A cloud-edge collaborative power network intelligent fault diagnosis positioning system and method

Abstract

The application provides a cloud-edge collaborative power network intelligent fault diagnosis positioning system and method, comprising: acquiring a data set in a complete collaborative transmission path, adopting a data integration method to fuse refined monitoring data set and historical network view data in a cloud control center, and obtaining an updated network view representation; according to an early symptom position, acquiring edge node related data with limited field of view, adopting a resource allocation mechanism to allocate cloud computing resources to the edge node, and obtaining enhanced computing capability configuration; based on the enhanced computing capability configuration, injecting compensation resources into a fault diagnosis process, adopting a time delay optimization method to adjust a diagnosis priority related to time delay, determining a rapid fault positioning sequence; through the rapid fault positioning sequence, integrating unified sensing information into the network view representation, if the positioning sequence covers multiple regional abnormalities, triggering a network repair protocol, and obtaining repaired monitoring information flow.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a cloud-edge collaborative intelligent fault diagnosis and location system and method for power networks. Background Technology

[0002] As a critical national infrastructure, the safe and stable operation of the power grid is directly related to the normal operation of the social economy and the protection of people's lives. With the continuous expansion of the power grid, the large-scale integration of distributed renewable energy sources, and the surge in various intelligent terminal devices, the operating status of the power grid is characterized by high complexity and dynamic changes. Any local anomaly can quickly evolve into a large-scale fault. Therefore, achieving rapid and accurate fault diagnosis and location is particularly urgent.

[0003] Currently, most power fault diagnosis methods rely on centralized cloud analysis or decentralized edge independent judgment. Both methods have revealed significant shortcomings in practical applications. While centralized solutions can gather global information, the data transmission from edge devices to the cloud has a significant latency, resulting in slow response to transient faults and even missing the optimal handling window. Decentralized solutions, although quick to respond, are limited by the field of view and computing power of edge nodes, making it difficult to accurately determine complex fault correlations involving multiple nodes and regions, which can easily lead to false alarms or missed alarms.

[0004] A deeper problem lies in the lack of smooth and consistent flow of monitoring information between the cloud and the edge. If critical information such as edge device operating parameters, communication link status, computing resource usage, and diagnostic task execution cannot be reported promptly and completely and form a consistent network-wide view in the cloud, the cloud will struggle to construct a true panoramic view of the power network's operational status. Conversely, the cloud cannot effectively monitor and coordinate edge nodes, making it difficult to detect edge-side anomalies in a timely manner and distorting overall network risk assessments. For example, when multiple edge nodes in a region simultaneously experience communication link fluctuations and computing resource shortages, if the monitoring data received by the cloud is excessively compressed or loses key details, it will be unable to promptly identify whether there is a causal relationship between these anomalies, thus missing early signs of cascading failures.

[0005] How to achieve effective collaboration of monitoring information between the cloud and the edge while ensuring the real-time performance and integrity of data, and form a unified situational awareness capability covering the entire network, has become a key issue that urgently needs to be addressed in the current cloud-edge collaborative power network intelligent fault diagnosis and location. Summary of the Invention

[0006] This invention provides a cloud-edge collaborative intelligent fault diagnosis and location system and method for power networks, mainly comprising: Monitoring information is obtained from edge nodes, including the computing load parameters and network link connection status of the edge nodes. Data processing methods are used to aggregate the monitoring information to remove redundant parts, resulting in a refined monitoring dataset. Based on the refined monitoring dataset, the dataset is uploaded to the cloud control center using a real-time transmission mechanism. If network link status fluctuations exceed a preset threshold during transmission, the remaining data is transmitted via a backup communication link to determine the complete collaborative transmission path. The dataset in the complete collaborative transmission path is obtained, and the refined monitoring dataset is fused with historical network view data using a data integration method in the cloud control center to obtain an updated network view representation. For the updated network view representation, an anomaly analysis model is used to detect dynamic change features. If the dynamic change features show an abnormal evolution trend, potential cascading failure areas are marked, and the location of early signs is determined. Based on the location of the early signs, relevant data of edge nodes with limited visibility are obtained, and cloud computing resources are allocated to the edge nodes using a resource allocation mechanism to obtain enhanced computing power configuration; Based on the enhanced computing power configuration, compensation resources are injected into the fault diagnosis process, and a time delay optimization method is used to adjust the time delay-related diagnostic priorities to determine a rapid fault location sequence. The unified perception information is integrated into the network view representation through the rapid fault location sequence. If the location sequence covers multiple abnormal areas, the network repair protocol is triggered to obtain the repaired monitoring information stream. The repaired surveillance information stream is obtained, and a verification method is used to check the matching degree between the surveillance information stream and the real-time uploaded data. If the matching degree is lower than a preset threshold, the mismatched part is adjusted through an iterative processing method to determine the final unified situational awareness model.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a network fault diagnosis and repair method for edge computing and cloud collaboration. Addressing issues such as redundant monitoring data at edge nodes, transmission link fluctuations, early fault symptom identification, and uneven resource allocation in business scenarios, it constructs a logically interconnected solution through data aggregation, real-time transmission, anomaly analysis, and resource optimization. First, the invention refines and uploads edge node monitoring information in real time to ensure efficient data transmission and switches to backup links during network fluctuations. Then, it merges historical and real-time data in the cloud to generate an updated network view and uses an anomaly analysis model to accurately locate potential fault areas. For edge nodes with limited visibility, the invention dynamically allocates cloud resources to enhance computing power and optimize fault diagnosis latency. Finally, through rapid location sequences and network repair protocols, it achieves accurate repair of fault areas and information flow verification, ensuring the consistency of the situational awareness model. This invention significantly improves the early warning and rapid response capabilities of network faults, ensuring the efficiency and stability of edge-cloud collaboration. Attached Figure Description

[0008] Figure 1 This is a flowchart of a cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to the present invention.

[0009] Figure 2 This is a schematic diagram of a cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to the present invention.

[0010] Figure 3 This is another schematic diagram of a cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] like Figures 1-3 This embodiment of a cloud-edge collaborative power network intelligent fault diagnosis and location system and method may specifically include: S101. Obtain monitoring information from edge nodes. The monitoring information includes the computing load parameters and network link connection status of the edge nodes. Use data processing methods to aggregate the monitoring information to remove redundant parts and obtain a refined monitoring dataset.

