An artificial intelligence-based digital transmission fault diagnosis method
By combining hierarchical decision trees, quantum annealing algorithm, pulse-coupled neural network and fuzzy reinforcement learning, the problems of slow fault diagnosis speed and poor repair path optimization in complex network environments are solved, realizing real-time and efficient fault identification and repair, and improving the stability and reliability of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAN BAIHONG INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2025-12-16
- Publication Date
- 2026-06-26
AI Technical Summary
Existing fault diagnosis methods are slow in fault identification and diagnosis in complex network environments, have low accuracy, poor optimization of repair paths, and are difficult to capture the nonlinear relationships and dynamic evolution of the system, thus failing to meet the needs of real-time response and multi-factor optimization.
A combination of hierarchical decision tree processing, quantum annealing optimization algorithm, multi-scale annealing strategy, pulse-coupled neural network and fuzzy reinforcement learning is used to collect and preprocess multi-dimensional data, extract key features, optimize fault repair paths, dynamically adjust repair strategies, and realize real-time fault diagnosis and repair.
It improves the accuracy of fault identification and repair efficiency, and can adaptively optimize repair strategies to ensure efficient and accurate fault diagnosis and repair in dynamically changing environments, significantly improving the reliability and stability of the system.
Smart Images
Figure CN121441776B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology, and in particular to a digital transmission fault diagnosis method based on artificial intelligence. Background Technology
[0002] With the widespread application of digital transmission systems in communications, energy, transportation, and other fields, the reliability and stability of these systems have become paramount to ensuring their efficient operation. Especially in complex transmission networks, rapid fault diagnosis and repair have become critical technologies. However, existing fault diagnosis methods largely rely on traditional rule matching and human experience, which presents several problems:
[0003] Slow fault identification and diagnosis speed: Existing fault diagnosis methods typically rely on manual methods or rule-based systems, making fault detection and localization slow, especially in large-scale, complex network environments, failing to meet real-time response requirements. Low accuracy of fault mode identification: Traditional methods rely heavily on static feature analysis, lacking in-depth mining of multi-dimensional system data, resulting in low accuracy in fault mode classification and a high risk of misjudgment or omission, particularly in complex and variable system states. Poor performance in repair path optimization: While existing methods allow for manual setting of repair strategies, they typically do not consider multi-factor optimization issues such as repair time and resource consumption, leading to imprecise repair paths that are difficult to dynamically adjust and optimize. Insufficient modeling of nonlinear and temporal relationships: Most existing fault diagnosis methods employ traditional statistical methods and simple time-series analysis models, which struggle to capture complex nonlinear relationships and dynamic evolution within the system, affecting the accuracy of fault prediction and repair decisions.
[0004] Therefore, how to provide an artificial intelligence-based digital transmission fault diagnosis method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose an artificial intelligence-based method for diagnosing digital transmission faults. This invention achieves real-time diagnosis and repair of faults in digital transmission systems through hierarchical decision tree processing, quantum annealing optimization algorithm, multi-scale annealing strategy, time series prediction using pulse-coupled neural networks, and a fuzzy reinforcement learning-based repair strategy. This method is not only efficient and accurate in fault diagnosis but also adaptively optimizes the repair strategy, improving the fault repair effect.
[0006] A digital transmission fault diagnosis method based on artificial intelligence according to an embodiment of the present invention includes the following steps:
[0007] Collect multi-dimensional data from the transmission system and preprocess it to form the original network dataset;
[0008] Hierarchical decision trees are used to fuse the original network dataset, extract key features, and generate a feature dataset.
[0009] The hierarchical decision tree is applied to the feature dataset to perform fault mode classification and recognition, generating a fault mode dataset.
[0010] Based on the fault mode dataset, the improved quantum annealing algorithm is used to optimize the fault repair path. By dynamically adjusting the cooling rate and introducing quantum state constraints, a multi-scale annealing strategy is adopted to explore the solution space at multiple levels, generate the optimal repair path and obtain the repair path dataset.
[0011] By optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping, time series prediction is performed on the repair path dataset to generate a fault prediction dataset.
[0012] Based on the fault prediction dataset, the repair strategy is optimized through improved fuzzy reinforcement learning. The method combines incremental learning and fuzzy inference to dynamically update the fuzzy rule base. Adaptive fuzzy control and fuzzy autoencoder are used to dynamically optimize the membership function and output the repair strategy dataset.
[0013] The generated repair strategy dataset is fed back to optimize the fault mode dataset and repair path dataset in real time, dynamically adjust the repair path, and adjust the system configuration according to the repair strategy dataset to generate a new optimized dataset.
[0014] Optionally, the multi-dimensional data collected and transmitted by the system is preprocessed to form an original network dataset, specifically including:
[0015] The multi-dimensional data includes device status data, network traffic data, system load data, environmental factor data, and fault history data;
[0016] Kalman filtering and mean filtering algorithms are used to remove noise from multidimensional data. Missing data are filled in by interpolation or nearest neighbor imputation. Z-score is used for standardization. Principal component analysis is used to remove redundant features and retain effective features to form a standardized original network dataset.
