Remote fault diagnosis method and system based on fttr network and storage medium

Abstract

The application relates to the technical field of network communication, in particular to a remote fault diagnosis method and system based on an FTTR network and a storage medium. The method comprises the following steps: a monitoring module for monitoring network state is arranged on each node of the FTTR network, each monitoring module is provided with a data acquisition unit and a state analysis unit, and each node collects network state data in real time through the data acquisition unit. The network state data comprises network delay, packet loss rate, equipment temperature and bandwidth utilization rate, the network state data is preliminarily analyzed through the state analysis unit, and historical data is compared to determine the change trend of the current state. Through the distributed management mode, the single-point fault problem in the traditional centralized management is avoided. Each FTTR node can independently monitor its own state and work cooperatively with other nodes, even if a certain node fails, other nodes can still work normally, and the stability of the whole network is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network communication, in particular to a remote fault diagnosis method and system based on an FTTR network and a storage medium. BACKGROUND

[0002] Fiber To The Room (FTTR) network is a high-speed fiber network that directly wires optical fibers to each room in a user's home, providing high-bandwidth, low-latency network connections. With the development of smart home and Internet of Things, the demand for FTTR network is growing, and the stability requirement is also increasing. Remote monitoring and fault diagnosis system is a key technology to ensure the stable operation of FTTR network. Remote monitoring system is a system that monitors and manages devices in real time through network, which can help administrators to find and solve problems in time, and improve the reliability of the system. Fault diagnosis technology is to analyze the system failure and find out the root cause of the failure, so as to solve the problem quickly and effectively.

[0003] At present, the remote monitoring and fault diagnosis system of FTTR network mostly adopts centralized management mode, that is, the status information of all network devices (such as optical modem, router, etc.) is collected to the center node, and the center node collects data and analyzes faults. Usually, the center node collects the running data of each device regularly, and judges the health status of the system by analyzing the data. If an abnormal device is detected, the center node will notify the administrator to repair, or trigger the preset repair strategy through automatic tools. Although the existing solution can ensure the stable operation of FTTR network to a certain extent, there are still some problems:

[0004] Firstly, the centralized management mode is prone to single point failure problem, if the center node fails, the management and monitoring of the whole network will be affected. Secondly, the existing system can only passively monitor the status of the network, and cannot actively find potential fault hidden dangers. In addition, the existing fault diagnosis technology often relies on experienced administrators, lacks intelligent degree, and cannot quickly and accurately diagnose the root cause of the fault. SUMMARY

[0005] The present application provides a remote fault diagnosis method and system based on FTTR network and a storage medium to solve the single point failure risk and passive monitoring problem in the existing centralized management mode, and improve the fault diagnosis and processing efficiency through intelligent means, while ensuring the scalability of the system in large-scale network environment.

[0006] The technical solution of the present application to solve the above technical problems is as follows:

[0007] In one aspect, a remote fault diagnosis method based on an FTTR network is provided, the method comprising the following steps:

[0008] A monitoring module for monitoring network status is deployed on each node of the FTTR network, each monitoring module being provided with a data collection unit and a status analysis unit, and each node collecting network status data in real time through the data collection unit;

[0009] The network status data includes network delay, packet loss rate, device temperature and bandwidth utilization, and the network status data is preliminarily analyzed by the status analysis unit and compared with historical data to determine the trend of the current status;

[0010] Each node publishes the status information of the current status to a shared topic through a message queue system and synchronizes with other nodes to form a global status view;

[0011] An unsupervised learning algorithm is used to analyze the synchronized global status view to identify abnormal patterns for detecting extreme fluctuations in traffic and sustained high latency behavior;

[0012] A supervised learning model is used to analyze historical fault data, extract features, and analyze abnormal patterns to predict potential faults and issue warnings;

[0013] When the warning is triggered, a deep learning model is used to analyze the current fault, infer the cause of the fault, and generate and execute a repair operation.

[0014] Further, the status analysis unit uses a double-threshold model for trend analysis and prediction, and the expression of the double-threshold model is:

[0015] Where ΔM(t) is the status change rate, and T1 and T2 are threshold values.

[0016] Further, the unsupervised learning algorithm includes a clustering method based on a dynamic discriminant boundary, and the expression is:

[0017] Where D is the distance from data point x i to cluster center c j , and p is a tuning parameter.

[0018] Further, the step of each node publishing the status information of the current status to a shared topic through a message queue system and synchronizing with other nodes includes:

[0019] Each node publishes JSON-formatted status information to a globally shared topic through a message queue system every 30 seconds, and after receiving, integrates information from other nodes to generate a global status table.

[0020] Further, the supervised learning model is based on an ensemble learning method, including using multi-model collaboration, the expression of the multi-model collaboration is:

[0021] f(x)=∑ k w k ·h k (x),wherein f(x) is the total prediction of the fault, h k (x) is the output of each sub-model, and w k is the weight corresponding to the sub-model.

[0022] Further, the prediction of the fault also uses a time series analysis method, using an autoregressive integrated moving average model, the expression of the autoregressive integrated moving average model is:

[0023] wherein Y(t) represents the observation value of the time series at time t, c represents the mean value of the time series, φ i (i=1,2,…,p) is the autoregressive coefficient, representing the influence of Y(t) at the past i time points on the current Y(t), Y(t-i) represents the observation value of the time series at time t-i, θ j (j=1,2,…,q) represents the influence of the error term ε at the past j time points on the current Y(t), ε(t-j) represents the random error term of the time series at time t-j, and ε(t) represents the random error term of the time series at time t.

[0024] Further, the deep learning model is used to analyze the current fault based on a long short-term memory (LSTM) deep learning pattern matching process, and the parameters are optimized through error back propagation to quickly respond to similar historical faults;

[0025] The repair operation generates an optimal repair path according to the fault mode library, and automatically adjusts parameters or triggers maintenance prompts in combination with device APIs.