[0013] Step 1: Acquire monitoring data from edge nodes. This data includes calculated load parameters and network link connection status. The data is initially categorized and organized using a pre-set acquisition tool to obtain a categorized monitoring data set. Step 2: For the categorized monitoring data set, aggregate the data using data processing methods, focusing on removing redundant information and duplicate records. By comparing against pre-set redundancy judgment rules, a deduplicated subset of monitoring data is obtained. Step 3: Based on the deduplicated subset, apply cluster analysis to group the data. If the load parameters of a group exceed a pre-set threshold, it is marked as a high-load group, resulting in grouped and marked data units. Step 4: Analyze the correlation between high-load groups and network link connection status using the grouped and marked data units. If the connection status of a high-load group shows an anomaly, it is classified as a priority processing object, determining the priority processing data units. Step 5: Obtain the priority processing data units. For the load parameters and connection status data within these units, execute status evaluation logic. If the evaluation result indicates that the node status is unstable, trigger resource scheduling instructions and generate a scheduling instruction set. Step Six: Based on the scheduling instruction set, automatically adjust the resource allocation scheme of the edge nodes, rebalance the computing load according to preset allocation rules, and obtain the adjusted resource configuration data. Step Seven: Update the operating parameters of the edge nodes using the adjusted resource configuration data, continuously track changes in node status using real-time monitoring tools, determine whether the resource configuration has reached a balanced state, and generate the final operating status data.

[0014] Obtaining monitoring data from edge nodes is a fundamental step in resource scheduling.

[0015] For example, the data acquisition tool pulls CPU utilization, memory usage percentage, and network link status indicators from each edge node every thirty seconds. These status indicators include four categories: normal, high packet loss rate, sudden increase in latency, and disconnection. Through preliminary classification and organization, the computational load parameters are grouped into one group, and the network link connection status is grouped into another group, forming a clear monitoring data set that facilitates subsequent targeted processing.

[0016] In one possible implementation, when performing aggregation operations on the categorized monitoring data set, the focus is on removing redundant information.

[0017] Preferably, the preset redundancy determination rule is that if the load parameter change of the same node within three consecutive minutes is less than 5%, it is considered a duplicate record. For example, if an edge node reports a CPU load of 62% to 64% five times consecutively, only the first record is retained, and the subsequent four are discarded. After processing with the comparison rule, the data volume can be reduced by about 30% to 50%, significantly reducing the burden of transmission and analysis.

[0018] Specifically, cluster analysis is used to group the deduplicated subsets of monitoring data.

[0019] In one embodiment, density clustering based on load parameters is used to group nodes with CPU utilization higher than 85% and memory usage higher than 80% into a single cluster. If the load parameters of most nodes in a group exceed a preset threshold of 90%, the entire group is marked as a high-load group.

[0020] It should be noted that this grouping method can quickly pinpoint areas of concentrated resource pressure.

[0021] For example, correlation analysis between high-load groups and network link connection status is a key judgment point.

[0022] In one possible implementation, if the link status of a node in a high-load group shows a packet loss rate exceeding 3% or a latency exceeding 200 milliseconds, it is determined to be an abnormal connection and directly classified as a priority processing object.

[0023] Preferably, data units that are processed first have higher priority than units that are simply under high load but have normal links, because abnormal connections often lead to a faster increase in task failure rate. When performing status evaluation logic for data units that are processed first...

[0024] Understandably, the evaluation considers both load parameters and connection status. If a node's CPU usage consistently exceeds 92% and the link experiences intermittent disconnections, the evaluation result indicates an unstable state, triggering resource scheduling instructions. The generated scheduling instruction set includes specific instructions such as migrating tasks to nearby low-load nodes, temporarily increasing the number of container instances, and rate limiting non-critical traffic. The resource allocation scheme is automatically adjusted based on the scheduling instruction set.

[0025] For example, for nodes with high load and abnormal links, 60% of the computing tasks are preferentially migrated to adjacent healthy nodes with latency below 50 milliseconds, while the weight of the original nodes in accepting new tasks is reduced. After rebalancing through preset allocation rules, the overall load variance of the cluster can be reduced from the original 22% to about 9%. After updating the operating parameters of the edge nodes with the adjusted resource configuration data, real-time monitoring tools continuously track changes.

[0026] For example, if the latest load and link metrics are collected every minute, and the CPU load remains stable below 75% and the packet loss rate is below 0.5% for five consecutive minutes, then the resource configuration is determined to be balanced, and the final running status data is considered stable.

[0027] Understandably, the above process, through progressive filtering, labeling, evaluation, and adjustment, effectively avoids resource waste and task interruption, improves the stability and task completion rate of the edge computing system, and significantly enhances response speed and self-healing capabilities, especially in scenarios with sudden traffic surges or occasional hardware failures.

[0028] S102. Based on the refined monitoring dataset, the dataset is uploaded to the cloud control center using a real-time transmission mechanism. If network link status fluctuations exceed a preset threshold during transmission, the remaining data is transmitted via a backup communication link, thus determining the complete collaborative transmission path.

[0029] A refined monitoring dataset is acquired, and data is transmitted to the cloud control center via a real-time transmission mechanism. During data transmission, the network link status is continuously monitored to obtain current link status parameters. It is determined whether network link status fluctuations exceed a preset threshold; if so, a link switching process is triggered; otherwise, transmission continues via the existing link. When a link switch is triggered, the amount of data already transmitted is obtained to determine the remaining untransmitted data. A backup communication link takes over the transmission task, sending the remaining untransmitted data to the cloud control center. Based on the connection between the transmitted and remaining data, the upload order and segment identifiers of the two links are recorded. Using the segment identifiers and order information of the data uploaded via the two links, the complete dataset is reconstructed to determine the final structure of the collaborative transmission path.

[0030] For example, in the field of edge computing, the implementation of real-time transmission mechanisms for the transmission and processing of refined monitoring datasets can be meticulously designed from multiple perspectives to ensure efficient data delivery to the cloud control center. For network link status monitoring during data transmission, lightweight monitoring tools can be deployed to continuously collect parameters such as link latency and packet loss rate. Assuming the link latency of an edge node is typically around 20 milliseconds, if a sudden increase to 50 milliseconds is detected, exceeding the preset threshold of 30 milliseconds, the system will automatically identify it as an abnormal fluctuation and trigger subsequent processes. This monitoring method can promptly capture potential problems, ensuring the stability of data transmission.

