A smart grid fault-tolerant communication system and method based on multi-operator channel switching

Abstract

The application discloses a smart grid fault-tolerant communication system and method based on multi-operator channel switching. The heterogeneous network and device state data are collected through the intelligent sensing layer. The AI decision engine realizes network quality prediction and multi-objective optimization. The resource collaborative control layer integrates spectrum resources and network slices. The edge execution layer executes hardware-level fast switching. The emergency disaster recovery layer provides multi-dimensional fault recovery. The secure and reliable layer guarantees data security. The application breaks through the traditional network switching bottleneck, realizes high-reliability and low-latency transmission of power business, and is suitable for smart substation, distribution network automation and other scenes, which significantly improves the fault-tolerant capability and resource utilization efficiency of smart grid communication.

Description

Technical Field

[0001] This application belongs to the field of smart grid communication technology, and in particular relates to a fault-tolerant communication system and method for smart grids based on multi-operator channel switching. Background Technology

[0002] With the construction of new power systems, smart grids place higher demands on the reliability and real-time performance of communication systems. Traditional power communication networks have the following shortcomings:

[0003] Insufficient coordination among heterogeneous networks: Multiple communication technologies such as 5G, fiber optics, and private wireless networks coexist, but there is a lack of a unified resource management mechanism, resulting in high channel switching latency (typical switching latency > 200ms) and frequent service interruptions.

[0004] Weak fault tolerance: In extreme weather (such as typhoons and ice storms) or equipment failure, the lack of a rapid self-healing mechanism can easily lead to the paralysis of the distribution network automation system and cause the risk of large-scale power outages.

[0005] Low resource utilization: Each operator deploys its network resources independently, resulting in issues such as idle spectrum resources (average utilization rate <40%) and redundant construction, which increases the cost of power grid communication.

[0006] In existing technologies, Chinese patent CN116963216A proposes an automatic switching method between 5G and WiFi, but it does not solve the problems of multi-operator cross-network coordination and power service priority guarantee; CN108632936A involves core network anchoring technology, but it does not design a dedicated QoS mechanism for time-sensitive data of power equipment. Therefore, there is an urgent need for a communication system that integrates resources from multiple operators and has intelligent decision-making and rapid fault tolerance capabilities. Summary of the Invention

[0007] Therefore, it is necessary to provide a smart grid fault-tolerant communication system and method based on multi-operator channel switching to address the above-mentioned technical problems. Through intelligent sensing, AI decision-making, resource coordination and hardware-level switching, it can achieve highly reliable transmission of power services and solve the problems of insufficient heterogeneous network coordination, weak fault tolerance and low resource utilization in the existing technology.

[0008] In a first aspect, this application provides a fault-tolerant communication system for smart grids with multi-carrier channel switching, the system comprising:

[0009] The intelligent sensing layer is configured with a multi-mode terminal access gateway and a power grid equipment status collector. The multi-mode terminal access gateway adopts an embedded multi-band radio frequency chip, which supports the simultaneous collection of real-time network indicators of multiple communication standards. The power grid equipment status collector is used to collect power grid equipment status data.

[0010] The AI ​​decision engine includes a federated learning collaboration module, an LSTM dynamic prediction module, and a multi-objective optimization decision-maker. The federated learning collaboration module is used to achieve differential privacy protection and generate a network quality prediction model jointly trained across operator edge nodes. The LSTM dynamic prediction module takes into input a multi-dimensional feature vector containing weather data including temperature, humidity, and precipitation probability, as well as equipment load data, and constructs a wireless channel attenuation compensation model to predict network load. The multi-objective optimization decision-maker is based on the NSGA-II multi-objective optimization algorithm, with the optimization objectives of minimizing total latency, minimizing energy consumption, and minimizing operator tariff costs, and iteratively generates a Pareto optimal solution set.

[0011] The resource coordination control layer includes a blockchain spectrum sharing ledger and an SDN network slice controller. The blockchain spectrum sharing ledger realizes spectrum resource auction and allocation through consortium blockchain and smart contracts. The SDN network slice controller sends flow tables to the operator's core network to reserve bandwidth and maximum latency for the corresponding slices for services.

[0012] The edge execution layer includes a local fast switching agent and a protocol conversion gateway. The local fast switching agent integrates a PCIe hardware switching card and realizes physical layer link switching based on dual-link pre-synchronization technology. The service interruption time during the switching process does not exceed 5ms, and TCP session continuity is maintained through the data plane development kit. The protocol conversion gateway has a built-in multi-standard protocol conversion engine and realizes protocol conversion between different network standards based on a pre-stored protocol template library.

[0013] The emergency disaster recovery layer is constructed by heterogeneously integrating a low-altitude UAV mesh relay network and a satellite link redundant channel to build an integrated air-ground communication network with fault self-healing capabilities. The UAV mesh relay network is equipped with micro base stations and uses the OLSRv2 routing protocol to build the shortest path mesh network. The satellite link redundant channel integrates a satellite communication module and dynamically adjusts the coding rate according to the satellite signal quality through adaptive coding and modulation technology.

[0014] The security and trust layer includes a national cryptographic encryption engine and a device fingerprint authentication module. The national cryptographic encryption engine is used to dynamically rotate the encryption key, and the device fingerprint authentication module extracts the radio frequency signal features of newly accessed devices to generate device fingerprints and completes device access authentication through feature matching algorithms.

[0015] In one embodiment, the multi-mode terminal access gateway of the intelligent sensing layer supports the simultaneous collection of real-time network indicators of at least three different communication standards, and the power grid equipment status collector supports the parsing of equipment data of at least two different protocols.

[0016] In one embodiment, the federated learning collaboration module aggregates historical data from no fewer than three operators and employs differential privacy technology to ensure that the original data does not leave the local machine.

[0017] The root mean square error (RMSE) of the prediction module of the LSTM dynamic prediction module does not exceed 15%.

[0018] In one embodiment, the smart contract of the blockchain spectrum sharing ledger allocates spectrum resources through a Dutch auction mechanism, and generates a digital certificate containing the resource usage period after the transaction is confirmed; the SDN network slicing controller reserves bandwidth resources of no less than 50Mbps for QoS Class 0 services and sets the maximum latency threshold to 20ms.

[0019] In one embodiment, the hardware switching card of the local fast switching agent supports a PCIe 3.0 interface, and the switching latency is tested by the IEEE 802.21 standard and does not exceed 8ms; the protocol conversion gateway supports real-time conversion of no less than 10 communication protocols.

[0020] In one embodiment, the relay node spacing of the UAV Mesh relay network does not exceed 1km, and the shortest path hop count constructed using the OLSRv2 protocol does not exceed 5 hops;

[0021] The adaptive coding and modulation technology of the satellite link supports dynamic switching between three modulation modes: QPSK, 16QAM, and 64QAM.

[0022] In one embodiment, the national cryptographic encryption engine generates a new SM4 key at regular intervals and distributes the key through an out-of-band channel;

[0023] The device fingerprint database is updated in real time and automatically removes terminal devices that fail to authenticate three times consecutively.

[0024] Secondly, this application also provides a fault-tolerant communication method for smart grids with multi-operator channel switching, comprising the following steps:

[0025] Data Acquisition and Priority Marking: Real-time quality data from heterogeneous networks such as 5G NR and LTE-V2X are collected through the multi-mode terminal access gateway. The power grid equipment status collector parses equipment data such as relay protection signals and SCADA control commands, and marks time-sensitive data as QoS Class 0 based on the IEC 61850 standard.

[0026] Dynamic network quality assessment: The federated learning model aggregates historical data from no fewer than three operators, uses differential privacy technology to protect the original data, and generates a cross-domain network quality prediction model; the LSTM model takes into input a multi-dimensional feature vector containing weather data and equipment load data, constructs a wireless channel attenuation compensation model, and outputs a latency reliability score matrix.

[0027] Multi-objective optimization decision-making: Based on the NSGA-II multi-objective optimization algorithm, with constraints of total latency ≤50ms, single device power consumption ≤100mW, and operator tariff cost reduction of 20%, the Pareto optimal network combination solution set is generated, supporting priority intervention decision-making by manual policy library;

[0028] Cross-domain resource collaborative scheduling: Spectrum resources are allocated through the auction mechanism of blockchain smart contracts, and resource lock certificates are generated after the transaction is confirmed; the SDN controller reserves no less than 50Mbps of bandwidth resources and sets a maximum latency of 20ms according to the QoS Class 0 service requirements to complete the network slice creation;

[0029] Seamless hardware-level handover execution: Triggers the PCIe hardware switching card to achieve physical layer link handover, with handover latency not exceeding 8ms as tested by the IEEE 802.21 standard. TCP session continuity is maintained through DPDK technology. A dual-link synchronous transmission mechanism is adopted, with parallel transmission time between the old and new links ≥100ms, ensuring a packet loss rate of less than 0.1% during handover.

