A distributed DTU unit management and scheduling system

Abstract

The present invention relates to the field of power system automation technology, and in particular to a management and dispatching system for distributed DTU units, comprising a distributed DTU unit cluster, a communication network layer, and a cloud management platform. The distributed DTU unit cluster is hierarchically connected by common units and interval units, the latter of which integrates a point table generation and self-diagnosis module; the communication network layer supports 4G / 5G, PLC power carrier, and Ethernet multimode transmission switching; and the cloud management platform includes modules such as intelligent point table management and dynamic resource scheduling. The system implements program upgrades via the TFTP protocol, utilizes a topology-aware graph convolution-enhanced network to dynamically allocate computing resources, employs a lightweight dynamic consensus algorithm to store point table files on the blockchain, and the intelligent point table management module implements version compatibility checking and conflict resolution. The present invention can effectively improve DTU unit management efficiency, resource utilization, and data security, and is suitable for scenarios such as the Industrial Internet of Things and smart grids.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and in particular to a management and dispatching system for distributed DTU units. Background Art

[0002] Against the backdrop of the rapid development of the Industrial Internet of Things (IIoT) and smart grids, distributed terminal units (DTUs) serve as key hubs connecting the perception layer and the cloud. The effectiveness of their management and scheduling directly determines the performance of system operations. Traditional DTU management models face numerous challenges, becoming a technical bottleneck restricting the development of the industry.

[0003] From an architectural perspective, early centralized management systems relied on a single server to process massive amounts of DTU data, which posed a high risk of single-point failure and poor scalability. As the number of devices grew exponentially, the computing and storage pressures of the centralized architecture increased sharply, making it difficult to meet real-time requirements. Simple distributed architectures lacked effective coordination mechanisms, with each DTU unit operating independently, resulting in low data exchange efficiency and an inability to achieve global optimal resource allocation.

[0004] In terms of communication transmission, existing systems mostly use a single communication method. For example, some industrial scenarios rely solely on Ethernet and cannot be deployed in areas where network cabling is difficult. Systems that rely solely on wireless communication have problems such as unstable signals and high data transmission delays. In addition, the lack of unified conversion standards between different communication protocols leads to poor data transmission compatibility in heterogeneous network environments, seriously affecting system reliability.

[0005] The areas of data management and security also present prominent problems. Traditional point table management relies on manual configuration and maintenance, and data conflicts and version incompatibilities are very likely to occur when equipment is upgraded or newly added. At the same time, data storage mostly uses centralized databases, which face the risk of data tampering and leakage, and cannot meet the strict requirements of the industrial field for data security and integrity.

[0006] At the resource scheduling level, traditional algorithms are mostly based on static rules and cannot adapt to the complex and changing load demands in industrial environments. For example, fixed sampling frequency strategies lead to untimely data processing under high load and waste of resources under low load. The lack of dynamic allocation capabilities for DTU computing resources often leads to an imbalance in which some devices are overloaded while others are idle.

[0007] Therefore, to address the above problems, a distributed DTU unit management and scheduling system is proposed. Summary of the Invention

[0008] The object of the present invention is to provide a management and scheduling system for distributed DTU units to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A distributed DTU unit management and scheduling system, including a distributed DTU unit cluster, a communication network layer, and a cloud management platform:

[0011] The distributed DTU unit cluster is composed of a public unit and multiple interval units connected hierarchically through an industrial switch. Each interval unit is equipped with a point table generation module and a self-diagnosis module.

[0012] The communication network layer integrates 4G / 5G wireless modules, PLC power carrier modules and Ethernet interfaces, supporting dynamic switching of multi-mode transmission paths;

[0013] The cloud management platform includes a point-to-table intelligent management module, a dynamic resource scheduling engine, and a security protection module;

[0014] The system performs the following process:

[0015] The cloud management platform sends program upgrade instructions to the public unit, which are then forwarded to each bay unit via the TFTP protocol.

[0016] After the bay unit is upgraded, an XML point table file containing the address and range of telesignaling, telemetering and remote control data is generated and transmitted back to the public unit;

[0017] The public unit performs conflict detection and merging of multi-point table files, and adds IP identifiers to generate a total point table template;

[0018] Dynamically allocate DTU computing resources based on topology-aware graph convolution-enhanced networks;

[0019] The hash value of the point table file is stored on the blockchain through a lightweight dynamic consensus algorithm.

[0020] As a preferred solution, the point table intelligent management module performs:

[0021] Version compatibility check: triggers an alarm when the software version difference between the point table file before and after the upgrade exceeds 0.5;

[0022] Point table conflict handling: XML parser is used to standardize and restructure the nested tag conflicts. The reorganization method is to select the point table structure with the smallest Euclidean distance to the standard template structure feature vector from the candidate structure set;

[0023] Blockchain evidence storage: The SHA-256 hash value of the total point table template is stored in the distributed ledger.

[0024] As a preferred solution, the resource dynamic scheduling engine includes:

[0025] Topology-aware graph convolutional enhanced network model:

[0026] Calculate the normalized topological relationship matrix by adding the adjacency matrix and degree matrix of self-loops;

[0027] Perform three-layer graph convolution operations on the node feature matrix containing load rate and task queue depth;

[0028] Perform global average pooling on the topological embedding vector output by the graph structure encoder;

[0029] The graph convolution result and the pooling result are added with a weight coefficient of 0.6, and the action value is output through the LeakyReLU activation function;

[0030] Real-time scheduler: Assign telemetry and telecontrol task priorities based on action value output results.

[0031] As a preferred solution, the communication network layer implements:

[0032] Multimode transmission channel selection:

[0033] When the data urgency level is greater than 0.8 and the signal strength is not less than -85dBm, select the 4G / 5G channel;

[0034] Select the PLC channel when the data urgency level is between 0.4 and 0.8;

[0035] In other cases, select Ethernet Channel;

[0036] Encrypted transmission: AES-256 algorithm is used to encrypt point table files and scheduling instructions.

[0037] As a preferred solution, the security protection module performs:

[0038] Lightweight dynamic consensus algorithm:

[0039] When calculating the node consensus weight, the historical credibility score is multiplied by a coefficient of 0.7;

[0040] Multiply the negative exponential function value of the node response delay mean by a coefficient of 0.3;

[0041] The sum of the above two items is multiplied by the ratio of the node's stake amount to the system's maximum stake amount;

[0042] The final weight is normalized by the sum of all node calculation values;

[0043] Tampering detection: An alarm is triggered when the Hamming distance between the new block hash value and the stored value is greater than zero.

