5GCpe dynamic networking optimization system and method based on SDN technology

Abstract

This invention discloses a 5GCPE dynamic networking optimization system and method based on SDN technology. The system includes an SDN controller, dynamic resource scheduling, QoS policy management, intelligent network configuration engine, fault self-healing, multicast routing management, and diagnostic analysis modules. The SDN controller collects data, the dynamic resource scheduling module achieves link load balancing and fast switching, and its integrated engine and accelerator reduce path calculation latency. The QoS policy management module allocates bandwidth and establishes queues according to service characteristics, and the holographic service perception unit reconstructs service flow characteristics. The fault self-healing module completes fault detection and switching within 50ms, the digital twin prediction system simulates link status, and the multicast routing management module updates the routing table to ensure transmission continuity. The diagnostic analysis module generates a diagnostic report. This invention improves the overall utilization efficiency of network resources, meets the diverse needs of multi-service concurrency, optimizes network performance, improves resource utilization, and ensures stable network operation.

Description

Technical Field

[0001] This invention relates to the field of communication network applications, and in particular to a 5GCPE dynamic networking optimization system and method based on SDN technology. Background Technology

[0002] With the rapid development of 5G communication technology, CPE devices, as a crucial bridge connecting users and 5G networks, directly impact user experience and service quality through their performance and network quality. While existing CPE devices possess basic network access and handover capabilities, challenges remain in dynamic handover between multiple operator networks, intelligent resource allocation during multi-service concurrency, and rapid fault recovery in distributed scenarios.

[0003] Common CPE devices rely on manual configuration for network switching, have fixed resource allocation strategies, and lack centralized fault monitoring and recovery mechanisms. Manually configured network switching cannot respond to changes in network quality in real time; static resource allocation is difficult to adapt to the differentiated needs of multiple concurrent services; and decentralized device management leads to low fault recovery efficiency, affecting the overall stability of the network and failing to meet the working requirements of communication network applications. To address this, a 5G CPE dynamic networking optimization system and method based on SDN technology is proposed. Summary of the Invention

[0004] This invention provides the following technical solution: a 5GCPE dynamic networking optimization system based on SDN technology, comprising: The system comprises an SDN controller module, a dynamic resource scheduling module, a QoS policy management module, an intelligent network configuration engine, a fault self-healing module, a multicast routing management module, and a diagnostic analysis module. The SDN controller module is used to collect real-time network traffic data, including the service traffic characteristics and network quality indicators of each CPE device. The dynamic resource scheduling module is used to realize 5G / 4G link load balancing and millisecond-level network switching decisions. The dynamic resource scheduling module integrates a Dijkstra algorithm optimization engine and a neuromorphic computing accelerator. The neuromorphic computing accelerator adopts a spiking neural network architecture and realizes in-memory computing through a memristor cross array, which will reduce path calculation latency. The QoS policy management module is used to configure differentiated bandwidth guarantee policies according to different service characteristics and establish priority queues. The QoS policy management module integrates a holographic service perception unit, which is used to reconstruct the three-dimensional features of service flows through terahertz holographic imaging technology. The intelligent network configuration engine is used to interact with the OpenWrt system through the northbound interface and dynamically adjust 5GSA / NSA network parameters, Wi-Fi 6 channel allocation and VPDN tunnel encryption parameters. The fault self-healing module monitors network link status based on the BFD protocol, enabling fault detection and backup link switching within 50ms. The fault self-healing module is equipped with a digital twin prediction system, which is used to construct a network digital twin through the LSTM network and simulate the status change trend of each link in real time. The multicast routing management module is used to dynamically update the multicast routing table when a fault occurs to ensure the continuity of data transmission paths. The diagnostic analysis module is used to generate an intelligent diagnostic report that includes a fault location heatmap. The diagnostic analysis module is equipped with a causal reasoning engine.

[0005] This invention provides a dynamic networking optimization method for 5GCPE based on SDN technology. Based on the aforementioned dynamic networking optimization system for 5GCPE based on SDN technology, the method includes the following steps: S1 Full Network Data Collection and Cleaning: The SDN controller module collects the service traffic characteristics and network quality indicators of all CPE devices in the network in real time, and uses a wavelet transform-based filtering algorithm to clean the collected data and remove outliers and noise data. S2 Intelligent Path Optimization Calculation: The data processed in step S1 is input into the dynamic resource scheduling module, which performs path calculation by integrating the Dijkstra algorithm optimization engine and the neuromorphic computing accelerator. S3 Service Flow Three-Dimensional Classification: The network status data generated in step S2 is transmitted to the QoS policy management module. The QoS policy management module reconstructs the three-dimensional features of the service flow using terahertz holographic imaging technology, classifies the service flow using the support vector machine algorithm, and establishes differentiated bandwidth guarantee policies and priority queues based on the classification results. S4 dynamic parameter configuration: Based on the QoS policy generated in step S3, the 5GSA / NSA network parameters, Wi-Fi 6 channel allocation, and VPDN tunnel encryption parameters are dynamically adjusted through the intelligent network configuration engine. S5 rapid fault self-healing: Based on the network parameters configured in step S4, the fault self-healing module monitors the link status in real time through the BFD protocol and predicts link status changes in conjunction with the network digital twin constructed by the digital twin prediction system. It completes fault detection and backup link switching within 50ms. At the same time, the multicast routing management module dynamically updates the multicast routing table. S6 Intelligent Diagnostic Analysis: The fault handling result of step S5 is transmitted to the diagnostic analysis module, which generates an intelligent diagnostic report containing a fault location heatmap and analyzes the root cause of network anomalies through a causal reasoning engine.