[0017] Optionally, the step of using a hierarchical decision tree to fuse the original network dataset, extract key features, and generate a feature dataset specifically includes:
[0018] The original network dataset is input into a hierarchical decision tree model. The first layer uses the decision tree to classify features of a single data source and generate preliminary features for each data source. The second layer inputs multiple preliminary features into the decision tree for integration and merges information from different data sources to generate a feature dataset.
[0019] A dynamic splitting criterion is used when splitting each node. At each node split, the splitting criterion is dynamically selected based on the data distribution and feature importance of the current node. The splitting effect of each feature on the current node is evaluated by calculating the Gini index or information gain of each feature to improve the node purity. Features are selected for splitting based on the splitting effect.
[0020] Adaptive tree depth control is adopted to automatically adjust the depth of the decision tree according to the complexity and sample size of the original network dataset. When the sample size of a node is lower than a predetermined threshold, or when the depth of the tree reaches a predetermined limit, the depth of the tree is automatically limited.
[0021] Redundant branches are removed using an adaptive pruning algorithm. After generating the decision tree, the contribution of each branch of the tree is evaluated based on the performance of the validation set. If the improvement of the decision tree's accuracy by a certain branch is less than a preset threshold, the current branch is pruned. During the pruning process, if the error of a node is greater than a certain threshold or the node's contribution to the overall performance is less than a preset threshold, the current branch is removed.
[0022] Optionally, the improved quantum annealing algorithm specifically includes:
[0023] The fault mode dataset is fed into the improved quantum annealing algorithm to optimize the fault repair path and generate the optimal repair path;
[0024] During quantum annealing, quantum state constraints are set to limit the search space. The constraints include repair time and resource consumption. Repair time thresholds and resource consumption thresholds are applied to the repair path.
[0025] A multi-scale annealing strategy is adopted to optimize the solution space at multiple levels in different annealing stages. In the initial stage, a coarse-grained global search is performed by setting a large step size, and a fine-grained local search is performed by gradually reducing the step size. The cooling rate and step size of each annealing stage are dynamically adjusted according to the search stage.
[0026] Each repair path is quantitatively evaluated through a path optimization evaluation mechanism. The evaluation indicators include the total repair time, resource consumption, and success rate of the repair path. The feasibility of the path is evaluated based on the time threshold and resource consumption threshold of the repair path. Each repair path is scored according to the evaluation indicators. The optimal repair path with a score greater than the score threshold is selected and a repair path dataset is generated. The repair path dataset contains the optimal repair path and other candidate paths.
[0027] Optionally, the step of optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping to perform time-series prediction on the repair path dataset and generate a fault prediction dataset specifically includes:
[0028] Time series features and multi-dimensional information are extracted from the repair path dataset and converted into vector form to obtain multi-dimensional feature vectors. The self-organizing map algorithm is used to organize and map the multi-dimensional feature vectors to two-dimensional or three-dimensional grid nodes according to similarity to form a feature mapping matrix. Each node corresponds to a specific feature in the repair path dataset, and adjacent nodes represent similar repair path features.
[0029] During the training of the self-organizing map algorithm, the weights of the nodes are updated through a competitive learning mechanism. Each time data is input, the Euclidean distance between the input data and each node is calculated, and the best matching unit closest to the input data is selected. The weights of the node and its corresponding neighboring nodes are updated according to the Gaussian neighborhood function.
[0030] The feature mapping matrix is input into the pulse-coupled neural network for optimization. Based on the feature mapping matrix, the coupling strength between each neuron is calculated. The coupling strength is measured by Euclidean distance to measure the similarity between each pair of neurons. Based on the similarity, the coupling strength between neurons and the propagation speed of the pulse signal are adjusted. The coupling strength and the propagation speed of the pulse signal are proportional to the coupling strength between neurons.
[0031] The feature mapping matrix is input into the optimized spiking coupled neural network. Through the propagation of spiking signals from neurons, the future change trend of repair path data is simulated, and the possible fault types, fault occurrence time, affected equipment, and execution order of repair paths are predicted. By utilizing the spiking propagation of neurons, the spiking coupled neural network captures the time dependence and evolution trend of each node in the repair path and generates a fault prediction dataset.
[0032] Optionally, the improved fuzzy reinforcement learning specifically includes:
[0033] Based on the fault prediction dataset, the repair strategy is optimized by improving fuzzy reinforcement learning. By analyzing the time series data of the repair path and repair task, each possible repair strategy is evaluated, the reward value of each repair strategy is calculated, the Q-learning mechanism is used to update the Q-value function, and the repair strategy selection is continuously adjusted by optimizing the Q-value function to select the optimal path.
[0034] Incremental learning is introduced to dynamically update and optimize the fuzzy rule base. After each repair path is executed, the existing rules are dynamically adjusted based on the actual execution effect and newly collected data, and the repair strategy is updated through fuzzy inference.