[0031] For example, a dynamic switching strategy can be designed during the implementation of the link switching process. When the primary link experiences fluctuations, the system quickly calculates the amount of data already transmitted. Assuming the total data is 1000 megabytes and 600 megabytes have been transmitted, the remaining 400 megabytes will be transmitted via the backup link. The backup link can be pre-configured as a low-load auxiliary network path to ensure uninterrupted transmission after the switch. This segmented transmission design allows for flexible responses to changes in the network environment without affecting overall data integrity.

[0032] For example, data segment identification and sequence recording can be achieved by adding a unique timestamp and sequence number to each data segment. Assuming the first 600 megabytes of data transmitted via the main link are marked as Sequence 1, and the remaining 400 megabytes transmitted via the backup link are marked as Sequence 2, the system will reassemble the data in the cloud based on these identifiers. This method ensures that data can be accurately pieced together after transmission across different links, avoiding data out-of-order issues or loss. Furthermore, this segment identification recording method facilitates subsequent tracing of the transmission path, improving the transparency of data management.

[0033] For example, during the process of reconstructing a complete dataset, a verification mechanism can be introduced to ensure data integrity and consistency. After receiving segmented data, the cloud control center checks the data packets for completeness according to preset verification rules. If a data segment is found to be missing, the system automatically requests the edge nodes to retransmit that portion. This verification mechanism effectively avoids data incompleteness issues caused by transmission interruptions, providing a reliable foundation for subsequent data analysis.

[0034] For example, to determine the final structure of the cooperative transmission path, an adaptive path optimization scheme can be designed. The system analyzes the performance of the primary and backup links based on historical transmission records and dynamically adjusts the future transmission path allocation ratio. Assuming the primary link has shown high stability in the past 10 transmissions, more data can be allocated to it. This adaptive adjustment method can optimize resource utilization efficiency based on the actual network environment.

[0035] For example, a real-time feedback mechanism can be set up for the interaction between edge nodes and the cloud control center throughout the entire transmission process. After switching links or completing data transmission, edge nodes send status updates to the cloud so that the cloud can promptly monitor the transmission progress. This feedback mechanism helps improve the overall system response speed, ensuring that data transmission tasks can still be completed efficiently even in complex network environments.

[0036] For example, from the perspective of expansion solutions, a multi-link parallel transmission strategy can be introduced. In addition to the main and backup links, an auxiliary link can be configured to share some of the data transmission tasks. This approach is suitable for scenarios with large data volumes, further distributing transmission pressure and improving overall transmission efficiency. Through the collaborative work of multiple links, the system exhibits stronger robustness in the face of sudden network fluctuations.

[0037] S103. Obtain the dataset in the complete collaborative transmission path, and use the data integration method in the cloud control center to merge the refined monitoring dataset with the historical network view data to obtain the updated network view representation.

[0038] Acquire refined monitoring datasets and historical network view data. Perform format unification processing on the refined monitoring datasets using data cleaning rules to obtain standardized monitoring data. Match the standardized monitoring data with the historical network view data using timestamp alignment to obtain time-aligned monitoring view pairs. Calculate the difference vector between the current state vector and the historical state vector of each network node based on the time-aligned monitoring view pairs, obtaining a state difference sequence. If the difference of a node in the state difference sequence exceeds a preset stable interval, the node is marked as an abnormally active node, obtaining a set of abnormally active nodes. Represent the current network topology structure containing the set of abnormally active nodes using graph embedding methods, obtaining the current network topology embedding vector. Concatenate the current network topology embedding vector with the topology embedding vector of the corresponding time period in the historical network view data to obtain the updated network view representation.

[0039] For example, in a drone swarm collaborative monitoring task, after obtaining a refined monitoring dataset, the raw monitoring data reported by each drone is first processed to unify the format through data cleaning rules.

[0040] Specifically, the time format, coordinate system unit, and data precision of the outputs from different UAV sensors are standardized and converted to obtain standardized monitoring data with consistent structure, thereby avoiding deviations in subsequent matching.

[0041] In one possible implementation, a timestamp alignment method is used to match standardized monitoring data with historical network view data.

[0042] Preferably, using millisecond-level timestamps as a benchmark, a one-to-one correspondence is established between monitoring records and historical view records falling within the same time window to obtain time-aligned monitoring view pairs. This alignment method ensures that the currently collected real-time information and historical trends are within the same time reference system.

[0043] Specifically, based on the time-aligned monitoring view pairs, the difference vector between the current state vector and the historical state vector of each network node is calculated, thus forming a state difference sequence. For example, if the historical packet loss rate of a relay node is stable in the range of 0.3% to 0.8%, but the current data shows a sudden increase in the packet loss rate to 4.2%, then the difference of this node exceeds the preset stable range and it is marked as an abnormally active node. This method can quickly filter out a set of abnormally active nodes, such as those containing 3 relay nodes and 1 edge sensing node.

[0044] It should be noted that the current network topology containing the set of abnormally active nodes is represented by a vector using a graph embedding method.

[0045] In one embodiment, node degree, adjacency, and anomaly markers are used as input features, and a 128-dimensional current network topology embedding vector is obtained by projection through a graph neural network. This vector encapsulates the anomaly features of both the local and global topology.

[0046] Understandably, the updated network view representation is obtained by concatenating the current network topology embedding vector with the topology embedding vector for the corresponding time period in the historical network view data. For example, if the current embedding vector is V_t and the historical corresponding segment embedding vector is V_{tk}, the concatenation results in a 256-dimensional vector. This updated representation incorporates anomalous evolution information, enabling a more comprehensive depiction of the network's dynamic change trends. The beneficial effect is that this representation method, based on time-aligned state difference detection combined with graph embedding, can identify abnormally active areas caused by electromagnetic interference, node overload, or malicious attacks in UAV swarm monitoring scenarios at an early stage. This provides accurate basis for subsequent path planning or resource reallocation, preventing the interruption of the entire collaborative monitoring link due to the abnormal spread of individual nodes, and improving the continuity and reliability of overall data acquisition.