[0006] Preferably, the data output terminal of the SDN controller module is connected to the control input terminal of the dynamic resource scheduling module via the OpenFlow protocol interface; the policy configuration terminal of the dynamic resource scheduling module is connected to the policy synchronization terminal of the QoS policy management module via the RESTAPI interface; the path calculation output terminal of the dynamic resource scheduling module is connected to the update trigger terminal of the multicast routing management module via a message queue; the status monitoring terminal of the intelligent network configuration engine is connected to the link detection terminal of the fault self-healing module via the SNMP protocol; the warning output terminal of the fault self-healing module is connected to the diagnostic analysis module via a bus; the route update terminal of the multicast routing management module is connected to the SDN controller module via the BGP protocol; and the optimization suggestion terminal of the diagnostic analysis module is connected to the parameter optimization terminal of the dynamic resource scheduling module via a feedback loop.

[0007] Preferably, the SDN controller module integrates a data cleaning unit, which is used to clean the collected network traffic data to remove outliers and noise data. The data cleaning unit adopts a wavelet transform-based filtering algorithm.

[0008] Preferably, the neuromorphic computing accelerator includes 128 neuronal cores and 1024 synaptic units, and the neuromorphic computing accelerator is equipped with an online learning mode.

[0009] Preferably, in addition to reconstructing the three-dimensional features of the service flow through terahertz holographic imaging technology, the holographic service perception unit is also equipped with a feature classifier. The feature classifier is based on the support vector machine algorithm and is used to classify the reconstructed three-dimensional features so as to configure differentiated bandwidth guarantee strategies for different types of service flows.

[0010] Preferably, the communication interface of the digital twin prediction system is connected to a meteorological data acquisition module via a data cable, and the meteorological data acquisition module is used to acquire real-time meteorological data.

[0011] Preferably, the intelligent network configuration engine is equipped with an encrypted interaction protocol enhancement unit, which adopts an asynchronous transmission method and has a data cache pool.

[0012] Preferably, the data cleaning process in step S1 specifically involves data purification through a three-stage filtering mechanism. The first stage of the three-stage filtering mechanism uses sliding window mean filtering to remove impulse noise. The second stage of the three-stage filtering mechanism applies the db4 wavelet basis function of wavelet transform for frequency domain filtering. The third stage of the three-stage filtering mechanism eliminates data offset through Z-score normalization.

[0013] Preferably, the neuromorphic computing accelerator in step S2 operates by mapping network topology data to a spiking neural network composed of 128 neuron cores, and extracting topological features in parallel through 1024 memristor synaptic units.

[0014] In summary, compared with the prior art, the present invention provides a 5GCPE dynamic networking optimization system and method based on SDN technology, which has the following beneficial effects: 1. The present invention can realize 5G / 4G link load balancing and millisecond-level network switching decisions through the added dynamic resource scheduling module. The Dijkstra algorithm optimization engine and neuromorphic computing accelerator integrated inside the dynamic resource scheduling module can reduce path calculation latency, enabling the network to quickly switch networks according to the real-time network quality status, ensuring that the network performance is always in the optimal state, and avoiding the network performance degradation caused by the lag of manual switching. 2. The present invention, through the added QoS policy management module, can configure differentiated bandwidth guarantee policies according to different service characteristics and establish priority queues. The holographic service perception unit integrated inside the QoS policy management module reconstructs the three-dimensional features of the service flow through terahertz holographic imaging technology, which can accurately identify the needs of different services. This allows network resources to be dynamically allocated according to the actual needs of different services. Whether it is a service with high bandwidth requirements or a service that is sensitive to latency, reasonable resource allocation can be obtained, which improves the overall utilization efficiency of network resources and meets the diverse needs of multiple services concurrently. 3. This invention, through the addition of a fault self-healing module and a multicast routing management module, can achieve fault detection and backup link switching within 50ms. The digital twin prediction system inside the fault self-healing module constructs a network digital twin through an LSTM network, simulates the status change trends of each link in real time, and predicts possible faults in advance, thereby enabling more rapid response measures. Furthermore, the added multicast routing management module can dynamically update the multicast routing table when a fault occurs, ensuring the continuity of data transmission paths, thereby improving the efficiency of fault recovery, reducing the impact of faults on network stability, and ensuring the overall stable operation of the network. Attached Figure Description

[0015] Figure 1 This is a system structure block diagram of the present invention.