[0044] As a preferred solution, it also includes a flexible resource allocation module:

[0045] Load forecasting model:

[0046] In the LSTM input gate, the historical load value is subjected to wavelet decomposition feature extraction;

[0047] The load change rate is used to extract residual features in the memory unit calculation;

[0048] Update the memory unit state based on the hidden state at the previous moment;

[0049] Dynamic resource adjustment:

[0050] When the predicted load value is higher than the preset upper limit, the sampling frequency is set to 1kHz;

[0051] When the predicted load value is lower than the preset lower limit, the sampling frequency is set to 100Hz.

[0052] As a preferred solution, the topology-aware graph convolution enhanced network training process includes:

[0053] The state space consists of load rate, network delay and task backlog;

[0054] The reward function is calculated as follows: 0.5 times the difference between the baseline scheduling delay and the actual delay, plus 0.3 times the energy saving rate, minus 0.2 times the overload indicator;

[0055] The training is terminated when the reward function variance is less than 0.01.

[0056] As a preferred solution, the self-diagnosis module performs:

[0057] Fault location: Calculate the ratio of the peak-to-peak current to the average current as the ripple factor. If the ripple factor exceeds 0.25, the capacitor is considered abnormal.

[0058] Collaborative control: After the faulty unit is isolated, the public unit takes over data collection, and the task transfer delay is controlled within 10 milliseconds.

[0059] As a preferred solution, the cloud management platform integrates a 3D visualization dashboard to dynamically display:

[0060] DTU unit topology connection relationship matrix;

[0061] Resource load heat map: green when the load rate is below 0.4, yellow when it is between 0.4 and 0.7, and red when it is 0.7 and above.

[0062] As a preferred solution, point-table merging works in conjunction with blockchain evidence storage:

[0063] Calculate the SHA-256 hash value of the total point table template immediately after it is generated;

[0064] Trigger the lightweight dynamic consensus algorithm to write the hash value and timestamp into the blockchain;

[0065] When the point table version is updated, it is necessary to verify that the new hash value is exactly the same as the stored value before activation.

[0066] It can be seen from the technical solutions provided by the present invention that the management and scheduling system of a distributed DTU unit provided by the present invention has the following beneficial effects:

[0067] Efficient resource management and scheduling: Based on the Topology-Aware Graph Convolutional Enhanced Network (TA-GCN-DQN) and the Lightweight Dynamic Consensus Algorithm (LDCA), dynamic allocation of DTU computing resources and efficient management of node table files are achieved. Through the TA-GCN-DQN model's in-depth analysis of DTU connection topology and operating status, telemetry and remote control task priorities can be accurately allocated, improving resource utilization and reducing task scheduling delays. At the same time, the LDCA algorithm ensures the secure storage and efficient processing of node table files, safeguarding data integrity and consistency.

[0068] Powerful communication and security performance: The communication network layer integrates 4G / 5G, PLC power carrier, and Ethernet interfaces, supports dynamic switching of multi-mode transmission paths, and intelligently selects the optimal transmission method based on signal strength and data urgency level to ensure stable and real-time data transmission. It uses AES-256 encryption technology to encrypt data transmission, and combines blockchain technology to store point table files, effectively preventing data leakage and tampering, and improving system security and credibility.

[0069] Intelligent fault diagnosis and self-healing: The self-diagnosis module configured in each bay uses current ripple characteristics and other technologies to quickly locate and isolate faults. When a fault is detected, the public unit can quickly take over the data collection task of the abnormal bay, with a task transfer latency of less than 10ms, ensuring the continuous and stable operation of the system. At the same time, the elastic resource allocation module uses multi-scale spatiotemporal feature fusion LSTM (MSTF-LSTM) to perform load forecasting, dynamically adjusting the sampling frequency based on the forecast results to optimize resource utilization efficiency.

[0070] Visualization and convenient operation and maintenance: The cloud management platform integrates a 3D visualization dashboard that dynamically displays the DTU unit topology connection relationship and real-time resource load heat map, allowing operation and maintenance personnel to intuitively understand the system operation status and quickly identify potential problems. The version compatibility check and conflict resolution functions of the point meter intelligent management module simplify the point meter management process, reduce operation and maintenance complexity, and improve system maintainability. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a schematic diagram of the overall structure of a distributed DTU unit management and scheduling system of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0073] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0074] like Figure 1 As shown, an embodiment of the present invention provides a management and scheduling system for a distributed DTU unit, including a distributed DTU unit cluster, a communication network layer, and a cloud management platform:

[0075] The distributed DTU unit cluster is composed of a public unit and multiple interval units connected hierarchically through an industrial switch. Each interval unit is equipped with a point table generation module and a self-diagnosis module.

[0076] The communication network layer integrates 4G / 5G wireless modules, PLC power carrier modules and Ethernet interfaces, supporting dynamic switching of multi-mode transmission paths;

[0077] The cloud management platform includes a point-to-table intelligent management module, a dynamic resource scheduling engine, and a security protection module;

[0078] The system performs the following process:

[0079] The cloud management platform sends program upgrade instructions to the public unit, which are then forwarded to each bay unit via the TFTP protocol.

[0080] After the bay unit is upgraded, an XML point table file containing the address and range of telesignaling, telemetering and remote control data is generated and transmitted back to the public unit;

[0081] The public unit performs conflict detection and merging of multi-point table files, and adds IP identifiers to generate a total point table template;

[0082] Dynamically allocate DTU computing resources based on topology-aware graph convolution-enhanced networks;

[0083] The hash value of the point table file is stored on the blockchain through a lightweight dynamic consensus algorithm.

[0084] In this embodiment, the self-diagnosis module performs:

[0085] Fault location: Calculate the ratio of the peak-to-peak current to the average current as the ripple factor. If the ripple factor exceeds 0.25, the capacitor is considered abnormal.

[0086] Collaborative control: After the faulty unit is isolated, the public unit takes over data collection, and the task transfer delay is controlled within 10 milliseconds;

[0087] Furthermore, the distributed DTU unit cluster, as the basic execution layer of the decentralized DTU unit management and scheduling system, is like the "nerve endings" of the Industrial Internet of Things, undertaking the core tasks of data collection, preliminary processing, and local control. Through a hierarchical interconnection architecture and intelligent module design, it achieves efficient coordination and autonomous management of equipment. The following is a detailed analysis from the architecture to the function:

[0088] 1. Overall architecture design:

[0089] The distributed DTU unit cluster adopts a tree-like hierarchical topology, consisting of a public unit and multiple bay units cascaded via industrial switches. This architecture offers high scalability and redundant fault tolerance, flexibly adapting to diverse scenarios such as power, energy, and industrial automation. The public unit serves as the cluster's core hub, responsible for global communication coordination, data aggregation, and command forwarding. Bay units are deployed at each subsystem node, responsible for data collection and local control. The two work together to establish an efficient "centralized management - decentralized execution" operational system.

