Slice switching method and apparatus, electronic device, and storage medium

Abstract

The disclosure provides a slice switching method and device, electronic equipment and storage medium, and relates to the technical field of automatic management. The method comprises the following steps: collecting hardware state data of each component on which a target slice depends on a B5G cloud platform; judging whether the component meets an adjustment condition based on the hardware state data; and in response to the adjustment condition being met, adjusting the component. Thus, the application collects the hardware state data of each component on which the target slice depends on the B5G cloud platform in sequence, intelligently judges whether the adjustment condition is met based on the data, and automatically performs targeted adjustment when the condition is met, forming a closed loop of "perception - decision - execution". The mechanism can identify the hardware performance bottleneck or sub-health risk in advance, guarantee the network slice service quality and service level agreement compliance, avoid unnecessary migration or expansion and contraction operation, and significantly improve the accuracy of adjustment and resource utilization efficiency.

Description

Technical Field

[0001] This disclosure relates to the field of automated management technology, and in particular to a slice switching method, apparatus, electronic device, and storage medium. Background Technology

[0002] As fifth-generation mobile communication (5G) evolves towards broadband and intelligence, and gradually integrates cloud computing, edge computing, and artificial intelligence technologies, the BeyondFifth-Generation Mobile Communication (B5G) cloud platform is becoming the core infrastructure supporting future network slicing, computing power scheduling, and service customization. Under the B5G architecture, network slices are no longer merely logically isolated communication channels, but end-to-end service entities integrating computing, storage, forwarding, and intelligent decision-making capabilities. Each slice typically consists of multiple heterogeneous components, including a Virtualized Radio Access Network (vRAN), User Plane Function (UPF), control plane network elements, an Artificial Intelligence (AI) inference engine, and distributed storage units. These components are widely deployed in cloud data centers, edge nodes, and even on the terminal side, and their operation is highly dependent on the real-time status of underlying hardware resources (such as the Central Processing Unit (CPU), Graphics Processing Unit (GPU), memory, network interfaces, and dedicated accelerators).

[0003] However, in actual operation, B5G cloud platforms face dynamically changing business loads, hardware performance fluctuations, energy efficiency constraints, and failure risks. For example, when a slice is performing a high-concurrency video analysis task, its graphics processing unit (GPU) may reduce its frequency due to overheating, leading to a sharp increase in inference latency; or the memory bandwidth of the edge server may become saturated, causing a bottleneck in the user plane function (UPF) processing. If a static resource configuration strategy is still used, it will not only cause a decline in the quality of service (QoS), but may also trigger a breach of the slice service level agreement (SLA), or even cause cascading failures. Summary of the Invention

[0004] This disclosure aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, one objective of this disclosure is to propose a slice switching method.

[0006] The second objective of this disclosure is to provide a slice switching device.

[0007] The third objective of this disclosure is to propose an electronic device.

[0008] The fourth objective of this disclosure is to provide a non-transitory computer-readable storage medium.

[0009] The fifth objective of this disclosure is to provide a computer program product.

[0010] To achieve the above objectives, a slice switching method is proposed in the first aspect of this disclosure, comprising: collecting hardware status data of each component on which the target slice depends in the B5G cloud platform; determining whether the component meets the adjustment conditions based on the hardware status data; and adjusting the component in response to meeting the adjustment conditions.

[0011] According to one embodiment of this disclosure, determining whether the component meets the adjustment conditions based on the hardware status data includes: obtaining the component type of the component; determining the adjustment conditions based on the component type; and determining whether the component meets the adjustment conditions based on the hardware status data.

[0012] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being an antenna array, comparing the beam gain data in the hardware status data with a gain judgment threshold; and in response to the beam gain data collected over N consecutive timestamps being less than the gain judgment threshold, determining that the component meets the adjustment condition.

[0013] According to one embodiment of this disclosure, adjusting the component includes: obtaining the slice type of the target slice; and adjusting the component based on the slice type and the hardware status data.

[0014] According to one embodiment of this disclosure, adjusting the component based on the slice type and the hardware status data includes: in response to the slice type being a vehicle-to-everything (V2X) slice, adjusting the UPF deployment location of the target slice to migrate towards the edge node; and in response to the slice type being an Internet of Things (IoT) slice, switching the antenna array to a wide-coverage beam mode and improving the signal coverage quality parameters in the edge area.

[0015] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being a liquid cooling device, comparing the coolant temperature in the hardware status data with a temperature judgment threshold; and in response to the coolant temperature being greater than the temperature judgment threshold, determining that the component meets the adjustment condition.

[0016] According to one embodiment of this disclosure, adjusting the component includes: determining a target node from the B5G cloud platform based on a particle swarm optimization algorithm, wherein the target node is a node whose coolant temperature is lower than a set temperature; and migrating the key network elements of the target slice to the target node.

[0017] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being a beam splitter, comparing the optical power attenuation value of the acquisition input / output port in the hardware status data with an optical attenuation judgment threshold; and in response to the optical power attenuation value acquired over N consecutive timestamps being greater than the optical attenuation judgment threshold, determining that the component meets the adjustment condition.

[0018] According to one embodiment of this disclosure, adjusting the component includes: migrating the current beam splitter port of the target slice to a spare port of the beam splitter.

[0019] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and judging whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being an industrial-grade optical modem, comparing the packet loss rate in the hardware status data with a packet loss rate judgment threshold; and in response to the packet loss rate collected over N consecutive timestamps being greater than the packet loss rate judgment threshold, determining that the component meets the adjustment condition.

[0020] According to one embodiment of this disclosure, adjusting the component includes: configuring a reset command and executing the reset command to reset the current network port of the target network port.

[0021] To achieve the above objectives, a second aspect of this disclosure provides a slice switching device, comprising: a data acquisition module for acquiring hardware status data of each component on which the target slice depends in the B5G cloud platform; a judgment module for judging whether the component meets the adjustment conditions based on the hardware status data; and an adjustment module for adjusting the component in response to meeting the adjustment conditions.

[0022] To achieve the above objectives, a third aspect of this disclosure provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to implement the slice switching method as described in the first aspect of this disclosure.

[0023] To achieve the above objectives, a fourth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the slice switching method as described in the first aspect of this disclosure.

[0024] To achieve the above objectives, a fifth aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the slice switching method as described in the first aspect of this disclosure.

[0025] Therefore, this invention forms a closed loop of "perception-decision-execution" by sequentially collecting hardware status data of each component on which the target slice depends in the B5G cloud platform, intelligently determining whether adjustment conditions are met based on this data, and automatically executing targeted adjustments when the conditions are met. This mechanism can identify hardware performance bottlenecks or sub-health risks in advance, ensuring the quality of service and service level agreement fulfillment of network slices while avoiding unnecessary migration or scaling operations, significantly improving the accuracy of adjustments and resource utilization efficiency. It not only enhances the automated operation and maintenance and self-healing capabilities of the B5G cloud platform, but also supports differentiated scheduling based on slice type, effectively balancing the stability of high-reliability services with the energy efficiency requirements of ordinary services, thereby achieving high-performance, high-availability, and low-power intelligent cloud-network collaborative operation. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of a slice switching method according to one embodiment of the present disclosure;

[0027] Figure 2 This is a schematic diagram of another slice switching method according to one embodiment of the present disclosure; Figure 3 This is a schematic diagram of a slice switching device according to one embodiment of the present disclosure; Figure 4 This is a schematic diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0028] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0029] The acquisition, storage, use, and processing of data in this disclosed technical solution all comply with the relevant provisions of relevant laws and regulations.

