Backhaul link monitoring method and system applied to 5G base station

Abstract

The invention relates to the technical field of backhaul monitoring, in particular to a backhaul link monitoring method and system applied to a 5G base station, and the method comprises the steps: collecting physical signal statistical data, backhaul service data and network topology data of a backhaul link in real time; according to physical signal statistical data, a physical layer entropy value is calculated through a Shannon entropy formula, and recessive degradation characteristics of a physical layer are quantified; a pulse neural network space-time coding mechanism is used to generate business layer anomaly prediction parameters, and cross-layer anomaly prediction is realized; according to service types and priorities, constructing a time-varying service awareness reference model by adopting a chaos kinetic equation to serve as a dynamic judgment reference; when the return state is abnormal, combining the physical layer entropy and the network topology data to construct a feedback correction loop, and generating a self-adaptive protection strategy; according to the method, advanced prediction and accurate positioning of recessive degradation of the backhaul link are realized, and the self-healing capability and the service quality guarantee level of the backhaul link in a multi-service scene are improved.

Description

Technical Field

[0001] This invention relates to the field of backhaul monitoring technology, and in particular to a backhaul link monitoring method and system applied to 5G base stations. Background Technology

[0002] Against the backdrop of large-scale deployment of 5G communication networks and diversified development of service scenarios, the base station backhaul link, as a key infrastructure connecting the access network and the core network, directly determines the user experience of services such as ultra-reliable low-latency communication and enhanced mobile broadband in terms of transmission quality and reliability. Traditional backhaul monitoring technologies focus on monitoring explicit states such as link connectivity and traffic congestion, which are insufficient to address complex challenges such as implicit degradation of physical layer signals, cross-layer fault propagation, and service quality assurance for differentiated services. First, implicit degradation characteristics such as physical layer signal fluctuations and transient bit error rates are difficult to capture effectively by traditional threshold alarm mechanisms, and there is a lack of quantitative correlation models between these characteristics and service layer performance degradation, leading to delayed fault prediction and difficulty in root cause location. Second, existing monitoring methods are mostly limited to single-layer indicator analysis and fail to integrate multi-dimensional data such as physical media layer, service flow, and network topology, making it impossible to achieve a holistic understanding of the backhaul status and dynamic deduction of abnormal propagation paths. In addition, traditional policy scheduling mechanisms are static and rigid, unable to dynamically generate protection policies based on real-time service types and priority requirements, resulting in low backhaul self-healing efficiency and difficulty in meeting the stringent fault recovery requirements of 5G.

[0003] For example, Chinese patent CN112052231B discloses a method and device for monitoring return records. This method receives monitoring requests to determine the target single-table identifier and abnormal monitoring statements, retrieves data from a single-table database synchronized with the online database according to rules, and then queries abnormal return records. This prior art mainly solves the problems of cumbersome configuration, limited number of single tables, and high maintenance costs in traditional wide-table solutions based on ElasticSearch, achieving configurable, near real-time data monitoring and anomaly detection.

[0004] Existing technologies still suffer from the deep-seated problems raised in the background: they focus on fault classification of service layer indicators, failing to incorporate deep features of physical layer signals, thus unable to achieve early detection and prediction of "hidden degradation," and failing to construct a comprehensive closed-loop linkage mechanism, resulting in passive and delayed strategy generation. Therefore, existing solutions still exhibit significant shortcomings in addressing the cascading service quality degradation caused by hidden physical layer faults in 5G backhaul links, including fragmented monitoring dimensions, insufficient predictive capabilities, and a lack of adaptive protection strategies. To solve these systemic technical challenges, this invention proposes a backhaul link monitoring method and system for 5G base stations. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a backhaul link monitoring method and system for 5G base stations. By statistically analyzing physical signal data, the method quantifies the implicit degradation characteristics of the physical layer, predicts service layer anomalies, and achieves cross-layer prediction from physical layer characteristics to service layer anomalies. Based on service type and priority, a time-varying service perception benchmark model is constructed as a dynamic evaluation benchmark. When the backhaul status is abnormal, a feedback correction loop is constructed by combining physical layer entropy value and network topology data to generate an adaptive protection strategy. This achieves advanced prediction and accurate location of implicit degradation of the backhaul link, improving the self-healing capability and service quality assurance level of the backhaul link in multi-service scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Methods for monitoring backhaul links applied to 5G base stations include: Acquire backhaul link monitoring data of 5G base stations, the backhaul link monitoring data including physical signal statistics, backhaul service data and network topology data; Based on the physical signal statistics, statistical analysis is performed using probability distribution methods. The physical layer entropy value is calculated using the Shannon entropy formula. Based on the physical layer entropy value and the backhaul service data, the service layer anomaly prediction parameters are calculated using the spatiotemporal coding mechanism of a spiking neural network. Based on the backhaul service data, the service type and priority are determined, and a time-varying service perception benchmark model is constructed using chaotic dynamics equations. By quantifying the matching deviation between the anomaly prediction parameters of the business layer and the time-varying business perception benchmark model, the hierarchical backhaul status assessment results are output. When the hierarchical backhaul status assessment result is abnormal, a feedback correction loop is constructed based on the change gradient of the physical layer entropy value and the network topology data, and an adaptive protection strategy is generated based on the feedback correction loop.

[0007] The backhaul link monitoring data includes physical signal statistics, backhaul service data, and network topology data, among which: The physical signal statistics include signal quality indicators, signal stability indicators, and signal transmission characteristic indicators. The signal quality indicators are obtained based on the signal-to-noise ratio and bit error rate. The signal stability indicators are obtained based on link jitter and interruption frequency. The signal transmission characteristic indicators are obtained based on signal strength and transmission delay. The backhaul service data includes service traffic characteristics, data packet loss characteristics, transmission rate, and service continuity characteristics. The network topology data includes link connectivity, node load characteristics, and link redundancy characteristics.

[0008] Based on the physical signal statistics, statistical analysis is performed using probability distribution methods, and the physical layer entropy value is calculated using the Shannon entropy formula, including: Based on the time series data and parameter characteristics of each parameter in the physical signal statistics within a preset time window, a probability distribution method adapted to the parameter characteristics is used for statistical analysis to obtain the probability distribution characteristics of signal quality, link stability, and transmission performance. Based on the signal quality probability distribution characteristics, link stability probability distribution characteristics, and transmission performance probability distribution characteristics, the signal quality entropy value, link stability entropy value, and transmission performance entropy value are calculated respectively using the Shannon entropy formula. The physical layer entropy value is calculated using a weighted fusion calculation formula based on the signal quality entropy value, link stability entropy value, and transmission performance entropy value.

[0009] The anomaly prediction parameters for the business layer are calculated using a spatiotemporal coding mechanism derived from a spiking neural network. This spatiotemporal coding mechanism includes both temporal and spatial coding mechanisms. Based on the time sequence of the physical layer entropy value, the physical layer entropy value fluctuation feature vector is calculated by a sliding window. Based on the parameters in the backhaul service data, the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector and time distribution feature vector are calculated. The traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector are dimensionally aligned to obtain an aligned feature vector. According to the time coding mechanism, the time series data corresponding to the aligned feature vector is converted into a pulse sequence. According to the spatial coding mechanism, the pulse sequence is converted into a spatiotemporal pulse feature matrix through neuron group allocation and feature association learning. Based on the spatiotemporal pulse feature matrix, a cross-layer anomaly prediction model is constructed through feature learning and cross-layer information fusion of a multi-layer spiking neural network. Based on the cross-layer anomaly prediction model, the predicted sequences of business layer anomaly probability, anomaly impact range, and anomaly duration are calculated through a probability normalization function. The predicted sequences of business layer anomaly probability, anomaly impact range, and anomaly duration are weighted and fused and mapped to a specified interval to obtain business layer anomaly prediction parameters.

[0010] Determining the service type and priority based on the returned service data includes: The traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector are quantized to generate standardized feature parameters, which include traffic pattern features, quality sensitivity features, bandwidth demand features, and time pattern features. The corresponding service type feature identifier is determined based on the standardized feature parameters. The service type feature identifier includes traffic type identifier, reliability type identifier, bandwidth type identifier, and time type identifier. The service type is determined through multi-dimensional feature fusion analysis based on the combined features of the service type feature identifier. Based on the standardized feature parameters and service type, a service quality impact index is calculated using a nonlinear mapping function. The corresponding weight adjustment criteria are determined based on the service type, including reliability weight, bandwidth requirement weight, and real-time requirement weight. The business quality impact index and the weight adjustment criteria are weighted and summed to obtain a comprehensive priority score. Based on the comprehensive priority score, the priority corresponding to the business type is determined by a preset priority division standard.

[0011] A time-varying business perception benchmark model is constructed using chaotic dynamics equations, including: Based on the business type and priority, the initial range of the control parameters of the chaotic dynamics equation is determined. The control parameters include nonlinear coefficients, coupling strength, and initial conditions. Based on the returned service data, a parameter adaptive adjustment mechanism is constructed. Within the initial range of the control parameters, the optimal control parameters are dynamically determined through the parameter adaptive adjustment mechanism. Based on the optimal control parameters, a chaotic system is constructed, and a time-varying reference trajectory is generated. Based on the time regularity features in the standardized feature parameters, data segments that meet the preset normal operation standards are selected from the backhaul service data to obtain the service normal state feature set. The service normal state feature set is then mapped to the chaotic phase space through a phase space reconstruction technology set to obtain the service feature trajectory. The similarity between the business feature trajectory and the time-varying reference trajectory is calculated by a dynamic time warping algorithm. Based on the distribution statistics of the similarity, a dynamic threshold adjustment mechanism is established. Based on the similarity and the dynamic threshold adjustment mechanism, a time-varying business perception benchmark model is constructed.

[0012] By quantifying the deviation between the anomaly prediction parameters of the business layer and the time-varying business perception benchmark model, the hierarchical backhaul status assessment results are output, including: Based on the time-varying service perception benchmark model, obtain the normal service status reference parameters corresponding to the current time point. The normal service status reference parameters include the abnormal probability threshold, the boundary of the abnormal impact range, and the abnormal duration range under the normal service status. The abnormal prediction parameters and normal business status reference parameters of the business layer are standardized and unified into the same feature space. The matching degree deviation is calculated by mapping to the same feature space. The matching degree deviation includes Euclidean distance, cosine similarity and information entropy difference. Weights are assigned to the matching degree deviations based on the business type, and a multi-dimensional deviation vector is constructed by weighted summation. Preset multi-level thresholds are determined based on the historical distribution data of the multi-dimensional deviation vectors. The multi-dimensional deviation vectors are compared with the preset multi-level thresholds, and the hierarchical feedback status evaluation results are output.