[0090] 2. Core components and functions:

[0091] (1) Analysis of public unit functions:

[0092] Communication hub management: Serving as a communication bridge between the cloud management platform and bay units, it supports the TFTP protocol to quickly distribute remote program upgrade instructions and receives XML point table files sent back by bay units to ensure cluster software version consistency and data synchronization.

[0093] Data aggregation and processing: Conflict detection and merging operations are performed on the multi-point table files reported by each interval unit. A total point table template is generated through IP identification and standardized reorganization, providing a unified data interface for cloud resource scheduling.

[0094] Collaborative control and scheduling: When a fault occurs, the public unit automatically takes over the data collection task of the abnormal interval unit, ensuring that the task transfer delay is less than 10ms and maintaining system monitoring continuity;

[0095] (2) Functional characteristics of spacer unit:

[0096] Point table generation module: Built-in intelligent algorithm dynamically generates XML point table files, accurately mapping the address and range parameters of telesignaling, telemetering, and telecontrol data, and providing standardized descriptions for data interaction;

[0097] Self-diagnosis module:

[0098] Real-time fault detection: through current ripple characteristic formula (in, is the current ripple characteristic value, which is used to characterize the current fluctuation characteristics; is the peak-to-peak value of the current, that is, the difference between the maximum and minimum values ​​of the current in one cycle; (current average value, reflecting the average level of current) to monitor the status of key components such as capacitors in real time. When an abnormal alarm is triggered;

[0099] Fault self-healing mechanism: After detecting a fault, the system automatically executes the isolation procedure and sends status information to the public unit to cooperate with task migration, realizing closed-loop management of the entire process of "detection-isolation-reporting";

[0100] Edge computing capabilities: Integrates lightweight data preprocessing algorithms to filter, reduce noise, and extract features from raw collected data, reducing invalid data transmission and improving communication network efficiency.

[0101] 3. Implementation principles of key technologies:

[0102] (1) Principles of hierarchical interconnection architecture:

[0103] Industrial switches use redundant link design and spanning tree protocol to avoid network loops, ensuring data transmission reliability and low latency. Industrial standard protocols such as Modbus / TCP and IEC61850 are used between public and bay units to achieve interoperability between devices from different manufacturers.

[0104] (2) Dynamic management mechanism of point tables:

[0105] The interval unit updates the point table data in real time based on changes in the equipment's operating status. The public unit uses XML parsing technology to align the structure of heterogeneous point tables and resolve conflicts, ultimately generating a globally unified master point table template. Blockchain evidence is used to ensure data integrity and traceability.

[0106] (3) Collaborative self-diagnosis technology:

[0107] The self-diagnosis module combines multi-dimensional detection methods such as current ripple analysis and device operating parameter threshold judgment to build a fault feature library. When a single indicator is abnormal, it links other monitoring data for cross-validation to reduce the false alarm rate. After a fault occurs, rapid isolation and task migration strategies are implemented to ensure continuous operation of the system in a "sub-healthy" state.

[0108] 4. Detailed explanation of the work process:

[0109] (1) Initialization phase:

[0110] The bay unit completes hardware self-tests, including functional tests of the sensor interface, communication module, and power module to ensure equipment readiness;

[0111] The public unit establishes a secure connection with the cloud management platform, synchronizes the latest configuration parameters and upgrade packages, and provides initial data support for cluster operation;

[0112] (2) Data collection and transmission stage:

[0113] The interval unit collects equipment data at a preset sampling frequency. The sampling frequency is dynamically adjusted according to the actual load situation. The data is packaged by the point table generation module and uploaded to the public unit.

[0114] The public unit performs preliminary verification and format conversion on the data, and forwards it to the cloud management platform through the communication network layer (4G / 5G, PLC or Ethernet);

[0115] (III) Collaborative management stage:

[0116] Program upgrade process: The cloud management platform initiates the upgrade command, and the public unit pushes it to the interval unit in batches through the TFTP protocol. After the upgrade is completed, the point table file is collected for version compatibility check;

[0117] Fault response mechanism: When the self-diagnosis module detects a fault, the bay unit immediately isolates the fault point and reports the status; the public unit initiates emergency dispatch and coordinates redundant resources to maintain data collection tasks;

[0118] (IV) Data feedback and optimization stage:

[0119] The public unit feeds back the processed data and status information to the cloud management platform for the resource dynamic scheduling engine to optimize computing resource allocation;

[0120] The cloud management platform adjusts the point table template and scheduling strategy based on global operation data, and synchronizes it to the interval unit through the public unit to achieve continuous optimization of the system;

[0121] 5. Application value and advantages:

[0122] (1) Improving system scalability:

[0123] The hierarchical architecture supports "plug-and-play" node expansion. When a new bay unit is connected, it only needs to configure the IP address and node table information to quickly integrate into the cluster, reducing the cost of large-scale deployment.

[0124] (2) Enhance operational reliability:

[0125] Self-diagnosis and redundant switching mechanisms ensure 99.99% system availability, shortening the mean time to repair (MTTR) to minutes, effectively ensuring industrial production continuity.

[0126] (3) Reducing communication costs:

[0127] Edge-side data preprocessing and dynamic sampling strategies reduce invalid data transmission by more than 60%. Combined with intelligent switching of multi-mode communication paths, this significantly reduces network traffic costs and bandwidth pressure.

[0128] 4. Strengthening data quality:

[0129] Point table standardization and blockchain evidence storage technology ensure the integrity and authenticity of data throughout its entire life cycle, providing a trusted data foundation for industrial Internet of Things data analysis and predictive maintenance.

