Multi-modal data source probe state monitoring and failover method and multi-modal data source probe state monitoring and failover system

Abstract

The invention discloses a multi-modal data source probe state monitoring and failover method and system, and relates to the field of Internet, the method comprises the following steps: respectively deploying data acquisition probes at a wireless access network node and a core network UPF node to obtain preprocessed network data of the wireless access network node and preprocessed network data of the core network UPF; obtaining abnormal risk prediction information through an AI monitoring model; adjusting parameter configuration of a wireless side and a core network UPF side according to the abnormal risk prediction information; and based on the adjusted parameter configuration of the wireless side and the core network UPF side, feeding back to the service awareness module so as to realize failover. According to the method, a cross-domain and cross-layer unified probe monitoring system is realized, a dynamic, explainable and closed-loop network fault analysis and regulation and control mechanism is constructed through AI technologies such as a graph neural network and reinforcement learning, and the network state sensing precision, the response timeliness and the optimization intelligent level can be remarkably improved.

Description

Technical Field

[0001] This specification relates to the field of the Internet, and more specifically, to a method and system for multimodal data source probe status monitoring and fault transfer. Background Art

[0002] With the development of 5G and future network systems, network structures are becoming increasingly complex. With the decoupling of the data and control planes, real-time awareness of network status and precise scheduling have become particularly important. This is especially true between the radio access network (RAN) and the core network user plane function (UPF), where service paths are long, user scale is large, and data flows are complex. Once a network failure occurs, it can easily cause widespread service interruption or performance degradation.

[0003] Existing network operations and maintenance mainly rely on manually set rules or traditional alarm systems to complete status monitoring and fault handling. These methods have the following shortcomings:

[0004] 1. Fragmented probe deployment and isolated information: Data collection on the wireless side and the core network UPF side is often managed separately, lacking end-to-end cross-domain status integration capabilities.

[0005] 2. Lagging network status monitoring: Traditional systems rely on threshold alarms or static indicator judgments, making it difficult to detect dynamic and sudden risks in a timely manner;

[0006] 3. Lack of intelligent tuning mechanisms: Even when anomalies are detected, existing systems often rely on fixed policy execution and are unable to perform adaptive optimization based on actual network evolution.

[0007] 4. Lack of explainability of control responses: Network parameter adjustments lack the identification and analysis of abnormal sources, which can easily lead to excessive or invalid configurations and affect network stability.

[0008] In the modern network environment with high-density access and complex business types, there is an urgent need for a fault detection and transfer method based on multi-source probe fusion, with intelligent analysis capabilities and closed-loop optimization, to support a more flexible, explainable and efficient network operation assurance system. Summary of the Invention

[0009] The Summary of the Invention introduces a series of simplified concepts that will be further described in the Detailed Description of the Invention. The Summary of the Invention is not intended to limit the key features and essential features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0010] In a first aspect, the present invention provides a multimodal data source probe status monitoring and fault transfer method, the method comprising:

[0011] Data collection probes are deployed at the radio access network nodes and core network UPF nodes respectively to collect network data of radio access network nodes and core network UPF network data;

[0012] Performing a preprocessing operation on the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data;

[0013] The pre-processed wireless access network node network data and the pre-processed core network UPF network data are input into the AI ​​monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module and a risk prediction module, the multi-source feature encoding module is used to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space respectively, the spatiotemporal collaborative modeling module is used to construct a dynamic graph and extract the spatiotemporal dependency of the dynamic graph based on a graph neural network; the risk prediction module is used to model the historical state sequence of the nodes in the dynamic graph to output the abnormal risk prediction information;

[0014] Adjusting parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information;

[0015] The adjusted parameter configurations on the wireless side and the core network UPF side are fed back to the service perception module to achieve failover.

[0016] In a feasible implementation manner, the network data includes network traffic, delay, packet loss, jitter, application connectivity and application network quality.

[0017] In a feasible implementation, the risk prediction module adopts a multi-layer Transformer structure.

[0018] In a feasible implementation manner, adjusting the parameter configurations of the wireless side and the core network UPF side according to the abnormal risk prediction information includes:

[0019] Constructing a policy knowledge graph, wherein the policy knowledge graph is used to store associations between anomaly types and network parameter items and mappings between network parameter items and control actions;

[0020] Based on the abnormal risk prediction information, retrieving a set of candidate parameter adjustment strategies that matches the current abnormality type from the strategy knowledge graph;

[0021] Obtain current network state information, historical parameter adjustment records, and network performance feedback information to construct a state representation for reinforcement learning;

[0022] Based on the state representation, performing an optimal strategy selection operation based on a deep reinforcement learning algorithm in the set of candidate parameter adjustment strategies to determine a target control action for adjusting the wireless side and / or the core network UPF side;

[0023] Execute corresponding parameter adjustment operations according to the target control action to achieve optimization of network performance indicators, wherein the target control action aims to maximize network performance improvement, and the network performance improvement is evaluated by a reward function, and the reward function is calculated based on risk score changes, network delay changes, packet loss rate changes and QoS feedback information.

[0024] In a feasible implementation, based on the state representation, performing an optimal strategy selection operation based on a deep reinforcement learning algorithm in the candidate parameter adjustment strategy set to determine a target control action for adjusting the wireless side and / or the core network UPF side includes:

[0025] Passing the state representation as input to a policy selection model built based on deep reinforcement learning, wherein the policy selection model is used to evaluate actions and select policies from a set of candidate parameter adjustment policies. The policy selection model is trained using a deep reinforcement learning algorithm to minimize a policy loss function and maximize a reward function based on network performance feedback. The reward function is evaluated based on changes in risk score, network latency, packet loss rate, and QoS indicators.

[0026] The strategy score of each candidate action is calculated based on the current state representation within the candidate parameter adjustment strategy set, and the target control action with the highest expected long-term benefit is selected, wherein the target control action is used as a network control instruction to adjust the parameter configuration of the corresponding wireless side and / or core network UPF side.

[0027] In a feasible implementation manner, the target control action includes one or more of performing dynamic adjustment operations of cell scheduling priority on the wireless side, performing transmission power control operations on the wireless side, performing switching threshold optimization operations on the wireless side, performing rerouting operations of user plane traffic paths on the core network UPF side, performing QoS identification switching operations on the core network UPF side, and performing routing table management operations on the core network UPF side.