[0026] Further, after the fault is repaired, the actual effect data is recorded and fed back to the deep learning model to adjust and optimize the parameter library, for improving the decision-making efficiency of subsequent operations.

[0027] On the other hand, a remote fault diagnosis system based on the FTTR network is provided, which is used to implement the remote fault diagnosis method based on the FTTR network as described above, and the system comprises:

[0028] A data acquisition module is deployed at each FTTR node, which is used to acquire network state information in real time;

[0029] A state analysis module acquires network state data provided by the data acquisition module, and performs preliminary analysis and distributed synchronization on the network state data;

[0030] Anomaly detection module, using unsupervised learning algorithm to detect abnormal patterns, identify potential network problems;

[0031] Fault prediction module, using supervised learning algorithm to predict possible network risks, providing early warning signals;

[0032] Fault diagnosis module, based on deep learning algorithm to analyze fault causes, and fault mode matching;

[0033] Repair suggestion and execution module, generating optimal repair suggestions, and through system interface for automatic repair or trigger maintenance prompt.

[0034] In still another aspect, a computer readable storage medium is provided, having stored thereon a computer program, which, when executed on an electronic device, causes it to perform the above-mentioned remote fault diagnosis method based on FTTR network.

[0035] The beneficial effects of the present application are:

[0036] The present application avoids the single point failure problem in traditional centralized management through a distributed management mode. Each FTTR node can independently monitor its own state and work cooperatively with other nodes. Even if a node fails, other nodes can still work normally, ensuring the stability of the entire network. The present application introduces unsupervised learning and supervised learning models to realize real-time analysis and prediction of network status, and can actively identify potential network anomalies and fault hazards. Before a fault occurs, the system can issue a warning signal and take measures in advance to reduce the impact of sudden failures on users and improve the network's fault prevention ability. In addition, through a deep learning model, the system can quickly and accurately locate the root cause of the fault and provide intelligent repair suggestions based on historical data. For common faults, the system can automatically repair, reducing the need for manual intervention and improving fault handling speed. For complex faults, the system can provide an optimized repair scheme, greatly shortening the fault troubleshooting and repair time and ensuring the rapid recovery of the network. Using a distributed computing framework and a distributed storage architecture, the system can efficiently process massive amounts of monitoring data from multiple FTTR nodes, maintaining fast response and analysis capabilities even in a high-concurrency environment. Through the storage mechanism of the distributed database, the system ensures the high availability of data and enables fast retrieval and analysis of historical data, improving data processing and query efficiency. The distributed architecture and efficient data processing method make the system have good scalability, which can adapt to the rapid growth of the number of smart home and Internet of Things devices in the future. The system can easily expand the node scale without affecting the overall performance, ensuring efficient operation in a large-scale network environment. Since the system can handle faults through intelligent diagnosis and automatic repair mechanisms, it reduces the dependence on human intervention and reduces operational costs and manpower requirements. In addition, the system has high automation, which can effectively reduce the workload of network administrators. BRIEF DESCRIPTION OF DRAWINGS DETAILED DESCRIPTION OF THE INVENTION BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a flowchart of the FTTR network-based remote fault diagnosis method of the present application;

[0038] Figure 2 is a node operation flowchart of the present application;

[0039] Figure 3 is a distributed state synchronization diagram of the present application;

[0040] Figure 4 is an AI-driven active monitoring and early warning flowchart of the present application;

[0041] Figure 5 is a smart fault diagnosis and repair system flowchart of the present application;

[0042] Figure 6 is an efficient large-scale data processing flowchart of the present application;

[0043] Figure 7 is the overall workflow diagram of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application.

[0045] In the embodiments of the present application, the words such as “exemplary” or “for example” are used to represent an example, illustration or description. Any embodiment or design scheme described as “exemplary” or “for example” in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as “exemplary” or “for example” are intended to present the relevant concept in a specific manner.

[0046] As shown in Figures 1 to 7 , the present application provides the following preferred embodiments:

[0047] Embodiment One

[0048] In order to solve the problems of accuracy and efficiency of remote fault diagnosis in the FTTR network, the embodiment proposes a remote fault diagnosis method based on the FTTR network. Specifically, the method comprises the following steps:

[0049] S100, a monitoring module for monitoring network status is deployed on each node of the FTTR network, each monitoring module is provided with a data acquisition unit and a state analysis unit, and each node collects network status data in real time through the data acquisition unit.

[0050] S200, the network status data includes network delay, packet loss rate, device temperature and bandwidth utilization, the network status data is preliminarily analyzed by the state analysis unit, and the historical data is compared to determine the change trend of the current state.

[0051] S300, each node publishes the state information of the current state to a shared topic through a message queue system, synchronizes with other nodes, and forms a global state view.

[0052] S400, the unsupervised learning algorithm is used to analyze the synchronized global state view, and the abnormal mode is identified, which is used to detect the extreme fluctuation of traffic and the sustained high delay behavior.

[0053] S500, a supervised learning model is used to analyze historical fault data, extract features, and analyze abnormal modes in combination with the extracted features, which is used to predict potential faults and give early warnings.

[0054] S600、When the early warning is triggered, the current fault is analyzed using a deep learning model, the fault reason is inferred, and a repair operation is generated and executed.

[0055] In this embodiment, a monitoring module is deployed on each FTTR network node, which includes a data collection unit and a state analysis unit, responsible for real-time collection of network state data. It should be understood that the sources of these data include network latency, packet loss rate, device temperature and bandwidth utilization, and these information is preliminarily analyzed by the data collection unit. The analysis involves comparison with historical data to determine the trend of the current state, and it should be noted that the identification of the trend is used to quickly respond to abnormal situations in network operation.

