightness. It is further subdivided into "SCADA Front-end Services", "Status Estimation Services", and "AGC Services"; Subdivided into "reporting services", "user authentication services", etc., the resulting microservice partitioning scheme naturally meets the "partitioning and hierarchical" security requirements of the power grid. After clustering to generate a preliminary set of microservices, an aggregation root identification and interface specification generation step based on Domain-Driven Design (DDD) is added. Using the data dependency edges E(d) and call edges E(c) between nodes, combined with the C(v) attribute marked in step S1, the "aggregate root" node (i.e. the node that is frequently depended on by other nodes or serves as the main external interface) in each microservice candidate set is automatically identified. Using this aggregate root as the core, the system automatically generates a preliminary domain model and RESTful or gRPC interface specification for the microservice, significantly improving the internal cohesion of the split service and the clarity of its external interfaces. Subsequently, it performs "DDD-based aggregate root identification and interface specification generation." For example, in the candidate set of "SCADA front-end services," by analyzing the dependencies between internal nodes, the "telemetry data receiver manager" class is identified as the aggregate root, which is depended on by multiple data parsing and verification classes. The system automatically uses this class as the core to generate a draft domain model of the microservice and, based on the main public methods of this class, generates preliminary RESTful interface specification documents such as POST / api / v1 / scadata / telemetry, providing clear guidance for developers.

[0078] S4. Data splitting and reconstruction based on adaptive sampling frequency

[0079] For data communication in the split microservices, this embodiment abandons the general full message queue scheme and adopts a frequency-based approach. The adaptive traffic splitting strategy specifically includes:

[0080] 1. Set frequency threshold

[0081] The system presets high-frequency thresholds. That is, 100ms.

[0082] 2. Regarding Construct a high-frequency closed-loop channel

[0083] For the identified "SCADA front-end service" and "AGC service", the two need to perform data interaction at the second or even millisecond level, specifically including:

[0084] Deployment strategy: Schedule these two microservices to the same physical node or the same Pod in Kubernetes;

[0085] Communication implementation: Automatic injection of the Sidecar proxy, configured with shared memory or UnixDomain Socket channels;

[0086] Results: The AGC service reads SCADA measurement point data directly from shared memory, achieving zero-copy, eliminating network serialization overhead, and ensuring closed-loop control latency of <5ms.

[0087] 3. Regarding Build low-frequency business channels

[0088] For the "Report Service" to obtain historical data, configure a topic subscription for the Kafka message queue. The SCADA service will send the sampled data to Kafka after reducing the frequency or batch packaging, and the Report Service will consume it asynchronously.

[0089] S5, "Shadow Proxy" Transaction Coordination Across Physically Isolated Devices

[0090] For the forward and reverse physical isolation devices (network gateways) between Zone I (control zone) and Zone III (management zone), this embodiment constructs a "shadow proxy" mechanism to handle cross-zone transactions, specifically including:

[0091] 1. Transaction scenario identification

[0092] The "AGC Action Reversal" service was identified. This service requires that after the AGC service in Zone I performs an action, the action record can be seen on the Web display terminal in Zone III.

[0093] 2. Deploy the shadow proxy component

[0094] Source Proxy: Deployed in Zone I, it acts as the local data receiver for the AGC service;

[0095] Ferry Channel: Configures a hardware penetration strategy for forward isolation devices, allowing only the source agent to send binary files of a specific format to the destination agent in one direction;

[0096] Sink Proxy: Deployed in Zone III, responsible for receiving and parsing files into the database.

[0097] 3. Execution Process

[0098] (1) The I zone AGC service executes the adjustment command, commits the transaction, and writes the log to the I zone local database.

[0099] (2) The source agent captures changes (CDC mode), generates encrypted data packets, and "ferrys" them to Zone III through the forward isolation device;

[0100] (3) The III zone destination agent receives data packets and writes them into the III zone's independent shadow database (Shadow DB).

[0101] (4) The III zone reporting service listens to the Binlog of the shadow database. Once a new record is detected, it is pushed to the Web front end for display.

[0102] The entire process strictly adheres to the power grid security standards of "source-initiated, one-way transmission, and physical isolation," and no TCP bidirectional connection is established.

[0103] S6, Microservice Decomposition Quality Assessment

[0104] The quality assessment module is used to quantitatively score the above reconstruction schemes, specifically including:

[0105] 1. Calculate the penetration entropy of the safety boundary ( )

[0106] By analyzing the call chains between all microservices, if it is found that the vast majority of cross-region calls converge to the "shadow proxy" node described in step five, rather than being scattered across various business code segments, then the calculated entropy value is extremely low (approaching 0). In this embodiment... This indicates that the safety boundary is clear and the penetration point is controllable.

[0107] 2. Calculate the real-time closed-loop delay tolerance (T) loop )

[0108] T loop =T max -T actual T max T is the upper limit of tolerance. actual Based on the actual measured values, the key closed-loop path SCADA -> AGC -> RTU was extracted, and the service deployment topology on this path was checked. Since S4 uses shared memory communication, the additional network hop count (Hop Count) is 0. The evaluation module gives this indicator a full score, and it is determined that the splitting scheme does not compromise the real-time performance of the power grid.

[0109] 3. Overall Score

[0110] Combining conventional metrics such as cohesion, decoupling, and data consistency, a weighted formula is used. Calculate the total score, W j Let I be the weight coefficient of the j-th evaluation indicator. j For the score of the j-th evaluation metric, if the total score exceeds the threshold (e.g., 85 points), a refactored code framework and deployment configuration file are automatically generated; otherwise, the security potential weight is adjusted. Return to S2 and retrain.