[0016] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This invention provides a technical solution: a 5GCPE dynamic networking optimization system based on SDN technology, comprising: The SDN controller module comprises a dynamic resource scheduling module, a QoS policy management module, an intelligent network configuration engine, a fault self-healing module, a multicast routing management module, and a diagnostic analysis module. The SDN controller module collects real-time network traffic data, including service traffic characteristics and network quality indicators for each CPE device. Internally, the SDN controller module integrates a data cleaning unit to clean the collected network traffic data, removing outliers and noise. The data cleaning unit employs a wavelet transform-based filtering algorithm. The dynamic resource scheduling module is used to realize 5G / 4G link load balancing and millisecond-level network switching decisions. The dynamic resource scheduling module integrates a Dijkstra algorithm optimization engine and a neuromorphic computing accelerator. The neuromorphic computing accelerator adopts a spiking neural network architecture and realizes in-memory computing through a memristor cross array, which will reduce path calculation latency. The neuromorphic computing accelerator contains 128 neuron cores and 1024 synaptic units. The neuromorphic computing accelerator has an online learning mode. The QoS policy management module is used to configure differentiated bandwidth guarantee policies according to different service characteristics and establish priority queues. The QoS policy management module integrates a holographic service perception unit, which is used to reconstruct the three-dimensional features of service flows through terahertz holographic imaging technology. In addition to reconstructing the three-dimensional features of service flows through terahertz holographic imaging technology, the holographic service perception unit is also equipped with a feature classifier. The feature classifier is based on the support vector machine algorithm and is used to classify the reconstructed three-dimensional features so as to configure differentiated bandwidth guarantee policies for different types of service flows. The intelligent network configuration engine is used to interact with the OpenWrt system through the northbound interface and dynamically adjust 5G SA / NSA network parameters, Wi-Fi 6 channel allocation and VPDN tunnel encryption parameters. The intelligent network configuration engine has an internal encryption interaction protocol enhancement unit, which adopts an asynchronous transmission method and has an internal data cache pool. The specific dimensions of the 3D feature reconstruction in this scheme are: Traffic volume: The instantaneous bandwidth usage or data volume of the business flow.

[0019] Delay sensitivity: The tolerance of a service for transmission latency (e.g., real-time video vs. file download).

[0020] Protocol type: The communication protocol used by the business flow (such as TCP, UDP, HTTP, etc.).

[0021] Additional explanation of feature extraction logic: Temporal characteristics: Analyze the temporal behavior of business flows (e.g., periodicity, burstiness). For example: Periodic heartbeat packets → IoT device data reporting.

[0022] Burst transmission → Live stream comments or multi-person collaborative editing.

[0023] Spatial characteristics: By combining network topology and terminal location information (such as the area where the CPE device is located), the service type can be inferred. For example: Industrial park CPE → Industrial Internet of Things protocols (MQTT, CoAP).

[0024] Home CPE → Video streaming (Netflix, YouTube).

[0025] Contextual features: Introduce external data to assist in classification (such as user subscription services, time-sensitive businesses). For example: High bandwidth demand during work hours → access to enterprise cloud services.

[0026] Low traffic at night → Home security monitoring data transmission.

[0027] Physical layer characteristics: Extracting finer-grained features using terahertz holographic imaging technology: Signal phase distribution: distinguishes the modulation method (e.g., QAM64 vs. QPSK).

[0028] Multipath effect intensity: Identify the type of service in a dense multipath environment (e.g., VR vs. voice call).

[0029] Coordination mechanism between asynchronous transmission and synchronous control logic of SDN controller I. Collaborative Architecture Design Dual-mode transmission channel Synchronous control channel: Based on the OpenFlow protocol, the SDN controller sends flow table rules to network devices at 10ms intervals to ensure QoS of critical services (such as VoNR voice).

[0030] Asynchronous event channel: Network devices (such as 5GCPE) report link status changes (such as channel attenuation caused by heavy rain) in real time through asynchronous metadata defined in P4 language, triggering the controller to dynamically adjust the route.

[0031] Dynamic management of buffers Switch-level buffer: Deploy a 128KB shared buffer to temporarily store asynchronously arriving data packets, waiting for the SDN controller to issue forwarding rules.

[0032] Controller-level buffer: The SDN controller maintains a global event queue and processes asynchronous events according to priority (fault alarm > traffic adjustment > configuration update) to ensure that high-priority events (such as base station downtime) are responded to within 5ms.

[0033] II. Details of the transmission mechanism Asynchronous event reporting process Triggering conditions: Asynchronous meta-data packets are generated when a network device detects the following events: Link quality degrades (e.g., Wi-Fi 6 signal-to-noise ratio <10dB). Equipment malfunction (e.g., 5GCPE antenna overheating) Security threats (such as a sudden surge in DDoS attack traffic) Metadata format: Includes fields such as event type, occurrence time, device ID, severity (level 1-5), encapsulated via UDP protocol, with a fixed port number of 6633 (compatible with OpenFlow).

[0034] Synchronous control logic adjustment Dynamic flow table update: After receiving an asynchronous event, the SDN controller adjusts the flow table through the following steps: Verify the legitimacy of events (e.g., prevent forgery through digital signatures). Query the global network topology and calculate alternative paths (using the Dijkstra algorithm optimization engine). New flow table rules are issued with higher priority than rules issued during the synchronization cycle (to ensure priority for failover).