[0030] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0031] Current technologies attempt to migrate or scale up / down slices by monitoring system metrics such as CPU utilization and packet loss rate. However, these solutions generally suffer from issues such as slow response times, coarse adjustment granularity, and a lack of comprehensive awareness of the collaborative status of multiple components. Especially in the complex environment of B5G multi-tenancy and multi-slice coexistence, single-dimensional threshold alarm mechanisms are insufficient to accurately determine whether adjustments should be triggered, let alone achieve fine-grained, adaptive hardware resource collaborative optimization for specific slices.

[0032] Therefore, there is an urgent need for a slice switching method that can collect hardware status data of each component on which the target slice depends in real time, make a comprehensive judgment on the necessity of adjustment based on multi-dimensional status, and accurately execute component-level adjustments when the conditions are met, so as to improve the service resilience, resource efficiency and automated operation and maintenance capabilities of the B5G cloud platform.

[0033] To address the aforementioned issues, this disclosure proposes a slice switching method. Figure 1 This is a schematic diagram of a slice switching method according to one embodiment of the present disclosure, as shown below. Figure 1 As shown, the slice switching method includes the following steps: S101 collects hardware status data of the components on which the target slice depends in the B5G cloud platform.

[0034] The slice switching method of this application embodiment can be applied to the scenario of slice management of B5G cloud platform. The slice switching execution subject of this application embodiment can be the slice switching device of this application embodiment, which can be installed on an electronic device.

[0035] It should be noted that target slices in the B5G cloud platform are typically implemented collaboratively by multiple functional modules, and their dependent components may include, but are not limited to: Computing components: Virtualized Radio Access Network (vRAN), User Plane Function (UPF), control plane network elements (such as AMF, SMF), and Artificial Intelligence (AI) inference service containers deployed in the central cloud or edge nodes.

[0036] Acceleration components: Dedicated hardware accelerators for baseband processing, encryption / decryption, or AI inference, such as field-programmable gate arrays (FPGAs), tensor processing units (TPUs), or smart NICs.

[0037] Resource components: Physical or virtual resource entities that support the above services, including central processing units (CPU), graphics processing units (GPU), memory, cache, storage devices (such as NVMe SSD), and network interface controllers (NIC).

[0038] For the aforementioned components, the method disclosed herein requires first collecting multi-dimensional, high-time-sensitivity hardware status data. This data may include various types, without any limitation herein. For example, it may include, but is not limited to, the following types of data: Performance metrics: CPU / GPU utilization, clock speed, core temperature; memory utilization, bandwidth throughput; NIC packet transmission and reception rate, packet loss rate, queue latency.

[0039] Health status data: hardware error count (such as ECC memory error correction count), fan speed, power supply voltage fluctuation, accelerator fault flag.

[0040] Energy efficiency metrics: power consumption at the whole machine or component level (Watt), performance per Watt.

[0041] Resource contention signal data: the intensity of contention among other slices on the same node for shared resources (such as last-level cache, memory bandwidth).

[0042] In this embodiment of the disclosure, there are various methods for collecting hardware status data of the components on which the target slice depends in the B5G cloud platform, and no limitation is made here. For example, the following methods may be included: Agent-based data collection: Deploy a lightweight monitoring agent within the host machine or container to read the underlying state.

[0043] Non-intrusive telemetry: Utilizes eBPF (extended Berkeley Packet Filter) technology to efficiently capture system calls and resource usage in kernel mode, avoiding application layer interference.

[0044] Platform Integration: Connect to the B5G cloud platform's built-in Infrastructure as a Service (IaaS) monitoring system or carrier-grade operation and maintenance support system for real-time data acquisition.

[0045] Contextual association: Hardware status data is tagged and bound with slice ID, tenant identifier, and service chain path to ensure that subsequent analysis can accurately locate the target slice.

[0046] In the disclosed embodiments, collecting hardware status data of the components upon which the target slice depends on the B5G cloud platform is the starting point for the entire slice switching method. This aims to construct a full-stack, real-time, and fine-grained hardware view of the target network slice's operating environment. It identifies and monitors all underlying software and hardware components invoked or bound to the slice throughout its lifecycle and continuously collects its key hardware status indicators.

[0047] In one possible implementation, a real-time monitoring system can be built based on Prometheus and Grafana through a multimodal data collection and monitoring module. This system can collect metrics such as CPU utilization, traffic, and energy consumption, and support full-dimensional monitoring of physical resources, virtual resources, and core hardware devices. It includes three parts: the collection layer architecture, data processing and storage, and visualization and alarms.

[0048] Specifically, the data acquisition layer architecture includes physical resource monitoring, virtual resource monitoring, business layer monitoring, and an environmental awareness sublayer.

[0049] Specifically, physical resource monitoring collects server BMC data (temperature, power consumption, fan speed, etc.) via the enhanced IPMIv2.0 protocol, with a sampling frequency of 1Hz and a monitoring accuracy of ±0.5%. It also monitors rack power consumption in real time via intelligent PDUs, supporting three-phase power parameter acquisition with an accuracy of ±0.8%. Vibration and temperature / humidity sensors are introduced to collect data on the server room environment, enabling collaborative monitoring of physical resources and the environment. For server fault prediction, a fault precursor feature library is built based on the collected power consumption and temperature data, supporting fault warnings for critical components such as hard drives and power supplies, with a warning accuracy of 92%. The fault precursor feature library includes hard drive SMART attributes (such as the number of sectors reallocated and seek error rates), power supply output voltage fluctuations, and abnormal CPU temperature increases, using a random forest algorithm for fault prediction, issuing warnings 72 hours in advance.

[0050] Specifically, virtual resource monitoring uses PrometheusNodeExporter to collect basic metrics of VMs such as CPU, memory, and disk I / O, with a sampling frequency of 5 seconds per instance. It captures container-level network traffic using eBPF probes, capturing detailed information such as TCP / UDP connection status, packet size distribution, and application layer protocol types, with a collection latency of <1ms. Through cross-layer collaboration between the IPMI protocol and eBPF probes, it solves the data fragmentation problem between "physical-virtual-service" in existing technologies. Cross-layer data uses the PTP precise time protocol for timestamp synchronization, with a synchronization accuracy of <1ms, ensuring time alignment of data at different levels.

[0051] Specifically, the business layer monitoring utilizes the 3GPP-defined N4 interface probe (compatible with Rel.15 / 16 standards) to collect core metrics such as session establishment latency between the SMF and UPF, QoS rule hit rate, and packet forwarding latency, with a sampling frequency of 1 second per sample. For different industry business characteristics, it extends the collection of industry-specific metrics, such as the periodic packet arrival interval in industrial control scenarios and the handover success rate in vehicle-to-everything (V2X) scenarios. By developing an adaptation interface between 3GPP N4 and industry applications, it achieves the association mapping between business layer metrics and application layer status. The adaptation interface adopts a RESTful API design, supporting industry applications to proactively report business status data in JSON format.

