Cloud-edge collaborative predictive maintenance method and system for communication network equipment

Abstract

The application discloses a cloud-edge collaborative predictive maintenance method and system for communication network equipment, and belongs to the technical field of communication network equipment operation and maintenance, which comprises the following steps: collecting operation data of the communication network equipment in real time through an edge node and preprocessing the operation data to obtain feature data; calculating the weight of the feature data and weighting the feature data in the edge node, constructing an improved isolated forest based on the weighted feature set, performing anomaly detection on the feature data, outputting a diagnostic report, triggering a local alarm when an anomaly is detected, and uploading the diagnostic result as an anomaly summary to the cloud; receiving the anomaly summary in the cloud, performing deep analysis on the historical data uploaded by the edge node, generating a maintenance strategy, and delivering the maintenance strategy; collecting the operation and maintenance results of the maintenance strategy, and iteratively optimizing the improved isolated forest on the edge side and the deep analysis in the cloud. The anomaly detection task is sunk to the edge node, and the accuracy and real-time performance of the anomaly detection are improved.

Description

Technical Field

[0001] This application belongs to the field of communication network equipment operation and maintenance technology, and specifically relates to a cloud-edge collaborative predictive maintenance method and system for communication network equipment. Background Technology

[0002] With the development of the Industrial Internet and 5G networks, communication network equipment (such as optical network units (ONUs), switches, 5G base stations, and wireless access points (APs)) are being deployed on a large scale in industrial scenarios, and their stable operation is crucial for business continuity. Traditional equipment maintenance methods mainly include periodic maintenance and post-fault repair, but these methods suffer from problems such as resource waste or service interruption. In recent years, predictive maintenance (PdM) has gradually become a research hotspot.

[0003] In existing technologies, commonly used anomaly detection algorithms such as Isolation Forest and One-Class SVM are general-purpose models that do not fully consider the high dimensionality and strong temporal sequence of communication network device data, as well as the differentiated impact of key performance indicators (such as CPU utilization, memory utilization, and network traffic) on device status. Because all features are treated equally, the model may ignore changes in key indicators, leading to low anomaly detection accuracy and high false positive and false negative rates.

[0004] Furthermore, existing technologies typically employ a centralized cloud analytics architecture, uploading all raw data to the cloud for processing. This architecture can lead to significant network bandwidth pressure in large-scale device deployment scenarios, and due to data transmission and cloud processing delays, it is difficult to meet the real-time requirements of fault early warning.

[0005] Meanwhile, traditional edge nodes typically only have data collection and forwarding functions, lacking local intelligent analysis capabilities. They cannot independently complete basic monitoring and alarm tasks in environments with no or weak network coverage, causing edge nodes to fail to fully leverage their advantage of being close to the data source.

[0006] Furthermore, existing anomaly detection methods typically output a binary judgment of "normal / abnormal," lacking differentiation between anomaly types (such as sudden anomalies and gradual anomalies) and severity. This makes it difficult for maintenance personnel to quickly locate the root cause of the problem, thus prolonging the troubleshooting time.

[0007] Therefore, improving the accuracy, real-time performance, and interpretability of anomaly detection in communication network equipment has become a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a cloud-edge collaborative predictive maintenance method and system for communication network devices, so as to solve at least one of the following technical problems existing in the prior art: 1. General anomaly detection algorithms suffer from insufficient detection accuracy when processing high-dimensional and time-series data from communication network devices because they fail to distinguish the importance of features; 2. The anomaly detection results are only binary outputs, lacking diagnostic information on the type and severity of the anomaly, which is not conducive to rapid fault location; 3. Centralized cloud analytics leads to high network bandwidth pressure and poor real-time performance; 4. Edge nodes lack intelligent analysis capabilities and cannot operate independently in environments with no or weak network coverage.

[0009] This application provides a cloud-edge collaborative predictive maintenance method for communication network devices, including: Feature data is obtained by collecting and preprocessing the operational data of communication network devices in real time through edge nodes. At the edge node, the weights of the feature data are calculated and weighted, and an improved isolated forest is constructed based on the weighted feature set. Anomaly detection is performed on the feature data, and a diagnostic report is output. When an anomaly is detected, a local alarm is triggered and the diagnostic result is uploaded to the cloud as an anomaly summary. The system receives the anomaly summary in the cloud, performs in-depth analysis in conjunction with historical data uploaded by edge nodes, generates maintenance strategies, and distributes them. Collect the operational results of the maintenance strategy, and iteratively optimize the improved isolated forest on the edge side and the deep analysis in the cloud.