[0056] Further, each node publishes the current state information to a shared topic through the deployed message queue system, and synchronizes with other nodes to form a global state view. It can be understood that this design enables the distributed network architecture to share real-time data, thereby having a unified understanding of the running state of the entire network.

[0057] Further, in the abnormal pattern recognition process, this embodiment uses an unsupervised learning algorithm to analyze the global state view. It should be understood that this algorithm identifies and monitors extreme fluctuations in traffic and sustained high latency behavior, thereby providing early warning of abnormalities. Through this anomaly detection mechanism, abnormal conditions of the network can be identified at an early stage, preventing the problem from spreading.

[0058] Further, a supervised learning model is used to analyze and extract features from historical fault data to more accurately analyze abnormal patterns by combining these features. This process is used to predict potential faults and issue warnings, which can be understood as a supervised learning model that, through analysis of past faults, establishes the ability to identify potential future problems, providing a theoretical basis for subsequent diagnosis.

[0059] When the early warning is triggered, the deep learning model is used to analyze the current fault, infer the fault reason and generate the repair operation. It should be understood that the deep learning model can learn complex correlations from current and historical data, thereby accurately identifying and handling faults.

[0060] The benefit of this embodiment is that by integrating monitoring, data analysis, anomaly detection and repair functions, a complete remote fault diagnosis method is formed. This method effectively improves the operational stability and maintenance efficiency of FTTR networks in complex environments, providing an efficient and reliable network management solution. Through such a system architecture, operators can quickly identify and respond to network problems, thereby reducing the impact on user services and enhancing user experience.

[0061] Embodiment Two

[0062] To address the accuracy of state change detection and the reliability of trend analysis in FTTR networks, the embodiment further optimizes the state analysis unit by adopting a dual-threshold model to analyze and predict the trend of network state changes. The expression of the dual-threshold model is:

[0063]

[0064] It can be understood that this model sets two different thresholds T1 and T2 by analyzing the state change rate AM(t) to provide more accurate state monitoring. In this dual-threshold model, the state function S(t) is defined as follows: when AM(t) is greater than the upper threshold T1, the function output is 1, which indicates a significant state change and suggests a possible abnormal situation; when AM(t) is less than the lower threshold T2, the function output is 0, indicating that the network state is within the normal range of variation. State changes in the middle region are considered uncertain or within the normal fluctuation range and require further analysis in combination with other technical features and context.

[0065] Further, AM(t) is a dynamically changing value calculated based on the continuous changes in network latency, packet loss rate, device temperature, and bandwidth utilization provided by the data acquisition unit. It should be understood that selecting appropriate T1 and T2 values is closely related to the node's environment, historical data, and the specific needs of network operators. Therefore, T1 and T2 are not fixed but are gradually adjusted based on historical operating data and expected sensitivity to adapt to changing network conditions.

[0066] In addition, the dual-threshold model of the embodiment can effectively filter out false positives caused by short-term fluctuations, i.e., it can be understood as avoiding false triggering due to transient changes that are not abnormal. This enhances the accuracy of network management and fault diagnosis. Furthermore, the flexible design of this threshold setting can improve the adaptability of abnormal detection in different scenarios in real-time monitoring, providing a strong guarantee for the complexity and diversity of networks.

[0067] Through sensitive tracking of the above-mentioned change trend, this model provides a simple and effective monitoring method for the network. This is not limited to collecting and simply analyzing existing data, but also provides accurate predictions of future state changes, thereby enhancing the ability of fault warning and prevention. In specific applications, information from other monitoring and analysis modules in the network can be combined to form a more comprehensive assessment of network operating status.

[0068] The benefits of this embodiment are that by adopting such a model with double thresholds, it can be more flexible and accurate to analyze the state changes in the FTTR network. Especially in complex and variable network environments, the double threshold model provides a balanced solution: while ensuring sensitivity, it can effectively reduce the false positive rate. This ability plays an important role in ensuring the stability and reliable operation of the FTTR network. Through this embodiment, operators can accurately identify trends and potential failure points over a larger time span, thereby improving the efficiency of maintenance and response strategies.

[0069] Embodiment three

[0070] To solve the problem of accuracy of anomaly detection and fault identification in the FTTR network, this embodiment introduces a clustering method based on dynamic discriminant boundary in the unsupervised learning algorithm, which is expressed as:

[0071]

[0072] The expression of this embodiment realizes the dynamic distance calculation of data points x i to clustering center c j , where p is the adjustment parameter. It should be understood that the clustering method based on dynamic discriminant boundary can automatically adjust the clustering boundary according to the dynamic changes of input data, so as to adapt to new emerging abnormal situations.

[0073] Further, the unsupervised learning algorithm involves a dynamic distance formula in mathematics, which can flexibly change the tightness and sensitivity of clustering under different network environment conditions by adjusting p, thereby improving the identification ability of abnormal samples. When facing different quantities and types of network data, it can still maintain high adaptability.

[0074] In practical application, the unsupervised learning algorithm monitors the changes of network state data, calculates the distance between each data point and multiple possible clustering centers at the current time. The distance D(x i , C) between data points and clustering centers is not only the geometric distance, but also a measure of the significant difference between abnormal samples and normal samples. This enables the system to timely classify abnormal data into a new or existing abnormal class after detecting abnormal data, thereby improving the efficiency of fault identification.

[0075] In addition, the flexibility of the clustering method in boundary adjustment of the present embodiment is also reflected in its ability to adaptively learn new data sets. Specifically, when a new set of abnormal data points appears in the network, it can automatically identify that these points do not belong to any existing normal cluster, and calculate that these points can form a new abnormal cluster according to the self-adjusting mechanism. This dynamic adaptive ability enables the system to automatically learn without human intervention when facing new abnormal patterns, thereby improving the intelligent level of the entire remote fault diagnosis method.