[0047] For example, when the set of abnormally active nodes is small, high-risk nodes are prioritized for labeling and isolation; as the set grows, the updated network view representation can further support global rerouting decisions. This progressive processing logic, from local anomaly detection to global topology representation, enables the system to maintain high adaptability in complex and dynamic network environments.

[0048] S104. For the updated network view representation, an anomaly analysis model is used to detect dynamic change characteristics. If the dynamic change characteristics show an abnormal evolution trend, potential cascading failure areas are marked, and the location of early signs is determined.

[0049] Obtain the updated network view representation data. Process the network view representation data using an anomaly analysis model to obtain dynamic change characteristics. Perform trend analysis on the dynamic change characteristics to determine if there are any abnormal evolution trends. If the dynamic change characteristics show an abnormal evolution trend, perform a marking operation to obtain potential cascading failure regions. Based on the positional relationships of the potential cascading failure regions, use a graph traversal method to determine the locations of early symptoms. Perform feature extraction on the network nodes associated with the early symptom locations to obtain a set of symptom nodes. Through connectivity analysis between the symptom node set and the network view representation, obtain the cascading propagation path.

[0050] For example, in the process of processing data represented by a network view, anomaly analysis models can be used to deeply mine the data and extract dynamic change characteristics. The core of anomaly analysis models lies in identifying unusual fluctuations or patterns in the network view.

[0051] Specifically, the fluctuation range of a time series can be used to determine whether the activity of a network node deviates from the normal range within a certain period. Suppose that in a network monitoring scenario, the number of connection requests to a node surges from an average of 10 times per minute to 50 times per minute in a short period of time. This sudden change will be captured by the model as a dynamic change feature, and then used as the basis for subsequent trend analysis.

[0052] In one possible implementation, trend analysis of dynamic changes can be performed by comparing historical data to determine if there are any abnormal evolution trends. For example, if historical data shows that a node's connection request growth rate has remained stable at around 5% over the past week, but the growth rate suddenly reaches 30% in the current monitoring period, this can be identified as an abnormal evolution trend. In this case, the system will automatically mark the area where that node is located as a potential cascading failure zone. This marking operation helps to quickly locate the core area that may trigger a wider range of problems, providing precise guidance for subsequent handling.

[0053] For example, graph traversal methods can be used to determine the location of early signs of problems by analyzing the positional relationships of potential cascading failure areas. Suppose that in a network topology consisting of 100 nodes, the marked potential failure area contains 5 core nodes. Graph traversal methods can reveal that 2 of these nodes have abnormally high connection density with other areas; these 2 nodes can be considered as early sign locations. This method can effectively analyze the dependencies between nodes and quickly pinpoint the source of the problem.

[0054] In one possible implementation, when extracting features from network nodes associated with early symptom locations, key metrics such as node traffic load and latency can be considered. For example, if the average latency of a symptom node increases from the normal 20 milliseconds to 100 milliseconds, while its traffic load exceeds twice the normal value, then this node would be included in the symptom node set. This feature extraction method can help identify key nodes that may trigger a chain reaction, providing data support for subsequent analysis.

[0055] For example, cascading propagation paths can be identified through connectivity analysis of the symptom node set and the network view representation. Suppose that in the network view, three nodes in the symptom node set are connected to ten other nodes via high-frequency interaction paths, and the traffic load on these paths all shows an upward trend; then potential cascading propagation paths can be inferred. This analytical approach helps predict the direction and scope of fault propagation, thus providing important reference for network management.

[0056] In one possible implementation, the identification of chain propagation paths can be verified by combining historical network view data. If a similar propagation path has appeared in historical data and ultimately led to local network congestion, the current path will be prioritized to allow for early intervention. This analysis method, which incorporates historical data, improves prediction accuracy and reduces the risk of misjudgment. Through the above series of processes, abnormal features in the network view can be effectively identified, potential problem areas can be accurately located, and possible fault propagation paths can be predicted. This method has significant value in the field of network monitoring, providing network administrators with clear decision-making support and contributing to the stable operation of network systems.

[0057] S105. Based on the location of the early signs, obtain relevant data of edge nodes with limited visibility, and use a resource allocation mechanism to allocate cloud computing resources to the edge nodes to obtain enhanced computing power configuration.

[0058] The specific area of ​​visual obstruction at the current edge node is determined by identifying early signs of limited field of view. Raw image and sensor data collected by the edge node within the obstructed area are acquired. The local computing power of the current edge node is assessed based on the location and severity of the obstruction to determine if it meets real-time processing requirements. If local computing power is insufficient, available computing resource instances are obtained from the cloud resource pool. A resource allocation mechanism is employed to schedule cloud computing resources to the edge node based on the amount of data in the obstructed area and processing latency requirements. The scheduled cloud computing resources are bound to the local resources of the edge node through cloud-edge collaboration. The enhanced edge node computing power configuration is then obtained and deployed to the corresponding edge node for subsequent task execution.

[0059] For example, in edge network environments, determining the specific area of ​​field-of-view obstruction by edge nodes can be achieved through analysis of the node's geographical location and sensor feedback. Suppose that in an urban edge network, an edge node is deployed at an intersection to monitor traffic flow, with a field-of-view coverage radius of 50 meters. However, due to obstruction by temporary buildings, the actual effective field of view is reduced to only 30 meters. By analyzing the blurred edges of the image captured by the node's camera and the distance data from the sensors, the obstructed area can be accurately located within 10 meters southeast of the node. This process relies on real-time calibration of the node's location and multi-dimensional comparison of sensor data to ensure accurate determination of the obstructed area.

[0060] In one possible implementation, when acquiring the raw image data and sensor data of edge nodes within the occluded area, data filtering can be performed by combining image resolution and sensor sampling frequency. Assuming the edge node camera acquires 30 frames per second at a resolution of 1080p, and the sensor acquires distance data 100 times per second, within the occluded area, data from regions where the brightness is 20% below the average value is prioritized. This data, combined with abnormal distance fluctuations reported by the sensor, is used to comprehensively determine the specific impact range of the occlusion. This method effectively reduces data redundancy and improves the efficiency of subsequent processing.