[0130] In this embodiment, the communication network layer implements:

[0131] Multimode transmission channel selection:

[0132] When the data urgency level is greater than 0.8 and the signal strength is not less than -85dBm, select the 4G / 5G channel;

[0133] Select the PLC channel when the data urgency level is between 0.4 and 0.8;

[0134] In other cases, select Ethernet Channel;

[0135] Encrypted transmission: AES-256 algorithm is used to encrypt point table files and scheduling instructions;

[0136] Furthermore, the communication network layer serves as the "information highway" of the distributed DTU unit management and scheduling system, and is the key link connecting the distributed DTU unit cluster and the cloud management platform. By integrating multi-mode communication technology, intelligent path selection algorithms, and data security protection mechanisms, it achieves efficient and stable data transmission, ensuring the reliable operation of the system in complex industrial environments. The following will elaborate on the functional architecture, core technologies, and workflow aspects:

[0137] 1. Overall functional architecture:

[0138] The communication network layer adopts a modular design, integrating three communication units: 4G / 5G wireless module, PLC power carrier module, and Ethernet interface, and supports dynamic switching of multi-mode transmission paths. Its core functions include:

[0139] Multi-mode communication adaptation: flexibly select 4G / 5G, PLC power line carrier or Ethernet as the transmission channel according to different application scenarios and data transmission requirements;

[0140] Dynamic path switching: Based on parameters such as signal strength and data urgency level, the transmission path is adjusted in real time to ensure the timeliness and stability of data transmission;

[0141] Data encryption transmission: AES-256 encryption algorithm is used to encrypt the transmitted data to prevent data leakage and tampering, and ensure communication security;

[0142] Network status monitoring: Real-time monitoring of the operating status of each communication module, including signal strength, link connection quality, etc., to provide a decision basis for path switching;

[0143] 2. Detailed explanation of core functional modules:

[0144] (1) Multimode communication unit:

[0145] 4G / 5G wireless module: supports high-speed wireless communication and is suitable for scenarios with high real-time requirements and difficult wiring. When the data urgency level is high and the signal strength meets the requirements, it is given priority as a transmission channel to ensure the rapid transmission of critical data.

[0146] PLC power line carrier module: Utilizes power lines as the transmission medium, eliminating the need for additional wiring and facilitating easy deployment. Suitable for scenarios with medium-level data urgency, it effectively reduces communication costs and is particularly well-suited for data transmission in industrial power environments.

[0147] Ethernet interface: Provides stable, high-speed wired communication, suitable for conventional data transmission and short-distance communication scenarios; serves as the default transmission channel when the data urgency level is low and the wired network conditions are good;

[0148] (2) Dynamic path switching module:

[0149] Transmission path selection algorithm: based on data urgency level and signal strength , dynamically switch the transmission path, the selection logic is: (in, is the selected transmission channel; The data urgency level is dimensionless and ranges from [0,1]. The larger the value, the more urgent the data. is the signal strength, measured in decibel milliwatts, used to measure the strength of wireless signals or power carrier signals). This algorithm quantifies the urgency of signals and data to achieve intelligent selection of transmission paths. For example, when a remote control command ( ) and the 4G / 5G signal is good ( ), give priority to 4G / 5G channels to ensure that instructions are issued quickly; for conventional telemetry data ( ), then use PLC power carrier channel to reduce costs;

[0150] Switching execution mechanism: When the path switching conditions are met, the communication network layer automatically switches the transmission channel to ensure uninterrupted data transmission. During the switching process, the data being transmitted is cached and retransmitted to ensure data integrity.

[0151] (3) Data encryption transmission module:

[0152] The AES-256 encryption algorithm is used to encrypt transmitted data, including key data such as point table files and scheduling instructions. The AES-256 encryption algorithm has high encryption performance and can effectively resist data theft and malicious tampering. At the data sending end, the data is encrypted using a key; at the receiving end, the same key is used for decryption, ensuring data security during transmission.

[0153] 3. Key technical principles:

[0154] (1) Multi-mode communication collaborative technology:

[0155] Through the collaborative design of hardware circuits and software protocols, seamless switching between 4G / 5G, PLC power carrier, and Ethernet communication modes is achieved. Each communication module operates independently, while simultaneously collaborating through a unified control interface and scheduling algorithm to dynamically select the optimal transmission channel based on actual needs.

[0156] (2) Principle of intelligent path selection:

[0157] Based on signal strength detection and data urgency assessment, a path selection decision model is constructed. The signal strength of each communication channel is monitored in real time. Combined with the data urgency level parameters, the transmission path is dynamically optimized through preset selection logic to ensure optimal data transmission.

[0158] (3) Principle of data encryption transmission:

[0159] The AES-256 encryption algorithm uses symmetric encryption and encrypts data in blocks using a 256-bit key. The encryption process includes operations such as byte substitution, row shifting, column mixing, and round key addition to convert the original data into ciphertext. The decryption process restores the data through the reverse operation. Key security is crucial for encrypted transmission, and the system ensures key confidentiality and integrity through a secure key management mechanism.

[0160] 4. Detailed explanation of the work process:

[0161] (1) Initialization phase:

[0162] After the communication network layer is started, the hardware self-test of the 4G / 5G wireless module, PLC power carrier module and Ethernet interface is completed to ensure the normal operation of each communication unit;

[0163] Load path selection algorithm parameters, encryption keys and other configuration information, and establish initial communication connections with the distributed DTU unit cluster and cloud management platform;

[0164] (2) Data transmission stage:

[0165] Data reception: Receives data to be transmitted from the distributed DTU unit cluster, including point table files, monitoring data, and control instructions;

[0166] Path evaluation: Real-time detection of the signal strength of each communication channel , and determine the data urgency level based on the data type and application requirements ;

[0167] Path selection: Determine the optimal transmission path based on the path selection algorithm;

[0168] Data encryption: Use AES-256 encryption algorithm to encrypt data;

[0169] Data transmission: The encrypted data is sent to the cloud management platform through the selected transmission channel;

[0170] (3) Network status monitoring and path switching stage:

[0171] Continuously monitor the operating status of each communication channel, including signal strength changes, link stability, etc.

[0172] When the signal strength of the current transmission channel decreases or the data urgency level changes, the transmission path is re-evaluated;

[0173] If the path switching conditions are met, the path switching mechanism is triggered to complete the switching of the transmission channel, and the data transmission is cached and retransmitted accordingly;

[0174] (IV) Data reception and decryption stage:

[0175] The cloud management platform receives encrypted data;

[0176] Use the corresponding key to decrypt the data and obtain the original data for subsequent processing and analysis;

[0177] 5. Application value and advantages:

[0178] (1) Improving communication reliability:

[0179] Multi-mode communication and dynamic path switching mechanisms ensure that the system maintains stable data transmission in complex industrial environments, effectively avoiding data interruptions caused by single communication link failures.