[0076] It should be noted that in the FTTR network, it is necessary to combine the clustering algorithm with historical data and trend analysis. This not only helps to identify new abnormal patterns, but also improves future prediction models with the identified abnormal data characteristics. In this way, deep support for abnormal event identification can be provided in a wider application context.

[0077] Through the implementation of the present embodiment, the operation and maintenance personnel of the FTTR network can achieve accurate anomaly detection and fault prediction only by relying on automated algorithms in the case of usually requiring a large amount of manual analysis. This not only improves the efficiency of daily network management, but also reduces the response time to faults. In addition, the use of dynamic boundary discrimination effectively enhances the recognition accuracy of fault patterns under varying network conditions, enabling the system to continuously adapt to changing network anomalies, thereby maintaining the stability of the FTTR network.

[0078] Embodiment Four

[0079] In order to solve the real-time and consistency problem of state information synchronization of each node in the FTTR network, the present embodiment further refines the specific steps of each node publishing current state information to a shared topic and synchronizing through a message queue system. In this process, each node publishes JSON-formatted state information to a globally shared topic through an embedded message queue system at a fixed period, i.e., every 30 seconds. This design ensures that the state information of all nodes in the network can be timely and orderly shared and synchronized, thereby providing a reliable data foundation for generating a global state table.

[0080] Specifically, JSON format is selected as the transmission format of state information due to its lightweight and easy-to-parse characteristics. The state information of each node includes but is not limited to network delay, packet loss rate, device temperature, and bandwidth utilization, etc. These key indicators are encapsulated in JSON objects and transmitted through the message queue system. It should be understood that the message queue system not only provides a reliable message delivery mechanism, but also ensures the orderliness and non-loss of messages.

[0081] Further, the message queue system is designed to support high concurrency and high throughput to cope with the large number of nodes and massive data that may exist in the FTTR network. Each node not only publishes its own state information but also receives state information from other nodes in the message queue. This process is bidirectional, ensuring comprehensive synchronization of information. After receiving the state information of all relevant nodes, each node integrates the information according to the pre-set logic to generate a global state table.

[0082] It should be understood that the generation of the global state table is not just a simple data aggregation, but also involves some data processing and cleaning steps. For example, for repeated or incorrect data, nodes will filter and correct them according to the built-in logic to ensure the accuracy and consistency of the global state table. In addition, to improve the response speed of the system and reduce resource consumption, nodes use incremental updating when integrating information, i.e., only update the changed part instead of regenerating the entire state table.

[0083] Further, considering the complexity and variability of the FTTR network, the message queue system also includes a fault recovery and message retry mechanism. If a node fails to successfully publish or receive state information within a specified time window, the system will automatically trigger the retry mechanism to ensure the integrity of the information. It should be understood that this mechanism helps maintain the real-time and consistency of the entire network state, especially in the case of network fluctuations or temporary failures.

[0084] In addition, the embodiment also optimizes the load balancing strategy of the message queue system to prevent performance bottlenecks during network peak periods or large-scale data transmission. By dynamically adjusting the load distribution of the message queue, the system can smoothly cope with various network conditions, ensuring the efficiency and stability of state information synchronization.

[0085] The benefits of this embodiment are that by using fixed cycles and JSON format for state information publishing and synchronization, combined with the high reliability and high concurrency characteristics of the message queue system, nodes in the FTTR network can share and update the global state table in real time and accurately. This mechanism not only improves the accuracy and response speed of network state monitoring, but also enhances the fault tolerance and stability of the entire network. Through this embodiment, the FTTR network can maintain efficient operation in complex and variable environments, providing strong support for network operation and management.

[0086] Embodiment Five

[0087] To address the accuracy and stability of fault diagnosis in FTTR networks, the present embodiment further optimizes the construction of supervised learning models by adopting a multi-model collaboration method based on ensemble learning. This method integrates the outputs of multiple sub-models and assigns corresponding weights based on the performance of each sub-model to achieve more accurate fault prediction. Specifically, the multi-model collaboration in the present embodiment is achieved through the following expression:

[0088] f(x) = ∑ k w k ·h k (x)

[0089] where f(x) represents the final total prediction result of the fault, h k (x) is the output of the kth sub-model, and w k is the weight corresponding to the sub-model. Through this mechanism, the present embodiment can provide more adaptive and stable fault diagnosis capabilities when facing complex and variable network environments.

[0090] Further, the selection of sub-model h k (x) in the present embodiment includes various machine learning algorithms such as decision trees, support vector machines, neural networks, etc. These algorithms each have different advantages and application scenarios, and by integrating these diverse sub-models, the system can analyze and predict network status from multiple perspectives. It should be understood that different sub-models may exhibit different accuracy when dealing with different types of faults, for example, decision tree models may perform well in handling discrete features, while neural networks have advantages in handling complex nonlinear relationships. Therefore, multi-model collaboration can provide more balanced and comprehensive diagnostic results in various situations.

[0091] In practical applications, the output h k (x) of each sub-model is first standardized to ensure the comparability and consistency of the outputs between different models. Subsequently, the system dynamically adjusts the corresponding weight w k of each sub-model based on its performance on historical data. The determination of weight w k is achieved by evaluating the error of the sub-model on the training set and validation set. It can be understood that the sub-model with better performance will be assigned a higher weight, while the sub-model with poor performance will have a relatively lower weight. This mechanism ensures that the final integrated prediction f(x) can maximize the use of the advantages of each sub-model, thereby improving the accuracy of the overall prediction.