[0035] Timed Synchronization Compensation Synchronization period adjustment: During periods of high incidence of asynchronous events (such as during rainstorm warnings), the SDN controller automatically shortens the synchronization period to 5ms to ensure state consistency.

[0036] State rollback mechanism: If an asynchronous event causes network inconsistency, the controller rolls back to the most recent consistent state using timestamps to avoid route oscillations.

[0037] Asynchronous transmission, through dual-mode channels, dynamic buffering, and timed synchronization mechanisms, forms a highly efficient synergy with the synchronization control logic of the SDN controller. Actual deployment verification demonstrates its stable operation in microsecond-level latency-aware networks with thousands of nodes, meeting the core requirements of high reliability and low interruption rate for dynamic 5GCPE networking.

[0038] The fault self-healing module monitors network link status based on the BFD protocol, enabling fault detection and backup link switching within 50ms. Internally, the module is equipped with a digital twin prediction system, which constructs a network digital twin using an LSTM network to simulate real-time trends in link status changes. The communication interface of the digital twin prediction system is connected to a meteorological data acquisition module via a data cable. This module acquires real-time meteorological data. A multicast routing management module dynamically updates the multicast routing table when a fault occurs, ensuring data transmission path continuity. Meteorological data profoundly influences network optimization strategies through the following key means: 1. Link quality prediction Dynamic parameter adjustment: Real-time meteorological data (such as rainfall and temperature) is input into the digital twin model to predict the link attenuation trend and dynamically optimize beamforming and frequency band switching (such as switching to low frequency bands when rain attenuation occurs).

[0039] 2. Environmental Fault Early Warning Proactive defense mechanism: When a weather warning (such as a typhoon or blizzard) triggers a link health assessment, the system will switch to an alternative path or activate an emergency communication plan in advance to ensure the continuity of critical business operations.

[0040] 3. Dynamic resource scheduling Service priority adaptation: Adjust QoS policy according to weather scenarios: In severe weather, priority will be given to ensuring the operation of high-priority services such as emergency communications and monitoring.

[0041] Dynamically allocate anti-interference frequency bands (such as enabling Sub-6GHz during sandstorms).

[0042] 4. Root cause analysis of the failure Meteorological correlation diagnosis: Correlate fault events with meteorological data (such as connection failure caused by lightning strikes) to optimize root cause location efficiency; historical data analysis can identify meteorologically sensitive areas and guide redundant design.

[0043] In this solution, meteorological data drives the network to shift from "passive response" to "active defense," achieving efficient resource utilization and ensuring business continuity.

[0044] The diagnostic analysis module is used to generate intelligent diagnostic reports that include fault location heatmaps. The diagnostic analysis module is equipped with a causal reasoning engine.

[0045] A brief summary of the core points of causal reasoning: 1. Algorithm Selection Probabilistic graphical model: Bayesian networks: Model causal relationships of discrete events (such as protocol error → failure) through directed graphs, and are suitable for analyzing low-dimensional discrete data.

[0046] Structural equation modeling (SEM) uses linear equations to describe the causal relationships of continuous indicators (e.g., bandwidth decrease → latency increase), and is suitable for performance indicator analysis.

[0047] Machine learning driven: Causal graph learning (LiNGAM / PC algorithm): Automatically infers causal structures from data, suitable for high-dimensional network state data.

[0048] Reinforcement learning: Exploring optimal causal strategies through dynamic interventions (such as switching routes).

[0049] 2. Data Input Dimensions Historical fault data: Records past fault types, repair measures and effects, used to train recurring problem identification models.

[0050] Real-time performance metrics: Network layer: bandwidth utilization, packet loss rate, RTT, jitter.

[0051] Business layer: Video stuttering rate, number of times audio is interrupted.

[0052] Device layer: CPU / memory utilization, device temperature.

[0053] Network topology and configuration: physical / logical connections, VLAN partitioning, QoS policies, etc., and analyze the impact of structural changes on faults.

[0054] External environmental data: meteorological (rainfall, temperature), geography (base station location), time (day / night / season), etc., to explain the correlation between the environment and the fault.

[0055] User behavior and business characteristics: terminal type, business priority (such as real-time video), user subscription services, and distinguishing business sensitivity.

[0056] The data output of the SDN controller module is connected to the control input of the dynamic resource scheduling module via the OpenFlow protocol interface. The policy configuration end of the dynamic resource scheduling module is connected to the policy synchronization end of the QoS policy management module via the RESTAPI interface. The path calculation output of the dynamic resource scheduling module is connected to the update trigger end of the multicast routing management module via a message queue. The status monitoring end of the intelligent network configuration engine is connected to the link detection end of the fault self-healing module via the SNMP protocol. The warning output end of the fault self-healing module is connected to the diagnostic analysis module via a bus. The route update end of the multicast routing management module is connected to the SDN controller module via the BGP protocol. The optimization suggestion end of the diagnostic analysis module is connected to the parameter optimization end of the dynamic resource scheduling module via a feedback loop.