[0028] In a second aspect, the present invention further provides a multimodal data source probe status monitoring and fault transfer system, which is used in any of the methods described in the first aspect, comprising:

[0029] A collection unit is configured to deploy data collection probes at radio access network nodes and core network UPF nodes respectively to collect radio access network node network data and core network UPF network data, wherein the network data includes network traffic, delay, packet loss, jitter, application connectivity, and application network quality;

[0030] A preprocessing unit, configured to perform a preprocessing operation on the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data;

[0031] A monitoring unit, configured to input the pre-processed wireless access network node network data and the pre-processed core network UPF network data into an AI monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module, and a risk prediction module; the multi-source feature encoding module is configured to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space, respectively; the spatiotemporal collaborative modeling module is configured to construct a dynamic graph and extract the spatiotemporal dependency of the dynamic graph based on a graph neural network; the risk prediction module uses a multi-layer Transformer structure to model the historical state sequence of nodes in the dynamic graph to output the abnormal risk prediction information;

[0032] An adjustment unit, configured to adjust parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information;

[0033] The optimization unit is used to feed back the adjusted parameter configuration on the wireless side and the core network UPF side to the service perception module to achieve failover.

[0034] In a feasible implementation, the multi-source feature encoding module includes a sliding window encoding unit, a heterogeneous domain fusion unit, a fusion mapping unit, and a connection unit;

[0035] The sliding window encoding unit is used to perform sliding window segmentation on the pre-processed radio access network node network data and the pre-processed core network UPF network data according to different time granularities to extract short-cycle mutation features and long-cycle trend features;

[0036] The heterogeneous domain fusion unit is used to calculate the feature attention weights of the wireless side data and the core network UPF side data respectively, and perform intra-domain feature weighting based on the attention weights;

[0037] A fusion mapping unit is used to map the weighted features into a unified semantic vector space through a deep neural network structure with shared weights, so as to achieve semantic alignment of data from different network domains;

[0038] The connection unit is used to introduce a residual connection structure in the mapping process and combine it with the channel attention mechanism to enhance the stability of feature propagation.

[0039] In a feasible implementation, the spatiotemporal collaborative modeling module includes a dynamic graph subunit, a graph neural network subunit, a graph evolution representation subunit, and a structure attribution subunit;

[0040] The dynamic graph subunit is used to map the feature vectors output by the multi-source feature encoding module into a graph structure of multiple consecutive time steps, wherein the graph structure of each time step includes nodes, edges and dynamic weights, and the weights of the edges are dynamically generated based on traffic coordination and delay correlation;

[0041] The graph neural network subunit is used to extract the graph structure representation of the nodes in the graph at each time step based on the graph attention mechanism. The graph neural network subunit includes a time-aware embedding mechanism, an edge attribute embedding mechanism, and a skip connection mechanism;

[0042] The graph evolution representation subunit is used to perform temporal modeling on the node representation of the continuous graph structure to obtain the dynamic embedding features of the nodes in the historical evolution process;

[0043] The structural attribution unit is used to interpret the node status based on the attention mechanism in the graph neural network, and output the impact path and key adjacent nodes of the high-risk node.

[0044] In a feasible implementation, the risk prediction module includes a historical state sequence subunit, a multi-layer Transformer subunit, a multi-scale risk perception subunit, and a risk causal explanation subunit;

[0045] The historical state sequence subunit is used to extract the node historical state vector from the dynamic graph output by the spatiotemporal collaborative modeling module and construct an input sequence in combination with the time position coding;

[0046] The multi-layer Transformer subunit includes multiple encoding layers, each of which contains a multi-head attention layer, a feedforward network, a residual connection, and a normalization structure to capture the temporal dependencies and abnormal evolution patterns in the node state sequence;

[0047] The multi-scale risk perception subunit is used to generate abnormal trend prediction results, risk type classification labels and confidence estimation information based on Transformer output;

[0048] The risk causal explanation subunit is used to identify key time segments and / or indicators that lead to high risk scores based on the attention weights, and output risk explanation information.

[0049] In summary, the multimodal data source probe status monitoring and fault transfer method provided by the present invention synchronously deploys data collection probes on the wireless access network nodes and the core network UPF nodes, and performs multi-source data fusion and spatiotemporal modeling based on the AI ​​monitoring model. This method realizes the unified probe deployment and data collection on the wireless side and the core network UPF side, and deeply integrates heterogeneous indicators through the multi-source feature coding module, which solves the problems of fragmented collection surfaces and isolated information in the existing system and improves the integrity of state perception. By using graph neural networks to model the spatiotemporal dependencies between nodes and deep learning of historical state sequences, early warning and evolution trend analysis of network anomalies can be achieved, which is significantly better than traditional hysteresis detection methods based on thresholds or sliding averages. By tracing back the source features or paths of high-risk scores through the attention mechanism, the system can provide a structural explanation of network risks, support more accurate local parameter adjustments, and avoid blind parameter adjustments or misadjustments. This application introduces a dual-driven control mechanism of strategy knowledge graph and deep reinforcement learning. After risk identification, it can automatically match candidate control strategies and adaptively select the optimal control action based on historical feedback to achieve joint intelligent regulation of the wireless and UPF sides, and has self-evolution capabilities. The proposed method can adapt to different network topologies and probe protocols of different operators. Both the AI ​​model and the policy module support modular replacement and online optimization, and have good engineering implementation and long-term evolution potential. In summary, compared with the existing technology, this application not only realizes a unified probe monitoring system across domains and layers, but also constructs a dynamic, explainable, and closed-loop network fault analysis and control mechanism through AI technologies such as graph neural networks and reinforcement learning. It can significantly improve the accuracy of network status perception, the timeliness of response, and the intelligence level of tuning, and has significant practical value and technological advancement.