[0111] Through the above specific implementation methods, the present invention realizes the automatic decomposition of a large monolithic scheduling system into a loosely coupled, highly available microservice architecture while meeting the strict physical isolation and real-time constraints of the power grid.

[0112] This invention is not limited to the embodiments described above. Any changes in shape or structure shall fall within the protection scope of this invention. The protection scope of this invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of this invention, but all such changes and modifications shall fall within the protection scope of this invention.

Claims

1. A method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks, characterized in that, Includes the following steps: S1. Construct a multi-channel program diagram with safety potential energy and frequency attributes; S2. Graph representation learning based on heterogeneous security potential loss function: Graph neural network is used to embed nodes into vectors. Heterogeneous security potential loss function is designed. Through this loss function, the model forces nodes of different security partitions to move far apart in the vector space during backpropagation, forming a high potential barrier. S3, Embedding Fusion and Microservice Clustering: The learned node embedding vectors H are fused and a clustering algorithm is executed to generate a set of microservice candidates that conform to the security partition. S4. Data splitting and reconstruction based on adaptive sampling frequency: In the data layer reconstruction stage, data frequency attributes of nodes are considered. Implement a hot / cold message tiering strategy to replace the single message queue model; S5, Cross-Physical Isolation Device "Shadow Proxy" Transaction Collaboration: For business transactions that cross forward and reverse physical isolation devices, a cross-regional shadow proxy mechanism is constructed; S6. Quality assessment of power grid characteristics, establishing safety boundary penetration entropy. Real-time closed-loop delay tolerance An evaluation index system is used to quantitatively score the refactoring solution. If the standard is met, a microservice architecture and deployment configuration are generated; if the standard is not met, the parameters are adjusted. Return to S2 and retrain; S1 specifically includes: performing static and dynamic analysis on the target monolithic application and constructing a multi-channel program dependency graph. For each program node Extract and label security domain attributes and data frequency attributes Let E be an edge, specifically: S11. Static structure analysis: Use static code analysis tools to scan the source code of a monolithic application; S12. Industry Attribute Feature Extraction and Labeling: Unlike general software analysis, this involves extracting and labeling features for each node. Add specific feature vectors Including security domain attributes and data frequency attributes Two power grid attribute fields.

2. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that, In S12: Security Domain Attributes To identify the physical security zone to which the node belongs, specifically Zone I - Real-time Control Zone, Zone II - Non-Control Production Zone, and Zone III - Management Information Zone; For data frequency attribute If a node is involved in data processing, mark the average sampling or update frequency of its data.

3. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that: The graph neural network in S2 uses GraphSAGE or GCN as its basic skeleton, and the network input is the initial features of the nodes. After K layers of neighbor aggregation and nonlinear transformation, the embedded representation of the output node is obtained. .

4. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that, The total loss function of the heterogeneous safety potential energy loss function in S2 is defined as follows: ,in For graph structure reconstruction loss, For the safety potential energy loss term, For hyperparameter weights; The specific formula for the safety potential energy loss term is: ； in, For nodes The embedding vector; d is the preset safety boundary distance threshold; For the safety potential energy penalty coefficient: if The coefficient is 0; if If there is no predefined compliance gateway interface between the two, then the coefficient is set to the maximum value. .

5. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that, S3 specifically includes: S31. Multi-channel embedding fusion: This involves fusing the node's embedding data obtained from the call relationship graph. Embedded with data dependency graph Weighted concatenation is performed to obtain the comprehensive embedding vector. ; S32. Unsupervised clustering generates microservices: for The K-Means clustering algorithm will be executed, and the clustering results will automatically form the control area service set. Management Area Service Collection Based on this, the algorithm further optimizes the control area service set according to the density of its internal functions. Management Area Service Collection Subdivide into smaller parts.

6. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that: The data layer reconstruction phase in S4 includes: Set frequency threshold ; In high-frequency real-time streaming channels, for For microservices, deploy a Sidecar proxy to directly transfer data between services on the same physical machine or within a Pod using shared memory or zero-copy technology, bypassing the network protocol stack and ensuring millisecond-level closed loop. In low-frequency service flow channels: For The microservices use standard message middleware for publish / subscribe decoupling.

7. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that: The cross-region shadow proxy mechanism in S5 includes: Source write: After the microservice in the high-density area completes the local transaction, it writes the changed data to the local "ferry buffer file / table"; Physical transfer: Using a hardware unidirectional transmission protocol of a forward isolation device, data is "transferred" to a low-density area; Shadow mirroring: Data falls into the shadow database on the low-density side, which is a read-only mirror of the data in the high-density area; Event triggering: The low-density microservice listens for changes in the shadow database, triggering subsequent business logic to achieve eventual consistency coordination based on the physical unidirectional link.

8. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that: Safety boundary penetration entropy in S6 The formula used to measure the dispersion of business processes across physical security partitions is as follows: ； in Entropy represents the probability that the i-th microservice involves cross-isolation device calls. The lower the entropy value, the more concentrated the cross-zone calls are in a few gateway services, and the higher the security.

9. The method for reconstructing and cross-isolation collaboration of power grid business microservices based on neural networks as described in claim 1, characterized in that: Real-time closed-loop delay tolerance in S6 It is used to identify strong real-time closed-loop paths in the system and calculate the additional network hops introduced by splitting the path. If the high-frequency closed-loop path is split to different physical nodes, a high penalty score is given to evaluate the impact of splitting on power grid stability.

Images