[0052] The environmental perception sublayer deploys an IoT sensor network to collect data on external factors affecting the network's environment, including meteorological data, electromagnetic interference intensity, and user density. This provides environmental data support for resource scheduling and fault diagnosis. Sensor data is transmitted to the gateway via the LoRaWAN protocol and then aggregated to the monitoring platform, with a transmission latency of less than 1 second.

[0053] Optionally, core hardware equipment monitoring collects status data through dedicated interfaces: antenna array beam gain, VSWR and other indicators are collected through the base station northbound interface, with a sampling frequency of 1 second / time; liquid cooling device temperature, flow rate and other data are collected through the Modbus protocol, with a monitoring accuracy of ±0.3℃; optical power attenuation value of the optical splitter is collected through the SNMP protocol, with a sampling frequency of 10 seconds / time; industrial-grade optical modem network port status, registration status and other data are collected through the device's built-in monitoring interface to ensure that hardware status data is time-series aligned with physical, virtual and service layer data.

[0054] Furthermore, data processing and storage adopt a "time-series database + message queue + distributed cache" architecture, consisting of Prometheus, InfluxDB, Kafka cluster and Redis.

[0055] Optionally, Prometheus stores 7 days of raw real-time metrics, employing an improved time-series data compression algorithm (compression ratio up to 10:1) to reduce storage overhead, with query latency <50ms. The improved compression algorithm is based on Delta encoding combined with LZ4 compression, first differentially encoding the time-series data and then performing LZ4 compression, achieving a 30% improvement in compression ratio compared to the native Prometheus compression algorithm. InfluxDB stores historical aggregated data using multi-granularity downsampling, with hot historical data (within one year) stored on SSDs and cold data stored on HDDs, retaining one year of raw data, and offering petabyte-level storage capacity scalability. A Kafka cluster buffers monitoring data, achieving a peak throughput of 500,000 records per second, supporting asynchronous communication between modules to prevent data loss. A Redis distributed cache is used to store hot metrics, achieving a cache hit rate of 95%, further improving query efficiency.

[0056] It's worth noting that the hybrid storage strategy of "Prometheus + InfluxDB + Kafka + Redis" in a heterogeneous data architecture breaks through the storage bottleneck of traditional single databases. Through layered management of time-series data, it improves the efficiency of real-time monitoring and historical trend analysis by more than 3 times and reduces storage costs by 50%. The KafkaConnect component enables real-time data synchronization with other storage systems, with a data synchronization latency of less than 1 second. Data synchronization employs at least-once semantics, ensuring no data loss through a message acknowledgment mechanism. If synchronization fails, it automatically retryes 3 times; if it still fails, an alarm is triggered.

[0057] S102, based on hardware status data, determines whether the component meets the adjustment conditions.

[0058] After collecting hardware status data of each component on which the target slice depends, the raw hardware indicators need to be converted into executable decision signals, that is, to determine whether the current component is in an abnormal, sub-healthy or inefficient state that "needs adjustment".

[0059] In this embodiment of the disclosure, the judgment is not a simple threshold comparison, but an intelligent evaluation process that integrates multi-dimensional states, contextual semantics and business objectives.

[0060] It should be noted that meeting the adjustment conditions means that the current hardware state of the component has posed a significant risk or actual damage to the target slice's Quality of Service (QoS), Service Level Agreement (SLA), or overall platform energy efficiency, and needs to be intervened through dynamic adjustments (such as migration, scaling up or down, downgrading, switching accelerators, reallocating resources, etc.).

[0061] To improve the accuracy and robustness of judgments, this disclosure may adopt a multi-level, multi-strategy integrated evaluation framework, for example as follows: (1) Static threshold rule (basic layer) Set hard alarm thresholds for key indicators, such as: GPU temperature >90°C → Triggers downclocking or migration; Memory usage > 95% for 10 seconds → Trigger expansion or slice migration.

[0062] It is suitable for clearly defined, high-risk scenarios, with a fast response but low flexibility.

[0063] (2) Dynamic baseline detection (enhancement layer) Build a component performance baseline (e.g., average normal CPU utilization ±2σ) based on historical operating data (e.g., a 7-day sliding window). If the current state deviates significantly from the baseline (e.g., Z-score > 3), it is considered abnormal even if the static threshold is not reached; It can effectively identify "slow degradation" or "hidden bottlenecks caused by environmental changes".

[0064] (3) Multi-indicator correlation analysis (intelligent layer) Introduce causal graphs or lightweight graph neural networks (GNNs) to model dependencies between components: Example: If UPF latency increases + NIC packet loss rate increases + other slices on the same node show no abnormalities → it is determined to be a network bottleneck specific to this slice; If GPU utilization is high, temperature is normal, and inference throughput meets the target, it is considered to be under high-efficiency full load and no adjustments are triggered.

[0065] Avoid "single-point misjudgment" and improve the accuracy of root cause localization.

[0066] (4) Business context awareness (decision level) A weighted judgment is made based on slice type, tenant priority, and current business stage: Autonomous driving slices: extremely sensitive to latency jitter, even slight fluctuations trigger adjustments; Video cache slicing: can tolerate short-term congestion, prioritizes energy saving rather than migration; Nighttime batch processing tasks: Allow downgraded operation and postpone adjustment to off-peak hours.

[0067] S103, in response to the availability of adjustment conditions, adjust the component.

[0068] In this embodiment, once the adjustment conditions are met, the most suitable adjustment method can be dynamically selected from a predefined strategy library based on the adjustment reason, component type, slice priority, and platform resource status. It should be noted that there can be multiple adjustment strategies, and no limitation is made here; the specific strategy can be determined according to the actual component type or current operating conditions. For example, the adjustment strategy may be as follows: Minimal changes preferred: Prioritize resource reconfiguration over migration.

[0069] Business continuity is the priority: hot migration or dual-active redundancy is used for high-reliability slices.

[0070] Cost-sensitive adaptation: Normal slicing allows for short-term SLA degradation to save cross-node bandwidth.

[0071] In this embodiment, after adjustment, continuous tracking and feedback can be provided to slices and components. Adjustment is not the end point, but rather the starting point for a new round of monitoring. The hardware status of the adjusted components and the QoS indicators of the slices can be continuously collected to verify whether the adjustment effect meets the standards. Simultaneously, the inputs (status data), decisions (reasons), actions (types), and results (effects) of this adjustment can be recorded in the operation and maintenance knowledge base for subsequent strategy optimization and root cause analysis.

[0072] In this embodiment, the hardware status data of each component on which the target slice depends in the B5G cloud platform is first collected. Then, based on the hardware status data, it is determined whether the component meets the adjustment conditions. Finally, in response to the condition being met, the component is adjusted. Thus, this invention forms a closed loop of "perception-decision-execution" by sequentially collecting the hardware status data of each component on which the target slice depends in the B5G cloud platform, intelligently determining whether the adjustment conditions are met based on this data, and automatically performing targeted adjustments when the conditions are met. This mechanism can identify hardware performance bottlenecks or sub-health risks in advance, ensuring the quality of service and service level agreement fulfillment of network slices while avoiding unnecessary migration or scaling operations. It significantly improves the accuracy of adjustments and resource utilization efficiency, enhancing the automated operation and self-healing capabilities of the B5G cloud platform. It also supports differentiated scheduling based on slice type, effectively balancing the stability of high-reliability services with the energy efficiency requirements of ordinary services, thereby achieving high-performance, high-availability, and low-power intelligent cloud-network collaborative operation.