[0050] The multimodal data source probe status monitoring and fault transfer method proposed in the present invention, and other advantages, objectives and features of the present invention will be reflected in part through the following description, and in part will be understood by technical personnel in this field through research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present description. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0052] Figure 1 A schematic diagram of a multimodal data source probe status monitoring and fault transfer method provided by an embodiment of the present invention;

[0053] Figure 2A schematic diagram of a process for adjusting parameter configurations on the wireless side and the core network UPF side according to abnormal risk prediction information provided by an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of a process for determining a target control action for adjusting a wireless side and / or a core network UPF side provided by an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of the structure of a multimodal data source probe status monitoring and fault transfer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments.

[0057] See also Figure 1 , which is a flow chart of a multimodal data source probe status monitoring and fault transfer method provided by an embodiment of the present invention, which may specifically include:

[0058] S110. Deploy data collection probes at radio access network nodes and core network UPF nodes to collect network data of radio access network nodes and core network UPF nodes, including network traffic, latency, packet loss, jitter, application connectivity, and application network quality.

[0059] S120: Preprocess the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data.

[0060] S130. Input the pre-processed wireless access network node network data and the pre-processed core network UPF network data into the AI ​​monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module and a risk prediction module. The multi-source feature encoding module is used to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space, respectively. The spatiotemporal collaborative modeling module is used to construct a dynamic graph and extract its spatiotemporal dependency based on a graph neural network; the risk prediction module is used to model the historical state sequence of the nodes in the dynamic graph to output abnormal risk prediction information.

[0061] S140. Adjust the parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information.

[0062] S150: Feedback the adjusted parameter configurations on the wireless side and the core network UPF side to the service perception module to achieve failover.

[0063] In a feasible implementation manner, the network data includes network traffic, delay, packet loss, jitter, application connectivity and application network quality.

[0064] In a feasible implementation, the risk prediction module adopts a multi-layer Transformer structure.

[0065] For example, data collection probes are deployed at wireless access network nodes (such as gNodeBs and cell base stations) and core network UPF (User Plane Function) nodes. These probes are used to collect multi-dimensional performance data related to network operation status in real time. The collected network data includes, but is not limited to: network traffic indicators (such as uplink and downlink speeds, number of connections); latency indicators (such as single-hop latency and end-to-end latency); packet loss rate and jitter information; application layer connectivity (such as HTTP success rate and DNS response); and application network quality (such as video freeze rate and MOS score).

[0066] The collected data then undergoes a unified preprocessing process, including timestamp alignment, noise removal, missing value filling, normalization, and feature structuring. This preprocessing process generates standardized time series feature data that can be used by the model, ensuring data consistency and modeling effectiveness from different network sides.

[0067] The preprocessed data is then fed into an AI monitoring model for intelligent analysis. This AI monitoring model consists of three parts: First, a multi-source feature encoding module performs sliding window segmentation, attention-weighted fusion, and unified semantic space mapping on both wireless access network data and core network UPF data, enabling deep fusion of data from different domains. Second, a spatiotemporal collaborative modeling module constructs a dynamic graph, using network nodes as vertices and traffic coordination and latency correlation as edge weights. A graph neural network extracts the spatiotemporal dependency features of each node at different time steps. Finally, a risk prediction module uses a multi-layer Transformer structure to model the historical state sequence of nodes in the dynamic graph, outputting a future risk score for each node. Combined with the attention mechanism, it provides explanations for the causes of high-risk nodes, enabling interpretable analysis of network anomalies.

[0068] After obtaining anomaly risk prediction information, the system dynamically adjusts key parameters on both the wireless and core network UPF sides based on the anomaly location and type. For example, on the wireless side, load or interference in risky areas can be mitigated through scheduling priority adjustment, transmit power control, or handover threshold optimization. On the core network UPF side, network traffic can be directed and controlled through path rerouting, QFI level switching, or forwarding table management strategies. These parameter configuration actions are provided by a policy knowledge graph with a set of candidate policies. A deep reinforcement learning model then selects the optimal control policy based on the current state, achieving multi-objective network optimization.

[0069] Ultimately, the adjusted parameter configuration results will be fed back to the service perception module, which will monitor the user service experience quality in real time and evaluate whether the adjustment effect meets the optimization goal, forming a closed-loop control mechanism of perception-prediction-adjustment-feedback.

[0070] The multimodal data source probe status monitoring and fault transfer method provided in this application is achieved by synchronously deploying data collection probes on the wireless access network nodes and the core network UPF nodes, and performing multi-source data fusion and spatiotemporal modeling based on the AI ​​monitoring model. This method realizes the unified probe deployment and data collection on the wireless side and the core network UPF side, and deeply integrates heterogeneous indicators through the multi-source feature coding module, which solves the problems of fragmented collection surfaces and isolated information in the existing system and improves the integrity of state perception. By using graph neural networks to model the spatiotemporal dependencies between nodes and combining multi-layer Transformers to perform deep learning on historical state sequences, early warning and evolution trend analysis of network anomalies can be achieved, which is significantly better than traditional hysteresis detection methods based on thresholds or sliding averages. By tracing back the source features or paths of high-risk scores through the attention mechanism, the system can provide a structural explanation of network risks, support more accurate local parameter adjustments, and avoid blind parameter adjustments or misadjustments. This application introduces a dual-driven control mechanism of strategy knowledge graph and deep reinforcement learning. After risk identification, it can automatically match candidate control strategies and adaptively select the optimal control action based on historical feedback to achieve joint intelligent regulation of the wireless and UPF sides, and has self-evolution capabilities. The proposed method can adapt to different network topologies and probe protocols of different operators. Both the AI ​​model and the policy module support modular replacement and online optimization, and have good engineering implementation and long-term evolution potential. In summary, compared with the existing technology, this application not only realizes a unified probe monitoring system across domains and layers, but also constructs a dynamic, explainable, and closed-loop network fault analysis and control mechanism through AI technologies such as graph neural networks and reinforcement learning. It can significantly improve the accuracy of network status perception, the timeliness of response, and the intelligence level of tuning, and has significant practical value and technological advancement.