[0061] For example, determining whether the local computing power of an edge node meets real-time processing requirements can be achieved by evaluating the node's hardware configuration and current task load. Suppose a node is equipped with a processor with a clock speed of 2.0 GHz and 4 GB of memory, but the current task requires processing 500 MB of data per second, while the node's actual processing power is only 300 MB per second, which is clearly insufficient. In this case, by analyzing the amount of data in the occluded area and the required processing latency, such as a latency requirement of less than 50 milliseconds, it can be concluded that the local computing power cannot meet the requirements.

[0062] In one possible implementation, if local computing power is insufficient, when obtaining computing resource instances from the cloud resource pool, the selection can be based on the pool's availability list and network latency. Assuming there are 10 computing instances in the cloud resource pool, and the network latency between the node and the cloud is 20 milliseconds, the instance with the lowest latency and twice the computing power of the local node can be selected for allocation. This approach ensures both timely resource scheduling and matching accuracy.

[0063] For example, when using a resource allocation mechanism to schedule cloud computing resources to edge nodes, priorities can be set based on data volume and latency requirements. Assuming the data volume in the obscured area is 800MB and the processing latency requirement is within 100 milliseconds, the resource allocation mechanism will prioritize scheduling cloud instances with a bandwidth of 1Gbps to ensure data transmission and processing speed. This mechanism can dynamically adapt to different task requirements and improve resource utilization.

[0064] In one possible implementation, when binding cloud resources with local edge node resources through cloud-edge collaboration, virtual resource mapping technology can be used. Assuming the cloud-allocated computing resources are 2GHz and the edge node's local computing power is 1GHz, mapping technology can integrate them into a unified 3GHz virtual computing power pool for node use. This binding method can seamlessly connect the two resources, ensuring the continuity of task execution.

[0065] For example, when obtaining and distributing enhanced edge node computing power configurations to corresponding nodes, this can be achieved through configuration files. Assuming the enhanced computing power configuration is 3GHz and memory is expanded to 6GB, after the configuration is distributed, the node can process image and sensor data within occluded areas more efficiently. This configuration distribution method can quickly improve node capabilities, ensuring the real-time performance and stability of task processing.

[0066] S106. Based on the enhanced computing power configuration, inject compensation resources into the fault diagnosis process, use a time delay optimization method to adjust the time delay-related diagnostic priorities, and determine a fast fault location sequence.

[0067] Obtain the current system load status and the real-time computational requirements of each diagnostic task. Based on the total enhanced computing resources and the current load status, determine the remaining computing power available for compensation. Calculate the current expected completion latency of each diagnostic task using a latency optimization strategy. If the expected completion latency of a diagnostic task exceeds a preset threshold, it is marked as a high-latency task. For high-latency tasks, computing resources are reallocated through compensation computing power injection. Based on the resource distribution after compensation computing power injection, recalculate the adjusted latency of each diagnostic task. Employ a dynamic adjustment mechanism for diagnostic priorities, reordering all diagnostic tasks from lowest to highest adjusted latency. Obtain a rapid fault location sequence, and execute the diagnostic tasks sequentially according to the sorting result.

[0068] For example, in the roadside perception system of autonomous vehicles, when the field of vision of edge nodes is partially limited due to rain, fog or temporary obstacles, the system first continuously monitors the load status of each edge node and the real-time computing power requirements of various diagnostic tasks being performed.

[0069] Specifically, the current total load rate of edge nodes is 75%, while the three diagnostic tasks of vehicle trajectory prediction, obstacle classification, and lane detection require approximately 30%, 25%, and 20% of the computing resources, respectively.

[0070] In one possible implementation, based on the fact that the total amount of enhanced computing resources previously obtained through cloud-edge collaboration is 1.8 times the original local computing power, and considering the current load rate of 75%, the remaining computing power available for dynamic compensation is calculated to be approximately 0.4 times the original local computing power. This remaining computing power serves as a compensation pool, used to prioritize alleviating the pressure on critical diagnostic tasks.

[0071] Preferably, the system employs a latency optimization strategy based on the longest remaining processing time, estimating the expected completion latency for each diagnostic task. Assuming that under the current load, the expected completion latency for vehicle trajectory prediction is 420 milliseconds, obstacle classification is 320 milliseconds, and lane detection is 280 milliseconds, while the system's preset high latency threshold is 350 milliseconds, the vehicle trajectory prediction task is marked as a high-latency task. For this high-latency task, a compensation computing power injection method is used to preferentially allocate cloud resource instances equivalent to 0.25 times the local computing power from the remaining computing power pool and inject them into the edge node executing trajectory prediction. After the injection, the effective computing power of this node increases to 1.25 times the original, and the corresponding recalculated adjusted latency is reduced to approximately 260 milliseconds, which is below the threshold.

[0072] Understandably, after the compensation computing power injection is completed, the system uniformly refreshes the adjusted latency of all diagnostic tasks: trajectory prediction 260 milliseconds, obstacle classification 310 milliseconds, and lane line detection 270 milliseconds. A dynamic adjustment mechanism for diagnostic priorities is adopted, reordering the tasks according to their adjusted latency from smallest to largest, resulting in a rapid fault location sequence of trajectory prediction, lane line detection, and obstacle classification.

[0073] In one embodiment, prioritizing trajectory prediction and diagnostic tasks according to this sequence allows for the earliest identification of motion estimation bias caused by limited field of view, thereby quickly pinpointing specific sensor obstruction areas or algorithm model degradation issues. This prioritization method ensures that the diagnostic steps most impactful on driving safety are completed as early as possible, significantly shortening the exposure window for potential collision risks and improving the overall safety response capability of the system in complex weather or obstructed scenarios.

[0074] S107. Through the rapid fault location sequence, the unified perception information is integrated into the network view representation. If the location sequence covers multiple abnormal areas, the network repair protocol is triggered to obtain the repaired monitoring information stream.

[0075] A preliminary scan of the fault location is performed using a fast sequence to obtain the distribution data of the abnormal areas. Based on the obtained abnormal area distribution data, key anomalies in the sensing information are extracted to determine their specific locations and ranges. If the anomalies cover multiple areas, a network repair protocol is initiated, prioritizing the anomalies across multiple areas to obtain a sorted list of repair tasks. Using the repair task list, the corresponding modules in the repair protocol are called one by one to obtain the repair progress data for each area. Based on the repair progress data, the representation in the network view is updated to determine if the monitoring information has returned to normal. If the monitoring information has not returned to normal, the information flow is re-extracted for the unrepaired areas to determine the specific source of the unresolved anomalies. Using the source data of the unresolved anomalies, the execution order of the repair protocol is adjusted to obtain the final monitoring information flow.