[0013] A feedback correction loop is constructed based on the gradient of the physical layer entropy and the network topology data, including: Based on the link connection relationships in the network topology data, a basic network connectivity graph is constructed. Based on the node load characteristics and link redundancy characteristics in the network topology data, the load weight and reliability weight of each node in the basic network connectivity graph are determined. Based on the basic network connectivity graph, load weight, and reliability weight, a weighted network connectivity graph is constructed. In the weighted network connectivity graph, the weighted betweenness centrality of each node is calculated as the weighted centrality index. The change gradient of the physical layer entropy value of each node is compared with the size of a preset threshold. Nodes that are greater than the preset threshold are identified as abnormal nodes, and the links associated with the abnormal nodes are identified as abnormal links. Starting from the abnormal node, the abnormal propagation path is determined by graph theory algorithm. The abnormal risk level is determined based on the number of abnormal links and the weighted centrality index. The node priority is determined based on the abnormal propagation path, the abnormal risk level, and the weighted centrality index. A feedback correction loop is constructed based on the node priority.

[0014] An adaptive protection strategy is generated based on the feedback correction loop, including: Based on the node priority and the abnormal propagation path, determine the priority protection nodes and priority protection links; For the priority protection nodes and priority protection links, the backup paths and their corresponding load capacities are calculated based on the link redundancy characteristics and node load characteristics in the network topology data. Based on the priority protection nodes, priority protection links, backup paths and their corresponding load capacities, a preliminary protection strategy is generated. The preliminary protection strategy includes at least link switching, resource reservation and traffic adjustment. Based on the business type and priority, the initial protection strategy is optimized and adjusted to generate the final adaptive protection strategy.

[0015] A backhaul link monitoring system for 5G base stations includes a data acquisition module, a status analysis module, a feedback correction module, and a policy generation module, wherein: The data acquisition module is used to collect physical signal statistics, backhaul service data, and network topology data of the base station backhaul link. The status analysis module is used to generate business layer anomaly prediction parameters based on the output of the data acquisition module, quantify the matching degree deviation between the parameters and the time-varying business perception benchmark model, and output hierarchical back-transmission status assessment results. The feedback correction module is used to construct a feedback correction loop by combining the change gradient of the physical layer entropy value with the network topology data, so as to realize dynamic monitoring and correction of the backhaul network status. The strategy generation module is used to generate an adaptive protection strategy based on the hierarchical backhaul status assessment results and the feedback correction loop.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention collects physical signal statistics, backhaul service data, and network topology data of the backhaul link, and calculates the physical layer entropy value using the Shannon entropy formula to quantify the implicit degradation characteristics of the physical layer, achieving advanced prediction and accurate location of implicit degradation in the backhaul link. Through the spatiotemporal coding mechanism of a spiking neural network, it generates service layer anomaly prediction parameters, completing cross-layer prediction from physical layer features to service layer anomalies, significantly improving service quality assurance capabilities in multi-service scenarios. By constructing a time-varying service perception benchmark model using chaotic dynamics equations, it enhances the adaptability and flexibility of the backhaul link to different service types. Combining physical layer entropy values ​​and network topology data, it constructs a feedback correction loop to generate an adaptive protection strategy, realizing the self-healing function of the backhaul link, effectively improving the security and operational efficiency of the backhaul link, breaking through the limitations of traditional backhaul monitoring that only focuses on link connectivity, and ensuring the stability and high availability of the 5G base station backhaul link. Attached Figure Description

[0017] Figure 1 This is a flowchart of the backhaul link monitoring method applied to 5G base stations in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the steps involved in calculating the anomaly prediction parameters of the business layer using the spatiotemporal coding mechanism of a spiking neural network in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the steps for generating a time-varying reference trajectory in an embodiment of the present invention; Figure 4 This is an overall framework diagram of the backhaul link monitoring system applied to 5G base stations in an embodiment of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof.

[0019] The term "and / or" in the following text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0020] This application provides a method for backhaul monitoring, including: Acquire backhaul link monitoring data, which includes physical signal statistics, backhaul service data, and network topology data; Based on physical signal statistics, the physical layer entropy value is calculated using the Shannon entropy formula. Based on the physical layer entropy value and the backhaul service data, the service layer anomaly prediction parameters are calculated using the spatiotemporal coding mechanism of a spiking neural network. Based on the backhaul service data, the service type and priority are determined, and a time-varying service perception benchmark model is constructed using chaotic dynamics equations. By quantifying the matching degree deviation between the service layer anomaly prediction parameters and the time-varying service perception benchmark model, the hierarchical backhaul status assessment results are output. When the hierarchical backhaul status assessment result is abnormal, a feedback correction loop is constructed based on the change gradient of physical layer entropy value and network topology data, and an adaptive protection strategy is generated based on the feedback correction loop.

[0021] For example, this embodiment uses a 5G base station backhaul network system as an example to specifically illustrate the above backhaul monitoring method.

[0022] In existing technologies, monitoring of 5G base station backhaul links typically employs two methods. Method one involves monitoring performance indicators with fixed thresholds. This involves periodically checking basic parameters such as signal-to-noise ratio (SNR) and bit error rate (BER) of the backhaul link, triggering alarms when these parameters exceed preset thresholds. This method ensures basic link status monitoring while maintaining computational complexity under relatively stable network conditions. However, this method suffers from a lag in responding to dynamic network changes, and the system's monitoring accuracy is primarily affected by the fixed threshold and the rate of environmental change. This lag impacts the system's monitoring performance in complex network environments. Method two involves using machine learning-based anomaly detection methods for backhaul link monitoring. This method identifies abnormal patterns by analyzing historical data to train a model. While theoretically, this method offers higher detection accuracy, in practical applications, it requires extensive feature engineering and model training, resulting in high computational complexity. Furthermore, it still suffers from slow detection speed and a high risk of false alarms when dealing with complex network topology changes.

[0023] Therefore, existing technologies often struggle to achieve a perfect balance between improving detection accuracy, reducing response latency, and minimizing computational complexity. This is especially true in application scenarios with extremely high network reliability requirements, such as 5G base stations, where existing backhaul monitoring methods are unable to meet the dual demands of rapid response and efficient detection.

[0024] To achieve efficient and stable monitoring of link status in 5G base station backhaul network systems and improve the adaptability of the monitoring system to complex network environments, while also maximizing system reliability through rapid response to network anomalies, this application provides a backhaul link monitoring method for 5G base stations based on the aforementioned backhaul monitoring method. Figure 1 As shown, the execution steps of this method include: S1: Obtain backhaul link monitoring data of 5G base stations, the backhaul link monitoring data including physical signal statistics, backhaul service data and network topology data; In this step, during the operation of the 5G base station backhaul network system, monitoring data from different dimensions of the backhaul link is acquired, including physical signal statistics at the physical layer, backhaul service data at the service layer, and network topology data at the network layer. Compared to traditional single-parameter monitoring methods, backhaul link monitoring data acquisition covers parameter monitoring at three dimensions: physical layer, service layer, and network layer. This significantly improves data integrity and the time synchronization accuracy of multi-source data, while also enhancing the system's environmental awareness and anomaly identification capabilities. High-quality, highly consistent data acquisition provides reliable support for subsequent control steps such as entropy calculation, anomaly prediction, and protection strategy generation, thereby ensuring the reliability and accuracy of the overall monitoring system.

[0025] The physical signal statistics include signal quality indicators, signal stability indicators, and signal transmission characteristic indicators.

[0026] Specifically, the original radio frequency signal and digital bitstream data of the backhaul link are collected by a signal quality detection device. Digital signal processing technology is used to perform frequency domain and time domain analysis on the collected original radio frequency signal to obtain signal power spectral density data and noise power spectral density data. Based on the signal power spectral density data and noise power spectral density data, the signal-to-noise ratio (SNR) is calculated using the SNR measurement principle. At the same time, the received digital bitstream data is checked bit by bit by a bit error rate (BER) detection circuit. The error bit count and total bit count within a preset time period are counted to calculate the BER. The signal quality index is determined based on the SNR and BER.

[0027] Furthermore, the system continuously monitors and collects the delay measurement data sequence of signal transmission, performs statistical analysis on the collected delay measurement data sequence, calculates the standard deviation of delay fluctuation based on the delay data samples within a preset time window, and obtains the link jitter. Furthermore, the system monitors and records the timestamp data of link interruption events in real time, performs statistical analysis on the recorded timestamp data of interruption events, calculates the interruption frequency based on the interruption event count and the total monitoring duration, and obtains the signal stability index.

[0028] Furthermore, a signal strength detector is used to collect the signal strength data of the received signal. Timestamp synchronization technology is used to collect the signal transmission and reception timestamps. The collected signal strength and timestamp data are then processed by a signal conditioning circuit to obtain a standardized signal strength value and transmission delay. Signal transmission characteristics are then obtained based on the standardized signal strength value and transmission delay.

[0029] The backhaul service data includes service traffic characteristics, data packet loss characteristics, transmission rate, and service continuity characteristics.

[0030] Specifically, deep packet inspection technology is used to analyze the data flow in the backhaul link, and the number of data packets, the number of bytes, and the traffic distribution pattern per unit time are statistically obtained. Based on the number of data packets, the number of bytes, and the traffic distribution pattern per unit time, the periodicity and burstiness characteristics of the traffic distribution pattern are extracted using time series analysis methods to obtain the business traffic characteristics; Furthermore, during data transmission, lost data packets are detected and identified by sequence number. Based on the number of lost data packets, the packet loss rate, packet loss distribution, and packet loss duration are statistically obtained. Based on the packet loss rate, packet loss distribution, and packet loss duration, the data packet loss characteristics are determined. Furthermore, multi-point measurement technology is used to acquire rate variation data at different time periods. Based on the rate variation data, the average rate, peak rate, and rate stability index are calculated using statistical analysis methods to obtain the transmission rate. Furthermore, by using business session tracking technology to monitor the duration of various services, analyzing the start time, end time, and activity level of the services, and combining this with business type identification algorithms, continuous characteristics of the services can be obtained.

[0031] The network topology data includes link connectivity, node load characteristics, and link redundancy characteristics. Specifically, the connection status of each node in the network is automatically detected through a network discovery protocol, and the connection information between nodes is collected using a link state protocol to construct a network topology map and update link status changes in real time to obtain link connectivity. Furthermore, the CPU utilization, memory usage, interface utilization, and other load information of each network node are periodically queried through a simple network management protocol, and a load balancing algorithm is used to analyze the processing capacity and current load level of the nodes to obtain node load characteristics. Further still, a redundant path discovery algorithm is used to identify backup links and redundant paths in the network, analyze the redundancy and reliability level of each link, and evaluate the redundancy protection capability of the links in conjunction with a failover mechanism to obtain link redundancy characteristics.