[0071] In one possible implementation, Figure 2 As shown, Figure 2 A schematic diagram of a process for adjusting parameter configurations on the wireless side and the core network UPF side according to abnormal risk prediction information provided in an embodiment of the present invention, wherein S140 adjusts the parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information, including:

[0072] S1401. Construct a policy knowledge graph, wherein the policy knowledge graph is used to store associations between anomaly types and network parameter items and mappings between network parameter items and control actions;

[0073] S1402: Based on the above-mentioned abnormal risk prediction information, a candidate parameter adjustment strategy set matching the current abnormality type is retrieved from the above-mentioned strategy knowledge graph;

[0074] S1403, obtaining current network state information, historical parameter adjustment records, and network performance feedback information to construct a state representation for reinforcement learning;

[0075] S1404. Based on the state representation, perform an optimal strategy selection operation based on a deep reinforcement learning algorithm in the candidate parameter adjustment strategy set to determine a target control action for adjusting the wireless side and / or the core network UPF side;

[0076] S1405. Execute corresponding parameter adjustment operations according to the above-mentioned target control action to optimize the network performance indicators, wherein the above-mentioned target control action aims to maximize the network performance improvement, and the above-mentioned network performance improvement is evaluated by a reward function, and the above-mentioned reward function is calculated based on the risk score change, network delay change, packet loss rate change and QoS feedback information.

[0077] In one possible implementation, Figure 3 As shown, Figure 3 A process diagram for determining a target control action for adjusting the wireless side and / or the core network UPF side provided in an embodiment of the present invention, wherein S1404 is based on the above-mentioned state representation, and performs an optimal strategy selection operation based on a deep reinforcement learning algorithm in the above-mentioned candidate parameter adjustment strategy set to determine the target control action for adjusting the wireless side and / or the core network UPF side, including:

[0078] S14041. Pass the state representation as input to a policy selection model built based on deep reinforcement learning, wherein the policy selection model is used to evaluate actions and select policies within a set of candidate parameter adjustment policies. The policy selection model is trained using a deep reinforcement learning algorithm to minimize a policy loss function and maximize a reward function based on network performance feedback. The reward function is evaluated based on changes in risk score, network latency, packet loss rate, and QoS indicators.

[0079] S14042. Calculate the strategy score of each candidate action based on the current state representation within the above-mentioned candidate parameter adjustment strategy set, and select the target control action with the highest expected long-term benefit, wherein the above-mentioned target control action is used as a network control instruction to adjust the corresponding parameter configuration of the wireless side and / or the core network UPF side.

[0080] For example, the system first constructs a policy knowledge graph to store the triplet relationships between network anomaly types, parameter items, and control actions. This graph is organized in the form of "anomaly type - parameter item - control action," for example, relationships such as "increased packet loss rate - UPF path - rerouting" or "latency jitter - scheduling weight - priority increase." The policy knowledge graph not only records the control actions corresponding to historical anomalies but also integrates expert experience and data-driven results to provide structured support for the subsequent generation of candidate actions.

[0081] After receiving the abnormal risk prediction information output by the AI ​​monitoring module, the system retrieves the candidate parameter adjustment strategy set that best matches the current situation from the strategy knowledge graph based on the predicted abnormality type, location and intensity, forming a limited action space for use in the strategy decision-making stage.

[0082] Subsequently, the system obtains the current network operation status information, including but not limited to node load, delay distribution, link utilization, historical risk level, etc., and extracts the parameter adjustment action sequence executed in the past and the corresponding network performance feedback. The above information constitutes the state representation, which is recorded as:

[0083]

[0084] in: Represents the multi-dimensional state characteristics of the current network node; A t-n:t-1 represents the sequence of actions executed in the past n time steps; R t-n:t-1 Indicates network performance feedback within a certain period of time (such as latency, packet loss, and changes in QoE indicators).

[0085] The state vector is used as input to the Actor-Critic structure based on the PPO algorithm to perform the policy selection operation. The policy network (Actor) scores and samples actions from the candidate policy set and outputs the optimal control action under the current state. Reward function R t The design is used to measure the degree of improvement in network performance brought about by a certain strategy, and its calculation formula can be set as:

[0086] R t =λ1·ΔP t +λ2·ΔS t -λ3·C(a t )+λ4·C t

[0087] Where: ΔP t is the prediction consistency term, It is used to measure the degree of verification of the model for abnormal predictions. For example, if it is predicted that the risk of a certain path will increase in the future, whether the strategic action successfully makes the node be judged as high risk at the next moment.

[0088] ΔS t is a comprehensive performance improvement item (which can be refined into a function),

[0089] ΔS t =f(ΔQoS t ,ΔDelay t ,ΔLoss t ,ΔUtil t )

[0090] f(·) uses a normalized dynamic weighted average function, where the weight is determined by the current network load state. For example, when the service is busy, ΔQoS t The weight is greater; during the network stability period, ΔDelay t and ΔUtil t More important.

[0091] C(a t ) is the control action cost item, C(a t ) = θ1 × instruction complexity + θ2 × impact range + θ3 × QoS disturbance risk, taking into account the "disturbance cost" required to perform the action, such as the execution load of cutting UPF and raising scheduling priority. A threshold value can be set, and a negative penalty is imposed when the control cost exceeds the threshold.

[0092] C t is the prediction confidence adjustment term, C t =φ(Confidence)·(ΔRiskScore t ), where φ(·) is a confidence enhancement function (such as sigmoid). If a high-confidence prediction result successfully avoids risk, the positive reward will be increased to strengthen the model's ability to "trust strong signals."

[0093] Through this design, the reward function not only reflects traditional performance improvements but also introduces predictive drive and control cost awareness, helping reinforcement learning models achieve a more balanced, stable, and interpretable policy learning process in complex network environments. This mechanism continuously adjusts the policy network during training, aligning it with optimal actions that effectively reduce network risk without incurring excessive control costs, thereby enhancing the intelligent tuning capabilities of the entire system.

[0094] This implementation method constructs a policy knowledge graph as expert experience support and combines it with a deep reinforcement learning policy network for dynamic policy selection. It not only improves the intelligence level of exception handling, but also realizes the joint optimization of multiple network indicators. It has the significant advantages of strong interpretability, high execution efficiency and good adaptability, and is suitable for fault prediction and self-recovery control tasks in various network scenarios.

[0095] In a feasible implementation, the above-mentioned target control action includes performing one or more of cell scheduling threshold optimization operations on the wireless side, performing user plane traffic path rerouting operations on the core network UPF side, performing QoS identification switching operations on the core network UPF side, and performing routing table management operations on the core network UPF side.