[0010] Optionally, the operational data includes, but is not limited to, CPU utilization, memory utilization, network traffic, temperature, packet loss rate, and latency.

[0011] Optionally, calculating and weighting the feature data includes: The importance score of the feature data is calculated using a random forest, and the weights are adjusted in conjunction with operation and maintenance expert rules.

[0012] Optionally, the construction of the improved isolated forest based on the weighted feature set includes: An improved isolated forest is constructed by prioritizing high-weight features from the weighted feature set for node cutting.

[0013] Optionally, the step of performing anomaly detection on the feature data and outputting a diagnostic report includes: Calculate the path length of feature data in each tree and the average path length in the isolated forest. Calculate the anomaly score based on the average path length and determine the anomaly type based on the anomaly score and path length. The output includes a diagnostic report with abnormality labels, abnormality types, and severity scores.

[0014] Optionally, receiving the anomaly summary in the cloud, performing in-depth analysis in conjunction with historical data, generating a maintenance strategy, and distributing it includes: The system uses knowledge graphs to match common failure modes and performs root cause analysis. It also uses a remaining useful life prediction model to predict the remaining useful life of critical components, formulate maintenance strategies, and distribute them to edge nodes or maintenance terminals.

[0015] Optionally, the iterative optimization of edge-side isolated forests and cloud-based deep analysis includes: The operation and maintenance results are used as new samples, together with the historical data, to retrain the improved isolated forest and the remaining lifespan prediction model. The model increment of the improved isolated forest is synchronized to the edge nodes through differential update.

[0016] Based on the same inventive concept, this application also provides a cloud-edge collaborative predictive maintenance system for communication network devices, comprising: The data acquisition module is used to collect and preprocess the operating data of communication network devices in real time through edge nodes to obtain feature data; The edge analysis module is used to calculate and weight the feature data at the edge nodes, construct an improved isolated forest based on the weighted feature set, perform anomaly detection on the feature data, output a diagnostic report, trigger a local alarm when an anomaly is detected, and upload the diagnostic result as an anomaly summary to the cloud. The cloud-based analysis module is used to receive the anomaly summary in the cloud, perform in-depth analysis in combination with historical data, generate maintenance strategies, and distribute them. The model optimization module is used to collect the operation and maintenance results of maintenance strategies and to iteratively optimize the improved isolated forest on the edge side and the deep analysis in the cloud.

[0017] Furthermore, this application also provides an electronic device, comprising: at least one processor and a memory; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, a cloud-edge collaborative predictive maintenance method for communication network devices, as described above, is implemented.

[0018] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the cloud-edge collaborative predictive maintenance method for communication network devices as described above.

[0019] Compared with the prior art, this application has the following advantages: This application provides a cloud-edge collaborative predictive maintenance method and system for communication network devices, comprising: real-time collection and preprocessing of operational data of communication network devices through edge nodes to obtain feature data; calculating and weighting the feature data at the edge nodes, constructing an improved isolated forest based on the weighted feature set, performing anomaly detection on the feature data, outputting a diagnostic report, triggering local alarms when anomalies are detected, and uploading the diagnostic results as an anomaly summary to the cloud; receiving the anomaly summary in the cloud, performing in-depth analysis in conjunction with historical data uploaded by the edge nodes, generating and distributing maintenance strategies; collecting the operation and maintenance results of the maintenance strategies, and iteratively optimizing the improved isolated forest on the edge side and the in-depth analysis in the cloud.

[0020] This application has the following beneficial effects: 1. Cloud-edge collaborative architecture: By offloading lightweight real-time anomaly detection tasks to edge nodes, local rapid response is achieved, the amount of raw data uploaded is reduced, network bandwidth pressure is effectively alleviated, and system real-time performance is improved. 2. Algorithm-specific optimization: Based on the characteristics of communication network equipment data, an improved isolated forest algorithm based on feature weighting and anomaly type differentiation is proposed, which increases the focus on key performance indicators and significantly improves the accuracy of anomaly detection; 3. Enhance diagnostic capabilities: By introducing an anomaly type differentiation mechanism, multi-dimensional diagnostic information containing anomaly type and severity is output, providing interpretable decision support for operation and maintenance personnel and shortening troubleshooting time; 4. Closed-loop self-learning mechanism: Through feedback of operation and maintenance results, the model is updated online and continuously optimized, enabling the system to have adaptive capabilities and continuously improve prediction performance over time; 5. High availability guarantee: Edge nodes have independent analysis capabilities and can maintain basic monitoring and alarm functions even in network outages or weak network environments, ensuring business continuity.

