Operator DICT project management and control system and method based on machine learning

Abstract

The invention discloses an operator DICT project management and control system and method based on machine learning, belongs to the technical field of communication network operation and maintenance, and aims to solve the technical problem of how to realize automatic monitoring and fault prediction of each stage process of a DICT project and improve the operation and maintenance efficiency and accuracy. Comprising a data acquisition layer which acquires multi-source data through an edge operation and maintenance box and uploads the preprocessed multi-source data to a data fusion layer; the data fusion layer is used for carrying out standardization processing and feature extraction on the preprocessed multi-source data; the intelligent analysis layer is used for performing fault prediction analysis, root cause positioning and optimization suggestion generation on the time series data based on the model in the model library; and the management and control display layer issues the early warning notice to the executor and is used for displaying an execution result log recorded by the executor and returning the execution result log to the data acquisition layer to form closed-loop management and control.

Description

Technical Field

[0001] This invention relates to the field of communication network operation and maintenance technology, specifically to a machine learning-based operator DICT project management system and method. Background Technology

[0002] With the rapid development of information technology, the scale of DICT projects (leased lines, IT systems, and smart IoT devices) undertaken by telecom operators is constantly expanding. However, during project implementation, operators face the challenge of effectively monitoring existing IT systems. Because customers often do not allow operators to remotely monitor systems, or lack sufficient monitoring tools, operators find it difficult to comprehensively supervise and manage project systems. This insufficient monitoring not only affects system stability and reliability but may also lead to the inability to promptly locate and handle faults, thereby impacting the normal operation of services.

[0003] Traditional monitoring methods primarily rely on manual acquisition of work order information, using work orders to drive feedback via telephone communication. This requires manually compiling leased line information and entering it into the operator's leased line operation and maintenance system. This is not only time-consuming and labor-intensive but also susceptible to human factors, leading to low operational efficiency and insufficient accuracy. Furthermore, the lack of comprehensive control and analysis of data across the entire DICT project results in low overall information utilization, with a significant amount of data that could improve operational efficiency remaining underutilized.

[0004] How to automate the monitoring and fault prediction of each stage of a DICT project, and improve operational efficiency and accuracy, is a technical problem that needs to be solved. Summary of the Invention

[0005] The technical objective of this invention is to address the above-mentioned shortcomings by providing a machine learning-based operator DICT project management system and method to solve the technical problem of how to achieve automated monitoring and fault prediction of each stage of the DICT project, and improve operation and maintenance efficiency and accuracy.

[0006] In a first aspect, the present invention provides a machine learning-based operator DICT project management and control system, comprising a data acquisition layer, a data fusion layer, an intelligent analysis layer, and a management and control display layer;

[0007] The data acquisition layer includes an edge operation and maintenance box deployed on the customer side. The edge operation and maintenance box supports multi-purpose protocol adaptation technology and is used to collect multi-source data, including leased line network management data, IT equipment performance indicators, IoT sensor data and application performance indicators. The multi-source data is preprocessed and then uploaded to the data fusion layer.

[0008] The data fusion layer is used to standardize and extract features from the preprocessed multi-source data to obtain time-series data with a predetermined structure. The time-series data is then persisted and sent to the intelligent analysis layer.

[0009] The intelligent analysis layer is equipped with a model library built on machine learning algorithms. This library is used to perform fault prediction analysis, root cause localization, and optimization suggestion generation on time series data based on the models in the library. It also optimizes the models in the library based on reinforcement learning.

[0010] The control and display layer is used to display multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. It is used to generate early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions and send the early warning notifications to the executors. It is used to display the execution result logs recorded by the executors and to feed the execution result logs back to the data acquisition layer to form a closed-loop control.

[0011] Preferably, the edge maintenance box collects leased network management data via the SNMP protocol, including port traffic and bit error rate of the leased network management system; it collects IT equipment performance indicators via IPMI / Redfish or proxy methods, including CPU and memory utilization of the IT equipment; and it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance indicators via browser simulation or traffic mirroring, including response time and error rate.