[0092] Further, to ensure the real-time and adaptability of the model, the embodiment also introduces an online learning mechanism. Specifically, when new failure modes or abnormal data appear in the network, the system will automatically trigger the retraining and weight adjustment of the model. By continuously updating the weights and outputs of the sub-models, the system can maintain high diagnostic accuracy in dynamically changing network environments. In addition, to prevent overfitting of the model and improve generalization ability, the embodiment also uses cross-validation and regularization techniques during training. These techniques can achieve a good balance between model complexity and prediction accuracy, further enhancing the practicality of the system.

[0093] It should be understood that the cooperation between sub-models is not limited to simple weighted summation. In some cases, the system can further enhance the performance of the model through more complex fusion strategies such as stacking or blending. These advanced fusion methods introduce meta-models to perform secondary learning on the outputs of sub-models, thereby obtaining more optimal prediction results in certain specific scenarios.

[0094] The benefit of this embodiment is that by using an integrated learning-based multi-model cooperation method, the fault diagnosis capability of the FTTR network has been significantly improved. This method not only effectively integrates the advantages of multiple sub-models, but also dynamically adjusts the weights of each sub-model according to the actual network conditions, thereby providing more accurate and reliable fault prediction. Through this embodiment, the FTTR network can maintain efficient and stable fault diagnosis capability when facing complex and variable operating environments, providing strong technical support for network operation and management.

[0095] Embodiment Six

[0096] To solve the timing correlation and precision problem of fault prediction in FTTR network, the embodiment further refines the time series analysis method in fault prediction and introduces the Autoregressive Integrated Moving Average (ARIMA) model. The introduction of this model aims to capture the timing dependence structure in time series data and improve the prediction ability of network faults. Specifically, the ARIMA model models the time series through the following expression:

[0097]

[0098] where Y(t) represents the observation value of the time series at time t, c represents the mean value of the time series, φ i (i = 1, 2, …, p) is the autoregressive coefficient, representing the influence of Y(t) values at the past i time points on the current Y(t), Y(t-i) represents the observation value of the time series at time t-i, θ j(j = 1, 2, …, q) represents the influence of the error term ε at the past j time points on the current Y(t), ε(t-j) represents the random error term of the time series at time t-j, and ε(t) represents the random error term of the time series at time t.

[0099] Further, before applying the ARIMA model, the time series data in the FTTR network needs to be preprocessed. This includes the stationarity test and difference processing of the data. Specifically, if the time series data presents a non-stationary trend, it can be transformed into a stationary sequence through difference operation. It needs to be understood that stationarity is a prerequisite for applying the ARIMA model, and only when the data is stationary, the model can accurately capture the time series dependence in the time series.

[0100] After determining the stationarity of the data, the parameters p, d, q of the model need to be estimated and selected. Among them, p is the order of the autoregressive part, d is the difference number, and q is the order of the moving average part. The selection of these parameters is usually analyzed and determined through the autocorrelation function (ACF) and partial autocorrelation function (PACF) graphs. By observing the characteristics of the ACF and PACF graphs, the most suitable parameter combination can be determined, and the best ARIMA model can be constructed.

[0101] Further, in the model training process, the maximum likelihood estimation (MLE) method is used to estimate the parameters of the ARIMA model. This process maximizes the likelihood function so that the model can best fit the training data. It needs to be understood that the fitting effect of the model directly affects the accuracy of fault prediction, so multiple experiments and tuning need to be performed during the training process to ensure the optimal performance of the model.

[0102] In order to improve the generalization ability of the model, this embodiment also introduces cross-validation and regularization techniques. Specifically, by dividing the time series data into sliding windows, multiple validations and evaluations of the model can be achieved. In addition, by introducing a regularization term, overfitting of the model can be effectively prevented, so that the model can still maintain high prediction accuracy when facing new data.

[0103] Further, considering the complexity and variability of the FTTR network, this embodiment also optimizes the online updating mechanism of the model. Specifically, when new fault data or abnormal patterns appear in the network, the system will automatically trigger the retraining and parameter adjustment of the model. This mechanism ensures that the ARIMA model can adapt to changes in network state in real time and maintain high prediction performance.

[0104] It should be understood that the application of ARIMA model is not limited to the prediction of a single time series, but can also be combined with other features for multivariate time series analysis. For example, multiple key indicators in the network (such as bandwidth utilization, packet loss rate, delay, etc.) can be used as multivariate input to build a multivariate ARIMA model. By considering multiple factors, the comprehensiveness of fault prediction can be further improved.

[0105] The benefit of this embodiment is that the fault prediction capability of the FTTR network is improved by introducing the autoregressive integrated moving average model. The ARIMA model can capture the time series dependence structure in the time series data and provide accurate fault prediction results through reasonable parameter selection and model optimization.

[0106] Embodiment Seven

[0107] To solve the real-time response and historical fault pattern reproduction problems in FTTR network fault diagnosis, this embodiment further optimizes the application of deep learning model and introduces a deep learning pattern matching process based on long short-term memory network (LSTM). This process aims to analyze the current fault features through LSTM network and optimize the model parameters using error backpropagation mechanism, so as to realize the rapid response to similar historical faults. At the same time, this embodiment also refines the fault repair operation, generates the optimal repair path through the fault pattern library, and automatically adjusts the parameters or triggers the maintenance prompt combined with the device API, to ensure that the fault can be handled in a timely and effective manner.

[0108] Specifically, LSTM network, as a special recurrent neural network (RNN), has the advantage of processing and predicting time series data, especially in dealing with long time dependence problems. In the fault diagnosis of FTTR network, LSTM network can effectively capture the time dependence of network state and realize effective learning and reproduction of historical fault features through its internal memory unit and gating mechanism. The input of LSTM network includes multi-dimensional features of current network state, such as bandwidth utilization, packet loss rate, delay, etc. These features are time series modeled through LSTM layer and feature fused and output through fully connected layer.