[0096] For example, when the system detects anomalies such as increased packet loss, scheduling congestion, or service queuing delays in a cell, it can dynamically adjust the cell's scheduling priority. This operation prioritizes scheduling resources and improves the user experience by increasing the scheduling weight of key services (such as VoIP and video) and reducing the scheduling priority of background services.

[0097] When abnormalities manifest as increased interference between neighboring cells, severe signal overlap, or uneven coverage, the system triggers transmit power control. Specifically, this includes reducing transmit power in the interfering cell to minimize signal leakage, or increasing transmit power in cells with insufficient edge coverage to enhance coverage, thereby improving overall cell quality and handover success rates.

[0098] To address frequent ping-pong handovers, handover failures, or abnormal handover delays, the system can optimize handover thresholds. This operation dynamically adjusts handover trigger thresholds such as RSRP and RSRQ, or the delay time (T), to suppress unnecessary handovers and improve handover success rates and network stability.

[0099] On the core network UPF side, when abnormal load, abnormal forwarding entries, or deteriorating path performance are detected on a specific UPF node, the system prioritizes rerouting the user plane traffic path. This operation updates PFCP rules to direct user traffic to alternative, lower-risk, lower-load UPF paths, achieving traffic bypass and load balancing.

[0100] In addition, to address QoS failures for specific services, the system can perform a QoS identification switch operation. This operation updates the 5QI (QoS Flow Identifier) ​​or QFI, assigning abnormal service flows to a higher-level service category to ensure their transmission latency, bandwidth, and packet loss rate requirements, thereby improving overall service level matching.

[0101] When UPF paths experience frequent anomalies, persistent entry accumulation, or rule aging, the system can also perform forwarding table management operations. These operations include shortening the lifecycle of abnormal path entries, accelerating the recycling of invalid path entries, or increasing the update frequency of high-priority service entries. This ensures the timeliness and efficiency of the UPF forwarding table, preventing the spread of anomalies and wasting resources.

[0102] The system automatically selects these control actions based on risk prediction results, knowledge graph recommendation strategies, and reinforcement learning evaluation results, and combines them as needed to form multi-action sequences, enabling joint control of complex scenarios. For example, when a cell experiences both scheduling congestion and path anomalies, the system can jointly execute scheduling priority adjustments and UPF path rerouting operations, achieving cross-domain, collaborative, adaptive network optimization.

[0103] In summary, the target control action provided by this embodiment has the characteristics of fine operation granularity, high response accuracy, and strong cross-domain linkage. It can realize closed-loop network self-optimization from predictive analysis to fine control, and is particularly suitable for complex communication network environments with multiple services concurrently and dynamically evolving structures.

[0104] Second, as Figure 4 As shown, the present invention further proposes a multimodal data source probe status monitoring and fault transfer system, which is used in any of the methods described in the first aspect, comprising:

[0105] The collection unit 21 is configured to deploy data collection probes at the radio access network nodes and the core network UPF nodes respectively to collect network data of the radio access network nodes and the core network UPF network data, wherein the network data includes network traffic, delay, packet loss, jitter, application connectivity and application network quality;

[0106] A preprocessing unit 22 is configured to perform a preprocessing operation on the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data;

[0107] A monitoring unit 23 is configured to input the pre-processed wireless access network node network data and the pre-processed core network UPF network data into an AI monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module, and a risk prediction module. The multi-source feature encoding module is configured to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space, respectively. The spatiotemporal collaborative modeling module is configured to construct a dynamic graph and extract the spatiotemporal dependency of the dynamic graph based on a graph neural network. The risk prediction module uses a multi-layer Transformer structure to model the historical state sequence of nodes in the dynamic graph to output the abnormal risk prediction information.

[0108] An adjustment unit 24 is configured to adjust parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information;

[0109] The optimization unit 25 is used to feed back the adjusted parameter configurations on the wireless side and the core network UPF side to the service perception module to achieve failover.

[0110] For example, the collection unit is deployed at radio access network nodes and core network UPF nodes, and uses data collection probes to collect network operation data in real time. This data includes, but is not limited to, key indicators such as network traffic, latency, packet loss, jitter, application connectivity, and application-layer network quality, comprehensively covering multiple dimensions of network performance.

[0111] The preprocessing unit performs preprocessing operations such as cleaning, denoising, and format unification on the above-collected original network data to obtain structured and standardized preprocessed data, which correspond to the wireless access network nodes and the core network UPF nodes respectively.

[0112] The monitoring unit is responsible for inputting preprocessed data into the AI ​​monitoring model to achieve intelligent prediction of abnormal risks. This AI model consists of three parts: a multi-source feature encoding module, which maps heterogeneous data from the wireless side and the core network side into the same feature space, eliminating representation bias caused by source differences; a spatiotemporal collaborative modeling module, which constructs a dynamic graph based on a graph neural network and extracts the spatial and temporal associations and evolution patterns of each node in the network topology; and a risk prediction module, which uses a multi-layer Transformer structure to model the historical state sequence of each node in the dynamic graph, thereby accurately predicting future abnormal risks.

[0113] The adjustment unit dynamically adjusts the relevant network parameters on the wireless side and the core network UPF side based on the abnormal risk prediction information output by the AI ​​model, so as to avoid potential failure risks in advance or mitigate their impact.

[0114] The optimization unit feeds back the adjusted parameter configuration results to the business perception module, enabling rapid reconfiguration at the network level and intelligent switching of business paths, thereby completing failover operations and ensuring the continuity and stability of network services.

[0115] In summary, the system proposed in this invention can achieve anomaly monitoring and rapid adaptive optimization of key nodes in the communication network through end-to-end data collection, AI-driven risk prediction and closed-loop fault response mechanism, significantly improving the system's intelligent operation and maintenance capabilities and business stability.