[0076] For example.

[0077] In one possible implementation, when performing a preliminary scan for fault location using a fast sequence, all critical nodes in the network can be divided into several scan groups, each containing approximately ten adjacent nodes. The scan process prioritizes covering the core routing area, quickly acquiring the distribution data of abnormal indicators for each group, such as the set of nodes with sudden increases in packet loss rate or abnormal latency, thereby forming a preliminary heatmap of the abnormal area.

[0078] Specifically, when extracting key anomalies from the sensing information based on the abnormal area distribution data obtained from scanning, attention should be paid to those points whose indicators deviate from the normal baseline by more than 30%. For example, if three consecutive nodes in a certain area experience traffic inversion, these nodes are marked as core anomalies. Their specific locations can be accurate to the rack number and port number, and the range is limited to within that subnet segment to reduce the subsequent positioning radius.

[0079] Preferably, if the anomalies cover multiple areas, such as involving both backbone links and multiple access layer subnets, a network repair protocol is initiated. In this case, the anomalies across multiple areas are prioritized based on their severity and the number of affected users. For example, backbone link anomalies have the highest priority, followed by access subnets affecting more than 500 users, and finally, less affected edge areas are addressed, resulting in a prioritized list of repair tasks.

[0080] In one embodiment, when calling the corresponding modules in the repair protocol one by one through the repair task list, for the highest priority backbone anomalies, the link rerouting module can be executed first to obtain repair progress data, such as a rerouting completion rate of 85%. Subsequently, for access layer anomalies, the port reset and traffic cleaning modules are called, with progress data showing that abnormal traffic has decreased to within 10% of the normal value. Based on this repair progress data, the color indicators and alarm icons in the network view are updated, with green indicating recovery, orange indicating repair in progress, and red indicating continued anomalies.

[0081] It should be noted that when determining whether the monitoring information has returned to normal, the main observation is whether the key performance indicators have returned to the threshold range. If the monitoring information has not returned to normal, for example, if a subnet view still shows intermittent packet loss, then the information flow should be re-extracted for the unrepaired area, and the upstream and downstream traffic paths should be analyzed to determine whether the specific source of the unresolved anomaly is upstream switch buffer overflow or downstream link micro-disruption.

[0082] Understandably, adjusting the execution order of the repair protocol based on the source data of unresolved anomalies—for example, prioritizing buffer overflow issues by inserting them at the front of the protocol queue, calling targeted cache clearing and rate limiting modules, and re-executing the repair loop until the monitoring information flow is fully restored to stability—ensures resources are concentrated on bottlenecks. The beneficial effect is that this approach significantly shortens the cycle from fault discovery to full recovery, while reducing the scope of business interruptions caused by anomaly propagation. Through multi-regional priority sorting and real-time progress updates, operations personnel can intuitively grasp the repair status, and precise anomaly extraction and source tracing further improve location accuracy, ultimately significantly improving the continuity and reliability of network monitoring information.

[0083] S108. Obtain the repaired surveillance information stream, and use a verification method to check the matching degree between the surveillance information stream and the real-time uploaded data. If the matching degree is lower than a preset threshold, adjust the mismatched part through an iterative processing method to determine the final unified situational awareness model.

[0084] By processing the monitoring information stream, a repaired data stream is obtained. A preliminary matching degree value is determined by comparing this data stream with the real-time uploaded data. If the preliminary matching degree value is lower than the preset threshold, an iterative processing method is triggered to correct the inconsistencies, resulting in a corrected data stream. The corrected data stream is then used to re-execute the verification method to obtain an updated matching degree value, and it is determined whether the preset threshold has been reached. If the updated matching degree value is still lower than the preset threshold, a second iterative process is performed on the corrected data stream to adjust the remaining inconsistencies, resulting in an optimized data stream. The optimized data stream is then used to execute the final verification method to obtain a final matching degree value, and it is determined whether it meets the preset threshold. After obtaining the final matching degree value, a unified situational awareness model is constructed for the data streams that meet the preset threshold, and the final perception result is determined. For the constructed situational awareness model, information stream detection is performed to obtain the model's output stability index, and it is determined whether the index meets the preset standard.

[0085] For example, in network monitoring information stream processing, the degree of matching can be quantified by comparing and analyzing the repaired data stream with the real-time uploaded data.

[0086] Specifically, key metrics of the repaired data stream, such as traffic rate and packet loss rate, are compared item by item with the corresponding positions in the real-time data. The overall matching degree is calculated and usually expressed as a percentage. If the value reaches 92% or above, it is considered to be initially consistent; if it is below the preset threshold of 85%, it indicates that there is still a significant deviation, and it needs to enter the iterative processing stage.

[0087] In one possible implementation, after triggering the iterative processing method, the system first identifies inconsistencies. For example, a link might show a packet loss rate of 0.8% after repair, while the real-time uploaded data shows 4.2%. To address this discrepancy, data correction is employed. This involves tracing back the historical information flow before and after the anomaly, smoothing the data forward through interpolation, or correcting it based on the weighted average of nearby normal nodes, resulting in a corrected data flow. At this point, the corrected packet loss rate may decrease to around 1.5%. Re-performing verification using this corrected data flow may improve the matching accuracy to 88%, but it remains below the threshold, thus initiating a second iterative process.

[0088] Preferably, the second iteration further refines the remaining deviation, for example, by filtering sudden peaks within a specific time window or adjusting weights based on protocol layer characteristics. After optimization, the packet loss rate of the data stream further converges to 0.9%. Through final verification, the final matching degree reaches 94%, meeting the preset threshold requirement. The beneficial effect of this process is a significant improvement in the reliability and continuity of the data stream, avoiding overall misjudgment caused by small-scale residual anomalies.

[0089] Understandably, after acquiring the data stream that meets the threshold, building a unified situational awareness model becomes a crucial step.

[0090] In one embodiment, the model integrates multi-source repair information, using node health, link connectivity, and traffic trends as core dimensions to form a multi-layered network view representation. For the constructed situational awareness model, information flow detection is performed to evaluate stability indicators; for example, a model is considered stable if its output fluctuation is controlled within 3% for five consecutive minutes. If the indicators meet preset standards, the current perception results are confirmed as reliable and used for subsequent decision support.