[0030] Emergency disaster recovery and self-healing: When a network failure is detected, the drone Mesh relay network is activated, and the shortest path forwarding route (relay node spacing ≤ 1km) is constructed within 3 minutes using the OLSRv2 protocol to restore SCADA data transmission; the satellite link uses adaptive coding and modulation technology to dynamically adjust the coding rate (QPSK / 16QAM / 64QAM) according to SNR to achieve bandwidth optimization;

[0031] Security hardening and model update: The national cryptographic encryption engine generates new SM4 keys at regular intervals and distributes them to terminal devices through an independent out-of-band channel; the device fingerprint database updates the radio frequency signal characteristics of access devices in real time and blocks access for terminals that fail to authenticate three times in a row; the prediction model is updated based on the incremental learning algorithm, and the mean square error of network quality prediction is reduced by no less than 15% after the update.

[0032] In one embodiment, the data acquisition and priority marking step includes:

[0033] The multi-mode terminal access gateway supports the simultaneous collection of network indicators for no less than three communication standards, with a sampling frequency of no less than 100Hz.

[0034] In one embodiment, the dynamic network quality assessment step includes:

[0035] The global aggregation cycle of the federated learning model is 5 minutes, and the input feature vector of the LSTM model has a dimension of no less than 8.

[0036] In one embodiment, the multi-objective optimization decision step includes:

[0037] The NSGA-II algorithm initializes a population of 100-200, performs at least 50 iterations, and outputs at least 10 Pareto optimal solutions.

[0038] In one embodiment, the cross-domain resource collaborative scheduling step includes:

[0039] The spectrum auction cycle for blockchain smart contracts is 1 minute, and the slice creation latency for the SDN controller is no more than 200ms.

[0040] In one embodiment, the hardware-level seamless switching execution step includes:

[0041] In the hardware-level seamless switching execution step, the parallel transmission time of the dual-cast mechanism is dynamically adjusted by the network load, with a minimum of 50ms. After the switching is completed, a performance report including latency and packet loss rate is generated.

[0042] In one embodiment, the emergency disaster recovery and self-healing steps include:

[0043] The coverage radius of the drone mesh relay network is no less than 5km, and the adaptive coding and modulation response time of the satellite link is no more than 50ms.

[0044] In one embodiment, the security hardening and model update step includes:

[0045] The out-of-band distribution channel of the SM4 key is encrypted with AES-256, and the feature matching accuracy of device fingerprint authentication is no less than 99.5%.

[0046] The aforementioned system and methods, through multi-standard data acquisition at the intelligent sensing layer (supporting 5G NR, LTE-V2X, etc., covering 30+ network indicators), AI decision engine federated learning and NSGA-II algorithm collaboration (aggregating data from ≥3 operators to achieve multi-objective optimization of latency, energy consumption, and cost, reducing tariffs by 20%), seamless hardware-level switching at the edge execution layer (PCIe card achieves ≤8ms physical layer switching, dual-cast mechanism packet loss rate <0.1%), air-space-ground fusion at the emergency disaster recovery layer (drone mesh network restores SCADA transmission in 3 minutes, satellite link adaptive modulation), and national cryptographic technology at the security and trust layer (SM4 key rotation in 4 minutes, device fingerprint authentication accuracy ≥99.5%), construct an integrated fault-tolerant communication system. This achieves a 35% reduction in end-to-end latency for critical services, a 99.99% system fault tolerance rate, and a 25% increase in spectrum resource utilization. It also overcomes communication blind spots in complex terrain and domestic security bottlenecks, providing a highly reliable, low-latency, and fully autonomous cross-domain communication solution for smart grids. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of the system architecture according to an embodiment of this application;

[0049] Figure 2 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] The smart grid fault-tolerant communication system based on multi-carrier channel switching provided in this application, as shown in Figure 1, includes six major modules: intelligent sensing layer, AI decision engine, resource collaborative control layer, edge execution layer, emergency disaster recovery layer, and security and trust layer. Each module realizes data interaction and collaborative control through standardized APIs (application programming interfaces).

[0052] The intelligent sensing layer includes a multi-mode terminal access gateway and a power grid equipment status acquisition device.

[0053] The multi-mode terminal access gateway is used to aggregate heterogeneous network signals such as 5G / 4G / WiFi / LoRa, and collect channel quality data such as RSSI (signal strength), latency, and packet loss rate in real time. It supports simultaneous access to ≥8 communication standards. The multi-mode terminal access gateway adopts an embedded multi-band radio frequency chip (such as Qualcomm X65), integrates a customized Linux driver, realizes parallel processing of multiple protocol stacks, and has a data acquisition cycle of ≤100ms.

[0054] While different operators' 5G / 4G networks differ in frequency bands and parameter settings, the Qualcomm X65 can independently tune and receive signals from different frequency bands through multiple internal RF front-end modules, accurately receiving 5G bands from operators like China Unicom, China Mobile, and China Telecom. This multi-band reception capability provides the foundation for aggregating network signals from different operators, allowing the gateway to simultaneously acquire multiple network signal resources. Customized Linux drivers are developed to address the communication protocols of different operator network signals. For example, for different operators' 4G networks, drivers adapted according to 3GPP standards are developed to parse and process the LTE protocol, ensuring correct identification and processing of 4G signals from different operators. These drivers provide a unified software interface for multi-band RF chips, enabling the gateway operating system to efficiently manage and control the chip, coordinating the reception and processing of different network signal standards. Based on customized Linux drivers, multi-mode terminal access gateways achieve parallel processing of multiple protocol stacks. Different operator network signals follow different communication protocols, requiring the gateway to process multiple protocol stacks simultaneously. At the same time, the gateway may simultaneously receive NR protocol signals from mobile 5G networks, LTE protocol signals from telecom 4G networks, and 802.11 protocol signals from WiFi networks. Through multi-protocol stack parallel processing technology, the gateway can independently process these different protocol signals, performing data unpacking, reassembly, and other operations according to their respective protocol specifications, thus achieving effective aggregation of signals from different operator networks. During the data transmission phase, transmission resources are rationally allocated based on data priority and network conditions to ensure stable transmission of various signals.

[0055] To improve signal reception, the multi-mode terminal access gateway is equipped with components such as signal amplifiers and filters. Signal amplifiers enhance weak network signals, ensuring sufficient signal strength for subsequent processing. In remote areas with weak signals, when smart grid devices communicate with the gateway, signal amplifiers amplify LoRa signals, enabling stable reception. Filters filter out useful signals, removing noise and interference. In complex electromagnetic environments, such as near substations, filters eliminate electromagnetic interference generated by power equipment, allowing the gateway to accurately receive operator network signals, improving signal purity and stability, and providing high-quality input for signal aggregation.

[0056] Among them, the power grid equipment status collector is used to acquire the working status of power equipment, such as transformer temperature and line current, associate communication demand priority, and mark time-sensitive data such as relay protection signals as QoS Class 0 (highest priority). The power grid equipment status collector is based on an industrial IoT gateway integrating the Modbus / 61850 protocol, supports the parsing of multiple power equipment protocols, and achieves data acquisition accuracy ≤1μs through hardware timestamps.

[0057] As a core component in power transmission, the temperature of a transformer is a crucial indicator of its operational status. Excessively high transformer temperatures may indicate underlying faults, such as winding short circuits or core overheating. If these issues are not addressed promptly, they can severely impact the transformer's lifespan and even trigger power outages. Line current directly affects the efficiency and safety of power transmission; overcurrent can lead to line overload, causing line burnout, tripping, and other faults. Power grid equipment status acquisition devices, connected to these devices, collect these parameters in real time, providing fundamental data for power system monitoring and management.

[0058] Different power equipment data have varying communication requirements. Power grid equipment status acquisition devices prioritize communication needs based on data importance and real-time requirements. During power system operation, different services have varying requirements for the timeliness and reliability of data transmission. For services with extremely high real-time requirements, such as the issuance and execution feedback of power grid dispatching instructions, priority must be given to ensuring communication bandwidth and transmission stability. Conversely, for some non-real-time equipment status monitoring data, such as the operating status information of certain auxiliary equipment, the timeliness requirement for communication is relatively lower. By prioritizing communication needs, the acquisition device can rationally arrange the data transmission order and resource allocation, ensuring that important data is transmitted first and improving the utilization efficiency of communication resources.

[0059] Time-sensitive data, such as relay protection signals, is crucial for ensuring the safe and stable operation of power systems. Power grid equipment status acquisition devices mark this type of data as QoS Class 0. Relay protection signals are used to quickly trigger protection devices and disconnect faulty circuits when a power system fault occurs, protecting power equipment and personnel. In the event of a short-circuit fault, the relay protection signal must be transmitted to the protection device within a very short time, enabling the device to act quickly, disconnect the faulty line, and prevent the fault from escalating. Marking it as the highest priority ensures that resources are allocated preferentially to this type of signal during data transmission, reducing transmission delays and guaranteeing the safe and reliable operation of the power system.

[0060] The AI ​​decision engine includes a federated learning collaboration module, an LSTM dynamic prediction module, and a multi-objective optimization decision maker.

[0061] The federated learning collaboration module is used to establish a cross-carrier jointly trained network quality prediction model, sharing knowledge while protecting data privacy and supporting model updates at edge nodes. This module employs the PySyft framework to implement asynchronous federated learning, using the FedAvg algorithm for model parameter aggregation, and setting ε-differential privacy parameters ε=10 and δ=1e-5 to ensure that the original data remains local while meeting privacy protection requirements. Lightweight models are deployed at edge nodes, ensuring that the original data does not leave the local machine during training, and parameter updates are transmitted via encrypted channels.