[0021] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This paper illustrates a flowchart of the cloud-edge collaborative predictive maintenance method for communication network devices provided in this application. Figure 2 A flowchart illustrating the improved isolated forest algorithm provided in this application is shown; Figure 3 The diagram shows the architecture of the cloud-edge collaborative predictive maintenance system for communication network devices provided in this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Example 1 This application provides a cloud-edge collaborative predictive maintenance method for communication network devices, such as... Figure 1 ,include: Step 1: Collect and preprocess the operational data of communication network devices in real time through edge nodes to obtain feature data; Step 2: At the edge node, calculate and weight the feature data, construct an isolated forest based on the weighted feature set, perform anomaly detection on the feature data, output a diagnostic report, trigger a local alarm when an anomaly is detected, and upload the diagnostic result as an anomaly summary to the cloud. Step 3: Receive the anomaly summary in the cloud, perform in-depth analysis in conjunction with historical data, generate maintenance strategies, and distribute them. Step 4: Collect the operation and maintenance results of the maintenance strategy, and iteratively optimize the isolated forest on the edge side and the cloud in-depth analysis.

[0026] Optionally, the runtime data mentioned in step 1 includes, but is not limited to, CPU utilization, memory utilization, network traffic, temperature, packet loss rate, and latency.

[0027] Optionally, the running data may be preprocessed, including data cleaning, noise reduction, missing value imputation, and feature extraction.

[0028] Optionally, calculating and weighting the feature data includes: The importance score of the feature data is calculated by random forest, and the weights are adjusted by combining the rules of operation and maintenance experts, so that key indicators (such as CPU and memory) are given priority in the segmentation during the tree construction process.

[0029] Optionally, the construction of the improved isolated forest based on the weighted feature set includes: Based on the weighted feature set, high-weight features are selected first for node cutting to construct an isolated forest.

[0030] Optionally, the step of performing anomaly detection on the feature data and outputting a diagnostic report includes: Calculate the path length of feature data in each tree and the average path length in the isolated forest. Calculate the anomaly score based on the average path length and determine the anomaly type based on the anomaly score and path length. The output includes a diagnostic report with abnormality labels, abnormality types, and severity scores.

[0031] Optionally, the anomaly type can be determined based on the anomaly score and path length, including: Combining path length distribution characteristics with domain knowledge, anomalies are classified into three types: sudden anomalies (extremely short paths), gradual anomalies (shorter paths but concentrated distribution), and periodic anomalies. In this context, a shorter path indicates that the feature data is more likely to be an outlier.

[0032] Optionally, receiving the anomaly summary in the cloud, performing in-depth analysis in conjunction with historical data, generating a maintenance strategy, and distributing it includes: The system uses knowledge graphs to match common failure modes and performs root cause analysis. It also uses a remaining useful life prediction model to predict the remaining useful life of critical components, formulate maintenance strategies, and distribute them to edge nodes or maintenance terminals.

[0033] Specifically, fault diagnosis includes combining historical data and knowledge graphs to identify potential fault modes; root cause analysis includes using causal reasoning models to trace the source of anomalies; and remaining useful life (RUL) prediction includes using time series models (such as Long Short-Term Memory, LSTM, and the time series forecasting tool Prophet) to predict the degradation trend and life end of key components of equipment.

[0034] Optionally, the maintenance strategy includes, but is not limited to, immediate restart, planned replacement, and remote diagnostics.

[0035] Optionally, the iterative optimization of the improved isolated forest on the edge side and the cloud-based depth analysis includes: The operation and maintenance results are used as new samples, together with the historical data, to retrain the improved isolated forest and the remaining lifespan prediction model. The model increment of the improved isolated forest is synchronized to the edge nodes through differential update.

[0036] Optionally, the operation and maintenance results include, but are not limited to, the actual operation and maintenance results (such as whether a fault occurred, the model of the replaced component, and the repair time) and equipment status update information.

[0037] This application provides a preferred embodiment of a cloud-edge collaborative predictive maintenance method for communication network devices. Step S101: Data Acquisition and Preprocessing Edge nodes collect operational data of communication network devices through Simple Network Management Protocol (SNMP), telemetry / remote telemetry, and System Logging Protocol (Syslog), with a sampling period of 1 second to 1 minute, depending on the device type and performance requirements.

[0038] Preprocess the raw running data: use the sliding window method to remove noise, use linear interpolation or forward filling to handle missing values, and extract statistical features (such as mean, variance, and maximum value) and time-domain features (such as rate of change) as feature data.