[0180] (2) Ensuring data security:

[0181] AES-256 encryption technology provides high-intensity security protection for data transmission, meeting the strict requirements of the Industrial Internet of Things for data confidentiality and integrity, and preventing data leakage and malicious attacks;

[0182] (3) Optimizing transmission efficiency:

[0183] The intelligent path selection algorithm dynamically adjusts the transmission channel according to the data urgency and network status, so that critical data can be transmitted first and quickly, improving the overall data transmission efficiency and reducing transmission delay;

[0184] (IV) Reduce deployment costs:

[0185] The application of PLC power line carrier and Ethernet and other communication methods reduces the dependence on dedicated communication lines and reduces system deployment costs. It is especially suitable for complex wiring environments in industrial sites.

[0186] In this embodiment, the cloud management platform includes a point table intelligent management module, a resource dynamic scheduling engine, a security protection module, a flexible resource allocation module, and a three-dimensional visualization dashboard, among which:

[0187] Point table intelligent management module performs:

[0188] Version compatibility check: triggers an alarm when the software version difference between the point table file before and after the upgrade exceeds 0.5;

[0189] Point table conflict handling: XML parser is used to standardize and restructure the nested tag conflicts. The reorganization method is to select the point table structure with the smallest Euclidean distance to the standard template structure feature vector from the candidate structure set;

[0190] Blockchain evidence storage: The SHA-256 hash value of the total point table template is stored in the distributed ledger;

[0191] The resource dynamic scheduling engine includes:

[0192] Topology-aware graph convolutional enhanced network model:

[0193] Calculate the normalized topological relationship matrix by adding the adjacency matrix and degree matrix of self-loops;

[0194] Perform three-layer graph convolution operations on the node feature matrix containing load rate and task queue depth;

[0195] Perform global average pooling on the topological embedding vector output by the graph structure encoder;

[0196] The graph convolution result and the pooling result are added with a weight coefficient of 0.6, and the action value is output through the LeakyReLU activation function;

[0197] Real-time scheduler: assigns telemetry and telecontrol task priorities based on action value output results;

[0198] Security protection module execution:

[0199] Lightweight dynamic consensus algorithm:

[0200] When calculating the node consensus weight, the historical credibility score is multiplied by a coefficient of 0.7;

[0201] Multiply the negative exponential function value of the node response delay mean by a coefficient of 0.3;

[0202] The sum of the above two items is multiplied by the ratio of the node's stake amount to the system's maximum stake amount;

[0203] The final weight is normalized by the sum of all node calculation values;

[0204] Tampering detection: triggers an alarm when the Hamming distance between the new block hash value and the stored value is greater than zero;

[0205] Flexible resource allocation module:

[0206] Load forecasting model:

[0207] In the LSTM input gate, the historical load value is subjected to wavelet decomposition feature extraction;

[0208] The load change rate is used to extract residual features in the memory unit calculation;

[0209] Update the memory unit state based on the hidden state at the previous moment;

[0210] Dynamic resource adjustment:

[0211] When the predicted load value is higher than the preset upper limit, the sampling frequency is set to 1kHz;

[0212] When the predicted load value is lower than the preset lower limit, the sampling frequency is set to 100Hz;

[0213] The cloud management platform integrates a 3D visual dashboard to dynamically display:

[0214] DTU unit topology connection relationship matrix;

[0215] Resource load heat map: green when the load rate is below 0.4, yellow when it is between 0.4 and 0.7, and red when it is 0.7 or above;

[0216] Furthermore, the cloud management platform, as the "smart brain" of the distributed DTU unit management and scheduling system, undertakes the core functions of global device management, resource optimization and scheduling, and data security control. By integrating multiple functional modules such as point table intelligent management, dynamic resource scheduling, and security protection, combined with advanced algorithm models and visualization technology, it realizes intelligent and intensive management of distributed DTU unit clusters. The following will elaborate on the architecture design, core modules, technical principles, etc.

[0217] 1. Overall architecture design:

[0218] The cloud management platform adopts a microservices architecture and consists of core modules such as the intelligent point table management module, the dynamic resource scheduling engine, the security protection module, the flexible resource allocation module, and the 3D visualization dashboard. Each module is independent and yet operates in collaboration, enabling data interaction and functional linkage through API interfaces. Built on cloud computing technology, the platform has powerful computing and storage capabilities and can be elastically expanded to meet the access and management needs of large-scale DTU devices.

[0219] 2. Detailed explanation of core functional modules:

[0220] (1) Point meter intelligent management module:

[0221] Version compatibility check: Compares the software version numbers of the point table files before and after the DTU unit upgrade. When the difference exceeds the threshold (set to 0.5), an alarm is triggered. Through strict version management, system failures caused by software version incompatibility are avoided, and the stability and consistency of the point table files are guaranteed.

[0222] Point table conflict handling: Use XML parser to standardize and reorganize the nested tag conflicts of multi-point table files, and use formula (in, is the point table after reorganization; is a candidate structure set, which includes all possible point table structure combinations; It is a structural feature vector used to quantify the structure of the point table. It forms a vector representation by extracting information such as node relationships and data field types in the point table. The standard template is a pre-set point table structure that meets the specifications) and automatically eliminates conflicts and generates a unified total point table template. By minimizing the feature distance between the candidate structure and the standard template, the conflicting point table can be quickly integrated and optimized.

[0223] Blockchain evidence storage: The reorganized total point table template generates a SHA-256 hash value and is written to the blockchain using the Lightweight Dynamic Consensus Algorithm (LDCA). The evidence storage process ensures the immutability and traceability of the point table data, providing reliable protection for the authenticity and integrity of the data.

[0224] (2) Dynamic resource scheduling engine:

[0225] TA-GCN-DQN model: Dynamic allocation of DTU computing resources is achieved based on the topology-aware graph convolutional enhanced network (TA-GCN-DQN); its action value function is defined as

[0226] (in, It is the normalized adjacency matrix, which is used to describe the normalized expression of the connection relationship between DTU nodes; Represents the DTU connection topology matrix. The elements in the matrix reflect whether there is a connection between DTU units and the connection weight; is the degree matrix, and the elements on its diagonal represent the connection degree of the corresponding node; is the identity matrix; It is a node feature matrix, including load rate (value range [0,1], indicating how busy the DTU unit is in processing tasks), task queue depth (a positive integer, indicating the number of tasks waiting to be processed), and other information. To train the convolution kernel weights, the parameters of the node features are learned during the model training process; The state space contains the current operating status information of the system, such as the load of each DTU, network delay, etc. Actions represent different operations for scheduling DTU computing resources; are model parameters, covering all trainable weights and biases; It is a graph topology feature function used to extract the features of the entire DTU network topology structure; The model fully exploits the DTU topology and operating status characteristics by integrating graph convolutional networks and deep Q networks to achieve efficient resource scheduling.