[0062] Different network operators vary in network coverage, signal strength, and bandwidth, making it difficult for data from a single operator to comprehensively reflect overall network quality. Integrating data from multiple operators to train a model would allow for more accurate network quality predictions. However, operator data involves sensitive information such as user privacy and trade secrets, and direct data sharing poses a serious risk of privacy breaches. Therefore, a federated learning collaborative module is used for predictive model training. Using the PySyft framework within this module to implement differential privacy federated learning means that the original data from each operator is strictly stored locally during training and is not directly uploaded elsewhere. Each edge node trains its model based on local data, and when uploading the generated model parameters, differential privacy techniques are used to process the parameters, adding appropriate noise. This makes it difficult for someone to deduce the original data from the uploaded parameters, effectively protecting data privacy.

[0063] The LSTM dynamic prediction module predicts network load fluctuations and device communication needs over the next 5 minutes based on 20 parameters, including historical traffic patterns, weather data, and device load, and outputs performance scores for each network standard. This LSTM dynamic prediction module uses TensorFlow Lite to deploy LSTM neural networks on edge nodes, and the input data is normalized, resulting in a prediction error rate of ≤5%.

[0064] The LSTM dynamic prediction module takes into account up to 20 dimensions of parameters, including historical traffic patterns, weather data, and device load. Historical traffic patterns record changes in network traffic over time and are crucial for predicting future network load. By analyzing the timing and magnitude of network traffic peaks and troughs over a past period, patterns and trends can be identified. During peak electricity consumption periods, the amount of power data transmitted in the network increases significantly. By learning from historical traffic patterns, the model can understand these periodic changes, thus making more accurate predictions of future traffic. Weather data is also a significant factor affecting network load and device communication demands. Severe weather, such as heavy rain and strong winds, may cause power equipment failures, increasing communication demands and potentially affecting network signal transmission quality, leading to network load fluctuations. By incorporating weather data into the input parameters, the model can account for the impact of these external factors on the network. Strong winds may cause transmission lines to sway, causing line sensors to send more monitoring data, increasing network traffic. Device load reflects the operating status of power equipment. When equipment is under high load, it means the equipment is handling a large amount of power transmission or conversion tasks, which may trigger more equipment status monitoring data transmission, thus increasing network load. Large transformers operating under high load need to send monitoring data such as temperature and pressure to the monitoring center more frequently. Using this multi-dimensional data as input allows the LSTM dynamic prediction module to gain a more comprehensive understanding of the network and equipment's operating status, improving prediction accuracy.

[0065] The 20-dimensional parameters cover data from multiple key aspects, including historical traffic, weather, equipment load, network status, and other environmental and power system operation data.

[0066] Historical traffic pattern parameters (6 dimensions): peak, trough, and average network traffic every 10 minutes over the past hour, reflecting the fluctuation range and overall trend of network traffic; standard deviation of traffic at different times over the past 24 hours, reflecting the dispersion of traffic and helping to judge traffic stability, such as the difference in standard deviation of traffic on weekdays and weekends, which can help predict traffic changes in different time periods; rate of change of traffic at the same time every day over the past week, capturing long-term periodic change patterns, such as the traffic change trend at 9 am every Monday; rate of change of traffic between adjacent time periods, showing the changes in network traffic within adjacent time periods, calculated as (current time period traffic - previous time period traffic) / previous time period traffic. By calculating the rate of change of traffic per hour over the past 24 hours, one can intuitively understand whether network traffic is increasing or decreasing, and the magnitude of the change.

[0067] Weather data related parameters (5 dimensions): real-time temperature, humidity, wind speed, and precipitation probability. These factors have direct or indirect impacts on power equipment and network signals. For example, high temperatures may cause heat dissipation problems in equipment, affecting its communication stability and thus increasing communication demand. A high probability of precipitation may interfere with wireless signal transmission, causing network load fluctuations. Weather warning information, such as warnings for heavy rain, strong winds, and lightning, can reflect potential risks in advance and prompt models to adjust their predictions. For example, receiving a lightning warning may indicate that some equipment will increase fault monitoring data transmission, thereby affecting network load.

[0068] Equipment load-related parameters (4 dimensions): The current active and reactive power of the power equipment reflects its actual workload. Active power determines the equipment's actual work capacity, while reactive power affects the power factor of the power grid; changes in both affect the equipment's communication needs and network load. The equipment's load factor, the ratio of actual load to rated load, directly reflects the equipment's workload. A load factor close to 1 indicates the equipment may generate more monitoring data, increasing network traffic. The equipment's cumulative operating time is used to assess the equipment's aging and potential for failure. Equipment operating for extended periods is more prone to failure, potentially triggering a large amount of fault data transmission and altering the network load.

[0069] Network status parameters (3 dimensions): The signal strength of each network standard (e.g., 5G, 4G, WiFi) directly affects data transmission quality and speed. Low signal strength may cause devices to retransmit data, increasing network load. Current network bandwidth utilization reflects network resource usage. High bandwidth utilization indicates network congestion, affecting the fulfillment of device communication needs. Real-time packet loss rate is a key indicator of network quality. An increased packet loss rate leads to more retransmissions, increasing network load.

[0070] Other environmental and operational parameters (2D): Peak and off-peak electricity consumption periods in the area, distinguishing differences in electricity demand during different times. During peak electricity consumption, the communication needs of power equipment increase, leading to a rise in network load. Recent maintenance plans for the power system may cause equipment to temporarily increase data transmission, affecting network load. For example, if a substation is scheduled for maintenance, equipment will frequently upload debugging data during this period, altering network traffic.

[0071] The LSTM dynamic prediction module uses an LSTM neural network for prediction. LSTM (Long Short-Term Memory) is a special type of recurrent neural network that effectively processes time-series data, overcoming the limitations of traditional neural networks in handling long-term dependencies. In network load and device communication demand prediction, the data has strong time-series characteristics; past network states and device operating conditions have a significant impact on future predictions. LSTM, by introducing gating mechanisms, including forget gates, input gates, and output gates, can selectively remember and update information, thereby better capturing long-term dependencies in the data. When predicting future network load fluctuations, LSTM can remember the trend of network traffic changes over a period of time and adjust and predict based on current input data. To efficiently run the LSTM neural network on edge nodes, this module is deployed using TensorFlow Lite. TensorFlow Lite is a deep learning framework specifically optimized for mobile and embedded devices, characterized by its lightweight and high efficiency. On edge nodes, computational and storage resources are typically limited. TensorFlow Lite can reduce model size and computational overhead while maintaining model performance, enabling the LSTM neural network to run quickly on these resource-constrained devices. At the edge nodes of substations, the LSTM dynamic prediction module deployed using TensorFlow Lite can process locally collected data in real time, quickly predict future network load and device communication needs, and provide timely decision support for local network resource scheduling.

[0072] Input data normalization is a crucial step in the LSTM dynamic prediction module. Input data of different dimensions often have different units and value ranges. Without normalization, features with larger value ranges may dominate during training, affecting the model's performance. Normalization maps all input data to a unified range, such as [0, 1] or [-1, 1]. This makes it easier for the model to learn the relationships between features, improving training efficiency and prediction accuracy. After normalizing network traffic and device load data, their weights in model training are more balanced, allowing the model to better consider these factors in prediction. Through continuous training and optimization, the prediction error rate of the LSTM dynamic prediction module can be controlled to ≤5%. This means that the module's prediction results have high accuracy, providing a reliable basis for the allocation and scheduling of network resources in smart grids. In practical applications, accurate predictions can help power systems prepare network resources in advance, avoiding data transmission delays or losses due to network congestion, ensuring normal communication of power equipment and stable operation of the power system.

[0073] The multi-objective optimization decision-maker uses the NSGA-II (Non-dominated Sorting Genetic Algorithm II) genetic algorithm to solve the Pareto front, achieving multi-objective optimization of latency, energy consumption, and cost, with a decision cycle of ≤50ms. The multi-objective optimization decision-maker employs FPGA hardware acceleration algorithms, pre-sets 10+ power service decision-making strategies (such as "low latency priority" and "cost priority"), and supports manual policy intervention (such as forcibly activating satellite links during extreme weather).

[0074] The NSGA-II genetic algorithm is the core algorithm of a multi-objective optimization decision-maker, optimizing for latency, energy consumption, and cost. Latency affects the real-time performance of power data transmission, such as the transmission of relay protection signals; excessive latency can lead to delayed protection actions, impacting grid safety. Energy consumption relates to the operating costs of power and network equipment; reducing energy consumption helps improve energy efficiency. Cost involves using different network resources, such as the fees for network services from different operators. These objectives often conflict; for example, pursuing low latency may increase energy consumption and cost. The NSGA-II genetic algorithm searches for the Pareto front by simulating natural evolution. It first randomly generates an initial population, where each individual represents a possible network resource allocation scheme, including information such as which networks to use and how much bandwidth to allocate. Then, the individuals in the population are evaluated, and the fitness of each individual is calculated based on the objectives of latency, energy consumption, and cost. Next, individuals are divided into different levels through non-dominated sorting, with non-dominated individuals (i.e., individuals for which no other individual is superior in all objectives) at higher levels. Simultaneously, the crowding degree of each individual is calculated, reflecting the distribution of the individual within its rank, to maintain population diversity. Then, new populations are generated through selection, crossover, and mutation operations. Selection tends to choose individuals with high ranks and high crowding, crossover simulates gene exchange in biological inheritance, and mutation introduces new genes, increasing population diversity. After multiple generations of evolution, the population gradually approaches the Pareto front. Individuals on the Pareto front represent the set of optimal solutions that achieve balance between different objectives. In a smart grid, these optimal solutions can provide suitable network resource allocation schemes for different power services.

