Optical cable immersion icing state detection method based on reverse attention mechanism

Abstract

The invention discloses an optical cable immersion icing state detection method based on a reverse attention mechanism, and aims to improve the accuracy and efficiency of optical cable fault detection. According to the method, firstly, multi-parameter data, including but not limited to key parameters such as temperature, humidity, mechanical stress and current, in the operation process of an optical cable are collected in real time through multiple sensors; then, a reverse attention mechanism is used to carry out automatic feature weighting and screening on the collected multi-dimensional data, important features related to optical cable faults are highlighted, interference of irrelevant data on detection results is reduced, and feature extraction precision is improved. On the basis, in combination with a multi-agent reinforcement learning algorithm, the agents continuously interact in a simulation environment, and through a strategy optimization and reward mechanism, a decision strategy is dynamically adjusted, an error value is minimized, and the robustness and generalization ability of a detection model are improved. According to the method, the optical cable state can be accurately identified in a complex and changeable environment, and efficient diagnosis of faults is realized.

Description

Technical Field

[0001] The present invention belongs to the field of electric power communications, and in particular relates to a method for detecting the water immersion and icing state of an optical cable based on a reverse attention mechanism. Background Art

[0002] With the rapid development of modern communications and power systems, optical cables, as an important communication medium, are widely used in fields such as data transmission and remote monitoring. However, over long-term use, optical cables may be affected by factors such as temperature, humidity, and mechanical stress, leading to performance degradation or even failure. Traditional methods for detecting optical cable faults rely primarily on manual inspections, periodic checks, or threshold-based detection. These methods are limited in effectiveness when dealing with high-dimensional data, complex environments, and multiple variables.

[0003] How to accurately detect the status of optical cables based on high-dimensional data to ensure the normal operation of the power system is an urgent problem to be solved by people in this field. Summary of the Invention

[0004] To address this issue, this study proposes a method for detecting the water-immersion and icing status of optical cables based on an inverse attention mechanism. By introducing an inverse attention mechanism to weight multiple perceived parameters, this method highlights the features most relevant to cable fault detection, thereby improving real-time detection of the cable's operating status. This method, combined with reinforcement learning, allows for rapid fault identification and fault level assessment.

[0005] The inventors discovered that a significant advantage of the inverse attention mechanism is its ability to automatically identify and weight key features in input data. In optical cable status monitoring tasks, sensor data is typically high-dimensional, containing multiple features (such as temperature, humidity, and stress), which may have complex relationships between them. Traditional methods typically require manual feature selection or simple statistical metrics, while the inverse attention mechanism automatically weights different features based on task requirements, highlighting those most relevant to optical cable fault detection. Using the SAC reinforcement learning algorithm, the fault detection strategy can be continuously adjusted and optimized in a dynamically changing environment, enhancing the system's adaptability. After reinforcement learning training, the model can achieve real-time fault detection and diagnosis, meeting the requirements of real-time optical cable status monitoring. This method can adapt to a variety of sensor data inputs and has strong versatility, making it widely applicable to fault detection and status monitoring in optical cables, power lines, communication networks, and other fields.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: A method for detecting the water-immersion and icing state of an optical cable based on an inverse attention mechanism comprises the following steps: S1 builds a fiber optic cable detection model to collect multi-parameter data on fiber optic cable operation in real time. It then performs denoising, standardization, outlier detection, and data enhancement on the collected raw data to ensure data quality and consistency. This improves the stability and accuracy of subsequent model training and reduces the interference of abnormal data on the model.

[0007] S2 designs the inverse attention mechanism and attention inference network. The inverse attention mechanism can automatically identify abnormal data that is ignored by the traditional attention mechanism, improving the model's ability to identify hidden fault characteristics. The attention inference network automatically infers the importance of each feature and dynamically adjusts the attention weight.

[0008] S3 introduces the inverse attention mechanism into the reinforcement learning algorithm, which is a soft actor-critic algorithm. It performs feature weighting on the input multi-parameter data, so that the algorithm pays more attention to the key data related to the fault and gives the corresponding fault type.

[0009] S4 initializes the parameters of the actor and critic networks and builds an experience replay pool to improve the model's data utilization and processing speed. The SAC algorithm, which incorporates the inverse attention mechanism, is iterated repeatedly until the loss function converges. During the optimization process, the weights of the inverse attention mechanism are dynamically adjusted based on different fiber optic cable operation scenarios.

[0010] S5, deploys the trained SAC model into the optical cable detection system to achieve real-time data collection, analysis and fault warning.