[0073] In one possible implementation, visual alarms can also be generated based on hardware status data.

[0074] Specifically, the visual alerts are built on Grafana Enterprise Edition to create a four-level visual dashboard, corresponding to the physical layer, virtual layer, business layer and global collaborative view.

[0075] Specifically, the physical layer view includes a server health status heatmap, real-time curves of data center environmental parameters, and PDU power consumption statistics; the virtual layer view includes VM / container resource utilization trend curves, CNF forwarding performance metrics, and a virtual network topology diagram; the service layer view includes a slice QoS metric waterfall chart, a service traffic distribution pie chart, and a slice resource usage bar chart; and the global collaborative view includes a cross-layer metric correlation analysis diagram of "physical-virtual-service-hardware," overall network resource utilization statistics, fault alarm distribution, and AI prediction results display, supporting one-click overview of the global status. Compared to existing technologies that independently display metrics at each layer, this design reduces the time for operations and maintenance personnel to locate faults by more than 50%. The cross-layer metric correlation analysis diagram uses a Sankey diagram format to show the causal relationship between hardware status, physical resource utilization, virtual resource performance, and service QoS.

[0076] Furthermore, the alarm strategy is based on PrometheusRule, configuring four levels of alarms: early warning, general alarm, critical alarm, and emergency, and defining a dynamic adjustment algorithm for alarm thresholds; an alarm aggregation algorithm is introduced to merge multiple alarms caused by the same root cause into a single alarm event, achieving an alarm noise reduction rate of 60%; an intelligent alarm routing algorithm pushes alarm information to the corresponding maintenance personnel, supporting multiple channels such as SMS, email, and WeChat Work, with an alarm response latency of <30 seconds; when the UPF packet loss rate is >5%, the latency is >20ms, or the hardware device status is abnormal, the work order system is automatically triggered to create a maintenance work order, achieving a work order closure rate of 98%.

[0077] The alarm threshold dynamic adjustment algorithm is based on the average value of the sliding window of the business load. It uses a linear regression algorithm to predict the load trend for the next 10 minutes and adjusts the alarm threshold to 120% of the predicted value. The sliding window size is 1 hour, and the step size is 10 minutes, ensuring that the threshold can adapt to changes in business load in real time.

[0078] The alarm aggregation algorithm uses the DBSCAN clustering algorithm based on time and spatial correlation, with a time window of 5 seconds and a spatial distance threshold of the same physical node or the same slice. Multiple clustered alarms are merged into a single root cause alarm, and possible causes of failure are labeled.

[0079] In the above embodiments, based on hardware status data, it can be determined whether the component meets the adjustment conditions, and further... Figure 2 To further explain, the method includes: S201, Get the component type of the component.

[0080] S202, determine the adjustment conditions based on the component type, and determine whether the component meets the adjustment conditions based on the hardware status data.

[0081] In this embodiment of the disclosure, in response to the component type being an antenna array, the beam gain data in the hardware status data is compared with the gain judgment threshold. In response to the beam gain data collected for N consecutive timestamps being less than the gain judgment threshold, it is determined that the component meets the adjustment conditions.

[0082] In this embodiment of the disclosure, the slice type of the target slice can be obtained, and then the components can be adjusted based on the slice type and hardware status data.

[0083] In one possible implementation, the antenna array in this disclosure employs a 32-channel massive MIMO modular structure, comprising 16 rectangular elements (50mm × 30mm in size), a reflector (200mm × 150mm × 3mm aluminum alloy), a feed network (microstrip line impedance 50Ω), and a mounting bracket (stainless steel, 8mm aperture), with an adjustable beamforming angle range of -60° to +60°.

[0084] In response to the slice type being a vehicle-to-everything (V2X) slice, the deployment location of the target slice's UPF is adjusted and migrated to the edge node. For example, the antenna array can dynamically track the UE's location through beamforming, and the monitoring module collects beam gain data every 100ms through the base station's northbound interface. When the gain is <-85dBm for three consecutive samplings, the intelligent decision-making module is automatically triggered to adjust the slice's UPF deployment location (migrate to the edge node).

[0085] In response to the IoT slice type, the antenna array is switched to wide-coverage beam mode, and the signal coverage quality parameters in the edge areas are improved. For example, a predictive tracking method based on UE measurement reports can be used according to the beam tracking algorithm. The UE's RSRP, RSRQ, and TA measurement values ​​are input, and Kalman filtering is used to predict the UE's position at the next moment, adjusting the beam pointing in advance with a tracking accuracy of ±1°.

[0086] In this embodiment of the disclosure, in response to the component type being a liquid cooling device, the coolant temperature in the hardware status data is compared with a temperature judgment threshold. In response to the coolant temperature being greater than the temperature judgment threshold, it is determined that the component meets the adjustment conditions.

[0087] To adjust the components, the target node can first be determined from the B5G cloud platform based on the particle swarm optimization algorithm. The target node is the node whose coolant temperature is lower than the set temperature. Then, the key network elements of the target slice are migrated to the target node.

[0088] It should be noted that the liquid cooling device in this disclosure adopts an integrated structure of distributed pipeline and cold plate, including a main coolant tank (10L capacity, PP material), distributed copper pipeline (15mm inner diameter, 2mm wall thickness), CPU / GPU dedicated cold plate (with chip surface roughness Ra≤0.8μm), micro water pump (rated flow rate 5L / min) and PT100 temperature sensor (M12 thread).

[0089] For example, when the temperature sensor detects a coolant temperature >40℃, the multimodal monitoring module inputs the temperature data into the intelligent collaborative decision-making module, triggering a "UPF node migration" decision. Based on the improved PSO algorithm, a target node with a temperature <30℃ is selected. The improved PSO algorithm uses real-number encoding, with the particle dimension being the number of nodes. The fitness function is f=0.6×(1-temperature / 100)+0.4×(1-node load), and the inertia weight uses a linear decreasing strategy (decreasing from 0.9 to 0.4). The learning factors c1=c2=2, and the convergence condition is that the fitness value change is less than 1e-5 for 10 consecutive generations. After pre-deploying slice instances, traffic switching occurs, with a service interruption time of <100ms. Compared to the traditional "manual migration after hardware failure" solution, this linkage reduces the service interruption rate caused by hardware failure by 80%, while also reducing the average server temperature by 8℃ and extending hardware lifespan by 30%.

[0090] In this embodiment of the disclosure, in response to the component type being a beam splitter, the optical power attenuation value of the acquisition input / output port in the hardware status data is compared with the optical attenuation judgment threshold. In response to the optical power attenuation value acquired over N consecutive timestamps being greater than the optical attenuation judgment threshold, it is determined that the component meets the adjustment conditions.