[0035] Adaptive fuzzy control and fuzzy autoencoder are used to dynamically optimize the membership function. Based on the real-time feedback of the repair task information, the adaptive fuzzy control adjusts the shape and control parameters of the membership function.
[0036] Fuzzy autoencoders are used to automatically optimize the parameters of membership functions. By automatically encoding and decoding, the membership functions in the fuzzy rule base are optimized. The fuzzy autoencoder learns the optimal repair path features through the process of minimizing errors.
[0037] The repair strategy dataset is output by dynamically optimized fuzzy rules and repair strategies. The repair strategy dataset includes optimization schemes for repair paths, resource allocation schemes, repair priorities, and timing arrangements.
[0038] Optionally, the generated repair strategy dataset is fed back to optimize the fault mode dataset and repair path dataset in real time, dynamically adjust the repair path, and adjust the system configuration according to the repair strategy dataset to generate a new optimized dataset, specifically including:
[0039] Using the repair strategy dataset as input, the repair path data is matched with the fault prediction dataset to generate a repair path execution sequence;
[0040] The execution repair strategy is optimized by calculating the execution error of the repair task based on real-time feedback data, optimizing the order of the repair task through a weighted delay mechanism, and adjusting the priority of repair tasks whose error exceeds a preset threshold.
[0041] Perform priority sorting and verification of the repair paths, calculate the repair success rate and latency of each repair path task, sort the repair paths, and demote high-latency repair tasks.
[0042] Perform resource scheduling optimization and adjust the resource allocation of each repair path in the repair strategy dataset according to the real-time resource status of the system.
[0043] The optimized repair path tasks are scheduled according to priority, repair time, and resource consumption to generate the optimal repair execution plan;
[0044] Using the repair strategy dataset as input, it is dynamically updated during the execution of the repair path. Based on the feedback data of the repair task execution, the repair path and repair strategy are adjusted, and incremental corrections are made on the repair path that does not achieve the preset execution effect.
[0045] After each incremental correction, the repair path sorting verification, resource scheduling optimization, and repair task execution error verification are re-executed. Repair path optimization continues in areas that have not reached the preset threshold until the verification results meet the preset requirements or the maximum number of iterations is reached, and a new optimized dataset is output.
[0046] The beneficial effects of this invention are:
[0047] This invention utilizes multi-dimensional data acquisition and preprocessing, along with hierarchical decision trees to fuse the original network dataset, enabling efficient extraction of key features and fault mode identification, thereby generating an accurate fault mode dataset. Next, an improved quantum annealing algorithm is employed to optimize the fault repair path. By dynamically adjusting the cooling rate, introducing quantum state constraints, and using a multi-scale annealing strategy to explore the solution space at multiple levels, the optimal repair path is effectively generated, improving the accuracy and efficiency of the repair operation. Building upon the optimized repair path, a self-organizing map and pulse-coupled neural network (PCNN) are combined to perform time-series prediction on the repair path dataset, accurately capturing the time nodes of fault occurrence and related impacts, providing more efficient decision support for fault prediction. Furthermore, an improved fuzzy reinforcement learning approach is used to optimize the repair strategy. By combining incremental learning and fuzzy inference methods, the fuzzy rule base is dynamically updated, and adaptive fuzzy control and fuzzy autoencoders are used to optimize the membership function, achieving real-time adaptive adjustment of the repair strategy, significantly improving the accuracy, execution efficiency, and system adaptability of the repair strategy. Ultimately, by feeding back the repair strategy dataset to the system, and combining it with real-time optimization of repair paths and system configuration adjustments, the repair strategy can be continuously updated and improved. This ensures that the system's fault diagnosis and repair operations remain efficient and accurate in a dynamically changing network environment, greatly improving the efficiency and accuracy of fault repair. It is highly practical and innovative. Attached Figure Description
[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0049] Figure 1 This is a flowchart of a digital transmission fault diagnosis method based on artificial intelligence proposed in this invention;
[0050] Figure 2 This is a schematic diagram of a digital transmission fault diagnosis method based on artificial intelligence proposed in this invention;
[0051] Figure 3 This is a flowchart of an improved quantum annealing algorithm in an artificial intelligence-based digital transmission fault diagnosis method proposed in this invention. Detailed Implementation
[0052] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0053] refer to Figure 1-3 A digital transmission fault diagnosis method based on artificial intelligence includes the following steps:
[0054] Collect multi-dimensional data from the transmission system and preprocess it to form the original network dataset;
[0055] Hierarchical decision trees are used to fuse the original network dataset, extract key features, and generate a feature dataset.
[0056] The hierarchical decision tree is applied to the feature dataset to perform fault mode classification and recognition, generating a fault mode dataset.
[0057] Based on the fault mode dataset, the improved quantum annealing algorithm is used to optimize the fault repair path. By dynamically adjusting the cooling rate and introducing quantum state constraints, a multi-scale annealing strategy is adopted to explore the solution space at multiple levels, generate the optimal repair path and obtain the repair path dataset.