[0032] S2: Based on the physical signal statistics, the physical layer entropy value is calculated using the Shannon entropy formula. Based on the physical layer entropy value and the backhaul service data, the service layer anomaly prediction parameters are calculated using the spatiotemporal coding mechanism of a spiking neural network. In this step, the physical signal statistical data is used as input. The uncertainty of the physical signal characteristics is quantified through probability distribution analysis, and the physical layer entropy value is further calculated using Shannon entropy theory. Based on the physical layer entropy value and the backhaul service data, service layer anomaly prediction parameters are obtained through the spatiotemporal coding mechanism of a spiking neural network. These parameters are used to achieve intelligent prediction of abnormal states at the service layer. The service layer anomaly prediction method, by integrating multidimensional features from the physical and service layers, significantly improves the sensitivity and prediction accuracy of network anomalies, providing a more accurate and stable anomaly early warning capability for the backhaul monitoring system.

[0033] S3: Determine the service type and priority based on the backhaul service data, and construct a time-varying service perception benchmark model through chaotic dynamics equations; In this step, firstly, based on the feature analysis of the returned service data, different service types are identified and classified, and the priority level corresponding to each service type is determined. Subsequently, a time-varying service perception benchmark model is constructed using chaotic dynamics theory. This model can dynamically adapt to the temporal changes in service characteristics, avoiding misjudgments caused by changes in service patterns in traditional static threshold methods by establishing a dynamic benchmark for service behavior. The construction of the time-varying service perception benchmark model provides a dynamic reference standard for subsequent anomaly detection, ensuring the adaptability and effectiveness of the monitoring strategy. It also plays a crucial role in establishing intelligent benchmarks throughout the entire solution, significantly improving the system's dynamic adaptability and monitoring accuracy.

[0034] S4: By quantifying the matching degree deviation between the anomaly prediction parameters of the business layer and the time-varying business perception benchmark model, output the hierarchical backhaul status assessment results. In this step, the backhaul network status is quantitatively assessed by calculating the matching deviation between the service layer anomaly prediction parameters and the time-varying service perception benchmark model. The quantitative assessment method uses multi-dimensional deviation analysis to classify the network status into different risk levels, forming a tiered backhaul status assessment result. This tiered backhaul status assessment result accurately reflects the current network health status and potential risks, providing a scientific basis for the formulation of subsequent protection strategies, effectively improving the accuracy and reliability of network status assessment, and ensuring that the monitoring system can promptly identify and respond to various anomalies.

[0035] Specifically, firstly, based on the time-varying service perception benchmark model, reference parameters for the normal service state corresponding to the current time point are obtained. These reference parameters include an anomaly probability threshold, an anomaly impact range boundary, and an anomaly duration interval under normal service conditions. The anomaly probability threshold is determined based on the anomaly probability distribution under normal service conditions; the anomaly impact range boundary is determined based on the defined spatial range of anomaly impact under normal service conditions; and the anomaly duration interval is determined based on the duration range of anomaly events under normal service conditions.

[0036] Furthermore, the anomaly prediction parameters and normal business state reference parameters of the business layer are standardized to obtain standardized parameter data. This standardized parameter data is unified into the same feature space to ensure the comparability of different parameters. Next, the matching degree deviation is calculated based on the standardized parameter data. The matching degree deviation includes Euclidean distance, cosine similarity, and information entropy difference. The Euclidean distance measures the linear distance between parameter vectors, the cosine similarity measures the angular similarity between parameter vectors, and the information entropy difference measures the information difference in parameter distributions. Then, the three matching degree deviation data (Euclidean distance, cosine similarity, and information entropy difference) are collected and processed using a weighted summation method to obtain a multidimensional deviation vector. This multidimensional deviation vector comprehensively reflects the overall deviation degree between the current state and the normal benchmark. Weights are assigned to Euclidean distance, cosine similarity, and information entropy difference according to the business type and priority. The higher the priority of the business type, the greater the assigned weight; the lower the priority of the business type, the smaller the assigned weight.

[0037] Furthermore, based on the historical distribution data of the multidimensional deviation vector, a preset multi-level threshold is determined using statistical analysis methods. Employing the distribution characteristic analysis method within the statistical analysis approach, statistical characteristic parameters are calculated based on the historical distribution data. These parameters include a central tendency parameter, a dispersion parameter, and a distribution shape parameter. The central tendency parameter reflects the central location of the data, the dispersion parameter reflects the range of data dispersion, and the distribution shape parameter reflects the distribution characteristics of the data. The multi-level threshold boundaries are determined based on the numerical distribution patterns of the statistical characteristic parameters. Based on the mathematical relationship between the dispersion parameter and the central tendency parameter, the threshold boundaries corresponding to different levels are determined using a standardized distance calculation method, forming a preset multi-level threshold based on statistical distance.

[0038] Furthermore, the probability density analysis method in the aforementioned statistical analysis can be employed. By constructing a probability density function for historical distribution data, multi-level threshold boundaries can be determined based on the characteristics of the cumulative probability distribution. The probability density function describes the probability of the multidimensional deviation vector occurring in different numerical intervals. The numerical boundaries corresponding to different probability levels are obtained through cumulative probability calculation methods. According to the preset probability level division principle, the cumulative probability interval is divided into multiple level segments, and the boundary value of each level segment serves as the threshold boundary for the corresponding anomaly level, forming a preset multi-level threshold based on probability distribution.

[0039] The preset multi-level thresholds include a minor anomaly threshold, a moderate anomaly threshold, and a severe anomaly threshold, forming a hierarchical anomaly detection system. The multi-dimensional deviation vector is compared with the preset multi-level thresholds, and based on the threshold range into which the multi-dimensional deviation vector falls, a corresponding hierarchical feedback status evaluation result is output, such as normal, minor anomaly, moderate anomaly, or severe anomaly.

[0040] S5: When the hierarchical backhaul status assessment result is abnormal, a feedback correction loop is constructed based on the change gradient of the physical layer entropy value and the network topology data, and an adaptive protection strategy is generated based on the feedback correction loop.

[0041] In this step, when an anomaly in network status is detected, the system constructs a dynamic feedback correction loop based on the gradient of physical layer entropy changes and network topology data. This feedback correction loop can track the propagation path and impact range of the anomaly in real time and generate targeted adaptive protection strategies. The introduction of the feedback correction mechanism realizes a shift from passive monitoring to active protection, effectively improving the network's self-healing capability and anti-interference performance, and ensuring the stable operation of the 5G base station backhaul network in complex environments.

[0042] The specific steps of S2 are as follows: S2.1: Based on the physical signal statistics, statistical analysis is performed using probability distribution methods, and the physical layer entropy value is calculated using the Shannon entropy formula; In this embodiment, the physical signal statistical data includes time-series data of signal quality indicators, signal stability indicators, and signal transmission characteristic indicators. An appropriate probability distribution method is selected to statistically model the above parameters, obtaining the probability distribution characteristics of signal quality, link stability, and transmission performance. Then, the Shannon entropy formula is applied to calculate the entropy values ​​for signal quality, link stability, and transmission performance. Subsequently, based on the signal quality entropy, link stability entropy, and transmission performance entropy values, a weighted summation is performed using a weighted fusion calculation formula to calculate the physical layer entropy value. This entropy value comprehensively reflects the overall uncertainty level of various physical layer indicators.

[0043] S2.2: The anomaly prediction parameters of the business layer are calculated through the spatiotemporal coding mechanism of the spiking neural network; In this embodiment, a spatiotemporal coding mechanism using a spiking neural network is employed to perform deep fusion analysis of physical layer entropy values ​​and backhaul service data, enabling intelligent prediction of service layer anomalies. This mechanism fully leverages the unique advantages of spiking neural networks in processing time-series information, achieving accurate modeling of complex network states by simulating the information processing mechanism of biological nervous systems. The spatiotemporal coding mechanism effectively integrates multidimensional time-series features from different levels and dimensions through the synergistic effect of temporal and spatial coding, transforming them into a spatiotemporal spiking feature matrix that can be processed by the spiking neural network. This matrix not only preserves the time-series characteristics of the original data but also establishes correlations between features through spatial coding. Based on the constructed spatiotemporal spiking feature matrix, a multi-layer spiking neural network architecture is used for deep feature learning. Through hierarchical information abstraction and feature combination, a cross-layer anomaly prediction model capable of capturing the complex correlations between the physical and service layers is constructed. This model possesses powerful nonlinear mapping and time-series modeling capabilities, enabling it to learn the potential patterns of anomaly modes from historical data and accurately predict future anomalies. Finally, by using a weighted fusion mechanism of multidimensional prediction results, and comprehensively considering prediction information from multiple dimensions such as the probability of anomaly occurrence, the scope of impact, and the duration, comprehensive business layer anomaly prediction parameters are calculated, providing a scientific basis for network operation and maintenance decisions.

[0044] Specific implementation of S2.1: S2.1.1: Based on the time series data and parameter characteristics of each parameter in the physical signal statistics within a preset time window, a probability distribution method adapted to the parameter characteristics is used for statistical analysis; In this embodiment, the physical signal statistics include time-series data of signal quality indicators, signal stability indicators, and signal transmission characteristic indicators. Specifically, time-series data of each parameter are first collected within a preset time window, the length of which is set comprehensively based on the network dynamic characteristics and monitoring accuracy requirements.

[0045] Then, based on the parameter characteristics of each parameter in the aforementioned time series data, statistical modeling is performed using a probability distribution method that adapts to the parameter characteristics. The parameter characteristics of each parameter include physical characteristics and numerical distribution types. Specifically, since the signal quality index is obtained based on the signal-to-noise ratio and bit error rate, its parameter characteristics include physical characteristics of continuity and symmetry, and numerical distribution types of normal or log-normal distributions. A normal or log-normal distribution is used for fitting, and the quality distribution parameters are determined using the maximum likelihood estimation method. The signal stability index is obtained based on link jitter and interruption frequency. Its parameter characteristics include physical characteristics of discreteness and sparsity, and numerical distribution types of Poisson or exponential distributions. A Poisson or exponential distribution is used for modeling, and stable distribution parameters are determined using the moment estimation method or least squares method. The signal transmission characteristic index is obtained based on signal strength and transmission delay. Its parameter characteristics include physical characteristics of boundedness and monotonicity, and numerical distribution types of beta or uniform distributions. A beta or uniform distribution is used for analysis, and the optimal fitting parameters are determined using a parameter estimation algorithm as the transmission distribution parameters. Next, based on the aforementioned quality distribution parameters, stable distribution parameters, and transport distribution parameters, the statistical characteristics of each distribution are calculated using theoretical formulas and numerical calculation methods to form a complete probability distribution characteristic description. The statistical characteristics include mean, variance, skewness, and kurtosis. The probability distribution characteristic description includes the distribution type identification results for each type of parameter, numerical estimation of the distribution parameters, quantitative calculation results of the statistical characteristics, and validity verification of the distribution fit.