[0227] Real-time scheduler: based on the action-value function output by the TA-GCN-DQN model , real-time allocation of telemetry and remote control task priorities; give priority to emergency tasks, balance the load of each DTU unit, and improve the overall operation efficiency of the system;

[0228] (3) Security protection module:

[0229] LDCA consensus algorithm: Lightweight dynamic consensus algorithm (LDCA) is used to achieve node consensus in the blockchain evidence storage process; The consensus weight calculation formula is (in, For nodes The consensus weight is used to measure the node's voice in the blockchain consensus process; For nodes The historical credibility score is based on the verification accuracy of a certain number of blocks in the past; For nodes The average response delay, in milliseconds (ms), reflects the speed at which the node processes and responds to data; For nodes The stake amount represents the resources invested by the node; The maximum pledge amount of the system; is the total number of nodes; Representation node Historical credibility score; Representation node By comprehensively considering node credibility, response delay, and stake amount, we ensure the security and fairness of the evidence storage process;

[0230] Tampering detection: Compare the Hamming distance between the new block hash value and the stored hash value. When the Hamming distance is greater than 0, an alarm is immediately triggered to effectively prevent data tampering.

[0231] (IV) Flexible resource allocation module:

[0232] Load forecasting model: A load forecasting model is constructed based on the multi-scale spatiotemporal feature fusion LSTM (MSTF-LSTM) algorithm; its gated iteration formula is: (in, It is the input gate, which controls the degree to which the current input information flows into the cell state; The sigmoid activation function maps the input to the [0,1] interval; is the input gate weight matrix, which is used to weight the input information; for The hidden state at each moment contains information from the previous moment; is the wavelet decomposition function, which is used to extract the periodic characteristics of the load signal; for Load value at each moment; is the candidate cell state, used to update the cell state; It is a hyperbolic tangent activation function that maps the input to the interval [-1,1]; is the cell state weight matrix, which weights the relevant information; is the residual extraction function, which is used to extract the residual characteristics of the load change rate; for Load change rate at any moment; for The state of cells at all times, storing long-term information; It is the forget gate, which controls the degree of retention of the cell state information at the previous moment; is element-level multiplication); the model accurately captures the spatiotemporal characteristics of load changes by integrating wavelet decomposition and residual extraction technology, and achieves accurate prediction of future loads;

[0233] Dynamic resource adjustment: based on load forecast results Adjust the sampling frequency of the DTU unit:

[0234] (in, is the sampling frequency, for The load forecast value at that moment, is the high load threshold, By dynamically adjusting the sampling frequency, we can optimize resource utilization efficiency while ensuring data accuracy;

[0235] (5) 3D Visualization Dashboard:

[0236] Topology display: Dynamically display the topological connection relationship matrix of DTU units in three-dimensional graphics, intuitively presenting the network structure and connection status between devices, making it easier for operation and maintenance personnel to quickly grasp the overall system architecture;

[0237] Real-time monitoring: Generates a real-time resource load heat map, using color to map the load ratio (value range [0,1]). When the load ratio is below 0.4, it is displayed as green, indicating low load; between 0.4 and 0.7, it is displayed as yellow, indicating medium load; and above 0.7, it is displayed as red, indicating high load. Operations and maintenance personnel can use the heat map to monitor the load of each DTU unit in real time and promptly identify potential performance bottlenecks.

[0238] 3. Key technical principles:

[0239] (1) Intelligent algorithm fusion technology:

[0240] Combining AI algorithms such as Graph Convolutional Networks (GCN), Deep Q Networks (DQN), and LSTM with blockchain technology enables intelligent resource scheduling, load forecasting, and data storage. Through the collaborative operation of algorithms, the system's autonomous decision-making and optimization capabilities are enhanced.

[0241] (2) Data-driven dynamic management:

[0242] Based on real-time collected equipment operation data, dynamic management of DTU units is achieved through modules such as point table intelligent management and flexible resource allocation. Potential patterns and needs are mined from the data to drive optimal configuration and policy adjustments of system resources.

[0243] (3) Visual interaction technology:

[0244] Utilize 3D modeling and visualization technology to transform abstract system data into an intuitive graphical interface. Through human-computer interaction design, operation and maintenance personnel can quickly understand the system operation status, improving management efficiency and decision-making accuracy.

[0245] 4. Detailed explanation of the work process:

[0246] (1) Initialization phase:

[0247] After the cloud management platform is started, the initial configuration of each functional module is completed, and basic data such as algorithm model parameters and security keys are loaded;

[0248] Establish connections with distributed DTU unit clusters and communication network layers to ensure smooth data transmission channels;

[0249] (2) Data reception and processing stage:

[0250] Receive point table files, monitoring data and status information uploaded by distributed DTU unit clusters;

[0251] The point table intelligent management module performs version checking, conflict resolution, and blockchain evidence storage on uploaded point table files;

[0252] The elastic resource allocation module analyzes monitoring data based on the load forecasting model to predict future load change trends;

[0253] (3) Resource scheduling and decision-making stage:

[0254] The resource dynamic scheduling engine dynamically allocates DTU computing resources and task priorities based on the calculation results of the TA-GCN-DQN model, combined with load forecasting and real-time task requirements;

[0255] The security protection module verifies and stores evidence data through the LDCA consensus algorithm and monitors data integrity in real time;

[0256] (IV) Visualization and feedback stage:

[0257] The 3D visualization dashboard converts system operation data into intuitive topology and heat maps, displaying the system status in real time.

[0258] The platform generates optimization strategies based on data analysis results and sends them to the DTU unit cluster through the communication network layer, achieving closed-loop management and continuous optimization of the system;

[0259] 5. Application value and advantages:

[0260] 1. Realizing Intelligent Operation and Maintenance

[0261] Through intelligent algorithms and dynamic management mechanisms, the cloud management platform can automatically identify problems in system operation, optimize resources and provide fault warnings, significantly reducing manual operation and maintenance costs and improving operation and maintenance efficiency and accuracy.