[0011] The present invention discloses the following technical effects: The present invention provides a method for detecting the water-immersion and icing state of an optical cable based on a reverse attention mechanism, comprising the steps of: constructing an optical cable detection model to collect data collected by different sensors and preprocessing the collected data to ensure data quality and consistency; introducing a reverse attention mechanism and an attention inference network design, introducing the reverse attention mechanism into the model to automatically identify abnormal data ignored by traditional attention mechanisms, thereby improving the model's ability to identify hidden fault features; designing an attention inference network to automatically infer the importance of each feature and dynamically adjust the attention weights, so that the model pays more attention to key and variable parameters; introducing the reverse attention mechanism into a SAC algorithm; and training and optimizing the SAC algorithm incorporating the reverse attention mechanism; deploying the trained SAC model into an optical cable detection system to achieve real-time data collection, analysis, and fault warning. In the detection of the water-immersion and icing state of an optical cable, the present invention introduces the reverse attention mechanism into the SAC algorithm. By introducing the reverse attention mechanism, weights are assigned to multiple parameters sensed, thereby improving the real-time detection of the operating state of the optical cable, wherein a reinforced school is used to quickly find and diagnose the fault. It solves the shortcomings of existing methods in complex environments and provides a new solution for optical cable fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0013] Figure 1 An embodiment of the present invention provides a method for detecting the water immersion and icing state of an optical cable based on a reverse attention mechanism. DETAILED DESCRIPTION

[0014] As described in the background technology section, how to accurately detect the status of optical cables based on high-dimensional data to ensure the normal operation of the power system is an urgent problem to be solved by people in this field.

[0015] The core idea of ​​this invention is to introduce an inverse attention mechanism to weight the multiple parameters sensed during the detection of optical cable immersion and icing, highlighting the features most relevant to optical cable fault detection, thereby improving the real-time detection of the optical cable's operating status. Reinforcement learning is then used to quickly locate faults and determine the fault level.

[0016] See also Figure 1 An embodiment of the present invention provides a method for detecting the icing state of an optical cable based on a reverse attention mechanism, the method comprising: S1 builds a fiber optic cable detection model to collect multi-parameter data on fiber optic cable operation in real time. It then performs denoising, standardization, outlier detection, and data enhancement on the collected raw data to ensure data quality and consistency. This improves the stability and accuracy of subsequent model training and reduces the interference of abnormal data on the model.

[0017] S2, the design of the inverse attention mechanism and attention inference network. The inverse attention mechanism can automatically identify abnormal data that is overlooked by traditional attention mechanisms, improving the model's ability to identify hidden fault characteristics. The attention inference network automatically infers the importance of each feature and dynamically adjusts the attention weight.

[0018] S3 introduces the inverse attention mechanism into the reinforcement learning (SAC) algorithm. This reinforcement learning algorithm is a soft actor-critic algorithm that performs feature weighting on the input multi-parameter data, allowing the algorithm to pay more attention to key data related to the fault and give the corresponding fault type.

[0019] S4 initializes the parameters of the actor and critic networks and builds an experience replay pool to improve the model's data utilization and processing speed. The SAC algorithm, which incorporates the inverse attention mechanism, is iterated repeatedly until the loss function converges. During the optimization process, the weights of the inverse attention mechanism are dynamically adjusted based on different fiber optic cable operation scenarios.

[0020] S5, deploys the trained SAC model into the optical cable detection system to achieve real-time data collection, analysis and fault warning.

[0021] This application takes into account that since traditional methods often only focus on a single or a few indicators, it is difficult to detect a variety of potential faults in a complex environment in a timely manner, and multi-parameter acquisition can improve the comprehensiveness and accuracy of detection. When the optical cable is running in a complex environment, the fault may be caused by a variety of factors, such as temperature, humidity, mechanical stress and electrical parameters. Therefore, the use of a variety of sensors to comprehensively collect multi-parameter data of the optical cable can ensure the comprehensiveness and accuracy of the fault information. The preprocessing step ensures high consistency and low noise of the data, reduces false alarms and missed reports, and enhances the effect of subsequent model training. Therefore, in the step S1, the multi-parameter data of the optical cable operation is collected in real time, and the collected raw data is subjected to denoising, normalization, outlier detection and data enhancement steps to ensure the quality and consistency of the data. This can improve the stability and accuracy of subsequent model training and reduce the interference of abnormal data on the model, specifically including: First, we design a fiber optic cable detection model. The state of the fiber optic cable can be detected by a variety of sensors, including temperature sensors, humidity sensors, and strain sensors. We can use the data of these sensors as multi-parameter inputs of the fiber optic cable state. is the multi-dimensional sensor data of the optical cable, where For the Observation data from a sensor.