[0012] The edge operation and maintenance box is used to clean, compress, encrypt, and securely isolate the collected multi-source data, and encapsulate it into a data stream with a fixed format. The data stream is then uploaded to the data fusion layer through a two-way TLS channel.

[0013] As a preferred option, the data fusion layer is used to perform data cleaning, data classification, and feature encoding on multi-source data. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data to obtain time-series data with a predetermined structure. The time-series data is then persisted to a relational database or NoSQL database.

[0014] As a preferred option, the model library deploys various models built based on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data, and artificial intelligence technologies including deep learning and natural language processing are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1, and the optimal model is determined through cross-validation technology.

[0015] The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models.

[0016] The optical module degradation prediction model takes the time series data corresponding to the leased line network management data as input, and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp and degradation prediction results for uncertain areas.

[0017] The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems.

[0018] The root cause localization model takes anomaly detection results and time series data as input, and is used to infer the fault propagation path based on graph neural networks and attention mechanisms. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization.

[0019] The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

[0020] As a preferred option, the control and management display layer is configured with cross-domain fault location service and contingency plan self-execution service. The cross-domain fault location service associates logs from multiple systems and data from multiple sources, and the contingency plan self-execution service generates contingency plans based on optimization suggestions and automatically executes the contingency plans.

[0021] Secondly, the present invention provides a machine learning-based method for managing and controlling DICT projects for telecom operators, including data collection, data fusion, intelligent analysis, and management and control display.

[0022] Data Acquisition: Collect multi-source data, including dedicated line network management data, IT equipment performance indicators, IoT sensor data, and application performance indicators, through the edge operation and maintenance box; preprocess the multi-source data; and upload the preprocessed multi-source data to the data fusion layer.

[0023] Data fusion: Standardize and extract features from preprocessed multi-source data to obtain time-series data with a predetermined structure, persist the time-series data and send it to the intelligent analysis layer;

[0024] Intelligent analysis: Based on models in the model library, it performs fault prediction analysis, root cause localization, and optimization suggestions on time series data, and optimizes models in the model library based on reinforcement learning;

[0025] Control and Management Display: Displays multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. Generates early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions, and sends the early warning notifications to the implementers. Displays the execution result logs recorded by the implementers and feeds the execution result logs back to the data acquisition layer, forming a closed-loop control.

[0026] Preferably, the edge maintenance box collects leased network management data via the SNMP protocol, including port traffic and bit error rate. It also collects IT equipment performance metrics via IPMI / Redfish or a proxy, including CPU and memory usage. Furthermore, it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance metrics, including response time and error rate, via browser simulation or traffic mirroring.

[0027] The edge operation and maintenance box cleans, compresses, encrypts, and securely isolates the collected multi-source data, and encapsulates it into a data stream in a fixed format. The data stream is then uploaded to the data fusion layer via a bidirectional TLS channel.

[0028] As a preferred approach, during data fusion, multi-source data undergoes data cleaning, data classification, and feature encoding. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data, resulting in time-series data with a predetermined structure. This time-series data is then persisted to a relational database or NoSQL database.

[0029] As a preferred option, the model library deploys various models built based on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data, and artificial intelligence technologies including deep learning and natural language processing are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1, and the optimal model is determined through cross-validation technology.

[0030] The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models.

[0031] The optical module degradation prediction model takes the time series data corresponding to the leased line network management data as input, and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp and degradation prediction results for uncertain areas.

[0032] The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems.

[0033] The root cause localization model takes anomaly detection results and time series data as input, and is used to infer the fault propagation path based on graph neural networks and attention mechanisms. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization.

[0034] The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

[0035] As a preferred option, the management and control display provides cross-domain fault location services and contingency plan self-execution services. The cross-domain fault location service associates logs from multiple systems and data from multiple sources, while the contingency plan self-execution service generates contingency plans based on optimization suggestions and automatically executes the contingency plans.