[0109] Further, the training process of the LSTM network is achieved through the error backpropagation mechanism to optimize the parameters. Specifically, the error between the model output and the actual fault label is measured by the loss function, and then the error is transmitted layer by layer through the backpropagation algorithm, and the network weight is adjusted according to the gradient descent method. This process is repeated until the loss function of the model converges, thereby optimizing the parameters of the LSTM network. It should be understood that the error backpropagation mechanism can effectively adjust the weights of the LSTM network, so that the model can quickly respond and make accurate diagnosis when facing similar historical faults.

[0110] To further improve the accuracy and efficiency of fault diagnosis, the embodiment also introduces a fault mode library. The fault mode library stores the features of historical faults and the corresponding repair operations, and through pattern matching with the current fault features, similar faults can be quickly located and the optimal repair path can be generated. It should be understood that the construction of the fault mode library not only includes historical fault data, but also involves the extraction and classification of fault features, which can be represented by the output of the LSTM network. By comparing the current fault features with the historical data in the fault mode library, the system can quickly identify the most similar fault mode and process it according to the pre-set repair path.

[0111] Further, the matching results of the fault mode library will be used to guide specific repair operations. According to the severity and type of the fault, the system will generate corresponding repair suggestions and automatically adjust the relevant parameters in combination with the device API. For example, for bandwidth bottleneck problems, the system can adjust the transmission power and channel selection of the router through the API; for device overheating problems, the system can trigger fan acceleration and load balancing operations. These operations are implemented through the device API to ensure the timeliness and effectiveness of the repair measures. In addition, for faults that require human intervention, the system will generate maintenance prompts to notify the maintenance personnel for corresponding processing.

[0112] It should be understood that the use of device API is not limited to parameter adjustment, but also includes real-time monitoring and feedback of device status. By continuously monitoring the running status of the device, the system can timely discover potential problems and perform preventive maintenance before the fault occurs. This mechanism helps to improve the stability and reliability of the FTTR network, reducing network interruption and performance degradation caused by device faults.

[0113] The benefits of this embodiment are that by introducing an LSTM-based deep learning pattern matching process, the fault diagnosis capability of the FTTR network is significantly improved. The LSTM network can effectively capture the time sequence dependence structure in the time sequence data and optimize the model parameters through the error back propagation mechanism, thereby realizing fast response to similar historical faults. At the same time, by introducing the fault mode library, the system can quickly locate similar faults and generate the optimal repair path, and combine with the device API to realize automatic parameter adjustment and maintenance prompt.

[0114] Embodiment Eight

[0115] To solve the problem of lack of feedback mechanism after fault repair in the FTTR network, thereby causing the deep learning model to be unable to continuously optimize, this embodiment further refines the effect feedback process after fault repair, and introduces a record and analysis mechanism for actual effect data. This mechanism aims to feed back the actual effect data after repair to the deep learning model to adjust and optimize the parameter library, thereby improving the decision-making efficiency of subsequent fault diagnosis and repair operations. Through this optimization, the fault diagnosis system of the FTTR network can realize self-learning and continuous improvement, gradually improving its adaptability and accuracy in complex network environments.

[0116] Specifically, after the fault is repaired, the system will automatically collect the actual effect data related to the fault. These data include the specific content of the repair operation, the network performance indicators after repair, and any possible user feedback. For example, if the fault involves a bandwidth bottleneck, the actual effect data may include bandwidth utilization, data transmission rate, and network latency indicators after repair. These data are obtained by the system through a pre-set data collection mechanism and are preliminarily cleaned and arranged to ensure the accuracy and consistency of the data.

[0117] Further, the collected actual effect data will be fed back to the deep learning model for parameter adjustment and optimization of the model. Specifically, the deep learning model (such as the aforementioned LSTM network) optimizes the model parameters not only in the training stage, but also in actual application through the error back propagation mechanism. The actual effect data as new training samples are used to update the weights and biases of the model, so that the model can better adapt to new changes and new patterns in the network. It should be understood that this feedback mechanism makes the deep learning model no longer a static decision-making tool, but an intelligent system that can dynamically learn and improve.

[0118] To implement this feedback mechanism, the present embodiment also introduces the concept of a parameter library. The parameter library is used to store all parameters and hyperparameters of the deep learning model, including the weights, biases, learning rate, optimizer state of the LSTM network, etc. By feeding the actual effect data back to the parameter library, the system can fine-tune and optimize the parameters of the model, gradually improving the prediction and decision-making capabilities of the model. It should be understood that the optimization of the parameter library does not rely solely on single feedback data, but through the continuous collection and analysis of a large amount of actual effect data, the overall optimization of the model is gradually realized.

[0119] Further, the optimization process of the parameter library also involves the adjustment of hyperparameters. For example, the number of layers, the number of hidden units, the learning rate, etc. of the LSTM network, all of which will affect the performance and convergence speed of the model. Through the feedback of actual effect data, the system can automatically adjust these hyperparameters to find the optimal model configuration. For example, if it is found that the convergence speed of the model is too slow, the system may try to increase the learning rate; if it is found that the model has overfitting phenomenon, the system may try to reduce the number of layers or the number of hidden units of the LSTM network. These adjustments are realized through automated scripts, ensuring the real-time and efficiency of model optimization.

[0120] It should be understood that the feedback of actual effect data and the optimization of parameter library not only aims to improve the accuracy of single fault diagnosis, but more importantly, to improve the generalization ability of the entire fault diagnosis system. Through continuous learning and optimization, the system can maintain high prediction and decision-making capabilities when facing new fault patterns and complex network environments. In addition, the optimization of the parameter library also involves retraining of historical data to ensure that the model can continuously learn and improve from historical data.