[0116] In a feasible implementation, the multi-source feature encoding module includes a sliding window encoding unit, a heterogeneous domain fusion unit, a fusion mapping unit, and a connection unit;

[0117] The sliding window encoding unit is used to perform sliding window segmentation on the pre-processed wireless access network node network data and the pre-processed core network UPF network data according to different time granularities to extract short-term mutation features and long-term trend features;

[0118] The above-mentioned heterogeneous domain fusion unit is used to calculate the feature attention weights of the wireless side data and the core network UPF side data respectively, and perform intra-domain feature weighting based on the above-mentioned attention weights;

[0119] A fusion mapping unit is used to map the weighted features into a unified semantic vector space through a deep neural network structure with shared weights, so as to achieve semantic alignment of data from different network domains;

[0120] The connection unit is used to introduce a residual connection structure in the mapping process and combine it with the channel attention mechanism to enhance the stability of feature propagation.

[0121] For example, the sliding window encoding unit divides the data on the wireless side and the core network side into sliding windows at different time granularities. By setting multiple time window lengths, this unit can extract both short-term mutation characteristics (such as burst traffic and jitter surges) and long-term trend changes (such as a gradual increase in packet loss rate and a continuous decrease in application connectivity). After segmentation, the data within the window forms characteristic segments of local time periods, laying the foundation for modeling the temporal evolution process.

[0122] Next, the heterogeneous domain fusion unit fuses data from different network domains. This unit calculates feature attention weights for both radio access network node data and core network UPF node data. Using a self-attention mechanism, it highlights key feature dimensions and strengthens focus on important metrics such as RTT fluctuations and changes in forwarding table hit rates. The weighted features better reflect the importance distribution of features within the domain, providing the prerequisite for unified mapping.

[0123] The fusion mapping unit then feeds these weighted features into a shared-weight deep neural network. This neural network maps the feature vectors from the wireless and core networks into a unified semantic vector space, eliminating representation differences between different network domains. This allows subsequent graph construction to perform topology modeling and information dissemination on the same embedding dimension. This ensures cross-domain consistency in data representation, improving the generalization and learning performance of the entire model.

[0124] Finally, the connection unit introduces a residual connection structure during feature propagation, combined with a channel attention mechanism to enhance the overall model's ability to retain deep features and identify important channels. The residual structure alleviates the vanishing gradient problem in deep networks, making model training more stable. The channel attention mechanism, through a dynamic weighting mechanism, further improves the model's ability to express important feature channels, thereby maintaining feature integrity while enhancing the model's sensitivity to abnormal behavior.

[0125] This implementation combines sliding windows, attention weighting, shared mapping, and channel enhancement to achieve time-series modeling and deep fusion of network data from both the wireless and core network UPF sides, building a unified multi-source feature representation and laying a solid data foundation for subsequent dynamic graph modeling and risk prediction. This approach balances the modeling of both short-term anomalies and long-term trends, offering excellent expressiveness, scalability, and adaptability.

[0126] In a feasible implementation, the spatiotemporal collaborative modeling module includes a dynamic graph subunit, a graph neural network subunit, a graph evolution representation subunit, and a structure attribution subunit;

[0127] The dynamic graph subunit is used to map the feature vectors output by the multi-source feature encoding module into a graph structure of multiple consecutive time steps, wherein the graph structure of each time step includes nodes, edges, and dynamic weights, and the weights of the edges are dynamically generated based on traffic coordination and delay correlation;

[0128] The above-mentioned graph neural network sub-unit is used to extract the graph structure representation of the nodes in the graph at each time step based on the graph attention mechanism. The above-mentioned graph neural network sub-unit includes a time-aware embedding mechanism, an edge attribute embedding mechanism, and a skip connection mechanism;

[0129] The graph evolution representation subunit is used to perform temporal modeling on the node representation of the continuous graph structure to obtain the dynamic embedding features of the nodes during the historical evolution process;

[0130] The above-mentioned structural attribution unit is used to interpret the node status based on the attention mechanism in the graph neural network, and output the impact path and key adjacent nodes of high-risk nodes.

[0131] Exemplarily, the dynamic graph sub-unit is responsible for organizing the node feature vectors output by the multi-source feature encoding module into a graph structure of multiple consecutive time steps. In each time step, the graph structure consists of nodes, edges and edge weights, where nodes represent network entities (such as gNodeB nodes, UPF nodes, etc.), and the existence of edges indicates that there are certain business connections or influences between network entities. Specifically, the weight of the edge is dynamically generated based on the traffic coordination between the two nodes (such as the trend of concurrent connection changes) and the delay correlation (such as the consistency of path RTT fluctuations) to characterize the intensity of interaction between entities in the network at this time step. The sub-unit finally outputs a series of weighted graph structures as the basis for subsequent modeling.

[0132] Subsequently, the graph neural network subunit extracts structural information from the graph structure at each time step and generates a graph embedding representation of the nodes. In order to improve the model's ability to model temporal evolution and network topology changes, this subunit integrates a number of key mechanisms. First, a time-aware embedding mechanism is introduced to fuse the timestamp information of the current time step with the node features, so that the node embedding not only contains structural information but also reflects the time change characteristics; secondly, an edge attribute embedding mechanism is adopted to encode information such as traffic coordination and delay correlation contained in the edge weight into the graph convolution operation process to achieve more accurate relationship modeling; finally, a jump connection mechanism is used to connect different graph convolution layers to avoid the over-smoothing problem that occurs in deep graph neural networks and enhance the stability and nonlinear expression ability of the model.

[0133] After the graph neural network extracts node representations at each time step, the graph evolution representation subunit performs temporal modeling on the node embedding sequences across all time steps, thereby capturing the dynamic feature trajectory of each node during its historical evolution. This subunit can employ recurrent neural networks (such as GRU) or Transformer-based temporal modeling structures to model the representation of the same node in a continuous graph structure, capturing its behavioral evolution patterns and abnormal development trends. Ultimately, it outputs a dynamic embedding representation with temporal dependencies for subsequent risk prediction or visualization analysis.

[0134] Finally, the structural attribution unit interprets node states based on attention weights within the graph neural network. By analyzing the attention distribution of high-risk nodes, the system can identify the impact paths and key adjacent nodes behind their high risk scores, specifically the "upstream nodes" or critical propagation links that caused the anomaly. This module outputs not only the risk scores of high-risk nodes but also their structural attribution paths, which can be used as a reference for subsequent parameter adjustments or network policy development modules, thereby improving the accuracy and interpretability of troubleshooting.