[0075] To meet the requirement of a decision cycle ≤50ms, the multi-objective optimization decision-maker employs FPGA hardware acceleration. FPGAs possess parallel processing capabilities, enabling them to handle multiple tasks simultaneously. During the multi-objective optimization decision-making process, many computational steps in the NSGA-II genetic algorithm, such as fitness calculation, non-dominated sorting, crossover, and mutation operations, can be implemented in parallel on the FPGA. Compared to traditional CPU serial computation, FPGA parallel processing significantly improves computation speed. When calculating fitness, the CPU needs to sequentially calculate the fitness value of each individual across all objectives, while the FPGA can perform calculations on multiple individuals simultaneously, thus significantly shortening the computation time. Through FPGA hardware acceleration, the multi-objective optimization decision-maker can complete complex multi-objective optimization calculations in a very short time, quickly providing the optimal decision scheme for network resource allocation, meeting the stringent real-time requirements of smart grids.

[0076] The multi-objective optimization decision-maker pre-sets 10+ power business decision-making strategies, such as "low latency priority" and "cost priority". These strategies are formulated according to the different characteristics and needs of power businesses, and help to achieve a balance among multiple objectives such as latency, energy consumption, and cost.

[0077] Low-latency priority strategy: Applicable to power services with extremely high real-time requirements, such as relay protection, rapid fault diagnosis and isolation in power grids. In these cases, ensuring the rapid and accurate transmission of control commands and fault information is crucial. The decision-maker will prioritize the path or combination of network resources with the lowest network latency, even if this may lead to increased energy consumption and costs. To ensure the rapid transmission of relay protection signals, low-latency 5G network slices or dedicated fiber optic links will be prioritized to ensure that protection devices can respond to faults in a timely manner, reducing the scope and duration of power outages.

[0078] Cost-first strategy: For some power services with low real-time requirements, such as data transmission during regular inspections of power equipment and status monitoring of non-critical equipment, cost becomes the primary consideration. The decision-maker will prioritize the lowest-cost network resources while meeting basic communication needs. Lower-cost LoRa networks or narrowband IoT networks are preferred for transmitting this type of data, even though their transmission speeds are relatively slower, as they significantly reduce communication costs.

[0079] Lowest energy consumption strategy: In some power business scenarios, especially for widely distributed power terminal equipment, such as smart meters and distributed energy monitoring equipment, energy consumption is a critical factor. The decision-maker will prioritize the network access method with the lowest energy consumption to extend the lifespan of the equipment and reduce overall energy consumption. For these devices, low-power short-range wireless communication technologies such as Bluetooth or Zigbee are preferred for data transmission. When long-distance transmission is required, low-power wide-area network (LPWAN) technologies are selected to reduce the frequency of battery replacements and energy consumption.

[0080] High reliability priority strategy: For critical services related to the stable operation of the power grid, such as the issuance and feedback of power grid dispatching instructions and real-time monitoring of large substations, reliability is paramount. The decision-maker will prioritize highly reliable network resources, such as fiber optic networks with redundant links or highly stable 5G private networks, to ensure that data is not lost or interrupted during communication. When constructing the power grid dispatching communication network, dual fiber optic links with redundancy are used, combined with a 5G private network as an emergency backup link, to improve communication reliability and ensure the safe and stable operation of the power grid.

[0081] Bandwidth maximization strategy: When power services require the transmission of large amounts of data, such as high-definition video surveillance in power systems and the transmission of power big data, the decision-maker will prioritize network resources that can provide the maximum bandwidth. For high-definition video surveillance services in substations, priority will be given to using 5G networks or high-speed fiber optic networks with larger bandwidth to ensure smooth transmission of video data and facilitate real-time monitoring of equipment operating status by maintenance personnel.

[0082] Real-time performance versus cost balance strategy: For some power services that require a certain level of real-time performance but also need to control costs, such as data interaction in the power marketing system and remote operation and maintenance of some equipment, the decision-maker will prioritize the selection of lower-cost network resources while ensuring that basic real-time requirements are met. In data interaction within the power marketing system, the cost-effective 4G network is given priority, and the allocation of network resources is flexibly adjusted according to changes in business traffic to control communication costs while ensuring real-time data transmission.

[0083] Reliability and Energy Consumption Balance Strategy: In certain power service scenarios, it is necessary to minimize energy consumption while ensuring reliability, such as for power monitoring equipment in remote areas. The decision-maker will select a network solution that meets reliability requirements while maintaining low energy consumption. Using solar-powered equipment paired with low-power, high-reliability satellite communication modules ensures reliable data transmission while reducing equipment energy consumption and improving equipment independence and stability.

[0084] Latency and Reliability Balancing Strategy: For services with high requirements for both latency and reliability, such as synchronous phasor measurement (PMU) data transmission in power systems, the decision-maker comprehensively considers latency and reliability factors to select the most suitable network resources. Priority is given to 5G-TSN (Time-Sensitive Networking) technology, which offers low latency and high reliability, or dedicated high-reliability, low-latency fiber optic networks to ensure timely and accurate transmission of PMU data, providing strong support for the stable operation of the power grid.

[0085] Dynamic Adaptive Strategy: Considering the dynamic nature of power services, the decision-maker dynamically adjusts its strategy based on real-time network conditions, changes in service demands, and other factors. During peak electricity consumption periods, when power data traffic increases significantly, the decision-maker automatically prioritizes ensuring the bandwidth and latency requirements of critical services. In the event of network failures or congestion, it dynamically switches to backup network resources to ensure service continuity.

[0086] Regional Differentiation Strategy: Develop differentiated decision-making strategies based on network coverage and the characteristics of power service needs in different regions. In urban areas with abundant network coverage, prioritize high-speed and stable network resources to meet the needs of urban-intensive power services. In remote rural or mountainous areas with relatively weak network coverage, prioritize adaptable network technologies, such as satellite communication or wireless ad hoc network technology, to ensure the normal operation of power services.

[0087] Differentiated Strategies for Different Types of Power Services: Develop specific decision-making strategies for different types of power services. For control-related services, such as remote switch control, prioritize low latency and high reliability; for data acquisition-related services, such as smart meter data acquisition, prioritize cost and energy consumption; for management-related services, such as office automation systems for power companies, prioritize network stability and bandwidth.

[0088] Multi-objective balancing strategy: In some cases, it is necessary to comprehensively balance multiple objectives such as latency, energy consumption, and cost, without favoring any particular objective. The decision-maker uses algorithms to calculate the optimal weights of each objective under current network conditions and service requirements, thereby selecting a network resource allocation scheme that performs well across multiple objectives. In typical power equipment condition monitoring services, the timeliness of data transmission is considered, while energy consumption and cost are also taken into account, selecting a suitable combination of network resources to achieve balanced optimization across multiple objectives.

[0089] The multi-objective optimization decision-maker also supports manual policy intervention, such as forcibly activating satellite links in extreme weather conditions. During the operation of a smart grid, various special circumstances may arise, such as typhoons and earthquakes, which may cause terrestrial networks like 5G, 4G, and WiFi to malfunction or experience significant performance degradation. In such situations, preset decision-making strategies may not meet the communication needs of power services. In these cases, manual intervention can be implemented to forcibly activate satellite links as a backup communication method. Satellite links offer advantages such as wide coverage and immunity to ground-based disasters, ensuring the transmission of power data even in extreme conditions. Through manual policy intervention, the multi-objective optimization decision-maker can more flexibly respond to complex and ever-changing realities, ensuring the reliability and stability of the smart grid communication system.

[0090] The resource coordination control layer includes a blockchain spectrum sharing ledger and an SDN network slicing controller.

[0091] The blockchain spectrum sharing ledger records the usage of spectrum resources by operators and uses smart contracts to automatically allocate and trade spectrum resources, ensuring transparency and trustworthiness in cross-network resource scheduling. This blockchain spectrum sharing ledger is built on the Hyperledger Fabric consortium blockchain and uses Solidity to write smart contracts, implementing a spectrum resource auction mechanism and usage permission management, with a transaction confirmation time of ≤2 seconds.

[0092] The blockchain-based spectrum sharing ledger is specifically designed to record the usage of spectrum resources by various operators. Each operator's spectrum resources are like their own "assets," and information such as the usage status, usage time, and user of these "assets" is recorded in detail in this ledger. If, during a specific time period, operator A uses a segment of spectrum resources in a particular area to transmit power data, this information will be accurately recorded. The advantage of this recording method is that all participating operators can view this information, achieving transparency. Moreover, due to the characteristics of blockchain, these records cannot be arbitrarily modified once written, ensuring the authenticity and reliability of the data and allowing each operator to clearly understand the overall usage of spectrum resources.

