Distributed Optical Fiber Temperature Anomaly Localization With CNN Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing distributed fiber-optics temperature sensing systems face challenges in accurately detecting and locating abnormal temperature events due to the cumbersome parameter adjustment and incompatibility issues arising from varying fiber lengths, leading to reduced positioning efficiency and accuracy.
Innovation Solution
A method utilizing a convolutional neural network trained with normalized anti-Stokes and Stokes data from a fiber-optic temperature sensing system, where data is segmented, labeled, and processed to identify abnormal temperature events by calculating cosine similarity between offset features and original data, effectively improving detection and localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If each signal in the RDTS is extended to a time series signal and individual models are established for each signal, then anomaly detection capability is improved, but the number of models increases with fiber length making parameter adjustment cumbersome and reducing positioning efficiency
Solution Approach 1:
The patent merges multiple individual time series models into a single unified model that processes all signals simultaneously. This is achieved by constructing a three-dimensional tensor from temperature data, spatial position data, and time series data, then applying a single anomaly detection model to the entire tensor, thereby maintaining high detection capability while eliminating the need to manage multiple separate models.
Solution Approach 2:
The patent creates a universal anomaly detection model that can handle signals from fiber lengths of different specifications through normalization processing. The model is designed to be specification-agnostic, accepting normalized input data and producing consistent anomaly detection results regardless of the original fiber length, thus making the system universally applicable across different deployment scenarios.
2Reliability
If individual models are established for each signal, then anomaly detection is performed, but parameter adjustment becomes cumbersome limiting application breadth and reducing positioning efficiency
Solution Approach 1:
The patent combines parameter management into a single unified structure. Instead of adjusting parameters for each individual model, the system uses one global parameter set for the unified model, dramatically simplifying operation and maintenance while ensuring consistent detection reliability across all signals.
3Length of stationary object
If the sensing fiber length increases, then measurement coverage is improved, but the number of models increases reducing positioning efficiency and accuracy
Solution Approach 1:
The patent implements a universal model architecture that maintains constant positioning efficiency regardless of fiber length. The model processes normalized data from any fiber length uniformly, eliminating the performance degradation that would normally occur as fiber length increases. This is achieved through normalization that scales data to a common reference frame.
Solution Approach 2:
The patent segments the processing into normalization (scaling to common reference) and unified model application, allowing the system to handle arbitrary fiber lengths by first normalizing the data structure then applying a single model, thereby decoupling processing complexity from physical fiber length.
4Measurement precision
If individual models are used for each signal, then anomaly detection is performed, but correlation between different measured values is split resulting in low positioning accuracy
Solution Approach 1:
The patent merges all temperature signals into a single three-dimensional tensor structure that preserves the inherent correlations between different spatial and temporal measurements. By processing the entire tensor simultaneously through a unified model, the system maintains and utilizes the correlation information that would be lost if signals were processed by separate models.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method enhances detection and localization accuracy of abnormal temperature events by reducing manual parameter adjustments and implicitly utilizing spatial relationships, improving compatibility across different fiber lengths.
Implementation Method 1
For distributed fiber-optics temperature sensing technology, Raman-based Distributed Temperature Sensing in English, and abbreviated as RDTS, each acquisition can obtain temperature information at different positions on the entire optical fiber.
Data Source
AI summary
A method of locating a temperature anomalies of a distributed optical fiber includes the steps of: (a) generating a training dataset having training samples; (b) setting labels for training samples; (c) building a convolutional neural network composed of multi-layer convolutional networks and a fully connected layer, training to form a convolutional neural network model; (d) utilizing a fiber-optic temperature sensing system for measurement of testing object; (e) sending acquired data into the convolutional neural network model to obtain output features, then processing mapping and binarization; (f) offsetting the binary feature to obtain an offset feature and calculating a cosine similarity; and (g) obtaining a location of the abnormal temperature event by identifying the offset feature with a largest cosine similarity and identifying its location in the sequence P.