[0091] When adjusting components, the current splitter port of the target slice can be migrated to the spare port of the splitter.

[0092] It should be noted that the beam splitter in this disclosure adopts a 1-to-8 beam splitter structure, including an input port (SC / APC interface, ceramic ferrule), a beam splitter chip (PLC planar waveguide material, size 10mm×8mm), 8 output ports (SC / UPC interface, evenly distributed with a spacing of 15mm) and a metal casing (size 80mm×50mm×20mm, heat dissipation hole diameter 3mm).

[0093] In this disclosure, for example, the monitoring module collects the optical power attenuation value of the input / output ports in real time using an optical power sensor, with a sampling frequency of 10 seconds per sampling. When the attenuation value > 0.3dB and persists for 30 seconds, a port switching decision is automatically triggered, migrating service traffic to the backup port. Port switching is achieved through an optical switch, and the switching command is sent to the optical switch device via the SNMP protocol, with a switching latency of < 5ms, ensuring the stability of optical signal transmission.

[0094] In this embodiment of the disclosure, in response to the component type being an industrial-grade optical modem, the packet loss rate in the hardware status data is compared with the packet loss rate judgment threshold. In response to the packet loss rate collected over N consecutive timestamps being greater than the packet loss rate judgment threshold, it is determined that the component meets the adjustment conditions.

[0095] When adjusting components, a reset command can be configured and executed to reset the current network interface of the target network interface.

[0096] It should be noted that the industrial-grade optical modem disclosed herein adopts a modular plug-in structure, including a power module (AC220V input, DC12V output), an optical module interface (SC interface with dust cover), four RJ45 gigabit ports, a main control board (size 120mm×80mm, integrating CPU and memory chips) and a cooling fan (40mm×40mm, speed 3000rpm).

[0097] For example, when the monitoring module detects a packet loss rate of >1% on the optical modem's network port for 10 seconds, it automatically triggers a configuration reset command. It then logs into the optical modem's management interface via Telnet and executes a network port restart command. The fault recovery time is less than 10 seconds, requiring no manual intervention. If the fault persists after three consecutive resets, a hardware replacement work order is automatically generated.

[0098] In another possible approach, future resource load trends can be predicted based on hardware status data.

[0099] In this disclosure, a cross-domain fusion prediction scheme based on "multimodal feature fusion + hybrid neural network + online adaptive optimization" can be adopted based on the cross-domain fusion prediction analysis module. Based on the enhanced prediction network of three-layer LSTM + bidirectional GRU + attention mechanism, historical traffic data is processed and trained in multiple steps to predict the resource load trend in the next 4 minutes with a prediction error rate of less than 5%. This algorithm is not a simple superposition of existing technologies, but achieves non-obvious technical effects through the collaborative design of features and models.

[0100] Specifically, the algorithm architecture adopts a three-tier architecture of "feature engineering + model training + inference service". Feature engineering uses Python scripts to extract time-series features (trends / seasonality / outliers) and includes three key steps: data preprocessing, feature extraction, and feature selection. The model training library improves prediction accuracy by integrating the LSTM master model. The inference service deploys models using TensorFlowServing, supports REST / gRPC interfaces, and has a response latency of <50ms.

[0101] Furthermore, the data preprocessing stage includes three steps: missing value imputation, outlier detection and correction, and data normalization. Missing value imputation uses a time-series interpolation algorithm based on an attention mechanism, combined with historical data of similar time periods and adjacent indicator data, achieving an imputation accuracy of 96%. Outlier detection and correction constructs an anomaly detection model combining isolated forest and DBSCAN clustering to identify outliers in the data, achieving an anomaly detection accuracy of 95%. For outliers, a correction algorithm based on moving average is used for replacement to ensure data integrity. Data normalization uses an improved Z-Score normalization algorithm, considering the temporal characteristics and periodicity of the data. This algorithm avoids the dependence of traditional Z-Score normalization on the global data distribution and is more adaptable to the periodic fluctuations of time-series data.

[0102] Indeed, the improved Z-Score normalization algorithm uses a sliding window of 24 hours for normalization based on the local mean and standard deviation, as shown in the formula:

[0103] in This represents the mean of the data within the sliding window. This represents the standard deviation of the data within the sliding window. This algorithm can effectively capture the daily periodic variations of time-series data, and compared with traditional global Z-Score normalization, the variance of the normalized data is reduced by 25%.

[0104] Furthermore, the feature extraction stage includes temporal feature extraction, cross-domain correlation feature extraction, and business semantic feature extraction: Temporal feature extraction uses the sliding window method (window size 168 hours) to extract trend, seasonality, and volatility features, and introduces wavelet transform to extract multi-scale features of the data, decomposing them into low-frequency trend components and high-frequency detail components to capture long-term trends and short-term fluctuations respectively; Cross-domain correlation feature extraction constructs a cross-layer feature correlation matrix of "physical-virtual-business-hardware", uses the mutual information algorithm to calculate the correlation degree between indicators at different levels, and selects feature pairs with a correlation degree > 0.6 to construct cross-domain correlation features; Business semantic feature extraction constructs one-hot encoded features based on business type, and constructs a weighted feature vector in combination with the QoS requirements of the business to realize the fusion of business features with resource indicators and hardware status.

[0105] It is confirmed that the wavelet transform uses the db4 wavelet basis function, with a decomposition level of 3, resulting in one low-frequency trend component and three high-frequency detail components. The low-frequency trend component reflects the long-term trend of the data, while the high-frequency detail components reflect the instantaneous fluctuation characteristics of the data.

[0106] The formula for calculating mutual information is as follows:

[0107] in Let X and Y be the joint probability density. and This represents the marginal probability density. When the mutual information value is >0.6, the two indicators are considered to have a strong correlation.

[0108] Specifically, feature selection employs a feature importance evaluation algorithm based on gradient boosting decision tree (GBDT) to calculate the contribution of each feature to the prediction result and remove redundant features with a contribution of <0.01. L1 regularization (Lasso) is used to sparsify features, reducing feature dimensions (dimensionality compression ratio up to 40%) and improving model training efficiency and generalization ability. The final feature set includes three major categories of 64-dimensional features: temporal features, cross-domain association features, and business semantic features, providing comprehensive support for model training.