[0046] Finally, based on the described probability distribution features, signal quality probability distribution features, link stability probability distribution features, and transmission performance probability distribution features are obtained. The signal quality probability distribution features include distribution type identifiers, parameter estimates, statistical feature values, and confidence interval information; the link stability probability distribution features include event occurrence probability, distribution shape parameters, statistical moment values, and goodness-of-fit indices; and the transmission performance probability distribution features include probability density function parameters, key points of the cumulative distribution function, statistical feature descriptions, and distribution validity verification results. These probability distribution feature descriptions provide an accurate probabilistic basis for subsequent entropy calculations, ensuring the scientific rigor and computational accuracy of the transformation process from raw data to probability features.

[0047] S2.1.2: Based on the signal quality probability distribution characteristics, link stability probability distribution characteristics, and transmission performance probability distribution characteristics, the signal quality entropy value, link stability entropy value, and transmission performance entropy value are calculated respectively using the Shannon entropy formula; In this embodiment, the range of values ​​for the signal quality index is first discretized based on the probability distribution characteristics of the signal quality. The number and width of the discretized intervals are determined according to the distribution type, statistical characteristics, and distribution parameters of the probability distribution characteristics of the signal quality. For signal quality indices conforming to a normal distribution or other continuous distributions, equal-probability or equal-width segmentation methods are selected based on the variance, and the effective segmentation range is determined using the 3σ principle; for discrete distribution signal quality indices, the original values ​​are used directly.

[0048] Based on the discretized intervals or original value set mentioned above, the set of signal quality index observation values ​​at multiple moments within a preset time window is used as the calculation object. The frequency of occurrence of the observation values ​​in each interval or value is counted, and the probability density of each interval or value is calculated. Then, according to the Shannon entropy formula, the signal quality entropy value is calculated. The signal quality entropy value is used to represent the degree of uncertainty of the signal quality. The higher the entropy value, the stronger the randomness of the signal quality and the more unstable the network state.

[0049] Then, the same processing method is applied to the link stability probability distribution characteristics. Based on the link stability index, the set of observed link stability index values ​​at multiple moments within a time window is used as the calculation object. Since link stability indices are typically discrete events with obvious discrete characteristics, entropy values ​​are calculated directly based on the probability of occurrence of each link stability index value, according to the distribution type of the link stability probability distribution characteristics. The frequency of occurrence of each link stability index value in the observation data is statistically analyzed, the corresponding probability density is calculated, and the Shannon entropy formula is applied to calculate the link stability entropy value. This link stability entropy value is used to quantify the randomness level of link stability and reflects the predictability of the link state.

[0050] Finally, the same processing method is applied to the transmission performance probability distribution characteristics. Based on the transmission performance index, the set of transmission performance index observations at multiple moments within a time window is used as the calculation object. Based on the distribution type and statistical characteristics of the transmission performance probability distribution characteristics, corresponding discretization segmentation is performed. The frequency of occurrence of observations in each interval or value is statistically analyzed, the probability density is calculated, and verified using the theoretical probability density function. The transmission performance entropy value is calculated using the Shannon entropy formula, reflecting the uncertainty and fluctuation of transmission performance.

[0051] S2.1.3: Based on the signal quality entropy value, link stability entropy value, and transmission performance entropy value, the physical layer entropy value is calculated using a weighted fusion algorithm; In this embodiment, firstly, based on the physical meaning and impact on the overall network performance of the signal quality entropy, link stability entropy, and transmission performance entropy, a weighting coefficient for each entropy value is determined. The weighting coefficient for the signal quality entropy value... The weighting coefficients of the link stability entropy value are set based on the decisive role of signal quality in the basic communication capability of the network. The weighting coefficients for the transmission performance entropy values ​​are set based on the critical impact of link stability on network reliability. The weighting coefficients are set according to their importance to network service quality. The specific values ​​of the weighting coefficients are determined through correlation analysis of historical network performance data, expert experience evaluation, or training with machine learning algorithms, and satisfy the constraint that the sum of the weighting coefficients equals 1.

[0052] Then, based on the aforementioned weighting coefficients, a weighted linear fusion algorithm is used to calculate the physical layer entropy value by fusing the signal quality entropy value, link stability entropy value, and transmission performance entropy value. The signal quality entropy value is... The link stability entropy value is The transmission performance entropy value is The corresponding weighting coefficients are respectively According to the weighted linear fusion calculation formula: The physical layer entropy value was calculated. The weighted fusion algorithm can comprehensively consider the uncertainty characteristics of various dimensions to form a unified quantitative assessment of the overall state of the physical layer. The higher the entropy value of the physical layer, the greater the uncertainty of the physical layer and the more unstable the network state.

[0053] Finally, the physical layer entropy value is subjected to numerical range verification and rationality validation. Through checking the numerical relationship with signal quality entropy, link stability entropy, and transmission performance entropy, comparing and analyzing historical data, and detecting outliers, the accuracy of the physical layer entropy calculation and the rationality of its physical meaning are ensured. If the physical layer entropy value passes the verification, it is used as a valid physical layer state quantification indicator; when the physical layer entropy value exceeds the specified range or exhibits abnormal fluctuations, the data re-acquisition and calculation process is initiated to ensure the reliability of the results.

[0054] Please see Figure 2 The figure shows a flowchart of the steps for calculating the anomaly prediction parameters of the service layer using the spatiotemporal coding mechanism of a spiking neural network, as provided in an embodiment of this application. The specific steps of S2.2 are as follows: S2.2.1: Based on the time sequence of the physical layer entropy value, the physical layer entropy value fluctuation feature vector is calculated through a sliding window. Based on the backhaul service data, the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector are calculated. In this embodiment, firstly, based on the time series of the physical layer entropy values, a sliding window technique is used for time series analysis. A fixed-length sliding window is set and gradually slides according to a set step size. Within the time range covered by each sliding window, the first-order and second-order differences of the physical layer entropy values ​​are calculated using a difference analysis method. The first-order difference is used to capture the rate of change and trend direction of the entropy values, and the second-order difference is used to capture the acceleration characteristics of the entropy value change trend. Statistical analysis methods are used to calculate the variance, peak value, and mean of the physical layer entropy values. The variance reflects the dispersion and stability of the physical layer entropy values, the peak value reflects the maximum fluctuation amplitude of the physical layer entropy values, and the mean reflects the average level of the physical layer entropy values. The first-order difference, second-order difference, variance, peak value, and mean are linearly combined according to preset weights using a feature combination method to form an entropy value fluctuation feature vector. This entropy value fluctuation feature vector quantitatively reflects the dynamic change law and fluctuation characteristics of the physical layer entropy values.

[0055] Then, based on the service traffic characteristics, data packet loss characteristics, transmission rate and service continuity characteristics in the backhaul service data, different feature extraction methods are used to calculate the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector and time distribution feature vector.

[0056] The traffic dynamic feature vector includes the average traffic, peak traffic, traffic change rate, and traffic growth trend of the backhaul service data. The average traffic is calculated using a moving average method, the peak traffic is identified and determined using a peak detection algorithm, the traffic change rate is calculated using a linear regression method, and the traffic growth trend is calculated using a trend analysis algorithm.

[0057] The transmission quality feature vector includes the packet loss rate, average latency, latency jitter, and bit error rate of the backhaul service data. The packet loss rate is obtained by statistical analysis using a packet counting method, the average latency is calculated by a timestamp analysis method, and the latency jitter is calculated by a variance analysis method. The bit error rate is obtained by statistical analysis using a bit comparison method. The bandwidth utilization feature vector includes the utilization rate, saturation, volatility, and peak-to-average power ratio of the backhaul service data. The utilization rate is calculated using a bandwidth monitoring method, the saturation is calculated using a peak value analysis method, the volatility is calculated using a coefficient of variation method, and the peak-to-average power ratio is calculated using a ratio analysis method. The time distribution feature vector includes the service continuity, service interval, periodicity, and burstiness of the backhaul service data. The service continuity is obtained by analyzing the duration statistics method, the service interval is obtained by the interval time analysis method, the periodicity is detected by the autocorrelation analysis method, and the burstiness is identified by the burst detection algorithm.

[0058] S2.2.2: Align the above feature vectors by dimensions to obtain aligned feature vectors, convert the time series data corresponding to the aligned feature vectors into pulse sequences according to the time coding mechanism, and convert the pulse sequences into spatiotemporal pulse feature matrices through the spatial coding mechanism. In this embodiment, the aforementioned feature vectors are the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector. Specifically, firstly, the aforementioned feature vectors are dimensionally aligned. A unified feature dimension is set using a dimension unification method. For feature vectors with insufficient dimensions, a zero-padding strategy is used to pad them. For feature vectors with excessive dimensions, a truncation strategy is used to prune them, ensuring that all feature vectors have the same length, resulting in aligned feature vectors.

[0059] Then, according to the time coding mechanism, a rate coding method is used to convert the aligned feature vector into a pulse sequence. Specifically, the values ​​of the aligned feature vector are mapped to a specified range using an eigenvalue normalization method. The pulse firing frequency is determined based on the magnitude of the eigenvalues ​​of the aligned feature vector. Random pulse events are generated using a Poisson process generation method. Finally, the continuous numerical signal of the aligned feature vector is converted into a discrete time pulse sequence using a time window division method. The pulse firing time and frequency of the pulse sequence carry the time and amplitude information of the aligned feature vector.

[0060] Next, based on the spatial encoding mechanism, a group vector encoding method is used for spatial mapping. Specifically, different types of features of the aligned feature vector are assigned to different neuron groups using a neuron grouping strategy. A Gaussian tuning method is used to determine the response strength of each neuron to different feature values ​​of the aligned feature vector. A mutual information calculation method is used to analyze the correlation and dependency relationships between features in the aligned feature vector. A weight allocation strategy is used to establish the connection strength between neurons. A topology construction method is used to establish the neuronal network structure, transforming the one-dimensional pulse sequence of the pulse sequence into a two-dimensional spatiotemporal pulse feature matrix with temporal and spatial dimensions.

[0061] S2.2.3: Input the spatiotemporal pulse feature matrix into a multilayer spiking neural network to construct a cross-layer anomaly prediction model. Calculate the prediction sequence of the business layer anomaly probability, anomaly impact range, and anomaly duration through a probability normalization function. Weight the prediction sequence and map it to a specified interval to obtain the business layer anomaly prediction parameters. In this embodiment, the spatiotemporal pulse feature matrix is ​​first input into a multilayer spiking neural network for feature learning, using a leaky integral firing neuron model as the basic computational unit. Specifically, a membrane potential accumulation mechanism is used to achieve time integration of the input signal to the spatiotemporal pulse feature matrix. When the membrane potential reaches a threshold, pulse firing is triggered and the membrane potential is reset. A multilayer network structure is used to abstract and extract the features of the spatiotemporal pulse feature matrix layer by layer. A residual connection mechanism is used to achieve cross-layer information fusion. A backpropagation learning algorithm is used to optimize network parameters, constructing a cross-layer anomaly prediction model capable of capturing the complex correlation between the physical layer entropy value and the business layer.