[0022] Then, perform data preprocessing and use filters (such as low-pass filters) to remove noise. In order to avoid detection differences between different sensors, the data is standardized: (1) in, and They are The mean and standard deviation of the sensor data are calculated. When missing data is detected during outlier detection, interpolation is used to fill in the missing data to enhance the usability of the data.

[0023] This method introduces the reverse attention mechanism and the attention inference network. Since traditional methods usually use static or manual feature selection, they are prone to missing hidden fault features. The reverse attention mechanism reversely strengthens weak single potential important features to ensure that the model will not miss important fault clues. Since the attention inference network can dynamically adjust the feature weights according to different environments and data, the generalization ability and real-time response ability of the model are improved. Therefore, compared with manual feature selection, automated weight inference reduces the dependence on professional knowledge. By introducing the reverse attention mechanism and the attention inference network into the intelligent agent, the reverse attention mechanism can automatically identify abnormal data that is ignored by the traditional attention mechanism, and improve the model's ability to recognize hidden fault features. Through the attention inference network, the importance of each feature is automatically inferred, and the reverse attention weight is dynamically adjusted. Therefore, in the step S2, the reverse attention mechanism and the attention inference network are introduced into the intelligent agent, specifically including: First, the reverse attention mechanism is designed. The goal of the reverse attention mechanism is to use attention weights The model pays more attention to the features that are useful for cable status detection and reduces the interference of useless features. Unlike traditional attention, the inverse attention mechanism uses the opposite direction of the original mechanism to adjust the weights to enhance the focus on useful data. The forward attention weight is calculated using the attention inference network. Expressed as: (2) in is the activation value after feature extraction. Then the reverse attention weight for:

[0024] Secondly, the goal of the inverse attention mechanism is to use attention weights Let the model pay more attention to the features that are useful for cable status detection and reduce the interference of useless features. Different from traditional attention, the inverse attention mechanism uses the opposite direction of the original mechanism to adjust the weights to enhance the focus on useful data. The update process is: (3) in, is the loss function, is the predicted cable status, is the learning rate, is the gradient of the loss function with respect to the attention weight.

[0025] Because the SAC algorithm has high sample efficiency and good stability, it is suitable for handling the complex and changing data environment of optical cable status detection. The introduction of the inverse attention mechanism can improve the focus on key features during the reinforcement learning process, allowing the agent to focus more on fault-related data during strategy learning. In step S3, the inverse attention mechanism is introduced into the reinforcement learning algorithm (soft actor-critic), specifically including: First, the optical cable status detection problem is modeled as a reinforcement learning problem. The agent is defined as a detection system that obtains rewards by adjusting the detection strategy through action selection. Therefore, the state space is defined as: the data set after preprocessing the data sensed by the optical cable sensor is used as the state space . Define the action space as: ,in Indicates that the perception data in the current state is normal. Indicates that there is a fault in the perception data in the current state. If it is judged to be a fault, the corresponding fault type will be output according to the corresponding state data. Reward function: If the agent makes a correct judgment, it will receive a positive reward; otherwise, it will receive a negative reward: (4) Secondly, reverse attention is introduced into the SAC algorithm, and reverse attention is combined with the perception data in the state space to obtain a new state space Expressed as: (5) in, Represents the state input after inverse attention optimization, which can enhance the main data features of the perception data in the state space and reduce the interference data features.

[0026] Since traditional static methods usually require a large amount of labeled data, SAC improves sample efficiency and reduces data dependence through experience replay. Through multiple iterations and strategy optimization, it can ensure that the model has high recognition capabilities in different scenarios. Initialize the parameters of the Actor and Critic networks, build an experience replay pool, and improve the model's data utilization and processing speed. The SAC algorithm that introduces the attention mechanism is iterated repeatedly until the loss function converges. During the optimization process, the weight of the inverse attention mechanism is dynamically adjusted according to different optical cable operation scenarios. Therefore, in the step S4, the SAC algorithm is trained and optimized. Specifically including: First, since the SAC algorithm consists of two Critic networks and one Actor network, the parameters are initialized. network and Actor Policy Network , value function network , and create the target value function network .

[0027] Secondly, the SAC algorithm of the inverse attention mechanism is trained to interact with the agent through the environment, generate data according to the current strategy, and generate an optimized state space for the data collected by the sensor through the inverse attention weight. , select an action , receive rewards and the next state , storage experience To the experience pool .