[0039] Step S102: Real-time analysis at the edge An improved isolated forest model can be deployed on edge nodes, with its training phase completed in the cloud and updated model parameters delivered periodically. The flowchart of the improved isolated forest algorithm is shown below. Figure 2 Specifically, it includes: Data input: Collect multi-dimensional metrics such as CPU, memory, and network traffic; Feature weighting: Dynamically assigning differentiated weights to key indicators, assuming the feature data set is F={ , , ···, The importance scores of each feature are calculated using a random forest. The weights are adjusted in conjunction with the rules of operation and maintenance experts to form the final weighted feature space; Forest construction: Multiple isolated trees are generated based on weighted feature sets. When constructing each tree, high-weight features are selected first for node segmentation to improve the ability to distinguish key indicators. Anomaly detection: Calculate the average path length to determine if an anomaly exists. Path length analysis is as follows: Sample x represents the value of the device's operating status at a certain moment across the feature set's dimensions, and the path length of sample x in each tree is... The average path length of sample x in the isolated forest is recorded and calculated. ; Exception type differentiation: combining path length and domain knowledge, based on... Calculating an anomaly score, wherein is the average correction factor of path lengths; anomaly type determination: if s>0.8 and the path length is extremely short (e.g., less than 2), it is determined as a sudden anomaly; if 0.6<s<0.8 and the path length distribution is relatively concentrated, it is determined as a gradual anomaly; if the anomaly presents periodicity, it is marked as a periodic anomaly; Result output: output anomaly labels / types / severity scores, and output a standardized diagnosis report including anomaly status, type and severity level.

[0040] Step S103: in-depth cloud analysis.

[0041] The cloud receives anomaly summaries and historical feature data uploaded by edge nodes, and performs in-depth analysis in combination with global historical data: Matching common fault modes using a knowledge graph; Performing root cause analysis by applying Granger causality test or Bayesian network; Adopting an LSTM neural network to predict the remaining useful life (RUL) of key components, output a confidence interval to quantify the reliability of the life prediction result, and divide the equipment maintenance level and intelligently adjust the work order distribution priority according to the range of the confidence interval.

[0042] Step S104: cloud-edge collaborative decision-making.

[0043] Generating a maintenance strategy according to the in-depth analysis result of the cloud, and delivering the strategy to edge nodes or operation and maintenance terminals through MQTT or HTTPS protocol; For example, if it is predicted that the memory of a certain ONU will continue to degenerate and RUL<7 days, a "planned replacement" instruction will be issued, and a maintenance window will be reserved.

[0044] Step S105: closed-loop feedback and model optimization.

[0045] After operation and maintenance personnel perform maintenance operations, record actual results (such as whether the fault is confirmed, the model of replaced parts, and time consumed for repair) as operation and maintenance results; Adding the operation and maintenance results as new samples to a training set, wherein the training set is historical feature data, regularly retraining the improved isolation forest model and the remaining useful life prediction model on the cloud, and incrementally synchronizing the improved isolation forest model to edge nodes through differential update, so as to realize continuous optimization of the improved isolation forest model and the remaining useful life prediction model.

[0046] Example 2 Based on the same inventive concept, the present application also provides a cloud-edge collaborative predictive maintenance system for communication network equipment, Figure 3 comprising: The data acquisition module is used to collect and preprocess the operating data of communication network devices in real time through edge nodes to obtain feature data; The edge analysis module is used to calculate and weight the feature data at the edge nodes, construct an improved isolated forest based on the weighted feature set, perform anomaly detection on the feature data, output a diagnostic report, trigger a local alarm when an anomaly is detected, and upload the diagnostic result as an anomaly summary to the cloud. The cloud-based analysis module is used to receive the anomaly summary in the cloud, perform in-depth analysis in combination with historical data, generate maintenance strategies, and distribute them. The model optimization module is used to collect the operation and maintenance results of maintenance strategies and to iteratively optimize isolated forests on the edge side and in-depth analysis in the cloud.

[0047] Optionally, the operational data includes, but is not limited to, CPU utilization, memory utilization, network traffic, temperature, packet loss rate, and latency.

[0048] Optionally, calculating and weighting the feature data includes: The importance score of the feature data is calculated using a random forest, and the weights are adjusted in conjunction with operation and maintenance expert rules.

[0049] Optionally, the construction of the improved isolated forest based on the weighted feature set includes: An improved isolated forest is constructed by prioritizing high-weight features from the weighted feature set for node cutting.

[0050] Optionally, the step of performing anomaly detection on the feature data and outputting a diagnostic report includes: Calculate the path length of feature data in each tree and the average path length in the isolated forest. Calculate the anomaly score based on the average path length and determine the anomaly type based on the anomaly score and path length. The output includes a diagnostic report with abnormality labels, abnormality types, and severity scores.