[0135] This implementation, through the synergistic integration of graph modeling and time series modeling, achieves high-fidelity modeling of the dynamic relationships between network entities and a detailed depiction of the evolution of node behavior. Furthermore, through a structural attribution mechanism, it outputs the abnormal impact paths of high-risk nodes, significantly enhancing insight into the operational status of complex networks and enabling intelligent intervention. This module is applicable to a variety of network anomaly analysis scenarios and offers strong scalability and engineering adaptability.

[0136] In a feasible implementation, the risk prediction module includes a historical state sequence subunit, a multi-layer Transformer subunit, a multi-scale risk perception subunit, and a risk causal explanation subunit;

[0137] The above-mentioned historical state sequence subunit is used to extract the node historical state vector from the dynamic graph output by the spatiotemporal collaborative modeling module and construct the input sequence in combination with the time position encoding;

[0138] The multi-layer Transformer subunit includes multiple encoding layers, each of which contains a multi-head attention layer, a feedforward network, a residual connection, and a normalization structure to capture temporal dependencies and abnormal evolution patterns in node state sequences.

[0139] The multi-scale risk perception subunit is used to generate abnormal trend prediction results, risk type classification labels, and confidence estimation information based on the Transformer output.

[0140] The risk causal explanation subunit is used to identify key time segments and / or indicators that lead to high risk scores based on the attention weights, and output risk explanation information.

[0141] For example, the historical state sequence subunit extracts the embedded feature vectors of the target node at multiple time steps from the dynamic graph output by the spatiotemporal collaborative modeling module. To preserve temporal order information, this subunit introduces a temporal position encoding mechanism. By encoding the timestamp corresponding to each time step, this subunit injects temporal information into the state representation, constructing a time-aware input sequence. This sequence, as input to the risk prediction model, reflects the dynamic changes of the node during its historical evolution and is a key foundation for subsequently capturing temporal patterns.

[0142] The input sequence then enters a multi-layer Transformer subunit for deep modeling. This subunit consists of multiple stacked Transformer encoding layers, each of which incorporates a multi-head attention mechanism, a feedforward network structure, residual connections, and layer normalization. The multi-head attention mechanism can globally model the dependencies between time segments in the sequence, capturing the evolutionary trends of node states and potential anomaly precursors; residual connections and normalization operations improve the network's training stability and nonlinear expression capabilities. By stacking multiple layers, the model can learn complex temporal evolution patterns and provide high-quality representations for downstream risk prediction tasks.

[0143] After obtaining the Transformer output, the multi-scale risk perception subunit further interprets it and produces multi-task outputs. This subunit uses a multi-channel structure to achieve multi-dimensional risk expression capabilities, assessing risks from different scales and perspectives. On the one hand, it generates abnormal trend prediction results to quantify the changes in the degree of abnormal risk over several future time steps. On the other hand, it classifies the risks into types, such as those caused by latency fluctuations, abnormal packet loss rates, or path congestion, and outputs corresponding labels. In addition, this unit also generates confidence estimates for each prediction result to help determine the credibility of the prediction conclusion and provide a risk reliability reference for subsequent control modules.

[0144] Finally, the risk causal explanation subunit performs interpretable analysis of anomaly prediction results based on the attention weights within the Transformer model. This subunit analyzes the time segments and metric dimensions that play a dominant role in generating anomaly scores, identifying which historical moments or performance indicators contribute most to high-risk scores. For example, if a region with a sudden increase in packet loss rate over a period of time is strongly highlighted by the attention weights, the system will label it as a "key anomaly driver" and output it as structured explanatory information for reference by the strategy selection module and operations personnel.

[0145] The risk prediction module provided by this implementation not only possesses sophisticated time series modeling and comprehensive risk perception capabilities, but also possesses strong interpretability and decision-making support capabilities. By introducing a multi-layer Transformer architecture and a multi-task risk output mechanism, the system enables early warning and hierarchical assessment of network status. Furthermore, through causal analysis driven by an attention mechanism, it improves the transparency and controllability of prediction results, significantly enhancing the application value of intelligent operation and maintenance systems in complex network environments.

[0146] The present invention also provides a computer program product, comprising a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer program or computer-executable instructions from the computer-readable storage medium and executes the computer program or computer-executable instructions, causing the electronic device to perform any step of the multimodal data source probe status monitoring and fault transfer method described above.

[0147] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multimodal data source probe status monitoring and fault transfer method, characterized in that: include: Data collection probes are deployed at the radio access network nodes and core network UPF nodes respectively to collect network data of radio access network nodes and core network UPF network data; Performing a preprocessing operation on the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data; The pre-processed wireless access network node network data and the pre-processed core network UPF network data are input into the AI ​​monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module and a risk prediction module, the multi-source feature encoding module is used to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space respectively, the spatiotemporal collaborative modeling module is used to construct a dynamic graph and extract the spatiotemporal dependency of the dynamic graph based on a graph neural network; the risk prediction module is used to model the historical state sequence of the nodes in the dynamic graph to output the abnormal risk prediction information; Adjusting parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information; The adjusted parameter configurations on the wireless side and the core network UPF side are fed back to the service perception module to achieve failover.

2. The multimodal data source probe status monitoring and fault transfer method according to claim 1, characterized in that: The network data includes network traffic, delay, packet loss, jitter, application connectivity and application network quality.

3. The multimodal data source probe status monitoring and fault transfer method according to claim 1, characterized in that: The risk prediction module adopts a multi-layer Transformer structure.

4. The multimodal data source probe status monitoring and fault transfer method according to claim 1, characterized in that: The adjusting the parameter configuration of the wireless side and the core network UPF side according to the abnormal risk prediction information includes: Constructing a policy knowledge graph, wherein the policy knowledge graph is used to store associations between anomaly types and network parameter items and mappings between network parameter items and control actions; Based on the abnormal risk prediction information, retrieving a set of candidate parameter adjustment strategies that matches the current abnormality type from the strategy knowledge graph; Obtain current network state information, historical parameter adjustment records, and network performance feedback information to construct a state representation for reinforcement learning; Based on the state representation, performing an optimal strategy selection operation based on a deep reinforcement learning algorithm in the set of candidate parameter adjustment strategies to determine a target control action for adjusting the wireless side and / or the core network UPF side; Execute corresponding parameter adjustment operations according to the target control action to achieve optimization of network performance indicators, wherein the target control action aims to maximize network performance improvement, and the network performance improvement is evaluated by a reward function, and the reward function is calculated based on risk score changes, network delay changes, packet loss rate changes and QoS feedback information.