[0036] The machine learning-based operator DICT project management system and method of this invention has the following advantages: It constructs a comprehensive monitoring system that simultaneously covers network leased lines, IT equipment, smart IoT devices, and application software, breaking down data silos; it designs an edge maintenance box as a unified data collection and security isolation carrier to solve customer-side access security risks and resource limitations; it establishes a machine learning-based fault prediction and root cause localization model library to achieve early warning and accurate diagnosis of system anomalies; it improves fault localization efficiency and reduces system downtime through multi-source data fusion analysis and collaborative judgment; it introduces a model self-optimization mechanism to continuously adjust and correct the judgment model based on historical data, improving prediction accuracy; and through layered processing and collaborative analysis, it reduces threat-assisted assessment efficiency from 10+ minutes to 5 minutes, increases average risk handling efficiency from 30 minutes to 6 minutes, and reduces system downtime by more than 30%. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] The invention will be further described below with reference to the accompanying drawings.

[0039] Figure 1This is a schematic diagram of the structure of a machine learning-based operator DICT project management system, as shown in Example 1.

[0040] Figure 2 Example 1 illustrates a complete process of a machine learning-based operator DICT project management system, from data collection, anomaly detection, root cause analysis to handling and feedback. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0042] This invention provides a machine learning-based operator DICT project management system and method to address the technical problem of how to achieve automated monitoring and fault prediction of each stage of a DICT project, and improve operational efficiency and accuracy.

[0043] Example 1:

[0044] This invention discloses a machine learning-based operator DICT project management and control system, comprising a data acquisition layer, a data fusion layer, an intelligent analysis layer, and a management and control display layer.

[0045] The data acquisition layer includes an edge operation and maintenance box deployed on the customer side. The edge operation and maintenance box supports multi-purpose protocol adaptation technology and is used to collect multi-source data, including leased line network management data, IT equipment performance indicators, IoT sensor data, and application performance indicators. The multi-source data is preprocessed and then uploaded to the data fusion layer.

[0046] As a concrete implementation of the edge operations and maintenance box, it collects leased network management data via the SNMP protocol, including port traffic and bit error rate. It also collects IT equipment performance metrics via IPMI / Redfish or proxy methods, including CPU and memory utilization. Furthermore, it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance metrics, including response time and error rate, via browser simulation or traffic mirroring. The edge operations and maintenance box performs data cleaning, compression, encryption, and security isolation on the collected multi-source data, uniformly encapsulating it into a fixed-format data stream, and uploading the data stream to the data fusion layer via a bidirectional TLS channel.

[0047] In this embodiment, the edge maintenance box serves as a secure carrier for customer-side data collection and preprocessing. This device employs multi-protocol adaptation technology, supporting various interface protocols such as SNMP for leased networks, MQTT for IoT devices, and APIs for application systems. It possesses edge computing capabilities, enabling local data cleaning, compression, and encryption. Secure isolation is achieved through firewalls and two-way certificate authentication to ensure secure data transmission. The data acquisition layer is responsible for collecting monitoring data from multiple dimensions, including network, IT equipment, and smart IoT devices. For monitoring leased network departments, it accesses network performance management data from the operational network and extracts the corresponding leased line information for customer analysis. By analyzing the traffic of downstream ports, such as significantly lower than long-term traffic or no traffic generation on downstream ports, leased line faults can be detected promptly. For monitoring customer-side IT equipment, smart IoT devices, and application software, the edge IT maintenance box serves as a carrier to achieve secure data transmission and analysis. The IT maintenance box uses hardware devices, including CPU, memory, storage, and network interfaces, and receives and transmits data through the network interface.

[0048] The data fusion layer is used to standardize and extract features from the preprocessed multi-source data to obtain time-series data with a predetermined structure. The time-series data is then persisted and sent to the intelligent analysis layer.

[0049] As a specific implementation of the data fusion layer, it is used to perform data cleaning, data classification, and feature encoding on multi-source data. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data to obtain time-series data with a predetermined structure. The time-series data is then persisted to a relational database or NoSQL database.