[0262] (2) Ensuring data security and reliability:

[0263] Blockchain evidence storage and security protection technology ensures the integrity and authenticity of data during storage and transmission, meeting the strict data security requirements of the industrial field and providing reliable data support for business decision-making;

[0264] (3) Improving resource utilization efficiency:

[0265] Dynamic resource scheduling and flexible allocation strategies enable the system to flexibly adjust resource configuration according to actual needs, avoid resource waste, improve overall resource utilization efficiency, and reduce operating costs;

[0266] (IV) Enhance system scalability:

[0267] Microservice architecture and cloud computing technology give the platform powerful scalability, which can easily cope with the growth of equipment scale and changes in business needs, providing strong guarantees for the long-term development of the system.

[0268] In this embodiment, the topology-aware graph convolution enhancement network training process includes:

[0269] The state space consists of load rate, network delay and task backlog;

[0270] The reward function is calculated as follows: 0.5 times the difference between the baseline scheduling delay and the actual delay, plus 0.3 times the energy saving rate, minus 0.2 times the overload indicator;

[0271] Terminate training when the reward function variance is less than 0.01;

[0272] Furthermore, the training process of the Topology-Aware Graph Convolutional Enhanced Network (TA-GCN-DQN) is a key step in enabling the network to achieve efficient resource allocation in a decentralized DTU unit management and scheduling system. It continuously optimizes network parameters through steps such as constructing a state space and designing a reward function. The following is a detailed breakdown of its training process:

[0273] 1. State space construction:

[0274] The TA-GCN-DQN model abstracts key indicators of system operation into a state space s, which is composed of three core dimensions: load rate, network latency, and task backlog:

[0275] Load rate: reflects the busyness of the DTU unit's current processing tasks. The value range is [0,1], where 0 indicates no task load and 1 indicates full load operation;

[0276] Network latency: measures the time loss for data transmission in a communication network, usually measured in milliseconds (ms). A lower value indicates higher data transmission efficiency.

[0277] Task backlog: counts the number of tasks waiting to be processed. The larger the value, the greater the current task processing pressure on the system.

[0278] By combining these three indicators into a state space, the model can fully perceive the system's operating status and provide a data basis for subsequent decision-making;

[0279] 2. Reward Function Design

[0280] The reward function R is used to evaluate the quality of the model's decisions and guide the model to learn the optimal resource allocation strategy. Its calculation formula is:

[0281] (in, is the benchmark scheduling delay, which is the preset ideal task scheduling time; is the new scheduling delay, which is the actual task scheduling time after the model takes action; Energy saving rate is used to measure the degree to which the resource allocation strategy optimizes the system energy consumption; is an overload indicator, which takes the value 1 when the DTU unit is overloaded and 0 otherwise); this function rewards and punishes the model decision from three dimensions:

[0282] Scheduling delay optimization: through Encourage the model to reduce task scheduling time; the greater the difference, the higher the reward;

[0283] Energy saving rewards: Some incentive models adopt energy-saving resource allocation strategies to improve system energy efficiency;

[0284] Overload penalty mechanism: Penalize decisions that lead to DTU overload, prompting the model to avoid uneven resource allocation;

[0285] 3. Model training and parameter update:

[0286] Data collection and sample generation: During system operation, real-time data on DTU unit load rate, network latency, task backlog, and other status data are collected, and the actions corresponding to each decision are recorded. and rewards , forming training samples ,in, To perform an action The new state after

[0287] Forward propagation and loss calculation: The training samples are input into the TA-GCN-DQN model, and the action value function is calculated through the graph convolution layer and the deep Q network. According to the predicted Values ​​and Goals Value (based on the reward and the maximum value of the next state The difference between the values ​​is calculated) and the loss function is calculated. ,The mean square error (MSE) is commonly used to measure the gap between the two;

[0288] Back propagation and parameter optimization: Using the back propagation algorithm to calculate the loss function About model parameters The gradient of , and use stochastic gradient descent (SGD) or its variant algorithms (such as Adam, RMSProp) to update the parameters , reduce the loss function value and make the model predict The value is closer to the true value;

[0289] Experience replay mechanism: To break the correlation between training samples and improve training stability, the model adopts an experience replay strategy. Training samples are stored in the experience replay buffer, and samples are randomly sampled for learning during training to avoid model overfitting due to continuous processing of similar samples.

[0290] 4. Training termination conditions:

[0291] When the variance of the reward function Var When , the model training is considered to be stable and the training is terminated at this time; this condition indicates that the reward value output by the model fluctuates less, which means that the resource allocation strategy learned by the model has converged and can stably output reliable decisions under different states, meeting the system's demand for dynamic resource scheduling.

[0292] In this embodiment, point table merging and blockchain evidence storage work together:

[0293] Calculate the SHA-256 hash value of the total point table template immediately after it is generated;

[0294] Trigger the lightweight dynamic consensus algorithm to write the hash value and timestamp into the blockchain;

[0295] When the point table version is updated, it is necessary to verify that the new hash value is completely consistent with the stored value before activation;

[0296] The coordinated work of point-table merging and blockchain evidence storage is a key mechanism to ensure data consistency, security, and traceability in the decentralized DTU unit management and scheduling system. The two work closely together to form a complete chain from data integration to secure storage. The following is a detailed explanation:

[0297] 1. Collaborative Workflow:

[0298] (1) Total point table template generation and hash calculation:

[0299] After the public unit collects the XML point table files uploaded by each interval unit, it performs conflict detection and merging operations on them, adds the IP identifier to generate a total point table template; after the total point table template is generated, its hash value is immediately calculated using the hash function. The formula is:

[0300] (in, The SHA-256 hash value of the total point table template is a fixed-length string generated by the SHA-256 hash algorithm from the total point table template data. It is used to uniquely identify the point table template. The reorganized point table, i.e., the complete point table data after conflict detection, normalization and reorganization, etc.); the calculation of the hash value ensures the integrity of the point table data, and any slight change in the data will cause a significant change in the hash value;

[0301] (2) Blockchain evidence trigger:

[0302] Calculate the hash value After that, the system triggers the Lightweight Dynamic Consensus Algorithm (LDCA) ) is written into the blockchain; wherein, The current timestamp records the time information of the point table template proof. During the proof process, each node calculates its own consensus weight through the LDCA consensus algorithm. By comprehensively considering the node credibility, response delay and stake amount, the security and fairness of the proof process are ensured. Only nodes with sufficient consensus weight can participate in block verification and writing.