5. The multimodal data source probe status monitoring and fault transfer method according to claim 4, characterized in that: The performing, based on the state representation, an optimal strategy selection operation based on a deep reinforcement learning algorithm in the candidate parameter adjustment strategy set to determine a target control action for adjusting the wireless side and / or the core network UPF side, including: Passing the state representation as input to a policy selection model built based on deep reinforcement learning, wherein the policy selection model is used to evaluate actions and select policies from a set of candidate parameter adjustment policies. The policy selection model is trained using a deep reinforcement learning algorithm to minimize a policy loss function and maximize a reward function based on network performance feedback. The reward function is evaluated based on changes in risk score, network latency, packet loss rate, and QoS indicators. The strategy score of each candidate action is calculated based on the current state representation within the candidate parameter adjustment strategy set, and the target control action with the highest expected long-term benefit is selected, wherein the target control action is used as a network control instruction to adjust the parameter configuration of the corresponding wireless side and / or core network UPF side.

6. The multimodal data source probe status monitoring and fault transfer method according to claim 4 or 5, characterized in that: The target control action includes one or more of dynamically adjusting the cell scheduling priority on the wireless side, performing transmission power control operations on the wireless side, performing switching threshold optimization operations on the wireless side, rerouting the user plane traffic path on the core network UPF side, performing QoS identification switching operations on the core network UPF side, and performing routing table management operations on the core network UPF side.

7. A multimodal data source probe status monitoring and fault transfer system for implementing the method described in any one of claims 1 to 6, characterized in that: include: A collection unit is configured to deploy data collection probes at radio access network nodes and core network UPF nodes respectively to collect radio access network node network data and core network UPF network data, wherein the network data includes network traffic, delay, packet loss, jitter, application connectivity, and application network quality; A preprocessing unit, configured to perform a preprocessing operation on the radio access network node network data and the core network UPF network data to obtain preprocessed radio access network node network data and preprocessed core network UPF network data; A monitoring unit, configured to input the pre-processed wireless access network node network data and the pre-processed core network UPF network data into an AI monitoring model to obtain abnormal risk prediction information, wherein the AI ​​monitoring model includes a multi-source feature encoding module, a spatiotemporal collaborative modeling module, and a risk prediction module; the multi-source feature encoding module is configured to map the pre-processed wireless access network node network data and the pre-processed core network UPF network data into a unified feature space, respectively; the spatiotemporal collaborative modeling module is configured to construct a dynamic graph and extract the spatiotemporal dependency of the dynamic graph based on a graph neural network; the risk prediction module uses a multi-layer Transformer structure to model the historical state sequence of nodes in the dynamic graph to output the abnormal risk prediction information; An adjustment unit, configured to adjust parameter configurations on the wireless side and the core network UPF side according to the abnormal risk prediction information; The optimization unit is used to feed back the adjusted parameter configuration on the wireless side and the core network UPF side to the service perception module to achieve failover.

8. The multimodal data source probe status monitoring and fault transfer system according to claim 7, characterized in that: The multi-source feature encoding module includes a sliding window encoding unit, a heterogeneous domain fusion unit, a fusion mapping unit and a connection unit; The sliding window encoding unit is used to perform sliding window segmentation on the pre-processed radio access network node network data and the pre-processed core network UPF network data according to different time granularities to extract short-cycle mutation features and long-cycle trend features; The heterogeneous domain fusion unit is used to calculate the feature attention weights of the wireless side data and the core network UPF side data respectively, and perform intra-domain feature weighting based on the attention weights; A fusion mapping unit is used to map the weighted features into a unified semantic vector space through a deep neural network structure with shared weights, so as to achieve semantic alignment of data from different network domains; The connection unit is used to introduce a residual connection structure in the mapping process and combine it with the channel attention mechanism to enhance the stability of feature propagation.

9. The multimodal data source probe status monitoring and fault transfer system according to claim 7, characterized in that: The spatiotemporal collaborative modeling module includes a dynamic graph subunit, a graph neural network subunit, a graph evolution representation subunit and a structure attribution subunit; The dynamic graph subunit is used to map the feature vectors output by the multi-source feature encoding module into a graph structure of multiple consecutive time steps, wherein the graph structure of each time step includes nodes, edges and dynamic weights, and the weights of the edges are dynamically generated based on traffic coordination and delay correlation; The graph neural network subunit is used to extract the graph structure representation of the nodes in the graph at each time step based on the graph attention mechanism. The graph neural network subunit includes a time-aware embedding mechanism, an edge attribute embedding mechanism, and a skip connection mechanism; The graph evolution representation subunit is used to perform temporal modeling on the node representation of the continuous graph structure to obtain the dynamic embedding features of the nodes in the historical evolution process; The structural attribution unit is used to interpret the node status based on the attention mechanism in the graph neural network, and output the impact path and key adjacent nodes of the high-risk node.

10. The multimodal data source probe status monitoring and fault transfer system according to claim 1, characterized in that: The risk prediction module includes a historical state sequence subunit, a multi-layer Transformer subunit, a multi-scale risk perception subunit and a risk causal explanation subunit; The historical state sequence subunit is used to extract the node historical state vector from the dynamic graph output by the spatiotemporal collaborative modeling module and construct an input sequence in combination with the time position coding; The multi-layer Transformer subunit includes multiple encoding layers, each of which contains a multi-head attention layer, a feedforward network, a residual connection, and a normalization structure to capture the temporal dependencies and abnormal evolution patterns in the node state sequence; The multi-scale risk perception subunit is used to generate abnormal trend prediction results, risk type classification labels and confidence estimation information based on Transformer output; The risk causal explanation subunit is used to identify key time segments and / or indicators that lead to high risk scores based on the attention weights, and output risk explanation information.

Images