[0062] Then, based on the output of the cross-layer anomaly prediction model, three specialized prediction branches are designed: a business layer anomaly probability prediction branch, a business layer anomaly impact range prediction branch, and a business layer anomaly duration prediction branch. The business layer anomaly probability prediction branch uses the Sigmoid normalization method for probability normalization, the business layer anomaly impact range prediction branch uses the Softmax activation method for probability normalization, and the business layer anomaly duration prediction branch uses the linear activation method for probability normalization, resulting in business layer anomaly probability prediction sequences, business layer anomaly impact range prediction sequences, and business layer anomaly duration prediction sequences, respectively. These sequences comprehensively describe the characteristics of business layer anomalies from three dimensions: the probability of anomaly occurrence, the degree of anomaly impact, and the persistence of anomalies.

[0063] Finally, a weighted fusion strategy is employed to synthesize the predicted sequences of business layer anomaly probability, anomaly impact range, and anomaly duration. Specifically, the weight coefficients of each prediction sequence are determined using statistical analysis of historical anomaly events and expert knowledge fusion methods. A linear weighted summation method is used to calculate the comprehensive anomaly prediction value. The fusion result is then mapped to a standard range using a Min-Max normalization method to obtain the final business layer anomaly prediction parameters. These parameters comprehensively and quantitatively reflect the overall risk level and early warning intensity of anomalies occurring at the business layer.

[0064] The specific steps for S3 are as follows: S3.1: Determine the service type and priority based on the returned service data; In this embodiment, the backhaul service data refers to the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector calculated based on service traffic characteristics, packet loss characteristics, transmission rate, and service continuity characteristics. Specifically, firstly, each of the above feature vectors is quantized. Through normalization transformation, the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector are mapped to a unified numerical range, eliminating the influence of dimensional differences between the feature vectors and generating standardized feature parameters. The standardized feature parameters include a traffic pattern feature generated based on the traffic dynamic feature vector, a quality sensitivity feature generated based on the transmission quality feature vector, a bandwidth demand feature generated based on the bandwidth utilization feature vector, and a time regularity feature generated based on the time distribution feature vector. The traffic pattern feature reflects the time distribution pattern and intensity changes of the backhaul service data; the quality sensitivity feature reflects the degree of transmission quality requirements of the backhaul service data; the bandwidth demand feature reflects the bandwidth resource requirements of the backhaul service data; and the time regularity feature reflects the time periodicity and continuity of the backhaul service data.

[0065] Then, based on the standardized feature parameters, the corresponding service type feature identifier is determined using threshold segmentation and pattern matching methods. The service type feature identifier includes a traffic type identifier determined based on the traffic pattern characteristics, a reliability type identifier determined based on the quality sensitivity characteristics, a bandwidth type identifier determined based on the bandwidth demand characteristics, and a time type identifier determined based on the time pattern characteristics. Based on the combined features of the traffic type identifier, the reliability type identifier, the bandwidth type identifier, and the time type identifier, the service type corresponding to the backhaul service data is determined using multi-dimensional feature fusion analysis and decision tree classification methods.

[0066] Finally, based on the standardized feature parameters and service type, a service quality impact index is calculated using a nonlinear mapping function. This index comprehensively considers the impact of service interruptions on user experience and network performance. Simultaneously, a corresponding weight adjustment criterion is determined based on the service type, including reliability weight, bandwidth requirement weight, and real-time requirement weight.

[0067] The business quality impact index and its weight adjustment criteria are weighted and summed to obtain a comprehensive priority score. Based on the comprehensive priority score, the priority corresponding to the business type is determined according to a preset priority classification standard.

[0068] S3.2: Construct a time-varying business perception benchmark model using chaotic dynamics equations; In this embodiment, a time-varying service perception benchmark model is constructed using chaotic dynamics equations to dynamically track the evolution trend of normal service behavior. First, the control parameter range of the chaotic dynamics equations is determined based on the service type and priority. Then, an adaptive parameter adjustment mechanism is constructed to dynamically optimize the control parameters and generate a time-varying reference trajectory. Next, normal operation data segments are selected from the returned service data, and service features are mapped into trajectory form using phase space reconstruction technology. Finally, a dynamic time warping algorithm is used to calculate trajectory similarity, and a dynamic threshold adjustment mechanism is established to construct a time-varying service perception benchmark model that can adapt to changes in service patterns.

[0069] The specific steps of S3.2 are as follows: S3.2.1 Based on the business type and priority, determine the initial range of control parameters for the chaotic dynamic equation, wherein the control parameters include nonlinear coefficients, coupling strength, and initial conditions; In this embodiment, the nonlinear coefficient controls the chaotic characteristics of the system, the coupling strength controls the interaction between different state variables, and the initial conditions determine the initial state of the system. Specifically, the corresponding characteristic requirements are determined according to the service type, and the initial range of the control parameters is determined by a service classification mapping method. For services belonging to the ultra-reliable low-latency category, a conservative parameter setting strategy is adopted to limit the range of the nonlinear coefficient to ensure system stability. The nonlinear coefficient is set within a low-sensitivity range to avoid excessive chaotic behavior. The coupling strength is set within a stable range to ensure the transmission stability of the returned service data. The initial conditions are set within a high-stability range to ensure reliable transmission of the returned service data.

[0070] For services classified as enhanced mobile broadband, a complex parameter setting strategy is adopted to expand the coupling strength range to reflect the complex traffic characteristics of the backhaul service data. The nonlinear coefficient is set in the medium sensitivity range to adapt to the dynamic changes of the backhaul service data; the coupling strength is set in the high dynamic range to fully capture the variability of the backhaul service data; and the initial conditions are set in the medium dynamic range to balance the performance requirements of the backhaul service data.

[0071] For services that fall under the category of large-scale machine communication, a moderate parameter setting strategy is adopted to balance system complexity and computational efficiency. The nonlinear coefficient is set within a moderately sensitive range; the coupling strength is set within a moderate range; and the initial conditions are set within a balanced range to ensure the processing efficiency of the backhaul service data.

[0072] Based on the business priority, the weight of the control parameter is adjusted using a business priority weight allocation method. The higher the business priority, the smaller the adjustment range of the control parameter weight.

[0073] S3.2.2: Based on the backhaul service data, construct a parameter adaptive adjustment mechanism. Within the initial range of the control parameters, dynamically determine the optimal control parameters through the parameter adaptive adjustment mechanism. Based on the optimal control parameters, construct a chaotic system and generate a time-varying reference trajectory. like Figure 3 The flowchart shown illustrates the steps for generating a time-varying reference trajectory according to an embodiment of this application. In this embodiment, the backhaul service data includes statistical characteristics of service traffic features, including the mean, variance, skewness, kurtosis, and autocorrelation coefficient of the service traffic features. Specifically, based on the mean and variance, a real-time load index of the backhaul service data is calculated using a sliding time window to reflect the service load level of the backhaul service. Based on the skewness and kurtosis, the uneven distribution of backhaul service traffic is quantified using a higher-order moment calculation method to reflect the service distribution characteristics of the backhaul service. Based on the autocorrelation coefficient, a periodic pattern of the backhaul service is determined using a delay correlation analysis method to reflect the service time dependence of the backhaul service.

[0074] Specifically, based on the statistical characteristics of the business traffic features, a parameter adaptive adjustment mechanism for chaotic systems is constructed. First, a parameter optimization mathematical model is built based on the statistical characteristics of the business traffic features. This model defines the control parameters of the chaotic system as three-dimensional optimization variables, including nonlinear coefficients, coupling strength, and an initial condition vector, forming a combination of control parameters. By finding the optimal combination of control parameters, the trajectory generated by the chaotic system constructed from this combination achieves the best statistical match with the target business traffic features.

[0075] Then, a multi-objective optimization function is established as the evaluation criterion for parameter optimization. This multi-objective optimization function includes three objectives: maximizing prediction accuracy based on backhaul service data, maximizing system stability based on backhaul service data, and minimizing computational complexity based on backhaul service data. A comprehensive objective function is constructed using a weighted summation method. Specifically, for highly complex service traffic, the weight of prediction accuracy is increased to prioritize prediction performance; for relatively stable service traffic, the weight of stability is increased to consider system stability; and in environments with limited computing resources, the weight of computational complexity is appropriately increased to moderately consider computational efficiency.

[0076] Furthermore, a multi-objective particle swarm optimization (PSO) algorithm is used as the core algorithm for parameter optimization, establishing a mapping relationship between PSO and the control parameter optimization problem. The position vector of each particle is designed as a five-dimensional vector, corresponding to the nonlinear coefficient, coupling strength, and three components of the initial conditions in the control parameter combination. The position vector represents a specific set of chaotic system control parameter configurations. The velocity vector of each particle is used as the adjustment direction and magnitude of the control parameters. Each component of the velocity vector represents the change of the corresponding control parameter in the next iteration, where positive values ​​indicate an increasing trend, negative values ​​indicate a decreasing trend, and the absolute value reflects the step size of the control parameter adjustment. Based on the velocity update formula of the standard PSO algorithm, the new particle velocity is calculated through a linear combination of inertia weight, individual optimal position, and global optimal position. Based on the updated particle velocity, the new position of the particle is calculated using the position update formula.

[0077] Based on the complexity characteristics of the three-dimensional parameter space and the convergence requirements of the algorithm, the maximum number of iterations is set to a moderate size to ensure that the algorithm converges under limited computing resources. The population size is set to a medium order of magnitude to control computational complexity while ensuring sufficient coverage of the parameter space, achieving a balance between search efficiency and solution quality. In each iteration, the algorithm first calculates the comprehensive objective function value of the control parameter combination corresponding to each particle position. By comparing the comprehensive objective function values ​​of all particles, the optimal particle position in the current iteration is identified. Subsequently, the algorithm updates the individual historical best position and the global historical best position of each particle, obtaining the updated individual best position and the updated global best position. The individual historical best position records the position of the [nth iteration]. The parameter combination corresponding to the maximum function value achieved by an individual particle in history, and the global historical best position record the parameter combination corresponding to the maximum function value achieved by the entire population in history.

[0078] The individual optimal position is used as the global optimal position. The updated velocity of each particle is calculated using the velocity update formula, and then the updated position of the particle is calculated using the position update formula. The particles move towards the individual optimal position and the global optimal position, achieving gradual convergence to the optimal parameter region. When the function value change corresponding to the global optimal position for several consecutive generations is less than the preset convergence threshold, or when the maximum number of iterations is reached, the algorithm terminates and outputs the global optimal position as the optimal control parameter.