[0050] In this embodiment, the data fusion layer performs standardization and feature extraction on the collected data, including data cleaning, data classification, and feature encoding. Data cleaning includes removing deactivated equipment, invalid data, outliers, etc.; data classification merges different types of data into a unified dimension; feature encoding converts non-numerical data into numerical data, such as converting string data into numeric codes. This layer stores the data in a data storage layer, using relational databases or NoSQL databases, to ensure data persistence and high availability.

[0051] The intelligent analysis layer is equipped with a model library built on machine learning algorithms. This library is used to perform fault prediction analysis, root cause localization, and optimization suggestion generation on time series data based on the models in the library. Furthermore, it optimizes the models in the library based on reinforcement learning.

[0052] As a specific implementation, the model library deploys various models built based on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data. Artificial intelligence technologies, including deep learning and natural language processing, are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1. The optimal model is determined through cross-validation.

[0053] The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models.

[0054] The optical module degradation prediction model takes time-series data corresponding to the leased line network management data as input and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp, and degradation prediction results for uncertain regions.

[0055] The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems.

[0056] The root cause localization model takes anomaly detection results and time series data as input, and uses graph neural networks and attention mechanisms to infer fault propagation paths. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization.

[0057] The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

[0058] This embodiment's model library includes various machine learning models tailored to the characteristics of DICT projects. For example, the optical module degradation prediction model can improve the prediction speed of traditional algorithms by 17 times; the business anomaly detection model identifies potential risks by analyzing historical performance data; and the intelligent root cause localization model uses graph algorithms to automatically generate service dependency topologies, enabling fault propagation path analysis. For each model, reinforcement learning techniques are employed to dynamically adjust model parameters based on historical handling results; model performance evaluation indicators are introduced to continuously monitor prediction accuracy and false alarm rate; and a knowledge base update process is established to incorporate expert experience and optimize decision-making logic. The fault analysis layer uses machine learning algorithms to perform fault analysis and prediction on processed data, including fault diagnosis, fault location, and fault cause analysis. Classification algorithms in supervised learning, such as decision trees and logistic regression, are used to train historical fault data and construct a fault prediction and analysis model library. Simultaneously, artificial intelligence technologies, such as deep learning and natural language processing, are introduced to correct and optimize the model's judgment results, improving accuracy. During model training, weight coefficients are used to measure the influence of different factors or variables on the results. The weight coefficients range from 0 to 1, and cross-validation techniques are used to determine the optimal model.

[0059] The control and display layer is used to display multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. It is used to generate early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions and send the early warning notifications to the executors. It is used to display the execution result logs recorded by the executors and to feed the execution result logs back to the data acquisition layer to form a closed-loop control.

[0060] In this embodiment, the management and display layer is configured with a cross-domain fault location service and a pre-defined plan execution service. The cross-domain fault location service associates logs from multiple systems with data from multiple sources, and the pre-defined plan execution service generates and automatically executes pre-defined plans based on optimization suggestions.

[0061] In this embodiment, the control and display layer is responsible for displaying fault information and optimization suggestions to operators and customers in an appropriate manner, including sending text messages, emails, mobile applications, etc.

[0062] This embodiment constructs a hierarchical intelligent monitoring architecture, which achieves closed-loop management of all components of the DICT project through unified edge acquisition devices, a multi-source data fusion engine, and an adaptive AI analysis platform.

[0063] Example 2:

[0064] This invention provides a machine learning-based method for managing and controlling DICT projects for telecom operators, comprising four steps: data collection, data fusion, intelligent analysis, and management and control visualization.

[0065] Step S100 Data Acquisition: Collect multi-source data, including dedicated line network management data, IT equipment performance indicators, IoT sensor data, and application performance indicators, through the edge operation and maintenance box. Preprocess the multi-source data and upload the preprocessed multi-source data to the data fusion layer.