[0093] In spectrum resource management, smart contracts can automatically execute spectrum resource allocation and trading processes according to preset rules. When a new power service needs to use spectrum resources, the smart contract will automatically find suitable spectrum resources and complete the allocation based on the service's needs (such as bandwidth requirements, usage duration, etc.), the resource idleness of various operators, and preset allocation strategies. If spectrum resource trading is involved, the smart contract can also automatically execute the trading process, ensuring fairness, impartiality, and efficiency. The smart contract will automatically verify the identities and permissions of both parties to the transaction, ensuring the legality and compliance of the transaction, and then complete the transfer of ownership or usage rights of the spectrum resources. This automated allocation and trading method greatly improves the efficiency of spectrum resource management and reduces the uncertainty and errors caused by human intervention.

[0094] A key characteristic of consortium blockchains is their collaborative management by multiple organizations or institutions. Only consortium members can join and access them; in smart grid communications, various operators are members of the consortium blockchain. Through the Hyperledger Fabric consortium blockchain, operators can jointly maintain this spectrum-sharing ledger, ensuring its security and reliability. The consortium blockchain's consensus mechanism guarantees that all members agree on the records in the ledger, preventing data inconsistencies. A new record can only be added to the ledger after verification by all participating nodes. Furthermore, the consortium blockchain's privacy protection mechanism allows operators to protect their sensitive data from leakage while sharing information, such as their trade secrets and user privacy information.

[0095] Smart contracts written in Solidity can precisely define the allocation rules, transaction processes, and usage permission management mechanisms for spectrum resources. When defining the spectrum resource auction mechanism, Solidity code can be written to determine details such as the starting price, bidding rules, and auction duration. Regarding usage permission management, Solidity code can set different operators' and services' usage permissions for spectrum resources, such as restricting certain services to using specific frequency bands of spectrum resources only during specific time periods. The auction mechanism implemented through smart contracts allows for more rational allocation of spectrum resources. When spectrum resources are available for auction, interested operators can participate in bidding within a specified time. The smart contract will automatically determine the winner based on preset rules, such as the highest bidder receiving the right to use the resource, and complete the resource allocation and transaction. In terms of usage permission management, smart contracts can precisely control the spectrum resource usage permissions for each operator and each service. Based on the priority and needs of the services, different spectrum resource usage permissions can be allocated to different power services, with high-priority services receiving priority access to high-quality spectrum resources and guaranteed usage time and bandwidth.

[0096] Transaction confirmation time is a crucial metric for evaluating the performance of a blockchain-based spectrum sharing ledger. In this system, transaction confirmation time can be controlled to ≤2 seconds, meaning that spectrum resource allocation and trading can be completed in an extremely short time. When an operator requests to use new spectrum resources or conducts a spectrum resource transaction, the smart contract quickly executes the relevant verification and allocation operations, providing confirmation to both parties within 2 seconds. This efficient transaction confirmation mechanism ensures rapid allocation of spectrum resources, meets the real-time requirements of smart grid communication, ensures that power services can promptly acquire the necessary spectrum resources, and guarantees stable transmission of power data.

[0097] The SDN network slice controller is used to create dedicated virtual network slices for the power industry, allocating dedicated bandwidth and latency guarantees for different services, and achieving traffic isolation and QoS classification. The SDN network slice controller extends the slice management API based on the Open-Daylight controller.

[0098] One of the main functions of the SDN network slice controller is to create dedicated virtual network slices for the power industry. A network slice can be thought of as multiple virtual "small networks" divided from a physical network. Each slice is an independent logical network. The SDN network slice controller creates specialized virtual network slices based on the needs of different power services. For relay protection services, a slice with high reliability and low latency is created; for power equipment condition monitoring services, a slice that meets its data transmission volume requirements but has relatively lower latency requirements is created. In this way, different power services can transmit data in their own dedicated network slices, avoiding mutual interference. The SDN network slice controller allocates appropriate bandwidth and latency guarantees to each slice according to the importance and requirements of the service. For power services with extremely high real-time requirements, such as the transmission of grid dispatch instructions, larger bandwidth and strict latency guarantees are allocated to ensure that instructions reach the execution end quickly and accurately. For some services with lower real-time requirements, such as the data transmission of periodic inspections of power equipment, relatively smaller bandwidth is allocated to meet basic transmission needs while improving network resource utilization. Through this fine-grained resource allocation method, each power service can obtain the best network service quality in its own network slice.

[0099] SDN network slicing controllers ensure that traffic in one slice does not interfere with the normal operation of other slices through QoS classification. Just as vehicles in different lanes can only travel in their own lanes and will not cross into other lanes and disrupt traffic, QoS classification categorizes services into different levels based on their importance and requirements, providing different levels of service quality. Relay protection services are set to the highest level, receiving the highest priority in network resource allocation; general equipment status monitoring services are set to a lower level. Through traffic isolation and QoS classification, SDN network slicing controllers can provide customized network services for different power services, improving the reliability and stability of the entire smart grid communication system.

[0100] The SDN network slice controller is developed based on the OpenDaylight controller, extending its slice management API (Application Programming Interface) to achieve the creation, management, and configuration of network slices. The API allows developers to create dedicated functional modules for network slice management on top of the OpenDaylight controller, enabling operations such as creating, deleting, and modifying network slices, as well as allocating and managing slice resources. This approach fully leverages the advantages of the OpenDaylight controller while meeting the specific needs of smart grids for network slice management.

[0101] The edge execution layer includes local fast switching agents and protocol conversion gateways.

[0102] The local fast switching agent is used to execute switching commands at edge nodes such as substations and distribution stations, bypassing cloud approval to achieve hardware-level fast switching and ensuring hot switching with 0ms interruption. The local fast switching agent is deployed as a lightweight agent using Docker containers (developed in Go), combined with a PCIe hardware switching card (such as Mellanox ConnectX-7), supporting a switching latency of ≤8ms between 5G and WiFi.

[0103] Traditional network switching often requires cloud approval, which introduces latency and can affect the real-time performance and stability of power systems. The local fast switching agent bypasses this cloud approval process, operating directly on the local machine to achieve hardware-level fast switching. This fast switching ensures 0ms interruption during hot switching, meaning communication remains uninterrupted during network switching. Similar to how power systems prevent power outages even when power lines are replaced, this ensures stable transmission of power business data and avoids failures or data loss due to network interruptions. The local fast switching agent is deployed using Docker containers, a lightweight deployment method. Docker containers package applications and their dependencies into a single, standardized "box" that can be quickly deployed and run in different environments. This lightweight agent is developed in Go, a language known for its efficiency, simplicity, and high concurrency performance, making it ideal for developing performance-critical agent programs. Through Docker containerization, the lightweight agent can quickly start and run on edge nodes, unaffected by the underlying operating system and other applications, improving deployment flexibility and maintainability.

[0104] The local fast switching agent integrates with PCIe hardware switching cards, such as the Mellanox ConnectX-7. PCIe is a high-speed serial computer expansion bus standard characterized by high data transfer rates and low latency. The hardware switching card is a key hardware device for network switching, enabling rapid switching between different network interfaces. Under the control of the local fast switching agent, the PCIe hardware switching card can quickly switch communication connections from one network (such as 5G) to another (such as WiFi) or vice versa as needed. This hardware-level switching is fast and reliable, providing strong hardware support for achieving rapid switching.

[0105] The protocol conversion gateway is used to convert control signaling for different network standards, supporting real-time conversion of over 100 communication protocols. Developed based on the Data Plane Development Kit (DPDK) high-performance framework, the gateway utilizes polling-driven mode and large-page memory technology to achieve zero-copy data processing. It pre-stores a protocol template library and employs a state machine algorithm for signaling parsing and conversion, achieving a processing rate of ≥10Gbps. The zero-copy mechanism of DPDK avoids data copying between kernel space and user space, reducing protocol conversion latency.

[0106] The primary task of a protocol conversion gateway is to convert control signaling from one network standard to another. Taking the conversion from 5G NAS (Non-Access Stratum) signaling to LoRaWAN MAC (Media Access Control) layer protocol as an example, 5G NAS signaling is used for non-access stratum control between the core network and terminal equipment in 5G networks, managing functions such as mobility, sessions, and security; while the LoRaWAN MAC layer protocol is the protocol for device access and data transmission control in low-power wide-area networks (LoRa). When devices in a 5G network need to communicate with devices in a LoRa network, the protocol conversion gateway "translates" the 5G NAS signaling into a protocol format that the LoRaWAN MAC layer can understand, and vice versa. This allows devices of different network standards to communicate across network differences, expanding the coverage and device compatibility of smart grid communication. The protocol conversion gateway is developed based on the high-performance DPDK (Data Plane Development Kit) framework. DPDK is a collection of libraries and drivers for fast packet processing, designed to improve the performance of network data processing. DPDK plays a crucial role in protocol conversion gateways. It provides an efficient packet processing mechanism capable of rapidly receiving, processing, and sending network packets. By optimizing memory management and employing polling-based driving techniques, DPDK reduces interrupt handling overhead and improves the parallelism of data processing. This enables protocol conversion gateways to quickly parse and convert packets when processing large amounts of network data, meeting the real-time and efficiency requirements of smart grid communication.