[0058] By optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping, time series prediction is performed on the repair path dataset to generate a fault prediction dataset.
[0059] Based on the fault prediction dataset, the repair strategy is optimized through improved fuzzy reinforcement learning. The method combines incremental learning and fuzzy inference to dynamically update the fuzzy rule base. Adaptive fuzzy control and fuzzy autoencoder are used to dynamically optimize the membership function and output the repair strategy dataset.
[0060] The generated repair strategy dataset is fed back to optimize the fault mode dataset and repair path dataset in real time, dynamically adjust the repair path, and adjust the system configuration according to the repair strategy dataset to generate a new optimized dataset.
[0061] In this embodiment, the multi-dimensional data from the acquisition and transmission system is preprocessed to form the original network dataset, specifically including:
[0062] The multi-dimensional data includes device status data, network traffic data, system load data, environmental factor data, and fault history data;
[0063] Kalman filtering and mean filtering algorithms are used to remove noise from multidimensional data. Missing data are filled in by interpolation or nearest neighbor imputation. Z-score is used for standardization. Principal component analysis is used to remove redundant features and retain effective features to form a standardized original network dataset.
[0064] In this embodiment, the step of using a hierarchical decision tree to fuse the original network dataset, extract key features, and generate a feature dataset specifically includes:
[0065] The original network dataset is input into a hierarchical decision tree model. The first layer uses the decision tree to classify features of a single data source and generate preliminary features for each data source. The second layer inputs multiple preliminary features into the decision tree for integration and merges information from different data sources to generate a feature dataset.
[0066] A dynamic splitting criterion is used when splitting each node. At each node split, the splitting criterion is dynamically selected based on the data distribution and feature importance of the current node. The splitting effect of each feature on the current node is evaluated by calculating the Gini index or information gain of each feature to improve the node purity. Features are selected for splitting based on the splitting effect.
[0067] Adaptive tree depth control is adopted to automatically adjust the depth of the decision tree according to the complexity and sample size of the original network dataset. When the sample size of a node is lower than a predetermined threshold, or when the depth of the tree reaches a predetermined limit, the depth of the tree is automatically limited to avoid overfitting or underfitting.
[0068] Redundant branches are removed using an adaptive pruning algorithm. After generating the decision tree, the contribution of each branch of the tree is evaluated based on the performance of the validation set. If the improvement of the decision tree's accuracy by a certain branch is less than a preset threshold, the current branch is pruned. During the pruning process, if the error of a node is greater than a certain threshold or the node's contribution to the overall performance is less than a preset threshold, the current branch is removed.
[0069] In this embodiment, the improved quantum annealing algorithm specifically includes:
[0070] The fault mode dataset is fed into the improved quantum annealing algorithm to optimize the fault repair path and generate the optimal repair path;
[0071] During quantum annealing, quantum state constraints are set to limit the search space to meet the actual operational requirements. The constraints include repair time and resource consumption. Repair time thresholds and resource consumption thresholds are applied to the repair path to ensure that the generated repair path is not only optimal, but also meets the actual requirements of system resource availability and repair operation.
[0072] A multi-scale annealing strategy is adopted to optimize the solution space at multiple levels in different annealing stages. In the initial stage, a coarse-grained global search is performed by setting a large step size to ensure extensive exploration of the solution space. The step size is gradually reduced to perform a fine-grained local search to optimize the accuracy of the solution. The cooling rate and step size of each annealing stage are dynamically adjusted according to the search stage to ensure the generation of the optimal repair path at different scales.
[0073] Each repair path is quantitatively evaluated through a path optimization evaluation mechanism. The evaluation indicators include the total repair time, resource consumption, and success rate of the repair path. The feasibility of the path is evaluated based on the time threshold and resource consumption threshold of the repair path. Each repair path is scored according to the evaluation indicators. The optimal repair path with a score greater than the score threshold is selected and a repair path dataset is generated. The repair path dataset contains the optimal repair path and other candidate paths.
[0074] In this embodiment, the step of optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping to perform time series prediction on the repair path dataset and generate a fault prediction dataset specifically includes:
[0075] Time series features and multi-dimensional information are extracted from the repair path dataset and converted into vector form to obtain multi-dimensional feature vectors. The self-organizing map algorithm is used to organize and map the multi-dimensional feature vectors to two-dimensional or three-dimensional grid nodes according to similarity to form a feature mapping matrix. Each node corresponds to a specific feature in the repair path dataset, and adjacent nodes represent similar repair path features.
[0076] During the training process of the self-organizing mapping algorithm, the weights of nodes are updated through a competitive learning mechanism, so that similar features are clustered on adjacent nodes as much as possible. Each time data is input, the Euclidean distance between the input data and each node is calculated, and the best matching unit closest to the input data is selected. The weights of the node and its corresponding neighboring nodes are updated according to the Gaussian neighborhood function to ensure that the low-dimensional space after mapping can effectively represent the key information of the repair path data.