[0028] Again, calculate the target Value: From the Actor network Generate Action , calculate the target Value (double minimum value of the network, reducing overestimation) (6) in, and is the status Next Actor network generates action of value, is the output of the policy network, is the entropy weight, is the discount factor.

[0029] Update the Critic network using the minimized mean squared error (MSE): (7) (8) Value function update: The value function estimates the long-term value of the policy, and the target value is: (9) Update the value function by minimizing the mean squared error: (10) Updating the Policy Network (Actor Update): Introducing Noise When Sampling Actions , ensuring gradient propagation: (11) in, Represents the action mean output by the Actor network, Indicates the standard deviation of the actions output by the Actor network.

[0030] By maximizing the entropy of the policy and Value function to optimize the policy network: (12) The goal of the policy network is to select those that maximize actions of high value while maintaining high exploratory potential.

[0031] Update the target network: The target network is usually updated using a soft update strategy: (13) (14) (15) in, is the soft update parameter, usually 0.005.

[0032] Finally, repeated training and multiple interactions allow the SAC agent to gradually optimize its strategy, so that under a given state, it can accurately determine whether the optical cable is faulty and maximize long-term rewards based on feedback from sensor data.

[0033] Sensors continuously collect multi-parameter data and input it into a trained model for status analysis. Through the inverse attention mechanism, the model can quickly identify abnormal patterns in the data and determine the health status of the optical cable in real time. The system also records abnormal data as a reference for subsequent model optimization and iteration. Therefore, the S5 deploys the trained SAC model into the optical cable detection system, enabling real-time data collection, analysis, and fault warning. Specifically, this includes: First, the trained policy network can receive sensor data in real time and determine the health of the optical cable based on the current state. The policy network outputs whether the action is normal or faulty. When the output is faulty, an alarm or maintenance operation is triggered.

[0034] Secondly, long-term optimization is carried out. Through reinforcement learning, the intelligent agent continuously explores the environment and optimizes the fault detection strategy to avoid false positives or missed positives.

[0035] Compared with the existing technology, the above technical solution has the following advantages: This invention utilizes an inverse attention mechanism and reinforcement learning to detect multi-parameter optical cable status. A significant advantage of the inverse attention mechanism is its ability to automatically identify and weight key features in input data. By dynamically calculating feature weights, the inverse attention mechanism can flexibly adjust the focused features based on data changes under varying environmental conditions. In multi-parameter sensor data, some features may have no direct correlation with the status of the optical cable fault, or their impact on the status may be very small. By assigning lower weights to these irrelevant features, the inverse attention mechanism can effectively reduce their interference with model decision-making. This not only improves the accuracy of fault detection, but also reduces the model's reliance on redundant data, improving efficiency. This enables real-time detection of the optical cable status.

[0036] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for detecting the icing state of an optical cable based on a reverse attention mechanism, characterized in that: The steps include: S1, by building an optical cable detection model, collects multi-parameter data of optical cable operation in real time, and performs denoising, standardization, outlier detection and data enhancement steps on the collected raw data to ensure data quality and consistency, thereby improving the stability and accuracy of subsequent model training and reducing the interference of abnormal data on the model; S2, the design of the inverse attention mechanism and attention inference network. The inverse attention mechanism can automatically identify abnormal data that is ignored by the traditional attention mechanism, improving the model's ability to identify hidden fault characteristics. The attention inference network automatically infers the importance of each feature and dynamically adjusts the attention weight. S3 introduces the inverse attention mechanism into the reinforcement learning algorithm, which is a soft actor-critic algorithm. It performs feature weighting on the input multi-parameter data, allowing the algorithm to pay more attention to key data related to the fault and give the corresponding fault type. S4 initializes the parameters of the Actor and Critic networks, builds an experience replay pool, improves the model's data utilization and processing speed, and repeatedly iterates the reinforcement learning algorithm with the inverse attention mechanism until the loss function converges. During the optimization process, the weight of the inverse attention mechanism is dynamically adjusted according to different optical cable operation scenarios. S5, deploys the trained optical cable detection model into the optical cable detection system to achieve real-time data collection, analysis and fault warning.

