Transformer substation optical fiber fault positioning method and system based on virtual-real fusion and deep learning

Abstract

The invention relates to the technical field of transformer substation optical fiber communication, in particular to a transformer substation optical fiber fault positioning method and system based on virtual-real fusion and deep learning. The transformer substation optical fiber fault positioning system based on virtual-real fusion and deep learning mainly comprises the following modules: (1) an offline topology construction module, (2) a multi-source data acquisition module, (3) a data preprocessing unit, (4) a deep learning processing module, (5) a fault accurate positioning module, and (6) an alarm and interaction module. The dynamic virtual-real fusion and D-S evidence theory verifies that the positioning error of the multimode optical fiber loop within 100m is less than or equal to 5m, and is greatly reduced compared with the traditional OTDR (Optical Time Domain Reflectometer). The mode dispersion hidden loss problem and the joint loss increase problem of the multimode optical fiber can be accurately identified, the omission ratio is reduced, and the protection maloperation caused by hidden faults is avoided.

Description

Technical Field

[0001] This invention relates to the field of fiber optic communication technology for substations, and in particular to a method and system for locating fiber optic faults in substations based on virtual-real fusion and deep learning. Background Technology

[0002] In smart substations, fiber optic loops are the core carriers for information transmission and real-time control. Fault location technologies are mainly divided into two major directions: physical modeling and data-driven approaches. Physical modeling technologies are based on the propagation characteristics of light, achieving location by collecting physical quantities such as optical power and reflected signals; a typical example is the traditional Optical Time-Domain Reflectometer (OTDR). Data-driven technologies rely on machine learning and deep learning algorithms to mine fault data features; common models include Bidirectional Long Short-Term Memory (BiLSTM) networks and Convolutional Neural Networks (CNNs). Furthermore, Substation Configuration Description (SCD) files are often used to construct virtual-physical loop mapping relationships, providing a topological foundation for fault location.

[0003] In physical modeling technologies, traditional OTDRs have a 15m test dead zone, and the positioning error of virtual loops within 100m exceeds 30m, requiring the interruption of fiber optic services for testing. The virtual-to-real loop mapping technology using SCD files relies on file accuracy; errors or missing data can reduce positioning accuracy by more than 30%, and it cannot handle latent faults such as modal dispersion and material aging in multimode fibers. Data-driven technologies, such as those based on Bidirectional Long Short-Term Memory (BiLSTM) networks and Convolutional Neural Networks (CNNs), require a large number of labeled fault samples. However, actual fault samples in substations are scarce, easily leading to model overfitting. Furthermore, relying solely on a single OTDR signal results in decreased accuracy under strong electromagnetic interference, making it unsuitable for scenarios involving physical damage and high environmental temperatures. Insufficient adaptability to multiple scenarios is also a limitation of existing technologies. Fiber optic virtual loops in smart substations are mostly short-distance transmissions within 100m, with a high proportion of multimode fibers. Existing technologies lack dedicated recognition models and cannot accurately characterize the coupling effects of modal dispersion, connector reflection, and material aging. Meanwhile, the online monitoring system has poor interface compatibility with the Supervisory Control and Data Acquisition (SCADA) system, with data interaction delays exceeding 40ms, resulting in missed fault warning windows.

[0004] In summary, existing fiber optic fault location technologies for smart substations have significant shortcomings in both core performance and practical application. Physical modeling technologies are limited by hardware defects and static topology, while data-driven technologies face challenges such as scarce samples and insufficient anti-interference capabilities. Furthermore, both types of technologies suffer from poor adaptability to multiple scenarios and high system interaction latency, making it difficult to meet the core requirements of smart substations for accurate fault location, business continuity, and efficient operation and maintenance. Therefore, an innovative solution that integrates the advantages of multiple technologies is urgently needed to overcome the existing bottlenecks. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for locating fiber optic faults in substations based on virtual-real fusion and deep learning, in order to address the above-mentioned shortcomings.

[0006] The technical problems solved by this invention are: 1) Overcoming the dead zone limitation of OTDR testing, reducing the positioning error of short-distance multimode fiber loops, and avoiding positioning failures caused by reliance on the accuracy of SCD files; 2) Reducing the model's need for labeled fault samples, improving the accuracy of fault positioning under strong electromagnetic interference environments, and adapting to multi-fault coupling scenarios; 3) Achieving co-fiber transmission of fault monitoring and service signals, avoiding interruption of fiber optic services during testing, and developing positioning solutions adaptable to multiple scenarios such as short-distance multimode fiber, mountain substations, and small and medium-sized substations; 4) Optimizing system interface design, reducing data interaction latency, and ensuring timely push of fault warnings.

[0007] The substation fiber optic fault location method and system based on virtual-real fusion and deep learning is implemented using the following technical solutions:

[0008] The substation fiber optic fault location system based on virtual-real fusion and deep learning is grounded in the deep mapping of the "physical circuit-logical circuit" of intelligent substations. It integrates multi-source sensing data and deep learning algorithms to construct a full-process location system encompassing "data acquisition - virtual-real modeling - intelligent diagnosis - visualization presentation," including the following modules:

[0009] 1) Offline Topology Construction Module: The offline topology construction module is the fundamental support module for the system to achieve "virtual-physical fusion" positioning. It consists of an SCD parser, a physical link parsing unit, a topology mapping engine, and a database, and is activated during the equipment installation and commissioning phase. This module imports SCD and physical link files, parses equipment configuration and optical cable parameters, establishes a "virtual loop ID - physical optical cable ID - connector location" mapping table, generates and stores a visual topology map, and provides virtual and physical anchor points for subsequent positioning.