[0066] As a concrete implementation of data collection, the edge operation and maintenance box collects leased network management data via the SNMP protocol, including port traffic and bit error rate. It also collects IT equipment performance metrics via IPMI / Redfish or proxy methods, including CPU and memory usage. Furthermore, it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance metrics, including response time and error rate, via browser simulation or traffic mirroring. The edge operation and maintenance box performs data cleaning, compression, encryption, and security isolation on the collected multi-source data, uniformly encapsulating it into a fixed-format data stream, and uploading the data stream to the data fusion layer via a bidirectional TLS channel.

[0067] In this embodiment, the edge maintenance box serves as a secure carrier for customer-side data collection and preprocessing. This device employs multi-protocol adaptation technology, supporting various interface protocols such as SNMP for leased networks, MQTT for IoT devices, and APIs for application systems. It possesses edge computing capabilities, enabling local data cleaning, compression, and encryption. Secure isolation is achieved through firewalls and two-way certificate authentication to ensure secure data transmission. During data collection, monitoring data is gathered from multiple dimensions, including network, IT equipment, and smart IoT devices. For monitoring of leased network departments, it accesses network performance management data from the operational network to extract and analyze the corresponding leased line information for customers. By analyzing the traffic on downstream ports, issues such as significantly lower than long-term traffic or no traffic generation on downstream ports can be detected promptly. For monitoring customer-side IT equipment, smart IoT devices, and application software, the edge maintenance box serves as a carrier to achieve secure data transmission and analysis. The IT maintenance box utilizes hardware components, including CPU, memory, storage, and network interfaces, and receives and transmits data through the network interface.

[0068] Step S200 Data Fusion: Standardize and extract features from the preprocessed multi-source data to obtain time-series data with a predetermined structure. Persist the time-series data and send it to the intelligent analysis layer.

[0069] As a specific implementation of data fusion, data cleaning, data classification, and feature encoding are performed on multi-source data. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data to obtain time-series data with a predetermined structure. The time-series data is then persisted to a relational database or NoSQL database.

[0070] In this embodiment, the collected data undergoes standardization and feature extraction during data fusion, including data cleaning, data classification, and feature encoding. Data cleaning removes deactivated equipment, invalid data, outliers, etc.; data classification groups different types of data into a unified dimension; feature encoding converts non-numerical data into numerical data, such as converting string data into numeric codes. This layer stores the data in a data storage layer, using a relational database or NoSQL database to ensure data persistence and high availability.

[0071] Step S300 Intelligent Analysis: Based on the models in the model library, perform fault prediction analysis, root cause localization, and optimization suggestions on time series data, and optimize the models in the model library based on reinforcement learning.

[0072] As a specific implementation, the model library deploys various models built based on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data. Artificial intelligence technologies, including deep learning and natural language processing, are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1. The optimal model is determined through cross-validation.

[0073] The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models.

[0074] The optical module degradation prediction model takes time-series data corresponding to the leased line network management data as input and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp, and degradation prediction results for uncertain regions.

[0075] The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems.

[0076] The root cause localization model takes anomaly detection results and time series data as input, and uses graph neural networks and attention mechanisms to infer fault propagation paths. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization.

[0077] The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

[0078] This embodiment's model library includes various machine learning models tailored to the characteristics of DICT projects. For example, the optical module degradation prediction model can improve the prediction speed of traditional algorithms by 17 times; the business anomaly detection model identifies potential risks by analyzing historical performance data; and the intelligent root cause localization model uses graph algorithms to automatically generate service dependency topologies, enabling fault propagation path analysis. For each model, reinforcement learning techniques are employed to dynamically adjust model parameters based on historical handling results; model performance evaluation indicators are introduced to continuously monitor prediction accuracy and false alarm rate; and a knowledge base update process is established to incorporate expert experience and optimize decision-making logic. The fault analysis layer uses machine learning algorithms to perform fault analysis and prediction on processed data, including fault diagnosis, fault location, and fault cause analysis. Classification algorithms in supervised learning, such as decision trees and logistic regression, are used to train historical fault data and construct a fault prediction and analysis model library. Simultaneously, artificial intelligence technologies, such as deep learning and natural language processing, are introduced to correct and optimize the model's judgment results, improving accuracy. During model training, weight coefficients are used to measure the influence of different factors or variables on the results. The weight coefficients range from 0 to 1, and cross-validation techniques are used to determine the optimal model.