[0077] The feature mapping matrix is input into the pulse-coupled neural network for optimization. Based on the feature mapping matrix, the coupling strength between each neuron is calculated. The coupling strength is measured by Euclidean distance to measure the similarity between each pair of neurons. Based on the similarity, the coupling strength between neurons and the propagation speed of the pulse signal are adjusted. The coupling strength and the propagation speed of the pulse signal are proportional to the coupling strength between neurons.
[0078] The feature mapping matrix is input into the optimized spiking coupled neural network. Through the propagation of spiking signals from neurons, the future change trend of repair path data is simulated, and the possible fault types, fault occurrence time, affected equipment, and execution order of repair paths are predicted. By utilizing the spiking propagation of neurons, the spiking coupled neural network captures the time dependence and evolution trend of each node in the repair path and generates a fault prediction dataset.
[0079] In this embodiment, the improved fuzzy reinforcement learning specifically includes:
[0080] Based on the fault prediction dataset, the repair strategy is optimized through improved fuzzy reinforcement learning. By analyzing the time series data of repair paths and repair tasks, each possible repair strategy is evaluated, the reward value of each repair strategy is calculated, the Q-learning mechanism is used to update the Q-value function, and the repair strategy selection is continuously adjusted by optimizing the Q-value function to select the optimal path and improve the efficiency and accuracy of fault repair.
[0081] Incremental learning is introduced to dynamically update and optimize the fuzzy rule base. After each repair path is executed, the existing rules are dynamically adjusted based on the actual execution effect and newly collected data, and the repair strategy is updated through fuzzy inference to adapt to new environments or failure modes.
[0082] Adaptive fuzzy control and fuzzy autoencoders are used to dynamically optimize membership functions. Based on real-time feedback of repair task information, adaptive fuzzy control adjusts the shape and control parameters of membership functions to adapt to different repair tasks and system states. For example, when certain operations in the repair path have high priority, the membership function is adjusted to prioritize the execution of those tasks. During the repair task, factors such as repair success rate and task latency are considered to dynamically adjust the responsiveness of the repair strategy.
[0083] Fuzzy autoencoders are used to automatically optimize the parameters of membership functions. By automatically encoding and decoding, the membership functions in the fuzzy rule base are optimized, making the parameters of each repair operation more accurate and better able to adapt to changes in system state. Fuzzy autoencoders learn the optimal repair path features through the process of minimizing errors, so that the membership functions can be adjusted more flexibly when faced with new repair paths.
[0084] By dynamically optimizing fuzzy rules and repair strategies, a repair strategy dataset is output. This dataset includes optimization schemes for repair paths, resource allocation schemes, repair priorities, and timing arrangements, providing a detailed execution plan for the system's repair operations. The generated repair strategy dataset can dynamically adjust the repair strategy under different environmental conditions and changes in repair tasks to ensure the smooth execution of repair tasks. The optimized repair strategy dataset also includes evaluations of resource consumption, repair duration, and fault handling effectiveness during the repair process. Based on this information, the system can adjust the repair strategy in real time and further optimize the repair path based on new fault prediction data and execution results, ensuring that the system continuously improves the efficiency of fault diagnosis and repair.
[0085] In this embodiment, the step of feeding back the generated repair strategy dataset, optimizing the fault mode dataset and repair path dataset in real time, dynamically adjusting the repair path, and adjusting the system configuration based on the repair strategy dataset to generate a new optimized dataset specifically includes:
[0086] Using the repair strategy dataset as input, the repair path data is matched with the fault prediction dataset to generate a repair path execution sequence;
[0087] The execution repair strategy is optimized by calculating the execution error of the repair task based on real-time feedback data, optimizing the order of the repair task through a weighted delay mechanism, and adjusting the priority of repair tasks whose error exceeds a preset threshold.
[0088] Perform priority sorting and verification of the repair paths, calculate the repair success rate and latency of each repair path task, sort the repair paths, and demote high-latency repair tasks.
[0089] Perform resource scheduling optimization. Based on the real-time resource status of the system, adjust the resource allocation of each repair path in the repair strategy dataset to ensure that resources can be prioritized for critical repair tasks and avoid over-allocation of resources.
[0090] The optimized repair path tasks are scheduled according to priority, repair time, and resource consumption to generate the optimal repair execution plan;
[0091] Using the repair strategy dataset as input, it is dynamically updated during the execution of the repair path. Based on the feedback data of the repair task execution, the repair path and repair strategy are adjusted, and incremental corrections are made on the repair path that does not achieve the preset execution effect.
[0092] After each incremental correction, the repair path sorting verification, resource scheduling optimization, and repair task execution error verification are re-executed. Repair path optimization continues in areas that have not reached the preset threshold until the verification results meet the preset requirements or the maximum number of iterations is reached, and a new optimized dataset is output.