[0051] Optionally, receiving the anomaly summary in the cloud, performing in-depth analysis in conjunction with historical data, generating a maintenance strategy, and distributing it includes: The system uses knowledge graphs to match common failure modes and performs root cause analysis. It also uses a remaining useful life prediction model to predict the remaining useful life of critical components, formulate maintenance strategies, and distribute them to edge nodes or maintenance terminals.

[0052] Optionally, the iterative optimization of the improved isolated forest on the edge side and the cloud-based depth analysis includes: The operation and maintenance results are used as new samples, together with the historical data, to retrain the improved isolated forest and the remaining lifespan prediction model. The model increment of the improved isolated forest is synchronized to the edge nodes through differential update.

[0053] Example 3 Based on the same inventive concept, this application also provides an electronic device. The electronic device of this application includes at least one processor and at least one storage medium electrically connected to each other. The storage medium is electrically connected to the processor, wherein the storage medium stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method described above.

[0054] Example 4 Based on the same inventive concept, this application also provides a storage medium storing instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the method described above.

[0055] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A cloud-edge collaborative predictive maintenance method for communication network devices, characterized in that, include: Feature data is obtained by collecting and preprocessing the operational data of communication network devices in real time through edge nodes. At the edge node, the weights of the feature data are calculated and weighted, and an improved isolated forest is constructed based on the weighted feature set. Anomaly detection is performed on the feature data, and a diagnostic report is output. When an anomaly is detected, a local alarm is triggered and the diagnostic result is uploaded to the cloud as an anomaly summary. The system receives the anomaly summary in the cloud, performs in-depth analysis in conjunction with historical data uploaded by edge nodes, generates maintenance strategies, and distributes them. Collect the operational results of the maintenance strategy, and iteratively optimize the improved isolated forest on the edge side and the deep analysis in the cloud.

2. The method according to claim 1, characterized in that, The operational data includes, but is not limited to, CPU utilization, memory utilization, network traffic, temperature, packet loss rate, and latency.

3. The method according to claim 1, characterized in that, The calculation and weighting of the feature data includes: The importance score of the feature data is calculated using a random forest, and the weights are adjusted in conjunction with operation and maintenance expert rules.

4. The method according to claim 1, characterized in that, The construction of the improved isolated forest based on the weighted feature set includes: An improved isolated forest is constructed by prioritizing high-weight features from the weighted feature set for node cutting.

5. The method according to claim 1, characterized in that, The step of performing anomaly detection on the feature data and outputting a diagnostic report includes: Calculate the path length of feature data in each tree and the average path length in the isolated forest. Calculate the anomaly score based on the average path length and determine the anomaly type based on the anomaly score and path length. The output includes a diagnostic report with abnormality labels, abnormality types, and severity scores.

6. The method according to claim 1, characterized in that, The process of receiving the anomaly summary in the cloud, performing in-depth analysis in conjunction with historical data, generating a maintenance strategy, and distributing it includes: The system uses knowledge graphs to match common failure modes and performs root cause analysis. It also uses a remaining useful life prediction model to predict the remaining useful life of critical components, formulate maintenance strategies, and distribute them to edge nodes or maintenance terminals.

7. The method according to claim 6, characterized in that, The iterative optimization of edge-side isolated forests and cloud-based deep analysis includes: The operation and maintenance results are used as new samples, together with the historical data, to retrain the improved isolated forest and the remaining lifespan prediction model. The model increment of the improved isolated forest is synchronized to the edge nodes through differential update.

8. A cloud-edge collaborative predictive maintenance system for communication network equipment, characterized in that, include: The data acquisition module is used to collect and preprocess the operating data of communication network devices in real time through edge nodes to obtain feature data; The edge analysis module is used to calculate and weight the feature data at the edge nodes, construct an improved isolated forest based on the weighted feature set, perform anomaly detection on the feature data, output a diagnostic report, trigger a local alarm when an anomaly is detected, and upload the diagnostic result as an anomaly summary to the cloud. The cloud-based analysis module is used to receive the anomaly summary in the cloud, perform in-depth analysis in combination with historical data, generate maintenance strategies, and distribute them. The model optimization module is used to collect the operation and maintenance results of maintenance strategies and to iteratively optimize the improved isolated forest on the edge side and the deep analysis in the cloud.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the cloud-edge collaborative predictive maintenance method for communication network devices as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the cloud-edge collaborative predictive maintenance method for communication network devices as described in any one of claims 1-7.

Images