[0079] Step S400 Control and Management Display: Display multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. Generate early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions, and send the early warning notifications to the executors. Display the execution result logs recorded by the executors and feed the execution result logs back to the data acquisition layer to form a closed-loop control.

[0080] In this embodiment, the management and control display provides cross-domain fault location service and contingency plan self-execution service. The cross-domain fault location service associates logs from multiple systems and data from multiple sources. The contingency plan self-execution service generates contingency plans based on optimization suggestions and automatically executes the contingency plans.

[0081] In this embodiment, the control and display mechanism is responsible for presenting fault information and optimization suggestions to operators and customers in an appropriate manner, including sending text messages, emails, and mobile applications.

[0082] The method in this embodiment is based on the system implementation disclosed in Embodiment 1.

[0083] The above provides a detailed description of the machine learning-based operator DICT project management system method provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A machine learning-based operator DICT project management system, characterized in that, It includes a data acquisition layer, a data fusion layer, an intelligent analysis layer, and a management and display layer; The data acquisition layer includes an edge operation and maintenance box deployed on the customer side. The edge operation and maintenance box supports multi-purpose protocol adaptation technology and is used to collect multi-source data, including leased line network management data, IT equipment performance indicators, IoT sensor data and application performance indicators. The multi-source data is preprocessed and then uploaded to the data fusion layer. The data fusion layer is used to standardize and extract features from the preprocessed multi-source data to obtain time-series data with a predetermined structure. The time-series data is then persisted and sent to the intelligent analysis layer. The intelligent analysis layer is equipped with a model library built on machine learning algorithms. This library is used to perform fault prediction analysis, root cause localization, and optimization suggestion generation on time series data based on the models in the library. It also optimizes the models in the library based on reinforcement learning. The control and display layer is used to display multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. It is used to generate early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions and send the early warning notifications to the executors. It is used to display the execution result logs recorded by the executors and to feed the execution result logs back to the data acquisition layer to form a closed-loop control.

2. The machine learning-based operator DICT project management system according to claim 1, characterized in that, The edge operation and maintenance box collects leased network management data via the SNMP protocol, including port traffic and bit error rate. It also collects IT equipment performance metrics via IPMI / Redfish or a proxy, including CPU and memory usage. Furthermore, it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance metrics, including response time and error rate, via browser simulation or traffic mirroring. The edge operation and maintenance box is used to clean, compress, encrypt, and securely isolate the collected multi-source data, and encapsulate it into a data stream with a fixed format. The data stream is then uploaded to the data fusion layer through a two-way TLS channel.

3. The machine learning-based operator DICT project management system according to claim 1, characterized in that, The data fusion layer is used to clean, classify, and encode features from multi-source data. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data, resulting in time-series data with a predetermined structure. The time-series data is then persisted to a relational database or NoSQL database.

4. The machine learning-based operator DICT project management system according to claim 1, characterized in that, The model library contains various models built on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data. Artificial intelligence technologies, including deep learning and natural language processing, are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1. The optimal model is determined through cross-validation. The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models. The optical module degradation prediction model takes the time series data corresponding to the leased line network management data as input, and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp and degradation prediction results for uncertain areas. The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems. The root cause localization model takes anomaly detection results and time series data as input, and is used to infer the fault propagation path based on graph neural networks and attention mechanisms. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization. The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

5. The machine learning-based operator DICT project management system according to claim 1, characterized in that, The management and control display layer is configured with cross-domain fault location service and contingency plan auto-execution service. The cross-domain fault location service associates logs from multiple systems and data from multiple sources. The contingency plan auto-execution service generates contingency plans based on optimization suggestions and automatically executes the contingency plans.