[0093] Example 1:
[0094] To verify the feasibility of this invention in practice, it was applied to a large-scale communication network environment, specifically a digital transmission system of a telecommunications operator. This system comprises multiple transmission nodes connected to different regional networks, providing data transmission services to millions of end users. In this application scenario, we face the problem of frequent transmission line failures, especially system performance degradation due to equipment failures, network congestion, and environmental factors. These failures not only affect service quality but may also lead to large-scale service interruptions, thereby impacting company revenue and customer satisfaction.
[0095] In traditional fault diagnosis methods, network administrators often rely on manual inspections or simple network monitoring tools to discover and locate faults. However, these methods cannot respond to the constantly changing state of the network in real time, and the diagnostic process is cumbersome and time-consuming. Therefore, network fault diagnosis is inefficient, especially when faced with multi-dimensional data and complex fault modes, where traditional methods fall short.
[0096] In this embodiment, an AI-based digital transmission fault diagnosis method is applied to address the problems of slow fault identification speed, poor repair path optimization, and insufficient nonlinear relationship modeling in traditional methods. The specific implementation steps include multiple stages such as data acquisition, preprocessing, feature extraction, fault mode recognition, repair path optimization, and real-time feedback.
[0097] First, multi-dimensional data is collected from multiple nodes in the transmission system. Data sources include device status information, network traffic, system load, environmental factors, and historical fault data. For example, at a given moment, data such as network device temperature, CPU utilization, and memory usage are collected from device status sensors, as well as data from traffic monitoring tools (e.g., upload / download bandwidth, packet loss rate). After collection, this data undergoes noise removal using a Kalman filter algorithm, and missing data is filled in using interpolation or nearest-neighbor imputation. Finally, Z-score standardization is applied to obtain a standardized raw dataset.
[0098] For example, the device status data of a certain transmission node is shown in Table 1:
[0099] Table 1. Example of Transmission Node Device Status and Network Performance Data
[0100] time CPU utilization (%) Memory usage (%) Upload bandwidth (Mbps) Download bandwidth (Mbps) Packet loss rate (%) 2025 / 10 / 1510:00 85 65 12 10 0.5 2025 / 10 / 1510:05 88 70 13 11 0.7 2025 / 10 / 1510:10 80 60 11 9 0.3 2025 / 10 / 1510:15 90 75 14 12 0.9 2025 / 10 / 1510:20 85 68 13 10 0.6
[0101] After preprocessing, the data in Table 1 forms a standardized raw network dataset, which is used for subsequent fault diagnosis.
[0102] A hierarchical decision tree is used to extract and fuse features from the processed data. The original dataset is processed through a two-layer structure of the hierarchical decision tree model. The first layer classifies features from a single data source (such as CPU utilization, memory usage, etc.), generating preliminary features for each data source. The second layer integrates multiple preliminary features into the decision tree, fusing information from different data sources to finally generate a feature dataset. Each node split uses a dynamic splitting criterion; at each node split, the optimal splitting criterion is dynamically selected based on the data distribution and feature importance of the current node.
[0103] For example, by combining data on CPU utilization and bandwidth uploads, the hierarchical decision tree identified the key feature of "excessive network load," indicating that the system may be experiencing a performance bottleneck.
[0104] Based on the aforementioned feature dataset, a hierarchical decision tree is applied for fault mode classification and identification. For example, based on historical fault data and current feature data, the system identifies several common fault modes, including equipment overload, network congestion, and insufficient bandwidth. During implementation, the system continuously analyzes the fault mode dataset, which not only improves the accuracy of fault identification but also allows for real-time adjustment of the fault mode classification strategy to adapt to changes in different network conditions.
[0105] Following fault mode identification, an improved quantum annealing algorithm is used to optimize fault repair paths. The quantum annealing algorithm dynamically adjusts the cooling rate, introduces quantum state constraints, and employs a multi-scale annealing strategy to explore the solution space at multiple levels, generating the optimal repair path. The optimization of the repair path considers not only repair time but also the consumption of system resources during the repair process, such as CPU and bandwidth usage. Through a path optimization evaluation mechanism, each repair path is quantitatively evaluated, ultimately selecting the optimal repair path.
[0106] For example, in a device overload failure, the quantum annealing algorithm determined the optimal repair path, quickly reducing the system load to a safe threshold, thereby avoiding a large-scale network outage.
[0107] By optimizing the neuron coupling mechanism of a pulse-coupled neural network through self-organizing maps, time-series prediction is performed on the repair path dataset to predict the types and timing of potential future faults. This step helps the system identify potential fault risks in advance, enabling repairs before faults occur and avoiding service interruptions. Based on this, an improved fuzzy reinforcement learning approach is used to optimize the repair strategy using the fault prediction dataset. Fuzzy reinforcement learning dynamically updates the fuzzy rule base through incremental learning and employs adaptive fuzzy control and a fuzzy autoencoder to dynamically optimize the membership function, achieving adaptive optimization of the repair strategy.
[0108] Finally, the generated repair strategy dataset is fed back into the system to optimize the fault mode dataset and repair path dataset in real time. The system dynamically adjusts the repair paths based on the real-time feedback data and optimizes the system configuration according to the repair strategy dataset. Through multiple rounds of optimization and feedback, the system continuously improves the efficiency and accuracy of fault repair, ensuring stable network operation.