[0109] Furthermore, the model training employs an LSTM model, which includes a 3-layer LSTM network structure, an attention mechanism, and a bidirectional GRU. Each LSTM network layer has 256 neurons, achieving an input window of 168 hours and a prediction step size of 4 minutes. This hybrid neural network uses fixed parameter configurations, constituting a unique technical hallmark of this approach. The model architecture includes an input layer, a feature encoding layer, a deep feature learning layer, an attention enhancement layer, and an output layer. The input layer receives 64-dimensional feature vectors and maps them to a 128-dimensional low-dimensional space through an embedding layer, enhancing feature representation. The feature encoding layer consists of a bidirectional GRU network, containing two layers of bidirectional GRU units, each with 128 neurons. The bidirectional GRU simultaneously encodes temporal features from both forward and backward directions, capturing the dependencies between data. High-frequency detail components from wavelet transform decomposition are input into the bidirectional GRU, and forward and backward encoding enhances the perception of instantaneous fluctuations. The forward GRU update formula is: ; The inverse GRU update formula reverses the time series, and the calculation process is the same as above. Finally, the outputs of the forward and inverse GRUs are concatenated into a 256-dimensional encoded vector. The deep feature learning layer consists of three LSTM network layers, each with 256 neurons. Dropout regularization (dropoutrate=0.3) and L2 regularization (λ=0.001) are used to prevent overfitting. The low-frequency trend components of wavelet transform decomposition are input into the LSTM network, and the gating mechanism effectively captures long-term temporal dependencies, making it suitable for long-term trend learning of historical data. The output formula for each LSTM layer is: ; The attention enhancement layer introduces a multi-head self-attention mechanism, dividing the output of the deep feature learning layer into eight heads for parallel attention computation. Each head captures feature dependencies of different dimensions. Heads 1-2 focus on temporal feature weights, heads 3-4 on cross-domain correlation feature weights, heads 5-6 on business semantic feature weights, and heads 7-8 on hardware state feature weights. This differentiated attention allocation strengthens the contribution of key features to the prediction results. Finally, the layers are concatenated into a 256-dimensional feature vector. The attention weight calculation formula is as follows:

[0110] ; in For learnable parameters, The dimension of the attention head is 256 / 8=32; the output layer consists of a fully connected layer (128 neurons) and a regression layer, which outputs the resource load prediction values ​​(CPU utilization, traffic, latency, etc.) for the next 4 minutes.

[0111] It is evident that the synergistic design of multimodal features and hybrid neural networks brings significant technical gains: compared to a single LSTM model, the prediction error is reduced by 30% after introducing wavelet transform multi-scale features; through differentiated feature weight allocation via a multi-head self-attention mechanism, the load prediction accuracy for latency-sensitive services is improved to 95%, far exceeding the 80% of traditional models, fully demonstrating that this combination is not a simple superposition, but rather achieves a performance breakthrough through synergy. Existing technologies typically input all features into a single LSTM model, failing to effectively capture both long-term trends and instantaneous fluctuations simultaneously, and neglecting the different contributions of different types of features to the prediction results. This solution specifically addresses these issues through wavelet transform decomposition and a differentiated attention mechanism.

[0112] Furthermore, the training strategy involves first normalizing the data, then splitting it into a training set (67%) and a test set (33%). The training set is used to train the LSTM network, while the test set is used to validate the effectiveness of each LSTM design. Time-series cross-validation is employed to avoid model overfitting caused by data leakage. The optimization objective combines minimizing MSE with maximizing MAPE, using a hybrid loss function, as shown in the formula:

[0113]

[0114] in (Determined through 5-fold cross-validation) For the true value, The predicted value is N, and the number of samples is N. The learning rate is set to 0.001, and a learning rate decay strategy is adopted (decaying to 0.9 every 50 rounds). The iteration is set to 200 rounds, and the batch size is set to 64. The above training parameters constitute the unique technical marker of this scheme, ensuring the repeatability of model training and the identifiability of infringement.

[0115] It is determined that the reasonable equivalent range of the above training parameters is: α=0.7±0.1, learning rate 0.001±0.0005, number of iterations 200±20 rounds, batch size 64±8. Any parameter adjustments made by the infringer within this range fall within the protection scope of this scheme.

[0116] Certainly, the LSTM model details need to be regularized, using L2 regularization (λ=0.001) + Dropout (0.3) to prevent overfitting.

[0117] Specifically, the online adaptive update mechanism is designed to adapt to dynamic changes in network status: the latest monitoring data is collected every hour to incrementally train the model and update the model parameters; a sliding window is used to update the training set, and historical data older than 168 hours is removed to ensure that the model always learns based on the latest data and maintains the stability of prediction accuracy; when the network undergoes significant changes, the model is triggered to retrain completely, with a retraining time of less than 30 minutes, ensuring that the model can quickly adapt to new scenarios.

[0118] Specifically, "significant network changes" are defined as: adding new network slice types, core network topology changes, large-scale hardware replacement (>30% of nodes), and significant changes in service traffic patterns (daily traffic changes exceeding 50% for 3 consecutive days). Full retraining uses historical data from the most recent month, with training parameters remaining consistent with the initial training.

[0119] Confirmed, model deployment and service adopt a "TensorFlowServing + Kubernetes" deployment scheme to achieve high availability and high concurrency of the model. Model deployment involves packaging the trained model into a Docker image and deploying it on a Kubernetes cluster, supporting horizontal scaling and a maximum concurrent processing capacity of 1000 QPS. The interface design provides a custom RESTful API and gRPC interface. The RESTful API request address is fixed at ` / api / v1 / b5g / predict / load`, and the request parameters include "prediction object, historical data window, prediction step size". The response format includes "predicted value, ... "Confidence level and recommended decision-making" are key aspects of this solution. The gRPC interface uses a custom protocol buffer definition, with the interface identifier fixed as "b5g.predict.v1.LoadPrediction". The above interface design constitutes the key basis for infringement identification in this solution. The RESTful API is suitable for low-concurrency scenarios (response latency <100ms), while gRPC is suitable for high-concurrency scenarios (response latency <50ms) and supports batch prediction. Service monitoring uses Prometheus to monitor the model service's response latency, prediction accuracy, request success rate, and other indicators. When the prediction accuracy is below 90% or the response latency exceeds 100ms, an alarm is automatically triggered to notify the operation and maintenance personnel to investigate.

[0120] Specifically, "significant network changes" are defined as: adding new network slice types, core network topology changes, large-scale hardware replacement (>30% of nodes), and significant changes in service traffic patterns (daily traffic changes exceeding 50% for 3 consecutive days). Full retraining uses historical data from the most recent month, with training parameters remaining consistent with the initial training.

[0121] Through the above steps, this module achieves the effects of data retrieval, data preprocessing, model training, and prediction. The prediction module is seamlessly integrated with the OSM orchestration platform, realizing full-process automation from load prediction to UPF instantiation, reducing the manual intervention rate to 0, and promoting the intelligent transformation of communication networks.

[0122] In another possible implementation, the slices can be scheduled based on hardware state data.

[0123] In this disclosure, the intelligent collaborative decision-making module can make intelligent decisions on resource allocation, slice scheduling, and fault handling based on cross-domain fusion prediction results, combined with multi-objective optimization algorithms and slice collaboration strategies, and output the optimal orchestration decision instructions, while also including hardware state-driven decision logic.

[0124] For the resource competition problem involving multiple slices and multiple services, a multi-objective resource allocation decision-making algorithm based on the non-dominated sorting genetic algorithm (NSGA-Ⅲ) is proposed. The optimization objectives include: maximizing slice resource utilization, maximizing slice QoS satisfaction, and maximizing resource allocation fairness.

[0125] Specifically, the decision variables are defined as follows: Assuming there are M network slices and N physical server nodes in the network, define the decision variables... Indicates physical server The Class resources (CPU, memory, bandwidth) are allocated to network slices The resource proportion, the value of this variable satisfies And for any server With any resource type The sum of the proportions of this type of resource used by all network slices must meet the following requirements. .