[0107] The protocol conversion gateway has a pre-stored protocol template library containing the formats, rules, and conversion logic for various communication protocols. When it receives signaling that needs conversion, the gateway quickly retrieves the corresponding template from the library based on the signaling type. For example, when converting 5G NAS signaling to the LoRaWAN MAC layer protocol, the gateway finds the corresponding template and performs the signaling format conversion according to the preset rules within the template. This approach significantly improves the speed and accuracy of protocol conversion, avoiding the tedious process of re-analyzing and processing protocol rules for each conversion.

[0108] The protocol conversion gateway employs a state machine algorithm to implement signaling parsing and conversion. A state machine algorithm is a state transition-based algorithm that divides the signaling parsing and conversion process into different states. Upon receiving signaling, the state machine performs state transitions according to preset rules based on the signaling content and the current state. When parsing 5G NAS signaling, the state machine starts from the initial state and gradually transitions states based on the signaling fields and format, extracting key information. Then, based on this information and the requirements of the target protocol (such as the LoRaWAN MAC layer protocol), it performs signaling conversion. The state machine algorithm ensures that the signaling parsing and conversion process proceeds systematically, improving the reliability and stability of the conversion, and enabling accurate completion of the conversion even when handling complex signaling.

[0109] The emergency disaster recovery layer includes a drone mesh relay network and redundant satellite link channels.

[0110] The drone mesh relay network is used to establish temporary communication links through a drone swarm self-organizing network in the event of failures such as fiber optic cable interruptions, maintaining critical data transmission. The drone mesh relay network is equipped with a micro base station (weighing <500g, with a flight time of ≥2 hours), adopts an optimized version of the OLSRv2 self-organizing network algorithm, supports dynamic route adjustment, and has a fault recovery time of ≤3 minutes.

[0111] When fiber optic communication fails, critical data that relies on fiber optic transmission (such as operational status monitoring data of power equipment and power grid dispatch instructions) cannot be transmitted normally. At this time, a drone mesh relay network activates, constructing a temporary communication network in the air through a self-organizing drone swarm. These drones act as mobile communication nodes, cooperating to reconnect the interrupted communication links and ensure the continued transmission of critical data. In the event of damage to fiber optic lines due to natural disasters such as earthquakes, the drone mesh relay network can respond quickly, providing real-time information on power equipment to repair personnel, helping them restore power grid operation promptly. The micro base station weighs less than 500g, which minimizes the burden on drones during transport, ensuring their flight performance and flexibility. Simultaneously, the micro base station has a flight endurance of ≥2 hours, meaning that drones can continuously support the temporary communication network during emergency communications, ensuring communication stability. The micro base station has signal transmission and reception capabilities and can forward data signals from ground power equipment, receiving data signals and forwarding them to other drones or ground receiving equipment, achieving reliable data transmission.

[0112] OLSRv2 (Optimized Link State Routing Protocol version 2) is a routing protocol suitable for mobile ad hoc networks. In UAV mesh relay networks, OLSRv2 is optimized to better suit the dynamic flight environment of UAVs. This algorithm calculates the optimal routing path by collecting link state information between nodes (i.e., UAVs) in the network. When a UAV's position changes or a link fails, the optimized OLSRv2 algorithm can quickly detect these changes and recalculate the routing path, ensuring data can be transmitted via the new path. This ad hoc networking algorithm enables UAVs to automatically establish and maintain communication links without manual intervention, greatly improving the efficiency and reliability of emergency communication. In practical applications, the flight environment of UAVs is complex and variable. For example, UAVs may change their flight position due to factors such as wind or building obstructions, affecting the original communication links. In this case, the UAV mesh relay network can use the optimized OLSRv2 algorithm to perceive changes in network topology in real time and dynamically adjust routes. If the signal quality of a link deteriorates, the algorithm will automatically select other reliable links to transmit data, ensuring communication continuity. This dynamic routing adjustment capability enables the UAV Mesh relay network to operate stably in complex environments, providing strong support for emergency communications in smart grids.

[0113] The satellite link redundancy channel is used to provide emergency communication support in extreme situations, supports adaptive coding and modulation, and reduces satellite communication bandwidth consumption. This redundancy channel integrates the Iridium NEXT module and uses ACM technology to dynamically adjust the coding rate, saving 30% of satellite bandwidth compared to traditional solutions, and supporting concurrent voice, data, and video services.

[0114] In extreme situations such as powerful earthquakes or tsunamis that severely damage terrestrial communication networks or cause widespread power outages that render communication equipment inoperable, the satellite link redundancy channel will immediately activate. This ensures that critical data of the power system, such as grid dispatch instructions and emergency status information of power equipment, can still be transmitted between various nodes. The satellite link redundancy channel integrates the Iridium NEXT module, a next-generation satellite communication module of the Iridium system. It features global coverage, meaning that satellite signals can be received and communication connections established wherever an Iridium NEXT module is present, whether in remote mountainous areas, vast oceans, or other regions where terrestrial communication is difficult to reach.

[0115] Adaptive coding and modulation (ACM) technology dynamically adjusts the coding rate based on the real-time signal quality of the satellite link. When the satellite signal quality is good, the coding rate is increased, allowing more data to be transmitted in the same amount of time, thus improving communication efficiency. Conversely, when signal quality deteriorates due to weather conditions (such as heavy rain or sandstorms) or other factors, the coding rate is reduced to ensure data transmission accuracy and reduce the bit error rate. In heavy rain, when satellite signals are significantly interfered with, ACM technology automatically reduces the coding rate. Although the transmission speed decreases, it ensures correct data transmission. This dynamic adjustment of the coding rate effectively reduces the bandwidth consumption of satellite communication.

[0116] The security and trust layer includes a national cryptographic encryption engine and a device fingerprint authentication module.

[0117] The national cryptographic encryption engine is used to perform end-to-end encryption of control commands and equipment status data to prevent man-in-the-middle attacks and data tampering. This national cryptographic encryption engine is based on a hardware encryption chip using the SM4 algorithm, supports dynamic key updates (update cycle < 5 minutes), and has an encryption throughput of ≥ 200 Mbps, meeting the security protection requirements of power secondary systems.

[0118] Control commands, such as power grid dispatch commands, directly affect the operation of power equipment; equipment status data, such as transformer temperature and line current, reflect the real-time operating status of power equipment. If this data is illegally obtained or tampered with during transmission, it may lead to power system failures or security risks. The national cryptographic encryption engine encrypts this data, converting plaintext into ciphertext. Only the recipient with the correct key can decrypt and obtain the true data, ensuring the safe and stable operation of the power system.

[0119] The device fingerprint authentication module is used to identify legitimate devices through radio frequency signal characteristics, establishing a unique fingerprint for each device to prevent unauthorized access. The device fingerprint authentication system utilizes a convolutional neural network to extract IQ signal features.

[0120] Every device transmits radio frequency (RF) signals with unique characteristics, much like a human fingerprint. This system creates a unique "fingerprint" profile for each legitimate device. When a device attempts to connect, the system extracts the RF signal characteristics emitted by that device and compares them to the existing fingerprint profiles of legitimate devices. If they match, the device is considered legitimate and allowed to connect; otherwise, it is deemed illegitimate and access is denied. This method effectively prevents unauthorized devices from entering the system, ensuring its security and stability. The device fingerprint authentication module uses a convolutional neural network (CNN) to extract IQ signal features, including the amplitude and phase information of the RF signal. The system first collects a large amount of IQ signal data from legitimate devices as a training set to train the CNN. During training, the network learns the characteristic patterns of the IQ signals from legitimate devices. When a new device connects, the CNN analyzes its IQ signal, extracts features, and compares them with the trained patterns. The system has an accuracy rate of ≥99.3% in extracting features, which means that in most cases, the system can accurately identify the features of legitimate devices, providing a reliable basis for subsequent authentication.

[0121] In one embodiment, a fault-tolerant communication method for smart grids based on multi-carrier channel switching is provided, comprising:

[0122] Step 1: Multi-dimensional data collection.

[0123] Collect real-time network indicators and power grid equipment status data.

[0124] The multi-mode terminal gateway collects heterogeneous network signals and transmits them to edge nodes via a bus.

[0125] The power grid equipment status acquisition device acquires equipment data, marks relay protection signals and SCADA control commands as QoS Class 0, and the remaining data as QoS Class 1-3.

[0126] Output a structured dataset containing metadata such as timestamps, geographic locations, business types, and priority tags.