[0079] Furthermore, a corresponding chaotic system dynamics model is constructed based on the optimal control parameters, generating a time-varying reference trajectory with specific statistical characteristics. Specifically, firstly, the chaotic system state is initialized according to the optimal control parameters; secondly, the state evolution sequence of the chaotic system within a preset time window is calculated through numerical integration; finally, the generated chaotic trajectory is normalized and its statistical characteristics are calibrated to ensure that the statistical characteristics of the reference trajectory are consistent with the target characteristics of the backhaul service data. The mean and variance of the time-varying reference trajectory match the load level of the backhaul service, the skewness and kurtosis of the time-varying reference trajectory reflect the unevenness of the service distribution, and the autocorrelation structure of the time-varying reference trajectory reflects the time dependence and periodicity of the service. By setting a threshold for statistical characteristic changes, when the mean, variance, skewness, kurtosis, or autocorrelation coefficient of the backhaul service data changes by more than a preset threshold, the parameter re-optimization process is automatically triggered.

[0080] S3.2.3: Based on the time regularity features in the standardized feature parameters, select data segments that meet the preset normal operation standards from the backhaul service data to obtain the service normal state feature set. Map the service normal state feature set to the chaotic phase space through phase space reconstruction technology to obtain the service feature trajectory. In this embodiment, data segments that meet preset normal operation standards are selected from the backhaul service data based on the temporal regularity characteristics in the standardized feature parameters. The temporal regularity characteristics reflect the periodicity and continuity of the backhaul service data. The periodicity is reflected in the regularity of daily fluctuations, weekly changes, and seasonal trends in service traffic, while the continuity reflects the stable existence of service traffic at different time scales and the reliability of its continuous operation. During the selection process, a sliding window technique combined with multi-scale entropy analysis is used to perform time-domain segmentation and evaluation of the backhaul service data. The selection criteria are determined based on a predefined normal operation indicator system. This system comprehensively considers multiple key dimensions of service operation, and a fuzzy comprehensive evaluation algorithm is used to fuse multi-dimensional indicators into a single evaluation value. Data segments with evaluation values ​​greater than a preset threshold are considered to meet the normal operation standards, thereby obtaining a normal operation status feature set.

[0081] Furthermore, the normal business state feature set is mapped to a chaotic phase space using phase space reconstruction technology to obtain the business feature trajectory. Phase space reconstruction technology, based on the embedding theorem, transforms a one-dimensional time series into a trajectory in a high-dimensional phase space using a time delay method. Specifically, the optimal time delay parameter is first selected at the position where the mutual information function first reaches a local minimum. Subsequently, a pseudo-nearest neighbor algorithm is used to determine the optimal embedding dimension. This algorithm calculates the rate of change of nearest neighbor points under different embedding dimensions to determine the minimum dimension that can fully unfold the system's dynamic characteristics. Based on the optimal time delay parameter and the optimal embedding dimension, a delay coordinate vector is constructed to form the trajectory representation of the business features in the phase space, serving as the business feature trajectory. This business feature trajectory can effectively capture the nonlinear dynamic characteristics and implicit patterns in the business data, providing a reliable feature representation for subsequent anomaly detection based on chaos theory. Compared to traditional statistical analysis methods, this method can better handle the non-stationarity and long-range correlation in the business data, improving the accuracy and sensitivity of anomaly detection.

[0082] S3.2.4: Calculate the similarity between the business feature trajectory and the time-varying reference trajectory using a dynamic time warping algorithm. Based on the distribution statistics of the similarity, establish a dynamic threshold adjustment mechanism. Based on the similarity and the dynamic threshold adjustment mechanism, construct a time-varying business perception benchmark model.

[0083] In this embodiment, the service feature trajectory is first preprocessed. Specifically, a multi-dimensional service feature trajectory is constructed based on the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector. The multi-dimensional service feature trajectory includes the time-series representations of the traffic dynamic feature vector, the transmission quality feature vector, the bandwidth utilization feature vector, and the time distribution feature vector. Simultaneously, the time-varying reference trajectory is subjected to dimensional expansion processing to obtain a processed time-varying reference trajectory. The processed time-varying reference trajectory maintains the same dimensions as the multi-dimensional service feature trajectory, where the fourth dimension is determined by a linear combination of chaotic system state variables.

[0084] Secondly, the similarity between the multidimensional service feature trajectory and the time-varying reference trajectory is calculated using a dynamic time warping algorithm. Specifically, trajectory alignment is performed based on the multidimensional service feature trajectory and the processed time-varying reference trajectory using a dynamic time warping algorithm. Weighted Euclidean distance is used as the metric, and the local distance is calculated by taking the square root of the weighted sum of the feature differences between the traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector. A cumulative distance matrix is ​​constructed using a dynamic programming recursive method, which is obtained by adding the local distance to the minimum cumulative distance of adjacent paths. The cumulative distance matrix records the minimum cumulative distance from the starting point to any point. Each element in the matrix is ​​calculated step-by-step using a dynamic programming method to ensure that the path satisfies monotonicity and continuity constraints.

[0085] Furthermore, based on the cumulative distance matrix, the optimal alignment path is extracted and the dynamic time-warped distance is calculated. The optimal alignment path is obtained through a backtracking algorithm. Specifically, starting from the bottom right corner of the cumulative distance matrix, the path is traced backward along the direction of the minimum cumulative distance, sequentially selecting the predecessor node that minimizes the cumulative distance, until the starting point is reached. The resulting path sequence is the optimal alignment path. The dynamic time-warped distance is calculated by normalization based on the path length of the optimal alignment path. The dynamic time-warped distance is converted into similarity using an exponential function, where the conversion scale parameter is determined based on historical distance distribution data. Similarity distribution statistics are then performed based on the similarity. First, similarity data within a preset time period is collected to construct a historical similarity sequence. Statistical analysis is performed on the similarity, calculating statistical parameters such as the mean, standard deviation, skewness, and kurtosis. The best-fit distribution type of the similarity data is identified using methods such as the Kolmogorov-Smirnov test or the Anderson-Darling test. Candidate distributions for the best-fit distribution type include normal distribution, beta distribution, and gamma distribution. The similarity distribution statistics are calculated based on the best-fit distribution type.

[0086] Next, a dynamic threshold adjustment mechanism is constructed based on the statistical distribution of similarity. First, multi-level benchmark thresholds are set according to the statistical distribution of similarity, including a severe anomaly threshold, a warning threshold, and a normal benchmark threshold. Based on these multi-level benchmark thresholds, dynamic adjustments are made through a first adjustment and a second adjustment. The first adjustment introduces a time decay factor to make the impact of recent data on the multi-level benchmark threshold adjustment more significant, while the impact of long-term data gradually weakens. The second adjustment combines a load factor and a load sensitivity coefficient to ensure that the multi-level benchmark thresholds adapt to changes in the business environment in real time. The load factor is calculated by comparing the collected real-time traffic data with the average traffic data within a past time window. The load sensitivity coefficient is determined based on the priority of the business scenario and the sensitivity to traffic changes, typically set according to the business type or expert experience.

[0087] Finally, a time-varying business perception benchmark model is constructed based on the similarity and dynamic threshold adjustment mechanism. Specifically, three threshold types are determined for the time-varying business perception benchmark model according to the dynamic threshold adjustment mechanism: normal, warning, and abnormal. The normal range is defined as from the normal benchmark threshold to the maximum similarity value; the warning range is defined as from the warning threshold to the normal benchmark threshold; and the abnormal range is defined as from the minimum similarity value to the warning threshold. Time-varying weights are assigned to the similarity of different dimensions according to the business type and priority. The time-varying weights are determined based on the business priority, time decay factor, and current load status. The current load status is determined based on the load factor and load sensitivity coefficient. A comprehensive similarity is calculated using a weighted fusion formula based on the time-varying weights. A state judgment logic is constructed by comparing the comprehensive similarity index with the threshold types. A time-varying business perception benchmark model is constructed based on the state judgment logic. The parameters of the time-varying business perception benchmark model are determined based on the real-time changes in similarity and the continuous updates of the dynamic threshold adjustment mechanism. By continuously updating historical similarity and statistical distribution, the model ensures that it can reflect the latest business state characteristics.

[0088] The specific steps for S5 are as follows: S5.1: When the hierarchical backhaul status evaluation result is abnormal, a feedback correction loop is constructed based on the change gradient of the physical layer entropy value and the network topology data. In this embodiment, a basic network connectivity graph is first constructed using graph theory based on the link connections in the network topology data. This basic network connectivity graph uses network nodes as vertices and link connections as edges, fully describing the physical connection structure of the network. Then, based on the node load characteristics and link redundancy characteristics in the network topology data, load weights and reliability weights are assigned to each node in the basic network connectivity graph. The load weight reflects the node's processing capacity and current load level, while the reliability weight reflects the node's failure probability and recovery capability. Based on the basic network connectivity graph, load weights, and reliability weights, a weighted network connectivity graph is constructed.

[0089] Next, in the weighted network connectivity graph, a network analysis algorithm is used to calculate the weighted betweenness centrality of each node as a weighted centrality index. Simultaneously, the gradient of change in the physical layer entropy value of each node is compared with a preset threshold; the gradient is calculated by the difference between the physical layer entropy values ​​at adjacent time points. Nodes exceeding the preset threshold are identified as abnormal nodes, and the links associated with these abnormal nodes are identified as abnormal links.

[0090] Then, starting from the anomalous node, a graph theory algorithm is used to determine the anomaly propagation path. This propagation path reflects the possible direction and scope of the anomaly's impact within the network. Based on the number of anomalous links and the weighted centrality index, the anomaly risk level is determined. The more anomalous links and the higher the weighted centrality index of the involved nodes, the more severe the anomaly risk level. Node priorities are determined based on the anomaly propagation path, anomaly risk level, and weighted centrality index; nodes with higher priorities require more frequent monitoring and protection.

[0091] Finally, a feedback correction loop is constructed based on the node priorities. The feedback correction loop includes an anomaly detection stage, an impact assessment stage, a correction strategy generation stage, and an execution feedback stage. In the anomaly detection stage, the feedback correction loop dynamically identifies abnormal nodes and links by real-time monitoring of the physical layer entropy values ​​and link connectivity of network nodes, combined with the node priorities in the weighted network connectivity graph. In the impact assessment stage, the feedback correction loop assesses the impact of the anomaly on the overall performance of the backhaul network based on the anomaly detection results, combined with the anomaly propagation path and anomaly risk level. In the correction strategy generation stage, the feedback correction loop formulates correction strategies for abnormal nodes and links based on the output of the impact assessment stage. The correction strategies include traffic redirection, link switching, and node load balancing. The execution feedback stage applies the correction strategies to the backhaul network, monitors the correction effect in real time, and dynamically adjusts the strategies.