6. A machine learning-based method for managing and controlling DICT projects for telecom operators, characterized in that, This includes data collection, data fusion, intelligent analysis, and management and control visualization; Data Acquisition: Collect multi-source data, including dedicated line network management data, IT equipment performance indicators, IoT sensor data, and application performance indicators, through the edge operation and maintenance box; preprocess the multi-source data; and upload the preprocessed multi-source data to the data fusion layer. Data fusion: Standardize and extract features from preprocessed multi-source data to obtain time-series data with a predetermined structure, persist the time-series data and send it to the intelligent analysis layer; Intelligent analysis: Based on models in the model library, it performs fault prediction analysis, root cause localization, and optimization suggestions on time series data, and optimizes models in the model library based on reinforcement learning; Control and Management Display: Displays multi-source data, fault prediction analysis results, root cause location results, and optimization suggestions. Generates early warning notifications based on fault prediction analysis results, root cause location results, and optimization suggestions, and sends the early warning notifications to the implementers. Displays the execution result logs recorded by the implementers and feeds the execution result logs back to the data acquisition layer, forming a closed-loop control.

7. The machine learning-based operator DICT project management method according to claim 6, characterized in that, The edge maintenance box collects leased network management data via the SNMP protocol, including port traffic and bit error rate. It also collects IT equipment performance metrics via IPMI / Redfish or a proxy, including CPU and memory usage. Furthermore, it collects IoT sensor data via the MQTT protocol, including temperature and humidity. Finally, it collects application system performance metrics, including response time and error rate, through browser simulation or traffic mirroring. The edge operation and maintenance box cleans, compresses, encrypts, and securely isolates the collected multi-source data, and encapsulates it into a data stream in a fixed format. The data stream is then uploaded to the data fusion layer via a bidirectional TLS channel.

8. The machine learning-based operator DICT project management method according to claim 6, characterized in that, During data fusion, multi-source data undergoes data cleaning, data classification, and feature encoding. Data cleaning removes deactivated equipment, invalid data, and outliers. Data classification merges different types of data into a unified dimension. Feature encoding converts non-numerical data into numerical data, resulting in time-series data with a predetermined structure. This time-series data is then persisted to a relational database or NoSQL database.

9. The machine learning-based operator DICT project management method according to claim 6, characterized in that, The model library contains various models built on machine learning algorithms. For each model, it is trained using a classification algorithm in supervised learning based on historical multi-source data. Artificial intelligence technologies, including deep learning and natural language processing, are introduced to correct and optimize the model's judgment results. During the training process, weight coefficients are used to measure the degree of influence of different factors or variables on the model's prediction results. The weight coefficients range from 0 to 1. The optimal model is determined through cross-validation. The model library includes optical module degradation prediction models, service anomaly detection models, root cause localization models, and optimization suggestion generation models. The optical module degradation prediction model takes the time series data corresponding to the leased line network management data as input, and is used to predict the optical power attenuation trend at a predetermined time based on CNNN and attention mechanism. The output includes degradation probability, expected failure timestamp and degradation prediction results for uncertain areas. The business anomaly detection model takes time-series data corresponding to application performance metrics as input and performs business anomaly detection based on LSTM. The output includes anomaly score, anomaly start and end time, and a list of affected business systems. The root cause localization model takes anomaly detection results and time series data as input, and is used to infer the fault propagation path based on graph neural networks and attention mechanisms. The output includes a root cause list, a fault propagation subgraph, and a confidence score for root cause localization. The optimization suggestion generation model takes root cause analysis results and time series data as input to analyze the frequency and severity of failures, generate early warning rules, and combine artificial intelligence algorithms to perform data analysis and generate optimization suggestions, including overall system configuration, database query optimization, modification of system security policies, and strengthening password management.

10. The machine learning-based operator DICT project management method according to claim 6, characterized in that, When managing and displaying data, the system provides cross-domain fault location services and custom contingency plan execution services. The cross-domain fault location service associates logs from multiple systems with data from multiple sources, and the custom contingency plan execution service generates contingency plans based on optimization suggestions and automatically executes them.

Images