[0121] The additional features of the present embodiment also include dynamic updating of the fault pattern library. The fault pattern library not only stores historical fault characteristics and repair operations, but also continuously updates and expands its content through the feedback of actual effect data. Specifically, the actual effect data after each fault repair will be used to update the fault pattern library, so that the fault patterns in the library are more rich and comprehensive. This dynamic updating mechanism ensures the real-time and effectiveness of the fault pattern library, so that the system can provide more accurate and efficient repair suggestions when facing similar faults.

[0122] The benefits of the present embodiment are that, by introducing the feedback mechanism of actual effect data, the fault diagnosis and repair system of the FTTR network realizes self-learning and continuous optimization. The deep learning model continuously adjusts and optimizes the parameter library through the feedback of actual effect data, thereby improving the decision-making efficiency of subsequent fault diagnosis and repair operations. At the same time, the dynamic updating mechanism of the fault pattern library enables the system to continuously expand and enrich its fault patterns, providing more comprehensive and accurate repair suggestions.

[0123] Example Nine

[0124] To address the issues of real-time performance, accuracy, and automation level of fault diagnosis systems in FTTR networks, the present embodiment proposes a remote fault diagnosis system based on FTTR networks. The system integrates multiple functional modules such as data acquisition, state analysis, anomaly detection, fault prediction, fault diagnosis, and repair suggestion and execution, achieving full-process automation and intelligentization from data acquisition to fault handling. Through close collaboration between modules, the system can promptly identify problems, predict risks, diagnose fault causes, and automatically perform repair operations or trigger maintenance prompts in complex network environments.

[0125] Specifically, the data acquisition module in the present embodiment is deployed at each FTTR node, aiming to acquire network state information in real time. These nodes are distributed throughout the FTTR network and can continuously monitor key network indicators such as bandwidth utilization, latency, and packet loss rate. The data acquisition module ensures the high accuracy and real-time performance of the acquired network state data through high-precision sensors and data acquisition algorithms. These data are then transmitted to the state analysis module for preliminary analysis and distributed synchronization. It should be understood that the distributed synchronization mechanism ensures data consistency between nodes, allowing the system to maintain a unified view of network state across different nodes.

[0126] Further, the state analysis module performs preliminary processing on the collected network state data, including data cleaning, outlier detection, and data normalization. These processing steps aim to improve data quality and consistency, providing a reliable data foundation for subsequent anomaly detection and fault diagnosis. The state analysis module also uses distributed computing technology to perform parallel processing of data on different nodes, improving the efficiency and scalability of data analysis. It should be understood that the output of the state analysis module will be provided as input to the anomaly detection module and the fault prediction module for further analysis.

[0127] Further, the anomaly detection module in the present embodiment employs an unsupervised learning algorithm to identify abnormal patterns from normal network behavior. The unsupervised learning algorithm learns the normal patterns of network state data and establishes a baseline behavior model. When the actual network state deviates from the model, the system identifies potential network problems and marks them as abnormal events. The advantage of this unsupervised learning method is that it does not require a large amount of labeled data and can automatically discover new abnormal behavior without knowing the fault patterns. Further, the output of the anomaly detection module will be passed to the fault prediction module for further risk prediction.

[0128] Further, the fault prediction module in this embodiment adopts a supervised learning algorithm to predict possible network risks and provide early warning signals. The supervised learning algorithm is trained through historical fault data, learning typical characteristics and patterns before the occurrence of faults. In actual operation, the module can predict the risk of future possible faults according to the current network state data and issue early warning signals in advance. These early warning signals help the operation and maintenance personnel to take timely measures to prevent the further expansion and deterioration of the fault. It should be understood that the accuracy of the fault prediction module depends on high-quality training data and reasonable model selection, so in this embodiment, model selection and parameter optimization is a continuous process.

[0129] Further, the fault diagnosis module is based on a deep learning algorithm to analyze the occurred faults in detail and perform fault pattern matching. The deep learning algorithm can handle complex nonlinear relationships and is suitable for modeling multi-dimensional and multi-variable network state data. In this embodiment, the fault diagnosis module uses deep learning models such as Long Short-Term Memory (LSTM) to analyze the causes of faults in depth and match patterns through a fault pattern library. The fault pattern library stores a large amount of historical fault data and corresponding repair operations, which are compared with the current fault characteristics to quickly locate similar faults and analyze the root cause of the fault.

[0130] Further, the repair suggestion and execution module aims to generate the optimal repair suggestion and automatically repair or trigger maintenance prompts through the system interface. Specifically, the repair suggestion and execution module generates the optimal repair path and operation suggestion based on the fault diagnosis results and historical data in the fault pattern library. If the fault can be automatically repaired through the system interface, such as adjusting router parameters or optimizing bandwidth allocation, the system will automatically perform the relevant operations; if the fault requires human intervention, the system will generate detailed maintenance prompts to notify the operation and maintenance personnel for processing. It should be understood that the automation level of the repair suggestion and execution module directly affects the efficiency and accuracy of fault handling, so in this embodiment, the stability and reliability of the system interface have been fully considered and optimized.

[0131] The additional features of this embodiment also include optimizing the data flow and control flow between modules. By introducing efficient data transmission protocols and distributed computing technologies, the system can achieve efficient collaboration and data sharing between modules. In addition, the system also introduces a fault backtracking mechanism, which automatically records the processing process and results after the completion of fault handling and feeds back to the fault prediction and diagnosis module for further optimization and adjustment of the model.