[0126] Furthermore, the objective function is constructed in three dimensions: maximizing resource utilization, maximizing slice QoS satisfaction, and maximizing resource allocation fairness. The formula for maximizing resource utilization is:

[0127] in Let the total capacity of resource type k be the resource type of server j. Let i be the utilization rate of resource type k for slice i; the formula for maximizing slice QoS satisfaction is: , Where Q is the set of QoS metrics. For slice i, the actual QoS metric. Let be the QoS requirement threshold for slice i; The formula for maximizing fairness in resource allocation is: .

[0128] Specifically, the constraints include resource constraints, QoS constraints, and isolation constraints. Compared with the traditional greedy algorithm, this algorithm improves resource utilization by more than 20%, slice QoS satisfaction by 30%, and resource allocation fairness by 25%, effectively balancing the resource needs of multiple slices with QoS guarantees.

[0129] Indeed, the intelligent scheduling decision for network slices is based on the predicted slice load trend and multi-objective resource allocation scheme to realize dynamic scheduling decisions for slices, including slice instantiation, scaling up and down, migration and termination.

[0130] Specifically, when a new service request is added, the slice instantiation decision automatically selects the optimal deployment mode and deployment node based on the service's QoS requirements and predicted resource requirements, and generates a network slice template (NST), which includes the slice's network service descriptor (NSD), virtual network function descriptor (VNFD), and resource configuration parameters. Through integration with the automated LCM module, zero-contact slice instantiation is achieved, with an instantiation time of less than 5 minutes.

[0131] Furthermore, the slice scaling decisions are based on the load trend for the next 4 minutes output by the cross-domain fusion prediction module. When the predicted load exceeds 80% of the slice's current resource capacity, a scaling decision is triggered, increasing the resource quota according to the multi-objective resource allocation scheme. When the predicted load is less than 30% of the slice's current resource capacity for 10 minutes, a scaling decision is triggered, releasing redundant resources. The execution of scaling decisions is linked with the automated LCM module through a standardized API, with scaling time <30 seconds and scaling time <20 seconds.

[0132] Specifically, the slice migration decision is triggered when the physical resource utilization of the slice deployment node exceeds 90%, or when the node shows signs of failure, or when the slice QoS indicators do not meet the requirements, or when the hardware device is in an abnormal state. During the migration process, a pre-deployment mechanism is adopted (the slice instance is deployed on the target node first, and then the service traffic is switched), and the service interruption time is less than 100ms to ensure service continuity.

[0133] The defined session state synchronization process for slice migration is as follows: The SMF sends a session pause command to the source UPF via the N11 interface. The source UPF synchronizes the UE context (including PDU session information, GTPU tunnel information, and QoS rules) to the target UPF via the N4 interface. After completing context synchronization, the target UPF sends a synchronization completion response to the SMF. The SMF sends a route update command to the UE to switch the UE's GTPU tunnel to the target UPF. After the switch is completed, the SMF sends a resource release command to the source UPF.

[0134] Optionally, the slice termination decision is triggered when the business corresponding to the slice terminates or the slice idle time exceeds 30 minutes, releasing all physical and virtual resources occupied by the slice with a resource release rate of 100% to avoid resource waste.

[0135] Specifically, intelligent fault decision-making is based on multimodal monitoring data and prediction results to build a closed-loop fault decision-making mechanism of "fault detection-fault location-fault repair", covering the handling of all scenarios of hardware faults, virtual resource faults, network faults and business faults.

[0136] Specifically, fault detection employs a combination of threshold and machine learning methods. Threshold detection identifies obvious anomalies, while the machine learning model (random forest) identifies early warning signs of faults, achieving a fault detection accuracy of 98% and a detection latency of less than 5 seconds. Fault localization utilizes a knowledge graph-based fault reasoning method, constructing a causal relationship graph encompassing "hardware faults - virtual resource faults - network faults - service faults." Root cause localization is located through reverse reasoning, achieving a fault localization accuracy of 95% and a localization time of less than 30 seconds. Fault repair automatically generates repair plans based on the fault type and root cause: in hardware faults, when the antenna array VSWR exceeds the limit, beamforming parameter adjustment or channel switching commands are automatically triggered. If adjustments are ineffective, a hardware replacement work order is generated; if the liquid cooling device's heat dissipation efficiency is insufficient, the water pump speed is automatically adjusted or a backup heat dissipation unit is activated; if the temperature still exceeds the threshold, UPF node migration is triggered; if the optical power attenuation of the splitter exceeds the threshold, a port switching or hardware replacement work order is automatically generated; if the industrial-grade optical modem fails to register, configuration reset or port allocation adjustment is automatically executed; virtual resource failure triggers resource restart or migration decisions; network failure triggers route adjustment or link switching decisions; service failure triggers QoS parameter adjustment or slice scaling decisions; fault repair solutions are executed through an automated LCM module, with an average fault repair time (MTTR) of <10 minutes, improving efficiency by 80% compared to traditional manual repair.

[0137] The hardware state-driven decision logic parameters are limited as follows: the antenna array beam parameter adjustment range is ±10°, with a maximum of 3 adjustments; the liquid cooling device water pump speed adjustment step is 0.5L / min, and the maximum speed does not exceed 120% of the rated speed; the splitter port switching can be attempted a maximum of 2 times, and if it still fails, a hardware replacement work order will be generated.

[0138] Corresponding to the slice switching methods provided in the above embodiments, one embodiment of this disclosure also provides a slice switching device. Since the slice switching device provided in this disclosure corresponds to the slice switching methods provided in the above embodiments, the implementation methods of the above slice switching methods are also applicable to the slice switching device provided in this disclosure, and will not be described in detail in the following embodiments.

[0139] Figure 3 This is a schematic diagram of a slice switching device according to one embodiment of the present disclosure. As shown in FIG3, the slice switching device 300 includes: The acquisition module 310 is used to acquire hardware status data of the components on which the target slice depends in the B5G cloud platform.

[0140] The judgment module 320 is used to determine whether the component meets the adjustment conditions based on the hardware status data.

[0141] Adjustment module 330 is used to adjust the component in response to the availability of the adjustment conditions.

[0142] According to one embodiment of this disclosure, determining whether the component meets the adjustment conditions based on the hardware status data includes: obtaining the component type of the component; determining the adjustment conditions based on the component type; and determining whether the component meets the adjustment conditions based on the hardware status data.

[0143] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being an antenna array, comparing the beam gain data in the hardware status data with a gain judgment threshold; and in response to the beam gain data collected over N consecutive timestamps being less than the gain judgment threshold, determining that the component meets the adjustment condition.

[0144] According to one embodiment of this disclosure, adjusting the component includes: obtaining the slice type of the target slice; and adjusting the component based on the slice type and the hardware status data.

[0145] According to one embodiment of this disclosure, adjusting the component based on the slice type and the hardware status data includes: in response to the slice type being a vehicle-to-everything (V2X) slice, adjusting the UPF deployment location of the target slice to migrate towards the edge node; and in response to the slice type being an Internet of Things (IoT) slice, switching the antenna array to a wide-coverage beam mode and improving the signal coverage quality parameters in the edge area.