[0092] S5.2: Generate an adaptive protection strategy based on the feedback correction loop; S5.2.1: Based on the node priority and the abnormal propagation path, determine the priority protection nodes and priority protection links; In this embodiment, in the abnormal propagation path, priority protection nodes are determined by comparing node priorities and quantile thresholds. Simultaneously, physical links directly connecting these priority protection nodes and with redundancy below a security threshold are marked as priority protection links. The quantile threshold is determined based on network scale and quality sensitivity characteristics. Furthermore, referencing link redundancy characteristics in network topology data, redundant link counts and fault recovery time assessments are used. If the number of redundant links on a link is less than a preset threshold or the recovery time after a fault exceeds a service tolerance threshold, it is also included in the priority protection link category to prevent priority protection nodes from becoming isolated or to avoid prolonged service interruptions due to link failures.

[0093] S5.2.2: For the priority protection node and priority protection link, calculate the backup path and its corresponding load capacity based on the link redundancy characteristics and node load characteristics in the network topology data. In this embodiment, based on the priority protection nodes and priority protection links, the corresponding link redundancy characteristics and node load characteristics are determined. A path search model is constructed using a constrained shortest path algorithm, starting from the service access end of the priority protection node and ending at the core network access node. The constraints of the path search model include: the backup path must not contain any abnormal links marked in the feedback correction loop; the current CPU utilization, memory usage, and interface bandwidth utilization of each node in the backup path must be lower than a preset safety threshold, which is determined based on the node hardware specifications and peak service load requirements; and the topological isolation between the backup path and the main link must satisfy the condition that the number of shared nodes is less than or equal to one, to avoid simultaneous failure of the main and backup paths due to sharing the same potential fault point.

[0094] Based on the path search model, multiple candidate backup paths that meet the constraints are selected. The load capacity of each path is calculated using bottleneck resource analysis. The load capacity includes node load capacity and link load capacity. The node load capacity is determined by calculating the maximum bandwidth each node can carry, based on the mapping relationship between the node's remaining processing capacity and service bandwidth. The link load capacity is determined by calculating the link's bandwidth that can carry, based on the difference between the link's maximum transmission bandwidth and its currently occupied bandwidth. The minimum value of the bandwidth that can carry for all nodes and links in the search path is taken as the load capacity of the backup path, ensuring that the backup path's carrying capacity can cover the total traffic demand of the services carried by the priority protection nodes and links.

[0095] S5.2.3: Based on the priority protection nodes, priority protection links, backup paths and their corresponding load capacities, generate a preliminary protection strategy, which includes at least link switching, resource reservation and traffic adjustment; In this embodiment, the link switching rapidly switches to a backup path to ensure service continuity when the priority protection link fails. When the physical layer entropy gradient of the priority protection link continuously exceeds a preset threshold, the feedback correction loop determines that the abnormal risk level associated with the priority protection link reaches the "switching required" level. When the backup path availability verification is triggered by conditions meeting the physical layer state abnormality or abnormal risk level, it first performs a secondary verification of the backup path's connectivity and current load capacity using a real-time link detection protocol. Based on the secondary verification result, the availability of the backup path is determined. Then, the link switching adopts a "connect first, disconnect later" switching mode, that is, a service transmission channel is first established on the backup path, and after the service traffic is stably transmitted on the backup path, the transmission of the main link is interrupted to avoid service interruption.

[0096] The resource reservation is used to pre-lock critical resources for the priority protection nodes, priority protection links, and backup paths, ensuring rapid service takeover in the event of anomalies. Once the priority protection nodes and links are determined, resource reservation is immediately initiated to pre-lock resources for core nodes and links, preventing resources from being occupied by non-priority services during sudden anomalies. Based on the temporal patterns in the backhaul service data, by predicting whether the service traffic carried by the priority protection nodes or links will reach its peak within a preset time window and whether the remaining resources of the current node or link are close to the threshold, it is determined whether to initiate resource reservation to meet the resource demands of traffic peaks. When the feedback correction loop determines that a priority protection node or link has a potential anomaly risk, and the anomaly risk level does not reach the switching level but resources need to be prepared in advance, resource reservation is initiated to prepare resource reserves before the switchover.

[0097] The traffic adjustment mechanism supports service migration in abnormal scenarios and improves resource utilization in normal scenarios by adjusting resource allocation. When a link switching command is executed, traffic adjustment is initiated simultaneously. Based on the load capacity ratio of the backup paths, the service traffic carried by the priority protection link is proportionally allocated to each available backup path. At the same time, traffic shaping is applied to low-priority services to limit peak rates, ensuring that high-priority services occupy backup path resources first. When it is detected that the current resource utilization rate of the priority protection node is greater than the balancing threshold, the bandwidth utilization rate of the priority protection link is greater than the upper limit, or the proportion of resources occupied by non-priority services is too high, traffic adjustment is initiated to increase the scheduling weight of high-priority services through a priority scheduling algorithm. When the service type or priority changes, and the current resource allocation cannot meet the needs of the new service, traffic adjustment is initiated to reallocate scheduling weights and adjust the resource reservation ratio to ensure that resource allocation matches service priority.

[0098] S5.2.4: Based on the business type and priority, optimize and adjust the preliminary protection strategy to generate the final adaptive protection strategy.

[0099] In this embodiment, firstly, based on the service type and priority, the core requirements corresponding to each service type are clarified. Specifically, high-reliability, low-latency services emphasize low latency and high reliability; enhanced broadband services emphasize high bandwidth and low packet loss rate; and massively connected services emphasize massive connectivity, low cost, and low power consumption. Then, based on these core requirements, the parameters for link switching, resource reservation, and traffic adjustment are optimized.

[0100] In link switching parameter optimization, for high-priority, high-reliability, low-latency service types, the anomaly confirmation period is determined by multiplying the service latency tolerance threshold by a first coefficient. The first coefficient is determined quantified based on the service type's latency sensitivity. Simultaneously, a pre-switching mechanism is enabled. When the gradient of the physical layer entropy value change in the main link reaches a preset trigger percentage of the anomaly threshold, a backup path connection is established in advance to ensure that the switching latency meets service requirements when an anomaly occurs. The preset trigger percentage is statistically determined based on service reliability requirements and the fluctuation pattern of physical layer entropy values. The service reliability requirements are quantified through the quality sensitivity feature in the standardized feature parameters. The higher the value of the quality sensitivity feature, the higher the service's reliability requirements, and the lower the preset trigger percentage. For low-priority, large-scale connection-type services, the anomaly confirmation period is extended to the product of the service latency tolerance threshold and a second coefficient. The second coefficient is determined jointly based on the service priority score and service continuity characteristics. The service priority score is obtained by weighted calculation using the standardized feature parameters and weight adjustment criteria. The service continuity characteristics are extracted from the backhaul service data. The lower the priority score, the longer the service duration, and the longer the allowed interval for interruption and reconnection, the larger the second coefficient will be. It must also satisfy the constraint that the anomaly confirmation period is greater than or equal to half of the service latency tolerance threshold. At the same time, a handover hysteresis is set. The hysteresis duration is determined based on the statistical analysis of the allowed interval for reconnection after service interruption in the service continuity characteristics. The longer the allowed reconnection interval, the longer the hysteresis duration, to avoid unnecessary handovers triggered by instantaneous signal fluctuations and reduce signaling overhead.

[0101] In optimizing resource reservation parameters, the resource reservation ratio for priority protection nodes associated with high-reliability, low-latency services is set as a high reservation ratio threshold determined based on service reliability requirements. This high reservation ratio threshold is calculated through the mapping relationship between the service quality impact index and the node's remaining processing capacity. The higher the service quality impact index, the higher the reservation ratio threshold, and it must be ensured that the remaining unreserved resources of the node after reservation are not lower than the basic requirements of non-priority services. The backup link bandwidth reservation ratio is set as the product of the service peak traffic and a preset redundancy coefficient. The preset redundancy coefficient is determined based on the probability distribution of service burst traffic to ensure coverage of service burst traffic requirements. The probability distribution of service burst traffic is statistically obtained based on the time regularity characteristics of the backhaul service data. The node resource reservation ratio for enhanced broadband services is set as a medium reservation ratio threshold determined based on service bandwidth demand characteristics. This medium reservation ratio threshold is calculated through matching analysis of service bandwidth demand characteristics and the node's remaining processing capacity to ensure that reserved resources can cover the fluctuation range of the service's average and peak bandwidth. The service bandwidth demand characteristics are determined based on the bandwidth demand characteristics in the standardized feature parameters. The link bandwidth reservation ratio is directly matched to the peak traffic of the service to ensure that bandwidth resources are consistent with the peak service demand. The peak traffic of the service is obtained by statistical analysis of the dynamic feature vector of traffic in the backhaul service data. The node resource reservation ratio for large-scale connection services is set as a low reservation ratio threshold determined based on the service priority score. The low reservation ratio threshold is calculated by mapping the service priority score to the proportion of the node's remaining processing capacity, with a low priority score corresponding to a low reservation ratio; its link bandwidth is not reserved separately, but only ensures basic transmission needs through traffic shaping mechanisms.

[0102] In traffic adjustment parameter optimization, high-reliability, low-latency services are assigned absolute priority in queue scheduling. This absolute priority is determined by service priority scores, with higher priority scores corresponding to the highest scheduling level, and no traffic shaping is performed. Enhanced broadband services are assigned a medium scheduling weight that matches their service priority. This medium scheduling weight is calculated by the weight ratio of the service priority score to that of other services. The peak rate limit for traffic shaping is determined based on a ratio matching the backup path link bandwidth utilization safety threshold. This ratio is determined by analyzing the difference between the backup path link bandwidth and the service peak traffic, ensuring that the link bandwidth utilization does not exceed a preset safety threshold. This safety threshold is statistically obtained based on link transmission characteristic indicators. The scheduling weight for large-scale connection services is determined based on the lowest weight level of the service priority score, which is determined by ranking the service priority scores. The peak rate limit for traffic shaping is determined based on the matching ratio of the remaining backup path bandwidth to the bandwidth demand ratio of low-priority services. This matching ratio is calculated by mapping the remaining backup path bandwidth to the bandwidth demand ratio of low-priority services. The remaining backup path bandwidth is determined based on the difference between the backup path load capacity and the allocated bandwidth. The bandwidth requirement ratio for low-priority services is determined by the ratio of the total bandwidth requirement for low-priority services calculated based on service priority scores to the total bandwidth of backup paths. Simultaneously, a batch transmission mode is employed to reduce the number of connection establishments.