[0303] (3) Point table version update verification:

[0304] When the system needs to update the point table version, first calculate the hash value of the new version point table , then compare The hash value of the certificate stored in the blockchain Hamming distance ;when When , it indicates that the new version point table data is consistent with the evidence data, and then the new version point table can be activated; if , it means that the point table data may be tampered with, the system triggers an alarm and prohibits the activation of the new version, thereby ensuring the authenticity and reliability of the data;

[0305] 2. Advantages of collaborative work:

[0306] (1) Data integrity protection:

[0307] The combination of hash calculation and blockchain evidence storage ensures that any unauthorized modification of point table data during its generation, transmission, and storage can be detected promptly. By comparing hash values, data integrity can be quickly verified, preventing system failures caused by data errors or tampering.

[0308] (2) Operation traceability:

[0309] The distributed ledger feature of the blockchain records every evidence storage operation of the point table template and the corresponding timestamp, forming a complete operation log. Operation and maintenance personnel can trace the historical versions and operation records of the point table data at any time, facilitating auditing and troubleshooting.

[0310] (3) Safe and reliable version management:

[0311] When the point table version is updated, the verification mechanism based on hash value comparison ensures that only verified legal versions can be activated, preventing malicious tampering or the use of incorrect versions, and providing solid protection for the stable operation of the system.

[0312] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A distributed DTU unit management and scheduling system, characterized by: Including distributed DTU unit cluster, communication network layer and cloud management platform: The distributed DTU unit cluster is composed of a common unit and multiple interval units connected hierarchically through an industrial switch, and each interval unit is equipped with a point table generation module and a self-diagnosis module; The communication network layer integrates 4G / 5G wireless modules, PLC power carrier modules and Ethernet interfaces, and supports dynamic switching of multi-mode transmission paths; The cloud management platform includes a point table intelligent management module, a resource dynamic scheduling engine and a security protection module. The resource dynamic scheduling engine includes: Topology-aware graph convolutional enhanced network model: Calculate the normalized topological relationship matrix by adding the adjacency matrix and degree matrix of self-loops; Perform three-layer graph convolution operations on the node feature matrix containing load rate and task queue depth; Perform global average pooling on the topological embedding vector output by the graph structure encoder; The graph convolution result and the pooling result are added with a weight coefficient of 0.6, and the action value is output through the LeakyReLU activation function; Real-time scheduler: assigns telemetry and telecontrol task priorities based on action value output results; The system performs the following process: The cloud management platform sends program upgrade instructions to the public unit, which are then forwarded to each bay unit via the TFTP protocol. After the bay unit is upgraded, an XML point table file containing the address and range of telesignaling, telemetering and remote control data is generated and transmitted back to the public unit; The public unit performs conflict detection and merging of multi-point table files, and adds IP identifiers to generate a total point table template; Dynamically allocate DTU computing resources based on topology-aware graph convolution-enhanced networks; The hash value of the point table file is stored on the blockchain through a lightweight dynamic consensus algorithm.

2. A distributed DTU unit management and scheduling system according to claim 1, characterized in that: The point table intelligent management module performs: Version compatibility check: triggers an alarm when the software version difference between the point table file before and after the upgrade exceeds 0.5; Point table conflict handling: XML parser is used to standardize and restructure the nested tag conflicts. The reorganization method is to select the point table structure with the smallest Euclidean distance to the standard template structure feature vector from the candidate structure set; Blockchain evidence storage: The SHA-256 hash value of the total point table template is stored in the distributed ledger.

3. A distributed DTU unit management and scheduling system according to claim 1, characterized in that: The communication network layer implements: Multimode transmission channel selection: When the data urgency level is greater than 0.8 and the signal strength is not less than -85dBm, select the 4G / 5G channel; Select the PLC channel when the data urgency level is between 0.4 and 0.8; In other cases, select Ethernet Channel; Encrypted transmission: AES-256 algorithm is used to encrypt point table files and scheduling instructions.

4. A distributed DTU unit management and scheduling system according to claim 1, characterized in that: The security protection module performs: Lightweight dynamic consensus algorithm: When calculating the node consensus weight, the historical credibility score is multiplied by a coefficient of 0.7; Multiply the negative exponential function value of the node response delay mean by a coefficient of 0.3; The sum of the two items is multiplied by the ratio of the node's stake amount to the system's maximum stake amount. The final weight is normalized by the sum of all node calculation values; Tampering detection: An alarm is triggered when the Hamming distance between the new block hash value and the stored value is greater than zero.

5. A distributed DTU unit management and scheduling system according to claim 1, characterized in that: Also includes elastic resource allocation module: Load forecasting model: In the LSTM input gate, the historical load value is subjected to wavelet decomposition feature extraction; The load change rate is used to extract residual features in the memory unit calculation; Update the memory unit state based on the hidden state at the previous moment; Dynamic resource adjustment: When the predicted load value is higher than the preset upper limit, the sampling frequency is set to 1kHz; When the predicted load value is lower than the preset lower limit, the sampling frequency is set to 100Hz.

6. A distributed DTU unit management and scheduling system according to claim 1, characterized in that: The topology-aware graph convolution enhanced network training process includes: The state space consists of load rate, network delay and task backlog; The reward function is calculated as follows: 0.5 times the difference between the baseline scheduling delay and the actual delay, plus 0.3 times the energy saving rate, minus 0.2 times the overload indicator; The training is terminated when the reward function variance is less than 0.

01.

7. The distributed DTU management and scheduling system according to claim 1, characterized in that: The self-diagnostic module performs: Fault location: Calculate the ratio of the peak-to-peak current to the average current as the ripple factor. If the ripple factor exceeds 0.25, the capacitor is considered abnormal. Collaborative control: After the faulty unit is isolated, the public unit takes over data collection, and the task transfer delay is controlled within 10 milliseconds.

8. The distributed DTU unit management and scheduling system according to claim 1, characterized in that: The cloud management platform integrates a 3D visual dashboard to dynamically display: DTU unit topology connection relationship matrix; Resource load heat map: green when the load rate is below 0.4, yellow when it is between 0.4 and 0.7, and red when it is 0.7 and above.

9. The distributed DTU management and scheduling system according to claim 1, characterized in that: Point table merging and blockchain evidence storage work together: Calculate the SHA-256 hash value of the total point table template immediately after it is generated; Trigger the lightweight dynamic consensus algorithm to write the hash value and timestamp into the blockchain; When the point table version is updated, it is necessary to verify that the new hash value is exactly the same as the stored value before activation.

Images