[0132] Embodiment Ten

[0133] To address the execution efficiency and reliability issues of the remote fault diagnosis method in FTTR networks during actual deployment, the present embodiment proposes a computer-readable storage medium. The storage medium stores a computer program that, when executed on an electronic device, can implement the remote fault diagnosis method based on FTTR networks as in the previous embodiments. Through this optimization, the system not only achieves efficient fault diagnosis and repair at the hardware level, but also ensures the stability and flexibility of program execution at the software level, thereby improving the overall performance of the entire fault diagnosis system.

[0134] Specifically, the computer program in the present embodiment is designed as a modular structure, with each module corresponding to a specific step in the fault diagnosis process. The additional features of the present embodiment also include exception handling and logging mechanisms during program execution. To ensure the stability and reliability of the program, various exception handling mechanisms are introduced into the system, which can capture and handle exceptions in time when errors occur during program execution, avoiding program crashes or data loss. At the same time, detailed logging mechanisms are used to record each key step of program execution, including data collection, state analysis, anomaly detection, fault prediction, fault diagnosis, and repair execution. These logs not only help maintenance personnel to troubleshoot, but also provide data support for subsequent program optimization and model updates.

[0135] The benefits of the present embodiment are that, by optimizing the execution flow and module design of the computer program, the remote fault diagnosis system of FTTR networks can achieve efficient, stable, and reliable execution at the software level. The modular program design and the introduction of distributed computing technology improve the scalability and execution efficiency of the system. At the same time, the optimization of exception handling and logging mechanisms enables the system to maintain good maintainability in the face of complex network environments and sudden failures.

Claims

1. A method for remote fault diagnosis based on an FTTR network, characterized in that, The method comprises the following steps: deploying a monitoring module for monitoring network status on each node of the FTTR network, each monitoring module being provided with a data acquisition unit and a status analysis unit, each node collecting network status data in real time through the data acquisition unit; the network status data including network delay, packet loss rate, device temperature and bandwidth utilization, the network status data being preliminarily analyzed by the status analysis unit and compared with historical data to determine the change trend of the current status; each node publishing the status information of the current status to a shared topic through a message queue system to synchronize with other nodes to form a global status view; using an unsupervised learning algorithm to analyze the synchronized global status view to identify abnormal patterns for detecting extreme fluctuations in traffic and sustained high delay behavior; using a supervised learning model to analyze historical fault data, extract features, and analyze the abnormal patterns combined with the extracted features to predict potential faults and provide early warnings; when the early warning is triggered, using a deep learning model to analyze the current fault, reasoning the fault cause to generate and perform a repair operation; the status analysis unit using a double threshold model for the analysis and prediction of the change trend, the expression of the double threshold model being: where ΔM(t) is the rate of state change, and T1 and T2 are threshold values. 2.The FTTR network-based remote fault diagnosis method of claim 1, wherein, the unsupervised learning algorithm including a clustering method based on a dynamic discriminant boundary, the expression being: where D is the distance of the data point x i to the cluster center c j p is a tuning parameter. 3.The FTTR network-based remote fault diagnosis method of claim 1, wherein, the step of each node publishing the status information of the current status to a shared topic through a message queue system to synchronize with other nodes comprises: each node publishing JSON format status information to a global shared topic through a message queue system every 30 seconds, and integrating other node information to generate a global status table after receiving. 4.The FTTR network-based remote fault diagnosis method of claim 1, wherein, the supervised learning model is based on an ensemble learning method, including using multiple model cooperation, the expression of the multiple model cooperation being: f(x) =∑ k w k ·h k (x), where f(x) is the total failure prediction, h k (x) is the output of each sub-model, and w k is the weight corresponding to the sub-model. 5.The FTTR network-based remote fault diagnosis method of claim 4, wherein, the prediction of the fault also uses a time series analysis method, using an autoregressive integrated moving average model, the expression of the autoregressive integrated moving average model being: where Y(t) denotes the observation of the time series at time t, c denotes the mean of the time series, φ i (i = 1, 2,..., p) are autoregressive coefficients, representing the influence of the Y(t) values at the past i time points on the current Y(t), Y(t-i) denotes the observation of the time series at time t-i, θ j (j = 1, 2,..., q) represent the influence of the error terms ε at the past j time points on the current Y(t), ε(t-j) denotes the random error term of the time series at time t-j, and ε(t) denotes the random error term of the time series at time t. 6.The FTTR network-based remote fault diagnosis method of claim 1, wherein, the use of a deep learning model to analyze the current fault is based on a long short-term memory (LSTM) deep learning pattern matching process, which optimizes parameters through error back propagation to quickly respond to similar historical faults; the repair operation is generated according to the fault mode library to generate an optimal repair path, and combined with the device API to automatically adjust parameters or trigger maintenance prompts. 7.The FTTR network based remote fault diagnosis method of claim 6, wherein, After the fault is repaired, the actual effect data is recorded and fed back to the deep learning model to adjust and optimize the parameter library for improving the decision-making efficiency of subsequent operations.

8. A remote fault diagnosis system based on FTTR network, for implementing the remote fault diagnosis method based on FTTR network according to any one of claims 1-7, characterized in that, The system comprises: a data acquisition module deployed on each FTTR node for real-time acquisition of network status information; a status analysis module that acquires network status data provided by the data acquisition module, preliminarily analyzes the network status data, and distributes the network status data; an anomaly detection module that uses an unsupervised learning algorithm to detect abnormal patterns and identify potential network problems; a fault prediction module that uses a supervised learning algorithm to predict possible network risks and provide early warning signals; a fault diagnosis module that uses a deep learning algorithm to analyze fault causes and perform fault pattern matching; The repair suggestion and execution module generates an optimal repair suggestion and performs automatic repair or triggers a maintenance prompt through a system interface.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program product comprises a computer readable storage medium having computer readable program code embodied therewith, the computer readable program code configured such that, upon execution by the electronic device, the electronic device is caused to perform the method of any one of claims 1 to 7.

Images