2. The optical cable immersion and icing state detection method based on the reverse attention mechanism according to claim 1 is characterized in that: Step S1 specifically includes: First, the optical cable detection model is designed. The state of the optical cable can be detected by a variety of sensors, including temperature sensors, humidity sensors, and strain sensors. The data of these sensors are used as multi-parameter inputs of the optical cable state. is the multi-dimensional sensor data of the optical cable, where For the Observation data of sensors; Then, the data is preprocessed and noise is removed using a low-pass filter. In order to avoid detection differences between different sensors, the data is standardized: (1) in, and They are The mean and standard deviation of the sensor data are calculated. When missing data occurs during outlier detection, interpolation is used to fill it in order to enhance the usability of the data.

3. The optical cable immersion and icing state detection method based on the reverse attention mechanism according to claim 1 is characterized in that: Step S2 specifically includes: First, design the reverse attention mechanism and use the attention inference network to calculate the forward attention weights Expressed as: (2) in is the activation value after feature extraction, then the reverse attention weight for: ; Secondly, the goal of the inverse attention mechanism is to inverse the attention weights Let the model pay more attention to the features that are useful for cable status detection and reduce the interference of useless features. Different from traditional attention, the inverse attention mechanism uses the opposite direction of the original mechanism to adjust the weights to enhance the focus on useful data. The update process is: (3) in, is the loss function, is the predicted cable status, is the learning rate, is the gradient of the loss function with respect to the attention weight.

4. The optical cable immersion and icing state detection method based on the reverse attention mechanism according to claim 1 is characterized in that: Step S3 specifically includes: First, the optical cable status detection problem is modeled as a reinforcement learning problem. The intelligent agent is defined as a detection system that adjusts the detection strategy by action selection to obtain rewards. Therefore, the state space is defined as: the data set after preprocessing the data perceived by the optical cable sensor is used as the state space , the action space is defined as: ,in Indicates that the perception data in the current state is normal. Indicates that there is a fault in the perception data in the current state. If it is judged to be a fault, the corresponding fault type will be output according to the corresponding state data. Reward function: If the agent makes a correct judgment, it will receive a positive reward; otherwise, it will receive a negative reward: (4) Secondly, the reverse attention is introduced into the SAC algorithm, and the reverse attention weight is combined with the perception data in the state space to obtain a new state space Expressed as: (5) in, Representing the state input after inverse attention optimization, the perception data in the state space can enhance the key data features related to the fault and reduce the interference data features that are not related to the fault.

5. The optical cable immersion and icing state detection method based on the reverse attention mechanism according to claim 1 is characterized in that: Step S4 specifically includes: First, since the reinforcement learning algorithm consists of two Critic networks and one Actor network, the parameters are initialized. network , Policy Network in Actor Network , and the value function network , and create the target value function network ; Secondly, the reinforcement learning algorithm of the inverse attention mechanism is trained to interact with the agent through the environment, generate data according to the current strategy, and generate an optimized state space for the data collected by the sensor through the inverse attention weight. , select an action , receive rewards and the next state , storage experience To the experience pool ; Again, calculate the target Value: From the Actor network Generate Action , calculate the target Value (double the minimum value of the network, reducing overestimation); (6) in, and is the status Next Actor network generates action of value, is the output of the policy network, is the entropy weight, is the discount factor; Update the Critic network using the minimized mean squared error (MSE): (7) (8) Value function update: The value function estimates the long-term value of the policy, and the target value is: (9) Update the value function by minimizing the mean squared error: (10) Updating the Policy Network (Actor Update): Introducing Noise When Sampling Actions , ensuring gradient propagation: (11) in, represents the action mean output by the Actor network, Represents the standard deviation of the action output by the Actor network; By maximizing the entropy of the strategy and Value function to optimize the policy network: (12) The goal of the policy network is to select those actions that maximize value while maintaining high exploratory potential; Update the target network: The target network is usually updated using a soft update strategy: (13) (14) (15) in, is the soft update parameter, usually 0.005; Finally, repeated training and multiple interactions allow the SAC agent to gradually optimize its strategy, so that under a given state, it can accurately determine whether the optical cable is faulty and maximize long-term rewards based on feedback from sensor data.

6. The optical cable immersion and icing state detection method based on the reverse attention mechanism according to claim 1 is characterized in that: Step S5 specifically includes: First, the trained optical cable detection model is deployed in the optical cable detection system, which receives sensor data in real time, determines the health status of the optical cable based on the current status, and acts as a normal or faulty output based on the policy network. When the output is faulty, an alarm or maintenance operation is triggered; Secondly, long-term optimization is carried out. Through reinforcement learning, the intelligent agent continuously explores the environment and optimizes the fault detection strategy to avoid false positives or missed positives.

Images