[0109] Verification using actual data shows that the digital transmission fault diagnosis method of this invention can significantly improve the efficiency of fault identification and repair. After applying this method, the system achieves real-time fault detection and optimized repair paths at multiple transmission nodes, reducing the average system fault response time by more than 30% and increasing the fault repair success rate by 20%. Furthermore, the accuracy of fault prediction is also improved, effectively reducing service interruptions caused by faults such as equipment overload and network congestion.
[0110] As illustrated by the above embodiments, the digital transmission fault diagnosis method of the present invention not only improves the accuracy of fault diagnosis and the efficiency of repair, but also greatly enhances the system's ability to cope with complex fault modes, thereby improving the reliability and stability of the entire network. These beneficial effects verify the feasibility and practicality of the present invention and demonstrate its broad application prospects.
[0111] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A digital transmission fault diagnosis method based on artificial intelligence, characterized in that, Includes the following steps: Collect multi-dimensional data from the transmission system and preprocess it to form the original network dataset; Hierarchical decision trees are used to fuse the original network dataset, extract key features, and generate a feature dataset. The hierarchical decision tree is applied to the feature dataset to perform fault mode classification and recognition, generating a fault mode dataset. Based on the fault mode dataset, the improved quantum annealing algorithm is used to optimize the fault repair path. By dynamically adjusting the cooling rate and introducing quantum state constraints, a multi-scale annealing strategy is adopted to explore the solution space at multiple levels, generate the optimal repair path and obtain the repair path dataset. By optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping, time series prediction is performed on the repair path dataset to generate a fault prediction dataset. Based on the fault prediction dataset, the repair strategy is optimized through improved fuzzy reinforcement learning. The method combines incremental learning and fuzzy inference to dynamically update the fuzzy rule base. Adaptive fuzzy control and fuzzy autoencoder are used to dynamically optimize the membership function and output the repair strategy dataset. The generated repair strategy dataset is fed back to optimize the fault mode dataset and repair path dataset in real time, dynamically adjust the repair path, and adjust the system configuration according to the repair strategy dataset to generate a new optimized dataset. The step of optimizing the neuron coupling mode of the pulse-coupled neural network through self-organizing mapping to perform time-series prediction on the repair path dataset and generate a fault prediction dataset specifically includes: Time series features and multi-dimensional information are extracted from the repair path dataset and converted into vector form to obtain multi-dimensional feature vectors. The self-organizing map algorithm is used to organize and map the multi-dimensional feature vectors to two-dimensional or three-dimensional grid nodes according to similarity to form a feature mapping matrix. Each node corresponds to a specific feature in the repair path dataset, and adjacent nodes represent similar repair path features. During the training of the self-organizing map algorithm, the weights of the nodes are updated through a competitive learning mechanism. Each time data is input, the Euclidean distance between the input data and each node is calculated, and the best matching unit closest to the input data is selected. The weights of the node and its corresponding neighboring nodes are updated according to the Gaussian neighborhood function. The feature mapping matrix is input into the pulse-coupled neural network for optimization. Based on the feature mapping matrix, the coupling strength between each neuron is calculated. The coupling strength is measured by Euclidean distance to measure the similarity between each pair of neurons. Based on the similarity, the coupling strength between neurons and the propagation speed of the pulse signal are adjusted. The coupling strength and the propagation speed of the pulse signal are proportional to the coupling strength between neurons. The feature mapping matrix is input into the optimized spiking coupled neural network. Through the propagation of spiking signals from neurons, the future change trend of repair path data is simulated, and the possible fault types, fault occurrence time, affected equipment, and execution order of repair paths are predicted. By utilizing the spiking propagation of neurons, the spiking coupled neural network captures the time dependence and evolution trend of each node in the repair path and generates a fault prediction dataset.
2. The method for diagnosing digital transmission faults based on artificial intelligence according to claim 1, characterized in that, The system collects and transmits multi-dimensional data, which is then preprocessed to form the original network dataset, specifically including: The multi-dimensional data includes device status data, network traffic data, system load data, environmental factor data, and fault history data; Kalman filtering and mean filtering algorithms are used to remove noise from multidimensional data. Missing data are filled in by interpolation or nearest neighbor imputation. Z-score is used for standardization. Principal component analysis is used to remove redundant features and retain effective features to form a standardized original network dataset.