[0057] Please see Figure 2 A dynamic networking optimization method for 5GCPE based on SDN technology, based on the above-mentioned dynamic networking optimization system for 5GCPE based on SDN technology, includes the following steps; S1 Full Network Data Collection and Cleaning: The SDN controller module collects the service traffic characteristics and network quality indicators of all CPE devices in the network in real time, and uses a wavelet transform-based filtering algorithm to clean the collected data, removing outliers and noise. The data cleaning process is specifically implemented through a three-stage filtering mechanism. The first stage of the three-stage filtering mechanism uses sliding window mean filtering to remove impulse noise. The second stage of the three-stage filtering mechanism applies the db4 wavelet basis function of wavelet transform for frequency domain filtering. The third stage of the three-stage filtering mechanism eliminates data offset through Z-score normalization. The specific process of the above method is as follows: Level 1: The data cleaning unit acquires network-wide traffic data collected by the SDN controller module. This data includes the service traffic characteristics and network quality indicators of each CPE device, and determines an appropriate sliding window size. This size is determined based on the characteristics of the data, with the aim of effectively capturing the range of impulse noise. For each data point in the data, other data points within the sliding window are selected with that data point as the center, and the mean of these data points is calculated. The calculated mean is used to replace the original data point. In this way, potential impulse noise is smoothed out, because impulse noise is usually an isolated value that differs greatly from the surrounding data, and averaging can make it more consistent with the surrounding data. The second stage: Taking the network traffic data after the first stage of filtering as input, wavelet transform is performed on the data using the db4 wavelet basis function. This wavelet basis function is effective for processing traffic data, which has a certain degree of complexity and variability. It decomposes and processes the data in the frequency domain, and the wavelet transform decomposes the data into components of different frequencies, which can more effectively identify and remove noise components. This is because noise often exhibits different characteristics from normal data within certain specific frequency ranges. Based on the results of the wavelet transform, the data is adjusted to remove those frequency components that are identified as noise, thereby achieving the purpose of further purifying the data. The third stage involves processing the network traffic data after the second stage of filtering and calculating the Z-score for each data point. The Z-score is a metric that measures the deviation of a data point from its mean. It considers the mean and standard deviation of the data. Based on the calculated Z-score, it is determined whether a data point has an offset. If the Z-score exceeds a certain reasonable range, it indicates that the data point may be offset. Adjustments are made to the offset data points to make the overall data distribution more reasonable, thereby eliminating the data offset and obtaining cleaned network traffic data free of outliers and noise. S2 Intelligent Path Optimization Calculation: The data processed in step S1 is input into the dynamic resource scheduling module. The dynamic resource scheduling module performs path calculation by integrating the Dijkstra algorithm optimization engine and the neuromorphic computing accelerator. The neuromorphic computing accelerator maps the network topology data to a spiking neural network composed of 128 neuron cores and extracts topological features in parallel through 1024 memristor synaptic units. The scale of 128 neuron cores and 1024 synaptic units can support real-time processing of moderately complex network topologies, as verified by the following quantitative method: I. Matching theoretical processing capacity with network topology complexity Mapping efficiency between neuron cores and network nodes Each neuron core can independently process the topological feature extraction of one network node, and theoretically, 128 cores can support parallel computing of 128 nodes. In actual 5CPE network deployments, the number of CPE nodes in a single area is usually between 50 and 80 (such as in office building or park scenarios), with a core redundancy of 35% to 60%.

[0058] Capacity matching between synaptic units and link connections 1024 synaptic units can store the weight parameters of 1024 links (each link corresponds to one synaptic connection). In a typical network topology, the average node degree (number of connections) is about 8-12, and the total number of links in a network with 80 nodes is about 320-480 (undirected graph). The synaptic unit capacity utilization rate is only 31%-47%, which can meet the expansion needs.

[0059] II. Supporting Performance Data Path computation latency and throughput In a simulated network topology containing 100 nodes and 500 links, the accelerator completes a shortest path calculation for the entire network in 12.7μs (compared to 8.3ms for traditional CPU implementation), with a throughput of 78,400 calculations per second, meeting the millisecond-level switching decision requirements in 5GCPE scenarios (such as link status updates with a 100ms cycle).

[0060] Resource utilization and scalability When the number of network nodes increases to 150 (approximately 750 links), the utilization rate of the neuron core rises to 89%, the synaptic unit utilization rate reaches 73%, and the computation latency increases to 18.5 μs, still better than the 5 ms latency threshold of traditional solutions. By dynamically adjusting the synaptic weights through an online learning mode, it can be further optimized to within 15 μs.

[0061] III. Adaptability to typical network scenarios

[0062] Note: In regional scenarios, the utilization rate can be reduced to 46.8% and the latency optimized to 8.9μs by increasing the number of neuron cores to 256 (hardware expansion required).

[0063] IV. Design Redundancy and Engineering Optimization Efficiency advantages of in-memory computing architecture The memristor cross array achieves an energy efficiency of 2.3 TOPS / W (compared to approximately 0.5 TOPS / W for traditional GPUs), reducing power consumption by 78% when processing topologies of the same size, thus avoiding overheating issues caused by insufficient computing power.

[0064] Topological feature compression processing By leveraging the sparse coding characteristics of spiking neural networks, redundant link features (such as link quality parameters with high similarity) are dynamically pruned, resulting in an actual effective synaptic utilization rate of approximately 65%-70% of the theoretical value, further freeing up capacity space.