[0146] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being a liquid cooling device, comparing the coolant temperature in the hardware status data with a temperature judgment threshold; and in response to the coolant temperature being greater than the temperature judgment threshold, determining that the component meets the adjustment condition.

[0147] According to one embodiment of this disclosure, adjusting the component includes: determining a target node from the B5G cloud platform based on a particle swarm optimization algorithm, wherein the target node is a node whose coolant temperature is lower than a set temperature; and migrating the key network elements of the target slice to the target node.

[0148] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and determining whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being a beam splitter, comparing the optical power attenuation value of the acquisition input / output port in the hardware status data with an optical attenuation judgment threshold; and in response to the optical power attenuation value acquired over N consecutive timestamps being greater than the optical attenuation judgment threshold, determining that the component meets the adjustment condition.

[0149] According to one embodiment of this disclosure, adjusting the component includes: migrating the current beam splitter port of the target slice to a spare port of the beam splitter.

[0150] According to one embodiment of this disclosure, determining the adjustment condition based on the component type and judging whether the component meets the adjustment condition based on the hardware status data includes: in response to the component type being an industrial-grade optical modem, comparing the packet loss rate in the hardware status data with a packet loss rate judgment threshold; and in response to the packet loss rate collected over N consecutive timestamps being greater than the packet loss rate judgment threshold, determining that the component meets the adjustment condition.

[0151] According to one embodiment of this disclosure, adjusting the component includes: configuring a reset command and executing a network port restart command to reset the current network port of the target network port.

[0152] Therefore, this invention forms a closed loop of "perception-decision-execution" by sequentially collecting hardware status data of each component on which the target slice depends in the B5G cloud platform, intelligently determining whether adjustment conditions are met based on this data, and automatically executing targeted adjustments when the conditions are met. This mechanism can identify hardware performance bottlenecks or sub-health risks in advance, ensuring the quality of service and service level agreement fulfillment of network slices while avoiding unnecessary migration or scaling operations, significantly improving the accuracy of adjustments and resource utilization efficiency. It not only enhances the automated operation and maintenance and self-healing capabilities of the B5G cloud platform, but also supports differentiated scheduling based on slice type, effectively balancing the stability of high-reliability services with the energy efficiency requirements of ordinary services, thereby achieving high-performance, high-availability, and low-power intelligent cloud-network collaborative operation.

[0153] To implement the above embodiments, this disclosure also proposes an electronic device 400. Figure 4 This is a schematic diagram of an electronic device according to one embodiment of the present disclosure, such as... Figure 4 As shown, the electronic device 400 includes: a processor 401 and a memory 402 communicatively connected to the processor. The memory 402 stores instructions executable by at least one processor. The instructions are executed by at least one processor 401 to achieve the functions described in this disclosure. Figures 1-2 Slice switching method of the embodiment.

[0154] To implement the above embodiments, this disclosure also proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to implement the present disclosure. Figures 1-2 Slice switching method of the embodiment.

[0155] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program, which, when executed by a processor, implements the features of this disclosure. Figures 1-2 Slice switching method of the embodiment.

[0156] It should be noted that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. Furthermore, such collection / sharing should only be conducted after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes authorization of relevant user information before the user uses the function. In addition, any necessary steps must be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy policies and procedures.

[0157] This application is intended to provide an implementation scheme for users to selectively prevent the use or access to their personal information data. Specifically, this disclosure is intended to provide hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, risks can be minimized by restricting data collection and deleting data. Furthermore, where applicable, such personal information is de-identified to protect user privacy.

[0158] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0159] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0160] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0161] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that contains, stores, communicates, propagates, or transmits programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0162] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0163] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0164] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0165] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A slice switching method, characterized in that, include: Collect hardware status data of each component on which the target slice depends in the B5G cloud platform; Based on the hardware status data, determine whether the component meets the adjustment conditions; In response to the availability of the adjustment conditions, the component is adjusted.

2. The method according to claim 1, characterized in that, The step of determining whether the component meets the adjustment conditions based on the hardware status data includes: Obtain the component type of the component; The adjustment conditions are determined based on the component type, and the hardware status data is used to determine whether the component meets the adjustment conditions.

3. The method according to claim 2, characterized in that, The step of determining the adjustment conditions based on the component type and determining whether the component meets the adjustment conditions based on the hardware status data includes: In response to the component type being an antenna array, the beam gain data in the hardware status data is compared with a gain judgment threshold. In response to the beam gain data collected over N consecutive timestamps being less than the gain judgment threshold, it is determined that the component meets the adjustment condition.

4. The method according to claim 3, characterized in that, The adjustment of the component includes: Obtain the slice type of the target slice; The components are adjusted based on the slice type and the hardware status data.

5. The method according to claim 4, characterized in that, The adjustment of the component based on the slice type and the hardware status data includes: In response to the fact that the slice type is a vehicle-to-everything (V2X) slice, the UPF deployment location of the target slice is adjusted and migrated to the edge node; In response to the slice type being an IoT slice, the antenna array is switched to a wide-coverage beam mode, and the signal coverage quality parameters in the edge area are improved.

6. The method according to claim 2, characterized in that, The step of determining the adjustment conditions based on the component type and determining whether the component meets the adjustment conditions based on the hardware status data includes: In response to the component type being a liquid cooling device, the coolant temperature in the hardware status data is compared with a temperature judgment threshold. In response to the coolant temperature being greater than the temperature judgment threshold, it is determined that the component meets the adjustment condition.

7. The method according to claim 6, characterized in that, The adjustment of the component includes: The target node is determined from the B5G cloud platform based on the particle swarm optimization algorithm. The target node is the node whose coolant temperature is lower than the set temperature. The key network elements of the target slice are migrated to the target node.

8. The method according to claim 2, characterized in that, The step of determining the adjustment conditions based on the component type and determining whether the component meets the adjustment conditions based on the hardware status data includes: In response to the component type being a beam splitter, the optical power attenuation value of the acquisition input / output port in the hardware status data is compared with the optical attenuation judgment threshold. In response to the optical power attenuation value collected over N consecutive times being greater than the optical attenuation judgment threshold, it is determined that the component meets the adjustment condition.

9. The method according to claim 8, characterized in that, The adjustment of the component includes: The current splitter port of the target slice is migrated to the spare port of the splitter.

10. The method according to claim 2, characterized in that, The step of determining the adjustment conditions based on the component type and determining whether the component meets the adjustment conditions based on the hardware status data includes: In response to the component type being an industrial-grade optical modem, the packet loss rate in the hardware status data is compared with the packet loss rate judgment threshold. In response to the packet loss rate being greater than the packet loss rate judgment threshold collected over N consecutive timestamps, it is determined that the component meets the adjustment condition.

11. The method according to claim 10, characterized in that, The adjustment of the component includes: Configure a reset command, and execute the reset command to reset the current network interface of the target network interface.

12. A slice switching device, characterized in that, include: The acquisition module is used to acquire hardware status data of the various components on which the target slice depends in the B5G cloud platform; The judgment module is used to determine whether the component meets the adjustment conditions based on the hardware status data. An adjustment module is configured to adjust the component in response to the availability of the adjustment conditions.

13. An electronic device, characterized in that, Including memory and processor; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-11.

Images