[0127] The collection of real-time network metrics covers key parameters of various network standards. For example, 5G's RSRP (Reference Signal Received Power) reflects the signal strength of the 5G network; WiFi's SNR (Signal-to-Noise Ratio) reflects the quality of the WiFi network; and 4G latency shows the delay in data transmission on the 4G network. The collection of power grid equipment status data includes circuit breaker location, which is related to the circuit on / off control of the power system; bus voltage, an important parameter for measuring power transmission stability; and equipment temperature, which reflects the operating status of the equipment; excessively high temperatures may indicate equipment failure. The multi-mode terminal gateway collects heterogeneous network signals at a period of 100ms and transmits the collected signals to edge nodes via the SPI bus to ensure the real-time performance and accuracy of the data. The power grid equipment status collector acquires equipment data via the Modbus TCP protocol and classifies and labels the data according to its importance. Relay protection signals and SCADA control commands, due to their extremely high requirements for real-time performance and reliability, are labeled as QoS Class 0 (highest priority), while other data are labeled as QoS Class 1-3 to ensure reasonable resource allocation during data transmission. The collected data will be output as a structured dataset, which includes a high-precision timestamp with an accuracy of 1μs to accurately record the time of data collection; latitude and longitude information of the geographical location to determine the location of the data source; business type, such as control or collection, to clarify the business scope to which the data belongs; and priority markers to reflect the importance of the data and provide basic data support for subsequent processing.

[0128] Step 2: Dynamic network quality assessment.

[0129] Input the structured dataset generated in step 1, which includes historical network performance data for the past 24 hours and weather forecasts for the next 2 hours; the federated learning collaborative module aggregates historical data from edge nodes of the three major operators, trains network quality prediction models, and outputs predicted latency and packet loss rates for each network for the next 5 minutes; the LSTM dynamic prediction module combines weather data to correct the wireless channel attenuation model; outputs a network quality scoring matrix, which includes latency scores, reliability scores, and bandwidth availability for 5G / 4G / WiFi / LoRa standards in different regions.

[0130] The input dataset contains historical network performance data from the past 24 hours, reflecting network performance trends over a period. It also includes weather forecasts for the next two hours (such as precipitation probability and wind speed), as weather factors can impact network signal transmission; for example, precipitation may cause signal attenuation, and high winds may interfere with wireless signals. The federated learning collaborative module aggregates historical data from edge nodes of the three major telecom operators. This data is anonymized to protect privacy. By learning from a large amount of historical data, a network quality prediction model is trained. This model can predict latency and packet loss rates for each network in the next five minutes, providing key indicators for network quality assessment. The LSTM dynamic prediction module incorporates weather data to correct the wireless channel attenuation model. For example, during thunderstorms, the predicted 5G signal attenuation increases by 20%. This weather-based correction more accurately assesses network performance under different environments. Finally, the output is a network quality score matrix, detailing latency scores, reliability scores, and bandwidth availability for different network standards such as 5G, 4G, WiFi, and LoRa in different regions, providing comprehensive network quality information for subsequent decision-making.

[0131] Optionally, this step can generate a network quality heatmap that visually displays the network reliability score (ranging from 0 to 100) for each region, intuitively showing the distribution of network quality in different regions.

[0132] Step 3: Multi-mode network switching decision.

[0133] Input the network quality scoring matrix, power grid service priorities, and cost constraints generated in step 2;

[0134] The population is initialized using the NSGA-II algorithm, with optimization objectives including minimizing total latency, minimizing energy consumption, and minimizing operator tariff costs. After multiple iterations, it converges to the Pareto front.

[0135] Output the optimal network combination scheme.

[0136] Based on the network quality score matrix generated in step 2, power grid service priorities, and cost constraints, a multi-objective optimization decision is made. Power grid service priorities are divided into control services (40% weight), data collection services (30% weight), and management services (30% weight), with the weights reflecting their importance in network resource allocation. Cost constraints include factors such as operator tariff packages and energy consumption budgets, ensuring reasonable cost control while meeting service needs. 100 populations are initialized using the NSGA-II algorithm, which aims to minimize total latency, energy consumption, and operator tariff costs. After 50 iterations, the algorithm converges to the Pareto front, finding a set of optimal solutions that balance different objectives. Finally, the optimal network combination scheme is output, clearly defining the primary link (e.g., a 5G slice from operator A, dedicated to QoS Class 0), backup links (e.g., a Mesh drone network), and auxiliary links (e.g., LoRaWAN for data collection services), providing a specific scheme for network resource allocation.

[0137] Optionally, this step can be intervened through a manual policy library. For example, during typhoon warnings, considering the special needs of coastal substations, satellite links can be forcibly allocated as backup channels to ensure communication reliability. During peak load periods, high-bandwidth 5G slices can be prioritized based on service priority and network quality to ensure the normal operation of important services.

[0138] Step 4: Cross-domain resource collaborative scheduling.

[0139] Input the network combination scheme generated in step 3, including the target operator, frequency band requirements, bandwidth threshold, and latency requirements; query the availability of the target frequency band through the blockchain spectrum sharing ledger, lock the operator's frequency band through a smart contract, and return a resource lock certificate after transaction confirmation; the SDN network slicing controller sends flow tables to the operator's core network, reserves bandwidth for control services, and sets the maximum latency; edge nodes preload switching configuration parameters; output a resource ready status signal, and trigger an emergency plan if resource allocation is not completed within 30 seconds.

[0140] The availability of target frequency bands is queried through the blockchain spectrum sharing ledger. For example, a smart contract can be used to lock the 3.5GHz band (10MHz bandwidth) of operator B, ensuring the reasonable allocation and use of frequency band resources. After transaction confirmation, a resource lock certificate is returned to ensure the legality and reliability of resource allocation. The SDN network slicing controller sends flow tables to the operator's core network, for example, reserving 50Mbps bandwidth for 5G slices for control services and setting a maximum latency of 10ms to meet the real-time and bandwidth requirements of control services. Edge nodes preload handover configuration parameters, such as APN: power-grid-ctrl, authentication keys, etc., to prepare for network handover. Finally, a resource readiness status signal, such as Ready or Timeout, is output. If resource allocation is not completed within 30 seconds, an emergency plan is triggered to ensure timely allocation of network resources and normal operation of services.

[0141] Step 5: Seamlessly switch execution.

[0142] Input resource readiness signal, including configuration parameters for the primary link and backup link;

[0143] Local fast agent switching triggers PCIe hardware switching card to achieve interrupt switching at the physical layer while maintaining TCP session continuity;

[0144] The protocol conversion gateway completes the signaling conversion and uses a duocast mechanism to ensure zero packet loss.

[0145] After the switch is complete, the edge nodes report the switch performance data back to the AI ​​decision engine;

[0146] The output switching is complete, accompanied by a real-time performance report.

[0147] When the input resource is ready, including configuration parameters for the primary and backup links, the local fast switching agent triggers the PCIe hardware switching card to achieve zero-ms interruption switching between 5G and WiFi links at the physical layer, ensuring communication continuity. Simultaneously, DPDK technology maintains TCP session continuity, preventing data transmission interruptions. The protocol conversion gateway completes signaling conversion within 50ms, for example, converting a 5G RRC connection release signaling into a WiFi association request, ensuring smooth conversion between communication protocols of different network standards. A dual-cast mechanism is employed, such as parallel transmission of data between the old and new links for 1 second, using both links to transmit data simultaneously during the handover process to ensure zero packet loss and improve handover reliability. After handover, the edge node feeds back handover performance data (e.g., 8ms latency, 0% packet loss) to the AI ​​decision engine, providing data support for subsequent network optimization. Finally, a handover completion confirmation command is output, along with a real-time performance report, including signal strength, latency, and bit error rate, enabling the system to promptly understand the network performance after the handover.

[0148] Step 6: Emergency disaster recovery and self-healing.

[0149] Input network anomaly alarms, including fault location, affected area, and service interruption type;

[0150] The drone mesh relay network automatically responds and / or the satellite link redundancy channel is activated;

[0151] Output the emergency network connectivity status.

[0152] When a network anomaly alarm is input (such as fiber optic link interruption or 5G base station outage), containing information such as fault location, cause, affected area, and service interruption type, the system will respond quickly. The drone mesh relay network automatically responds, with three drones within 5km of the fault location taking off to construct a shortest-path mesh network using the OLSRv2 optimization algorithm, with relay node spacing ≤1km. SCADA data transmission is restored within 3 minutes, ensuring normal transmission of monitoring data for the power system. The satellite link redundancy channel is activated, requesting temporary bandwidth from the Iridium system (1Mbps of free basic bandwidth provided for the first 10 minutes). Adaptive coding modulation (ACM) technology is used to dynamically adjust the coding rate based on satellite signal quality. For example, QoS modulation is used when SNR > 15dB, and BPSK modulation is switched when SNR < 10dB, ensuring effective data transmission under different signal quality conditions. Finally, the emergency network connectivity status is output, such as "Mesh link RSSI: -75dBm, satellite link latency: 250ms, service recovery rate 100%", reflecting the real-time operation of the emergency network.

[0153] In an optional embodiment, step 7 is also included: security hardening and feedback optimization.

[0154] Input real-time running data;

[0155] The national cryptographic encryption engine dynamically rotates session keys;

[0156] Device fingerprint authentication module updated;

[0157] Incremental updates of federated learning models;

[0158] Output system health score and optimization suggestions.