3. The method for diagnosing digital transmission faults based on artificial intelligence according to claim 1, characterized in that, The process of fusing the original network dataset using a hierarchical decision tree to extract key features and generate a feature dataset specifically includes: The original network dataset is input into a hierarchical decision tree model. The first layer uses the decision tree to classify features of a single data source and generate preliminary features for each data source. The second layer inputs multiple preliminary features into the decision tree for integration and merges information from different data sources to generate a feature dataset. A dynamic splitting criterion is used when splitting each node. At each node split, the splitting criterion is dynamically selected based on the data distribution and feature importance of the current node. The splitting effect of each feature on the current node is evaluated by calculating the Gini index or information gain of each feature to improve the node purity. Features are selected for splitting based on the splitting effect. Adaptive tree depth control is adopted to automatically adjust the depth of the decision tree according to the complexity and sample size of the original network dataset. When the sample size of a node is lower than a predetermined threshold, or when the depth of the tree reaches a predetermined limit, the depth of the tree is automatically limited. Redundant branches are removed using an adaptive pruning algorithm. After generating the decision tree, the contribution of each branch of the tree is evaluated based on the performance of the validation set. If the improvement of the decision tree's accuracy by a certain branch is less than a preset threshold, the current branch is pruned. During the pruning process, if the error of a node is greater than a certain threshold or the node's contribution to the overall performance is less than a preset threshold, the current branch is removed.
4. The method for diagnosing digital transmission faults based on artificial intelligence according to claim 1, characterized in that, The improved quantum annealing algorithm specifically includes: The fault mode dataset is fed into the improved quantum annealing algorithm to optimize the fault repair path and generate the optimal repair path; During quantum annealing, quantum state constraints are set to limit the search space. The constraints include repair time and resource consumption. Repair time thresholds and resource consumption thresholds are applied to the repair path. A multi-scale annealing strategy is adopted to optimize the solution space at multiple levels in different annealing stages. In the initial stage, a coarse-grained global search is performed by setting a large step size, and a fine-grained local search is performed by gradually reducing the step size. The cooling rate and step size of each annealing stage are dynamically adjusted according to the search stage. Each repair path is quantitatively evaluated through a path optimization evaluation mechanism. The evaluation indicators include the total repair time, resource consumption, and success rate of the repair path. The feasibility of the path is evaluated based on the time threshold and resource consumption threshold of the repair path. Each repair path is scored according to the evaluation indicators. The optimal repair path with a score greater than the score threshold is selected and a repair path dataset is generated. The repair path dataset contains the optimal repair path and other candidate paths.
5. The method for diagnosing digital transmission faults based on artificial intelligence according to claim 1, characterized in that, The improved fuzzy reinforcement learning specifically includes: Based on the fault prediction dataset, the repair strategy is optimized by improving fuzzy reinforcement learning. By analyzing the time series data of the repair path and repair task, each possible repair strategy is evaluated, the reward value of each repair strategy is calculated, the Q-learning mechanism is used to update the Q-value function, and the repair strategy selection is continuously adjusted by optimizing the Q-value function to select the optimal path. Incremental learning is introduced to dynamically update and optimize the fuzzy rule base. After each repair path is executed, the existing rules are dynamically adjusted based on the actual execution effect and newly collected data, and the repair strategy is updated through fuzzy inference. Adaptive fuzzy control and fuzzy autoencoder are used to dynamically optimize the membership function. Based on the real-time feedback of the repair task information, the adaptive fuzzy control adjusts the shape and control parameters of the membership function. Fuzzy autoencoders are used to automatically optimize the parameters of membership functions. By automatically encoding and decoding, the membership functions in the fuzzy rule base are optimized. The fuzzy autoencoder learns the optimal repair path features through the process of minimizing errors. The repair strategy dataset is output by dynamically optimized fuzzy rules and repair strategies. The repair strategy dataset includes optimization schemes for repair paths, resource allocation schemes, repair priorities, and timing arrangements.
6. The method for diagnosing digital transmission faults based on artificial intelligence according to claim 1, characterized in that, The process of feeding back the generated repair strategy dataset, optimizing the fault mode dataset and repair path dataset in real time, dynamically adjusting the repair path, and adjusting the system configuration based on the repair strategy dataset to generate a new optimized dataset specifically includes: Using the repair strategy dataset as input, the repair path data is matched with the fault prediction dataset to generate a repair path execution sequence; The repair strategy is optimized by calculating the execution error of the repair task based on real-time feedback data, optimizing the order of the repair task through a weighted latency processing mechanism, and adjusting the priority of repair tasks whose errors exceed a preset threshold. Perform priority sorting and verification of the repair paths, calculate the repair success rate and latency of each repair path task, sort the repair paths, and demote high-latency repair tasks. Perform resource scheduling optimization and adjust the resource allocation of each repair path in the repair strategy dataset according to the real-time resource status of the system. The optimized repair path tasks are scheduled according to priority, repair time, and resource consumption to generate the optimal repair execution plan; Using the repair strategy dataset as input, it is dynamically updated during the execution of the repair path. Based on the feedback data of the repair task execution, the repair path and repair strategy are adjusted, and incremental corrections are made on the repair path that does not achieve the preset execution effect. After each incremental correction, the repair path sorting verification, resource scheduling optimization, and repair task execution error verification are re-executed. Repair path optimization continues in areas that have not reached the preset threshold until the verification results meet the preset requirements or the maximum number of iterations is reached, and a new optimized dataset is output.