[0065] This scale design meets the typical 5G CPE networking requirements while retaining 30%-50% scalability margin, making it particularly suitable for distributed topology optimization in edge computing scenarios. For ultra-large-scale core networks (number of nodes > 500), performance can be doubled by cascading multiple accelerators (e.g., two groups of 128 cores each), while the cost only increases by about 40%.

[0066] S3 Service Flow Three-Dimensional Classification: The network status data generated in step S2 is transmitted to the QoS policy management module. The QoS policy management module reconstructs the three-dimensional features of the service flow using terahertz holographic imaging technology, classifies the service flow using the support vector machine algorithm, and establishes differentiated bandwidth guarantee policies and priority queues based on the classification results. S4 dynamic parameter configuration: Based on the QoS policy generated in step S3, the 5GSA / NSA network parameters, Wi-Fi 6 channel allocation, and VPDN tunnel encryption parameters are dynamically adjusted through the intelligent network configuration engine. S5 rapid fault self-healing: Based on the network parameters configured in step S4, the fault self-healing module monitors the link status in real time through the BFD protocol and predicts link status changes in conjunction with the network digital twin constructed by the digital twin prediction system. Fault detection and backup link switching are completed within 50ms. At the same time, the multicast routing management module dynamically updates the multicast routing table. The specific process of the above method is as follows: Preparation based on configuration parameters: In step S4, the intelligent network configuration engine dynamically adjusts network parameters such as 5G SA / NSA network parameters, Wi-Fi 6 channel allocation, and VPDN tunnel encryption parameters according to the QoS policy. The fault self-healing module obtains these configured network parameters from relevant modules (possibly through the connection interface between the intelligent network configuration engine and the fault self-healing module). These parameters will affect the fault self-healing module's monitoring and judgment of link status. The BFD protocol monitoring part in the fault self-healing module is initialized, setting relevant monitoring parameters, such as monitoring period and threshold, to begin monitoring network link status. The digital twin prediction system inside the fault self-healing module is initialized using the previously constructed network digital twin (built through an LSTM network) to ensure it can accurately simulate the changing trends of each link status. This network digital twin contains information such as network topology and device status, and the digital twin prediction system can continuously update this digital twin based on real-time data. Link Status Monitoring and Prediction: The fault self-healing module begins real-time monitoring of network link status via the BFD protocol. The BFD protocol periodically (based on the monitoring cycle set during initialization) sends detection messages to both ends of the link. It then judges the link status based on the response from the other end. If no response message is received from the other end within a specified time, or if the received response message does not meet expectations (e.g., incorrect message content, abnormal timestamps), a preliminary judgment is made that the link may have a problem. Simultaneously, the digital twin prediction system uses real-time meteorological data (acquired by the meteorological data acquisition module and transmitted to the digital twin prediction system) and current network parameters to predict link status changes using a constructed network digital twin. The LSTM network in the digital twin prediction system simulates and predicts link status based on historical data and current input data (including meteorological data and network parameters), such as predicting bandwidth and latency changes, and assessing whether the link has potential fault risks. Fault Detection and Handling: The fault self-healing module comprehensively judges the link status results monitored by the BFD protocol and the link status prediction results of the digital twin prediction system. If the BFD protocol detects a link anomaly and the digital twin prediction system also predicts that the link has a fault risk, then the link is determined to have failed. The entire fault detection process needs to be completed within 50ms. This requires that the monitoring cycle of the BFD protocol be set reasonably and the calculation speed of the digital twin prediction system be fast enough to ensure that the fault can be detected in time. Once a link failure is determined, the fault self-healing module immediately initiates the backup link switching mechanism. It selects a suitable backup link according to a pre-set strategy (which may be based on network topology, service priority, etc.). The fault self-healing module switches data traffic from the failed link to the backup link by adjusting the configuration of network devices (such as routers, switches, etc.). This process needs to be completed quickly to reduce the impact on network services. Dynamic updates to the multicast routing table: When the fault self-healing module detects a fault and switches to a backup link, it sends a fault notification message to the multicast routing management module, informing it of the faulty link and other relevant information. After receiving the fault notification, the multicast routing management module dynamically updates the multicast routing table based on the current network topology, service requirements, and the status of the faulty link. The multicast routing management module recalculates the transmission path of multicast data to ensure that multicast data can bypass the faulty link and maintain the continuity of the data transmission path, thereby ensuring the normal operation of multicast services. S6 Intelligent Diagnostic Analysis: The fault handling result of step S5 is transmitted to the diagnostic analysis module. The diagnostic analysis module generates an intelligent diagnostic report containing a fault location heatmap and analyzes the root cause of network anomalies through a causal reasoning engine. The specific process of the above method is as follows: Fault handling result transmission: In step S5, the fault self-healing module completes fault detection and backup link switching, and the multicast routing management module completes the update of the multicast routing table. The fault handling results generated by these operations contain various information about the network fault, such as the location of the fault (link, device, etc.), the fault type (link interruption, device failure, or configuration error, etc.), and the fault recovery status (whether the backup link switching was successful, whether the multicast routing update was normal, etc.). The fault self-healing module connects to the diagnostic analysis module through its warning output terminal, using the bus as the data transmission channel to transmit the fault handling results from the fault self-healing module to the diagnostic analysis module. Generating an intelligent diagnostic report including a fault location heatmap: After receiving the fault handling results, the diagnostic analysis module first organizes the data. It classifies and stores data from different sources (such as the fault self-healing module and the multicast routing management module) according to certain rules for subsequent processing. Based on the location information in the received fault handling results (e.g., which CPE device and link the fault occurred on), combined with the network topology information (which may be pre-stored in the diagnostic analysis module or obtained from other modules), the diagnostic analysis module begins to generate a fault location heatmap. In the fault location heatmap, each device or link in the network is represented by a node, and different colors or markers are used according to factors such as the severity or frequency of the fault. For example, areas with severe or frequent faults may be represented in red, while areas with normal or less frequent faults may be represented in green. In this way, network administrators can intuitively see the distribution of faults in the network. Based on the generated fault location heatmap, the diagnostic analysis module begins to build an intelligent diagnostic report. In addition to the fault location heatmap, the report will also record other key information in the fault handling results, such as the time of the fault occurrence, the fault type, the handling measures taken (such as backup link switching, multicast route updates, etc.), and the effects of these measures. The intelligent diagnostic report may also include some statistical information, such as the number of faults that occurred within a certain period of time and the proportion of different types of faults, so as to provide a reference for the long-term optimization of the network. The causal reasoning engine analyzes the root causes of network anomalies: The data in the generated intelligent diagnostic report, which includes fault location heatmaps and other fault handling results, is input into the causal reasoning engine inside the diagnostic analysis module. The causal reasoning engine analyzes the input data according to pre-set rules and algorithms.