[0159] The system receives real-time operational data, including switchover logs, encrypted traffic statistics, and device authentication records, which include security events from the past hour (such as three unauthorized access attempts). The national cryptographic encryption engine dynamically rotates session keys, generating a new SM4 key every 4 minutes and distributing it to terminal devices via out-of-band channels (such as satellite SMS). Upon detecting a replay attack, a key update is immediately triggered to ensure data transmission security. The device fingerprint authentication module is updated, adding the IQ signal characteristics of newly connected legitimate devices to the fingerprint database while removing abnormal terminals that have failed authentication three times consecutively, ensuring the accuracy and security of the fingerprint database. The fingerprint database update time is ≤30 seconds, improving authentication efficiency. The federated learning model performs incremental updates, collecting the latest prediction error data from edge nodes and fine-tuning the model hourly to reduce the mean squared error by ≤10%, improving network quality prediction accuracy. The final output is a system health score (range 0-100) and optimization suggestions, such as "Suggest adding weather data input dimensions to the LSTM model; current health score: 92 points," providing guidance for continuous system optimization and ensuring the safe and stable operation of the smart grid communication system.

[0160] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 2As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for calculating the lifespan of surgical instruments and a method for controlling surgical instruments. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0161] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0162] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0163] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0164] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0165] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It is understood that this invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of this invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of this invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are protected by this invention.

Claims

1. A fault-tolerant communication system for smart grids based on multi-carrier channel switching, characterized in that, include: The intelligent sensing layer is configured with a multi-mode terminal access gateway and a power grid equipment status collector. The multi-mode terminal access gateway adopts an embedded multi-band radio frequency chip, which supports the simultaneous collection of real-time network indicators of multiple communication standards. The power grid equipment status collector is used to collect power grid equipment status data. The AI ​​decision engine includes a federated learning collaboration module, an LSTM dynamic prediction module, and a multi-objective optimization decision-maker. The federated learning collaboration module is used to achieve differential privacy protection and generate a network quality prediction model jointly trained across operator edge nodes. The LSTM dynamic prediction module takes into input a multi-dimensional feature vector containing weather data including temperature, humidity, and precipitation probability, as well as equipment load data, and constructs a wireless channel attenuation compensation model to predict network load. The multi-objective optimization decision-maker is based on the NSGA-II multi-objective optimization algorithm, with the optimization objectives of minimizing total latency, minimizing energy consumption, and minimizing operator tariff costs, and iteratively generates a Pareto optimal solution set. The resource coordination control layer includes a blockchain spectrum sharing ledger and an SDN network slice controller. The blockchain spectrum sharing ledger realizes spectrum resource auction and allocation through consortium blockchain and smart contracts. The SDN network slice controller sends flow tables to the operator's core network to reserve bandwidth and maximum latency for the corresponding slices for services. The edge execution layer includes a local fast switching agent and a protocol conversion gateway. The local fast switching agent integrates a PCIe hardware switching card and realizes physical layer link switching based on dual-link pre-synchronization technology. The service interruption time during the switching process does not exceed 5ms, and TCP session continuity is maintained through the data plane development kit. The protocol conversion gateway has a built-in multi-standard protocol conversion engine and realizes protocol conversion between different network standards based on a pre-stored protocol template library. The emergency disaster recovery layer is constructed by heterogeneously integrating a low-altitude UAV mesh relay network and a satellite link redundant channel to build an integrated air-ground communication network with fault self-healing capabilities. The UAV mesh relay network is equipped with micro base stations and uses the OLSRv2 routing protocol to build the shortest path mesh network. The satellite link redundant channel integrates a satellite communication module and dynamically adjusts the coding rate according to the satellite signal quality through adaptive coding and modulation technology. The security and trust layer includes a national cryptographic encryption engine and a device fingerprint authentication module. The national cryptographic encryption engine is used to dynamically rotate the encryption key, and the device fingerprint authentication module extracts the radio frequency signal features of newly accessed devices to generate device fingerprints and completes device access authentication through feature matching algorithms.

2. The system according to claim 1, characterized in that, The multi-mode terminal access gateway of the intelligent sensing layer supports the simultaneous collection of real-time network indicators of at least three different communication standards, and the power grid equipment status collector supports the parsing of equipment data of at least two protocols.

3. The system according to claim 1, characterized in that, The federated learning collaboration module aggregates historical data from no fewer than three operators and employs differential privacy technology to ensure that the original data does not leave the local machine. The root mean square value of the prediction error of the LSTM dynamic prediction module does not exceed 15%.

4. The system according to claim 1, characterized in that, The smart contract of the blockchain spectrum sharing ledger allocates spectrum resources through a Dutch auction mechanism, and generates a digital certificate containing the resource usage period after the transaction is confirmed; the SDN network slicing controller reserves no less than 50Mbps of bandwidth resources for QoS Class 0 services and sets the maximum latency threshold to 20ms.

5. The system according to claim 1, characterized in that, The hardware switching card for the local fast switching agent supports a PCIe 3.0 interface; the protocol conversion gateway supports real-time conversion of no less than 10 communication protocols.

6. The system according to claim 1, characterized in that, The relay nodes of the UAV Mesh relay network are spaced no more than 1km apart, and the shortest path constructed using the OLSRv2 protocol has no more than 5 hops. The adaptive coding and modulation technology of the satellite link supports dynamic switching between three modulation modes: QPSK, 16QAM, and 64QAM.

7. The system according to claim 1, characterized in that, The national cryptographic encryption engine generates new SM4 keys at regular intervals and distributes the keys through out-of-band channels. The device fingerprint authentication module updates in real time and automatically removes terminal devices that fail to authenticate three times consecutively.

8. A fault-tolerant communication method for smart grids based on multi-operator channel switching, characterized in that, Includes the following steps: Data Acquisition and Priority Marking: Real-time quality data from various heterogeneous networks are collected through a multi-mode terminal access gateway. The power grid equipment status collector parses data from various devices and marks time-sensitive data as QoS Class 0 based on the IEC 61850 standard. Dynamic network quality assessment: The federated learning module aggregates historical data from no fewer than three operators, uses differential privacy technology to protect the original data, and generates a cross-domain network quality prediction model; the LSTM model takes a multi-dimensional feature vector containing weather data and device load data as input, constructs a wireless channel attenuation compensation model, and outputs a latency reliability score matrix. Multi-objective optimization decision-making: Based on the NSGA-II multi-objective optimization algorithm, with constraints of total latency less than or equal to 50ms, single device power consumption less than or equal to 100mW, and operator tariff cost reduction of 20%, the Pareto optimal network combination solution set is generated, supporting priority intervention decision-making by manual policy library; Cross-domain resource collaborative scheduling: Spectrum resources are allocated through the auction mechanism of blockchain smart contracts, and resource lock certificates are generated after the transaction is confirmed; the SDN controller reserves no less than 50Mbps of bandwidth resources and sets a maximum latency of 20ms according to the QoS Class 0 service requirements to complete the network slice creation; Seamless hardware-level switching execution: Triggers the PCIe hardware switching card to realize physical layer link switching, maintains TCP session continuity through DPDK technology; adopts a dual-link synchronous transmission mechanism, and the parallel transmission time of the old and new links is greater than or equal to the preset time; Emergency disaster recovery and self-healing: When a network failure is detected, the drone Mesh relay network is activated, and the shortest path forwarding route is built within 3 minutes using the OLSRv2 protocol to restore SCADA data transmission; the satellite link uses adaptive coding and modulation technology to dynamically adjust the coding rate according to SNR to achieve bandwidth optimization; Security hardening and model update: The national cryptographic encryption engine generates new SM4 keys at regular intervals and distributes them to terminal devices through an independent out-of-band channel; the device fingerprint database updates the radio frequency signal characteristics of access devices in real time, and blocks access for terminals that fail to authenticate three times in a row; the prediction model is updated based on the incremental learning algorithm.

9. The method according to claim 8, characterized in that, In the data acquisition and priority marking steps: The multi-mode terminal access gateway supports the simultaneous collection of network indicators for no less than three communication standards, with a sampling frequency no lower than the preset frequency.

10. The method according to claim 8, characterized in that, In the dynamic network quality assessment steps: The global aggregation cycle of the federated learning model is 5 minutes, and the input feature vector dimension of the LSTM model is no less than 8 dimensions.

11. The method according to claim 8, characterized in that, In the multi-objective optimization decision-making steps: The NSGA-II algorithm initializes a population of 100-200, performs at least 50 iterations, and outputs at least 10 Pareto optimal solutions.

12. The method according to claim 8, characterized in that, In the cross-domain resource collaborative scheduling steps: The spectrum auction cycle for blockchain smart contracts is 1 minute, and the slice creation latency of the SDN controller does not exceed the threshold.

13. The method according to claim 8, characterized in that, In the hardware-level seamless switching execution step: The parallel transmission time of the dual-cast mechanism is dynamically adjusted by the network load, and a performance report including latency and packet loss rate is generated after the switch is completed.

14. The method according to claim 8, characterized in that, In the aforementioned emergency disaster recovery and self-healing steps: The coverage radius of the drone mesh relay network is no less than 5km, and the adaptive coding and modulation response time of the satellite link is no more than 50ms.

15. The method according to claim 8, characterized in that, In the security hardening and model update steps: The out-of-band distribution channel for the SM4 key is encrypted using AES-256.

Images