[0103] Based on the results of link switching parameter optimization, resource reservation parameter optimization, and traffic adjustment parameter optimization, an optimization strategy is generated. This optimization strategy is then substituted into the time-varying service awareness benchmark model, and the corresponding service quality indicators are calculated through simulation testing. The service quality indicators are compared with the reference parameters for normal service states. If the service quality indicators meet the preset standards for that service type, the strategy is deemed effective; if the service quality indicators do not meet the standards, the parameters are readjusted until the requirements are met. Finally, the validated optimization strategies are integrated into adaptive protection strategies for different service types and priorities.

[0104] Furthermore, such as Figure 4 As shown in the figure, this application provides a backhaul link monitoring system for 5G base stations, including: The data acquisition module is used to collect physical signal statistical data, backhaul service data, and network topology data; The status analysis module is used to generate business layer anomaly prediction parameters based on the output of the data acquisition module, quantify the matching degree deviation between the parameters and the time-varying business perception benchmark model, and output hierarchical back-transmission status assessment results. The feedback correction module is used to combine the change gradient of the physical layer entropy value with the network topology data to construct a feedback correction loop, thereby realizing dynamic monitoring and correction of the backhaul network status. The strategy generation module is used to generate targeted adaptive protection strategies based on the hierarchical backhaul status assessment results and the feedback correction loop.

[0105] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0107] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0108] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A backhaul link monitoring method applied to 5G base stations, characterized in that, include: Acquire backhaul link monitoring data of 5G base stations, the backhaul link monitoring data including physical signal statistics, backhaul service data and network topology data; Based on the physical signal statistics, statistical analysis is performed using probability distribution methods, and the physical layer entropy value is calculated using the Shannon entropy formula. Based on the physical layer entropy value and the backhaul service data, the service layer anomaly prediction parameters are calculated using the spatiotemporal coding mechanism of a spiking neural network. Based on the backhaul service data, the service type and priority are determined, and a time-varying service perception benchmark model is constructed using chaotic dynamics equations. By quantifying the matching deviation between the anomaly prediction parameters of the business layer and the time-varying business perception benchmark model, the hierarchical backhaul status assessment results are output. When the hierarchical backhaul status assessment result is abnormal, a feedback correction loop is constructed based on the change gradient of the physical layer entropy value and the network topology data, and an adaptive protection strategy is generated based on the feedback correction loop.

2. The backhaul link monitoring method for 5G base stations according to claim 1, characterized in that, The backhaul link monitoring data includes physical signal statistics, backhaul service data, and network topology data, among which: The physical signal statistics include signal quality indicators, signal stability indicators, and signal transmission characteristic indicators. The signal quality indicators are obtained based on the signal-to-noise ratio and bit error rate. The signal stability indicators are obtained based on link jitter and interruption frequency. The signal transmission characteristic indicators are obtained based on signal strength and transmission delay. The backhaul service data includes service traffic characteristics, data packet loss characteristics, transmission rate, and service continuity characteristics. The network topology data includes link connectivity, node load characteristics, and link redundancy characteristics.

3. The backhaul link monitoring method for 5G base stations according to claim 2, characterized in that, Based on the physical signal statistics, statistical analysis is performed using probability distribution methods, and the physical layer entropy value is calculated using the Shannon entropy formula, including: Based on the time series data and parameter characteristics of each parameter in the physical signal statistics within a preset time window, a probability distribution method adapted to the parameter characteristics is used for statistical analysis to obtain the probability distribution characteristics of signal quality, link stability, and transmission performance. Based on the signal quality probability distribution characteristics, link stability probability distribution characteristics, and transmission performance probability distribution characteristics, the signal quality entropy value, link stability entropy value, and transmission performance entropy value are calculated respectively using the Shannon entropy formula. The physical layer entropy value is calculated using a weighted fusion calculation formula based on the signal quality entropy value, link stability entropy value, and transmission performance entropy value.

4. The backhaul link monitoring method for 5G base stations according to claim 3, characterized in that, The anomaly prediction parameters for the business layer are calculated using a spatiotemporal coding mechanism derived from a spiking neural network. This spatiotemporal coding mechanism includes both temporal and spatial coding mechanisms. Based on the time sequence of the physical layer entropy values, a physical layer entropy value fluctuation feature vector is calculated using a sliding window. Based on the backhaul service data, a traffic dynamic feature vector, a transmission quality feature vector, a bandwidth utilization feature vector, and a time distribution feature vector are calculated. The traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector and time distribution feature vector are aligned in dimension to obtain an aligned feature vector. The time series data corresponding to the aligned feature vector is converted into a pulse sequence according to the time coding mechanism. The pulse sequence is converted into a spatiotemporal pulse feature matrix through a spatial coding mechanism. The spatiotemporal pulse feature matrix is ​​input into a multilayer spiking neural network to construct a cross-layer anomaly prediction model. The prediction sequence of the anomaly probability, the scope of the anomaly impact, and the duration of the anomaly at the business layer is calculated by a probability normalization function. The prediction sequence is weighted and fused and mapped to a specified interval to obtain the anomaly prediction parameters at the business layer.

5. The backhaul link monitoring method for 5G base stations according to claim 4, characterized in that, Determining the service type and priority based on the returned service data includes: The traffic dynamic feature vector, transmission quality feature vector, bandwidth utilization feature vector, and time distribution feature vector are quantized to generate standardized feature parameters, which include traffic pattern features, quality sensitivity features, bandwidth demand features, and time pattern features. The corresponding service type feature identifier is determined based on the standardized feature parameters. The service type feature identifier includes traffic type identifier, reliability type identifier, bandwidth type identifier, and time type identifier. The service type is determined through multi-dimensional feature fusion analysis based on the combined features of the service type feature identifier. Based on the standardized feature parameters and service type, a service quality impact index is calculated using a nonlinear mapping function. The corresponding weight adjustment criteria are determined based on the service type, including reliability weight, bandwidth requirement weight, and real-time requirement weight. The business quality impact index and the weight adjustment criteria are weighted and summed to obtain a comprehensive priority score. Based on the comprehensive priority score, the priority corresponding to the business type is determined by a preset priority division standard.

6. The backhaul link monitoring method for 5G base stations according to claim 5, characterized in that, A time-varying business perception benchmark model is constructed using chaotic dynamics equations, including: Based on the business type and priority, the initial range of the control parameters of the chaotic dynamics equation is determined. The control parameters include nonlinear coefficients, coupling strength, and initial conditions. Based on the returned service data, a parameter adaptive adjustment mechanism is constructed. Within the initial range of the control parameters, the optimal control parameters are dynamically determined through the parameter adaptive adjustment mechanism. Based on the optimal control parameters, a chaotic system is constructed, and a time-varying reference trajectory is generated. Based on the time regularity features in the standardized feature parameters, data segments that meet the preset normal operation standards are selected from the backhaul service data to obtain the service normal state feature set. The service normal state feature set is then mapped to the chaotic phase space through a phase space reconstruction technology set to obtain the service feature trajectory. The similarity between the business feature trajectory and the time-varying reference trajectory is calculated by a dynamic time warping algorithm. Based on the distribution statistics of the similarity, a dynamic threshold adjustment mechanism is established. Based on the similarity and the dynamic threshold adjustment mechanism, a time-varying business perception benchmark model is constructed.

7. The backhaul link monitoring method for 5G base stations according to claim 6, characterized in that, By quantifying the deviation between the anomaly prediction parameters of the business layer and the time-varying business perception benchmark model, the hierarchical backhaul status assessment results are output, including: Based on the time-varying service perception benchmark model, obtain the normal service status reference parameters corresponding to the current time point. The normal service status reference parameters include the abnormal probability threshold, the boundary of the abnormal impact range, and the abnormal duration range under the normal service status. The abnormal prediction parameters and normal business status reference parameters of the business layer are standardized and unified into the same feature space. The matching degree deviation is calculated by mapping to the same feature space. The matching degree deviation includes Euclidean distance, cosine similarity and information entropy difference. Weights are assigned to the matching degree deviations based on the business type, and a multi-dimensional deviation vector is constructed by weighted summation. Preset multi-level thresholds are determined based on the historical distribution data of the multi-dimensional deviation vectors. The multi-dimensional deviation vectors are compared with the preset multi-level thresholds, and the hierarchical feedback status evaluation results are output.

8. The backhaul link monitoring method for 5G base stations according to claim 7, characterized in that, A feedback correction loop is constructed based on the gradient of the physical layer entropy and the network topology data, including: Based on the link connection relationships in the network topology data, a basic network connectivity graph is constructed. Based on the node load characteristics and link redundancy characteristics in the network topology data, the load weight and reliability weight of each node in the basic network connectivity graph are determined. Based on the basic network connectivity graph, load weight, and reliability weight, a weighted network connectivity graph is constructed. In the weighted network connectivity graph, the weighted betweenness centrality of each node is calculated as the weighted centrality index. The change gradient of the physical layer entropy value of each node is compared with the size of a preset threshold. Nodes that are greater than the preset threshold are identified as abnormal nodes, and the links associated with the abnormal nodes are identified as abnormal links. Starting from the abnormal node, the abnormal propagation path is determined by graph theory algorithm. The abnormal risk level is determined based on the number of abnormal links and the weighted centrality index. The node priority is determined based on the abnormal propagation path, the abnormal risk level, and the weighted centrality index. A feedback correction loop is constructed based on the node priority.

9. The backhaul link monitoring method for 5G base stations according to claim 8, characterized in that, An adaptive protection strategy is generated based on the feedback correction loop, including: Based on the node priority and the abnormal propagation path, determine the priority protection nodes and priority protection links; For the priority protection nodes and priority protection links, the backup paths and their corresponding load capacities are calculated based on the link redundancy characteristics and node load characteristics in the network topology data. Based on the priority protection nodes, priority protection links, backup paths and their corresponding load capacities, a preliminary protection strategy is generated. The preliminary protection strategy includes at least link switching, resource reservation and traffic adjustment. Based on the business type and priority, the initial protection strategy is optimized and adjusted to generate the final adaptive protection strategy.

10. A backhaul link monitoring system for 5G base stations, used to implement the backhaul link monitoring method for 5G base stations as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, a status analysis module, a feedback correction module, and a strategy generation module, among which: The data acquisition module is used to collect physical signal statistics, backhaul service data, and network topology data of the base station backhaul link. The status analysis module is used to generate business layer anomaly prediction parameters based on the output of the data acquisition module, quantify the matching degree deviation between the parameters and the time-varying business perception benchmark model, and output hierarchical back-transmission status assessment results. The feedback correction module is used to construct a feedback correction loop by combining the gradient of the physical layer entropy value change with the network topology data; The strategy generation module is used to generate an adaptive protection strategy based on the hierarchical backhaul status assessment results and the feedback correction loop.

Images