[0091] It should be noted that the entire process, from initial matching assessment to multiple iterative corrections and model stability verification, forms a closed-loop logic, ensuring that the monitoring information flow maintains high-precision situational awareness even after anomalies in multiple regions have been repaired. This progressive processing method effectively reduces the probability of missed and false alarms, and improves the overall network's visualization management capabilities and rapid response level.

[0092] If the technical solution of this application involves the collection, storage, use, processing, transmission, provision, disclosure, or deletion of personal information, the products using this technical solution have clearly and understandably informed the users of the personal information processing rules before processing personal information, and have obtained the individuals' voluntary consent in accordance with the law. If the technical solution of this application involves sensitive personal information (such as biometrics, religious beliefs, specific identities, medical and health information, financial accounts, and location tracking), the products using this solution have obtained the individuals' separate consent before processing sensitive personal information, and have also met the requirement of "express consent," ensuring that individuals make authorization decisions voluntarily based on full knowledge.

[0093] Specific implementation methods include, but are not limited to, the following: setting up clear and prominent signs at personal information collection devices such as cameras and sensors to inform relevant personnel that they have entered the scope of personal information collection and that their personal information will be collected and processed. If an individual voluntarily enters the collection scope after being informed, it is deemed that they have agreed to the collection of their personal information; or using obvious icons, text descriptions, or other means on the terminal device or system interface for personal information processing to inform them of the rules for personal information processing, and obtaining the individual's explicit authorization through interactive methods such as pop-up prompts, check confirmation boxes, or asking the individual to upload their personal information themselves.

[0094] The aforementioned personal information processing rules should include, but are not limited to, the name and contact information of the personal information processor, the specific purpose of personal information processing, the processing method, the types of personal information processed, the retention period, and the methods and procedures for individuals to exercise their relevant rights.

[0095] The above are only some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and modifications can be made. Any improvements and modifications made based on the basic principles of the present invention should be considered to fall within the protection scope of the present invention.

Claims

1. A cloud-edge collaborative intelligent fault diagnosis and location system and method for power networks, characterized in that, The method includes: Monitoring information is obtained from edge nodes, including the computing load parameters and network link connection status of the edge nodes. Data processing methods are used to aggregate the monitoring information to remove redundant parts, resulting in a refined monitoring dataset. Based on the refined monitoring dataset, the dataset is uploaded to the cloud control center using a real-time transmission mechanism. If network link status fluctuations exceed a preset threshold during transmission, the remaining data is transmitted via a backup communication link to determine the complete collaborative transmission path. The dataset in the complete collaborative transmission path is obtained, and the refined monitoring dataset is fused with historical network view data using a data integration method in the cloud control center to obtain an updated network view representation. For the updated network view representation, an anomaly analysis model is used to detect dynamic change features. If the dynamic change features show an abnormal evolution trend, potential cascading failure areas are marked, and the location of early signs is determined. Based on the location of the early signs, relevant data of edge nodes with limited visibility are obtained, and cloud computing resources are allocated to the edge nodes using a resource allocation mechanism to obtain enhanced computing power configuration; Based on the enhanced computing power configuration, compensation resources are injected into the fault diagnosis process, and a time delay optimization method is used to adjust the time delay-related diagnostic priorities to determine a rapid fault location sequence. The unified perception information is integrated into the network view representation through the rapid fault location sequence. If the location sequence covers multiple abnormal areas, the network repair protocol is triggered to obtain the repaired monitoring information stream. The repaired surveillance information stream is obtained, and a verification method is used to check the matching degree between the surveillance information stream and the real-time uploaded data. If the matching degree is lower than a preset threshold, the mismatched part is adjusted through an iterative processing method to determine the final unified situational awareness model.

2. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The monitoring information obtained from the edge nodes includes the computational load parameters and network link connection status of the edge nodes. Data processing methods are used to aggregate the monitoring information to remove redundancy, resulting in a refined monitoring dataset, including: Step 1: Obtain monitoring data from edge nodes. The monitoring data includes calculated load parameters and network link connection status. The data is initially classified and organized using a preset acquisition tool to obtain a classified monitoring data set. Step 2: For the classified monitoring data set, use data processing methods to perform aggregation operations, focusing on removing redundant information and duplicate records. By comparing with the preset redundancy judgment rules, obtain a subset of deduplicated monitoring data. Step 3: Based on the deduplicated subset of monitoring data, apply cluster analysis to group the data. If the load parameter of a certain group of data exceeds the preset threshold, it is marked as a high load group, and the grouped and marked data units are obtained. Step 4: Analyze the correlation between high-load groups and network link connection status through the grouped and labeled data units. If the connection status of a certain high-load group is abnormal, it is classified as a priority processing object, and the priority processing data units are determined. Step 5: Obtain the data unit to be processed first, and execute the status evaluation logic for the load parameters and connection status data. If the evaluation result shows that the node status is unstable, trigger the resource scheduling instruction and generate a scheduling instruction set. Step 6: Based on the scheduling instruction set, automatically adjust the resource allocation scheme of the edge nodes, rebalance the computing load through preset allocation rules, and obtain the adjusted resource configuration data; Step 7: Update the operating parameters of the edge nodes using the adjusted resource configuration data, continuously track changes in node status using real-time monitoring tools, determine whether the resource configuration has reached a balanced state, and generate the final operating status data.

3. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The process involves uploading the refined monitoring dataset to the cloud control center using a real-time transmission mechanism. If network link status fluctuations exceed a preset threshold during transmission, the remaining data is transmitted via a backup communication link. This process determines the complete collaborative transmission path, including: Acquire a refined monitoring dataset and begin sending data to the cloud control center via a real-time transmission mechanism; During data transmission, the network link status is continuously monitored to obtain the current link status parameters; Determine whether the network link status fluctuation exceeds a preset threshold. If it does, trigger the link switching process; otherwise, continue the original link transmission. When a link switch is triggered, obtain the amount of data that has been transmitted and determine the remaining untransmitted data portion; The backup communication link takes over the transmission task and sends the remaining untransmitted data to the cloud control center; Based on the connection between the transmitted data and the remaining data, record the order of upload and segmentation identifiers of the two links; By using the segment identifiers and sequence information of data uploaded through two links, the complete dataset is reassembled to determine the final structure of the cooperative transmission path.

4. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The process involves acquiring the dataset from the complete collaborative transmission path, and then using a data integration method at the cloud control center to fuse the refined monitoring dataset with historical network view data to obtain an updated network view representation, including: Acquire refined monitoring datasets and historical network view data; Standardized monitoring data is obtained by performing format unification processing on the refined monitoring dataset through data cleaning rules. The standardized monitoring data is matched with the historical network view data by using a timestamp alignment method to obtain time-aligned monitoring view pairs; Based on the time-aligned monitoring view, the difference vector between the current state vector and the historical state vector of each network node is calculated to obtain the state difference sequence. If the difference of a node in the state difference sequence exceeds the preset stable interval, then the node is marked as an abnormally active node, and a set of abnormally active nodes is obtained. The current network topology, which contains a set of abnormally active nodes, is represented by a vector using a graph embedding method, resulting in the current network topology embedding vector. The current network topology embedding vector is concatenated with the topology embedding vector of the corresponding time period in the historical network view data to obtain the updated network view representation.

5. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The updated network view representation employs an anomaly analysis model to detect dynamic change characteristics. If these dynamic change characteristics exhibit an abnormal evolution trend, potential cascading failure areas are marked, and early warning signs are identified, including: Retrieve the updated network view representation data; By processing the network view representation data using an anomaly analysis model, dynamic change characteristics are obtained; Perform trend analysis on dynamic change characteristics to determine whether there are any abnormal evolution trends; If the dynamic change characteristics show an abnormal evolution trend, then a marking operation is performed to obtain the potential cascading failure area; Based on the location relationships of potential cascading failure areas, a graph traversal method is used to determine the location of early signs; Feature extraction is performed on the network nodes associated with the early signs to obtain a set of sign nodes; The chain propagation path is obtained by analyzing the connectivity between the symptom node set and the network view representation.

6. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The process of acquiring relevant data of edge nodes with limited visibility based on the location of the early signs, and allocating cloud computing resources to the edge nodes using a resource allocation mechanism to obtain enhanced computing power configuration includes: The specific area where the current edge node's field of vision is obstructed is determined by the location information of early signs of limited field of vision. Acquire raw image data and sensor data collected from edge nodes within the occluded area; Determine whether the local computing power of the current edge node meets the real-time processing requirements based on the location and degree of occlusion. If local computing power is insufficient, obtain available computing resource instances from the cloud resource pool; A resource allocation mechanism is adopted to schedule cloud computing resources to edge nodes based on the amount of data in the obscured area and the processing latency requirements. By using a cloud-edge collaboration approach, the scheduled cloud computing resources are bound to the local resources of the edge nodes; The enhanced computing capabilities of the edge nodes are configured and distributed to the corresponding edge nodes to execute subsequent tasks.

7. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, Based on the enhanced computing power configuration, compensation resources are injected into the fault diagnosis process, and a latency optimization method is used to adjust the latency-related diagnostic priorities to determine a rapid fault location sequence, including: Obtain the current system load status and the real-time computing requirements of each diagnostic task; Based on the total amount of enhanced computing resources and the current load status, determine the remaining computing power available for compensation; A latency optimization strategy is used to calculate the current expected completion latency of each diagnostic task; If the expected completion time of a diagnostic task exceeds a preset threshold, it will be marked as a high-latency task. For high-latency tasks, computing resources are reallocated through a compensation-based computing power injection method; Based on the resource distribution after the compensation computing power injection, the adjusted latency of each diagnostic task is recalculated; A dynamic adjustment mechanism for diagnostic priorities is adopted, and all diagnostic tasks are reordered according to the adjusted latency from smallest to largest. Obtain the fault location sequence and execute the diagnostic tasks sequentially according to the sorting results.

8. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The unified sensing information is integrated into the network view representation through the rapid fault location sequence. If the location sequence covers multiple abnormal areas, a network repair protocol is triggered to obtain a repaired monitoring information stream, including: A preliminary scan of the fault location is performed using fast sequences to obtain distribution data of the abnormal area; Based on the abnormal area distribution data obtained from the scan, key abnormal points in the sensing information are extracted to determine the specific location and range of the abnormal points. If the anomaly points cover multiple areas, the network repair protocol is activated to prioritize the anomalies in multiple areas and obtain a sorted list of repair tasks. By calling the corresponding modules in the repair protocol one by one through the repair task list, the repair progress data for each area can be obtained; Based on the repair progress data, update the representation in the network view to determine whether the monitoring information has returned to normal. If the monitoring information does not return to normal, the information flow will be re-extracted for the unrepaired area to determine the specific source of the unresolved anomaly. By analyzing the source data of unresolved anomalies, the execution order of the remediation protocol is adjusted to obtain the final monitoring information stream.

9. The cloud-edge collaborative power network intelligent fault diagnosis and location system and method according to claim 1, characterized in that, The process of acquiring the repaired surveillance information stream involves using a verification method to check the matching degree between the surveillance information stream and the real-time uploaded data. If the matching degree is lower than a preset threshold, an iterative processing method is used to adjust the mismatched parts, and the final unified situational awareness model is determined, including: By processing the monitoring information stream, the repaired data stream is obtained. By comparing and analyzing the data stream with the real-time uploaded data, a preliminary matching degree value is determined. Based on the comparison between the initial matching degree value and the preset threshold line, if the matching degree value is lower than the preset threshold line, the iterative processing method is triggered to correct the data for the inconsistent parts and obtain the corrected data stream. Using the corrected data stream, the verification method is re-executed to obtain the updated matching degree value and determine whether the preset threshold line has been reached. If the updated matching degree value is still lower than the preset threshold, a second iteration is performed on the corrected data stream to adjust the remaining inconsistent parts and determine the optimized data stream. The optimized data stream is used to perform the final verification method to detect and obtain the final matching degree value, and to determine whether the preset threshold line is met. After obtaining the final matching degree value, a unified situational awareness model is constructed for the data stream that meets the preset threshold line to determine the final perception result; For the constructed situational awareness model, information flow detection is performed to obtain the stability index output by the model and determine whether the index meets the preset standard.

Images