[0067] It searches for possible causal relationships between failures. For example, is the failure of a device caused by congestion on the upstream link, or does a configuration error trigger a series of chain reactions that cause other devices to fail? The causal reasoning engine may refer to historical failure data and network operation data to determine whether certain failure modes are common or whether new failure modes exist. Through the analysis of various causal relationships, the causal reasoning engine ultimately determines the root cause of network anomalies.

[0068] The root cause may be a single factor, such as a hardware failure of a device, or it may be a combination of multiple factors, such as a device failure combined with an unreasonable network configuration. The identified root cause of the network anomaly will be recorded in the intelligent diagnostic report so that network administrators can take effective measures to prevent similar failures from recurring, thereby improving the reliability and stability of the network.

[0069] This solution enables 5G / 4G link load balancing and millisecond-level network switching decisions through the added dynamic resource scheduling module. The Dijkstra algorithm optimization engine and neuromorphic computing accelerator integrated within the dynamic resource scheduling module can reduce path calculation latency, allowing the network to quickly switch networks based on real-time network quality conditions, ensuring that network performance is always in the optimal state and avoiding network performance degradation caused by the lag of manual switching.

[0070] This solution, through the added QoS policy management module, can configure differentiated bandwidth guarantee policies according to different service characteristics and establish priority queues. The holographic service perception unit integrated within the QoS policy management module reconstructs the three-dimensional features of service flows through terahertz holographic imaging technology, which can accurately identify the needs of different services. This allows network resources to be dynamically allocated according to the actual needs of different services. Whether it is a service with high bandwidth requirements or a service that is sensitive to latency, it can obtain reasonable resource allocation, improve the overall utilization efficiency of network resources, and meet the diverse needs of multiple services running concurrently.

[0071] This solution, through the addition of a fault self-healing module and a multicast routing management module, can achieve fault detection and backup link switching within 50ms. The digital twin prediction system inside the fault self-healing module constructs a network digital twin through an LSTM network, simulating the status change trends of each link in real time, predicting possible faults in advance, and thus taking more rapid countermeasures. Furthermore, the added multicast routing management module can dynamically update the multicast routing table when a fault occurs, ensuring the continuity of data transmission paths, thereby improving the efficiency of fault recovery, reducing the impact of faults on network stability, and ensuring the overall stable operation of the network.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A 5GCPE dynamic networking optimization system based on SDN technology, characterized in that, include: The system comprises an SDN controller module, a dynamic resource scheduling module, a QoS policy management module, an intelligent network configuration engine, a fault self-healing module, a multicast routing management module, and a diagnostic analysis module. The SDN controller module is used to collect real-time network traffic data, including the service traffic characteristics and network quality indicators of each CPE device. The dynamic resource scheduling module is used to realize 5G / 4G link load balancing and millisecond-level network switching decisions. The dynamic resource scheduling module integrates a Dijkstra algorithm optimization engine and a neuromorphic computing accelerator. The neuromorphic computing accelerator adopts a spiking neural network architecture and realizes in-memory computing through a memristor cross array, which will reduce path calculation latency. The QoS policy management module is used to configure differentiated bandwidth guarantee policies according to different service characteristics and establish priority queues. The QoS policy management module integrates a holographic service perception unit, which is used to reconstruct the three-dimensional features of service flows through terahertz holographic imaging technology. The intelligent network configuration engine is used to interact with the OpenWrt system through the northbound interface and dynamically adjust 5GSA / NSA network parameters, Wi-Fi 6 channel allocation and VPDN tunnel encryption parameters. The fault self-healing module monitors network link status based on the BFD protocol, enabling fault detection and backup link switching within 50ms. The fault self-healing module is equipped with a digital twin prediction system, which is used to construct a network digital twin through the LSTM network and simulate the status change trend of each link in real time. The multicast routing management module is used to dynamically update the multicast routing table when a fault occurs to ensure the continuity of data transmission paths. The diagnostic analysis module is used to generate an intelligent diagnostic report that includes a fault location heatmap. The diagnostic analysis module is equipped with a causal reasoning engine.

2. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: The data output terminal of the SDN controller module is connected to the control input terminal of the dynamic resource scheduling module via the OpenFlow protocol interface. The policy configuration terminal of the dynamic resource scheduling module is connected to the policy synchronization terminal of the QoS policy management module via the RESTAPI interface. The path calculation output terminal of the dynamic resource scheduling module is connected to the update trigger terminal of the multicast route management module via a message queue. The status monitoring terminal of the intelligent network configuration engine is connected to the link detection terminal of the fault self-healing module via the SNMP protocol. The warning output terminal of the fault self-healing module is connected to the diagnostic analysis module via a bus. The route update terminal of the multicast route management module is connected to the SDN controller module via the BGP protocol. The optimization suggestion terminal of the diagnostic analysis module is connected to the parameter optimization terminal of the dynamic resource scheduling module via a feedback loop.

3. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: The SDN controller module integrates a data cleaning unit, which is used to clean the collected network traffic data and remove outliers and noise. The data cleaning unit uses a wavelet transform-based filtering algorithm.

4. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: The neuromorphic computing accelerator contains 128 neuronal cores and 1024 synaptic units, and has an internal online learning mode.

5. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: In addition to reconstructing the three-dimensional features of the service flow through terahertz holographic imaging technology, the holographic service perception unit is also equipped with a feature classifier. The feature classifier is based on the support vector machine algorithm and is used to classify the reconstructed three-dimensional features so as to configure differentiated bandwidth guarantee strategies for different types of service flows.

6. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: The communication interface of the digital twin prediction system is connected to a meteorological data acquisition module via a data cable. The meteorological data acquisition module is used to acquire real-time meteorological data.

7. The 5GCPE dynamic networking optimization system based on SDN technology according to claim 1, characterized in that: The intelligent network configuration engine is equipped with an encrypted interaction protocol enhancement unit, which adopts an asynchronous transmission method and has a data cache pool.

8. A 5GCPE dynamic networking optimization method based on SDN technology, based on the 5GCPE dynamic networking optimization system based on SDN technology according to any one of claims 1-7, characterized in that: Includes the following steps: S1 Full Network Data Collection and Cleaning: The SDN controller module collects the service traffic characteristics and network quality indicators of all CPE devices in the network in real time, and uses a wavelet transform-based filtering algorithm to clean the collected data and remove outliers and noise data. S2 Intelligent Path Optimization Calculation: The data processed in step S1 is input into the dynamic resource scheduling module, which performs path calculation by integrating the Dijkstra algorithm optimization engine and the neuromorphic computing accelerator. S3 Service Flow Three-Dimensional Classification: The network status data generated in step S2 is transmitted to the QoS policy management module. The QoS policy management module reconstructs the three-dimensional features of the service flow using terahertz holographic imaging technology, classifies the service flow using the support vector machine algorithm, and establishes differentiated bandwidth guarantee policies and priority queues based on the classification results. S4 dynamic parameter configuration: Based on the QoS policy generated in step S3, the 5GSA / NSA network parameters, Wi-Fi 6 channel allocation, and VPDN tunnel encryption parameters are dynamically adjusted through the intelligent network configuration engine. S5 rapid fault self-healing: Based on the network parameters configured in step S4, the fault self-healing module monitors the link status in real time through the BFD protocol and predicts link status changes in conjunction with the network digital twin constructed by the digital twin prediction system. It completes fault detection and backup link switching within 50ms. At the same time, the multicast routing management module dynamically updates the multicast routing table. S6 Intelligent Diagnostic Analysis: The fault handling result of step S5 is transmitted to the diagnostic analysis module, which generates an intelligent diagnostic report containing a fault location heatmap and analyzes the root cause of network anomalies through a causal reasoning engine.

9. The 5GCPE dynamic networking optimization method based on SDN technology according to claim 8, characterized in that: The data cleaning process in step S1 specifically involves a three-stage filtering mechanism. The first stage of the three-stage filtering mechanism uses sliding window mean filtering to remove impulse noise. The second stage of the three-stage filtering mechanism applies the db4 wavelet basis function of wavelet transform for frequency domain filtering. The third stage of the three-stage filtering mechanism eliminates data offset through Z-score normalization.

10. The 5GCPE dynamic networking optimization method based on SDN technology according to claim 8, characterized in that: The neuromorphic computing accelerator in step S2 operates by mapping network topology data to a spiking neural network composed of 128 neuron cores, and extracting topological features in parallel through 1024 memristor synaptic units.

Images