[0010] 2) Multi-source data acquisition module: This module is the core of the system for real-time monitoring. It includes a wavelength division multiplexer, an optical monitoring unit (OMU), a distributed temperature sensor (DTS), a bit error rate detector, and an Ethernet transmission unit, and operates continuously during routine monitoring. The multi-source data acquisition module uses wavelength division multiplexing to achieve simultaneous transmission of service and monitoring signals along the same fiber. It acquires optical power difference ΔP, optical cable temperature, and GOOSE / SV message bit error rate at 100ms intervals, and transmits the data to the preprocessing unit via Ethernet.

[0011] 3) Data Preprocessing Unit: This unit is crucial for connecting multi-source data acquisition and intelligent computing. It consists of a 3σ outlier removal module, a Min-Max standardization module, and an early warning triggering unit. The data preprocessing unit removes electromagnetic interference outliers and standardizes the data to the [0,1] range. When ΔP exceeds the -10 to -28 dB threshold, a fault warning is immediately triggered.

[0012] 4) Deep Learning Processing Module. The deep learning processing module is the intelligent core module for the system to achieve initial fault localization. It consists of a denoising convolutional autoencoder (DCAE) and a bidirectional long short-term memory network (BiLSTM), and is activated after the fault warning is triggered.

[0013] 5) Precise Fault Location Module. The precise fault location module is a key module for achieving high-precision positioning in the system. It consists of an OTDR and DS evidence theory fusion unit, and is activated after the deep learning processing module outputs preliminary results. This module uses the OTDR to acquire the optical cable reflection peak characteristics, fuses the BiLSTM results, reflection peak characteristics, ΔP, and bit error rate, and outputs the final fault location with a positioning error ≤5m.

[0014] 6) Alarm and Interaction Module. The alarm and interaction module is the terminal module for fault information transmission and maintenance guidance. It includes a visual topology display unit, an IEC61970 protocol interface, an audible and visual alarm unit, an SMS alarm unit, and a fault handling report generation unit. It is activated after accurate fault location is completed. The alarm and interaction module is used to overlay fault information onto the topology map, push it to the SCADA system via protocol (delay ≤20ms), trigger on-site and SMS alarms, and generate an maintenance suggestion report.

[0015] A fault location method for substation fiber optic fault location system based on virtual-real fusion and deep learning is proposed. This method utilizes the deep collaborative design of virtual-real fusion and deep learning technologies to overcome the bottlenecks of traditional location technologies through their bidirectional empowerment. The specific working process is as follows:

[0016] First, during the equipment installation and commissioning phase, offline topology initialization is performed. The SCD file and physical link configuration file of the target substation are imported. The SCD parser extracts the IED device list and virtual terminal link table, and parses the virtual terminal links and network communication configurations of the secondary equipment. The physical link configuration file extracts the optical cable type, length, connector coordinates, and laying path. The topology mapping engine establishes a mapping table of "virtual loop ID - physical optical cable ID - connector location" and generates a visual topology map, which is then stored in the database. For detailed procedures, please refer to [reference needed]. Figure 1 .

[0017] After entering the daily monitoring phase, the multi-source data acquisition module achieves same-fiber transmission through a wavelength division multiplexer. The optical monitoring unit (OMU) acquires the optical power at both ends of the optical cable in real time and calculates the power difference ΔP. The distributed temperature sensing (DTS) acquires the temperature along the entire optical cable. The bit error rate detector monitors the GOOSE / SV message bit error rate with an acquisition cycle of 100ms. After the data is transmitted to the preprocessing unit via Ethernet, outliers caused by electromagnetic interference are removed using the 3σ criterion (μ-3σ≤x≤μ+3σ). Then, the data is processed to the [0,1] range using the Min-Max normalization formula (x'=(x-x_min) / (x_max-x_min)). If ΔP exceeds the preset threshold (-10~-28dB), a fault warning is triggered. The flowchart for this phase can be found in the diagram. Figure 2 .

[0018] Then, the initial fault localization stage begins. The standardized dataset is input into the denoising convolutional autoencoder (DCAE) module to remove noise caused by 60dB strong electromagnetic interference. The denoised data is then fed into the BiLSTM module, which captures the temporal features of the data through two hidden layers (128 units per layer, Dropout=0.2) and outputs the initial fault location and corresponding confidence level.

[0019] Finally, the process enters the precise fault location and alarm stage. The OTDR module is triggered (test time < 10ms) to acquire the reflection peak characteristics of the target optical cable. Using DS evidence theory, four sets of evidence are fused: BiLSTM output results, OTDR reflection peak characteristics, optical power difference, and bit error rate. The location corresponding to the highest confidence level is calculated and selected as the final fault location (error ≤ 5m). Virtual loop information regarding the fault type, location, and impact is overlaid onto the visualized topology map and pushed to the SCADA system via the IEC61970 protocol. Simultaneously, on-site audible and visual alarms and SMS alarms for maintenance personnel are triggered. A fault handling report is generated, including priority activation of backup optical cables and repair during off-peak hours. This forms a complete closed-loop process from topology construction, data acquisition, fault identification to result output and alarms. For details of this stage, please refer to [link / reference needed]. Figure 3 The overall system flowchart can be referenced. Figure 4 .

[0020] The core synergistic effect of this invention lies in the deep integration of three major technologies: First, the virtual-physical fusion topology provides precise anchor points for multi-source data acquisition. Through a mapping table of "virtual loop ID - physical optical cable ID - connector location," scattered data such as OMU optical power difference, DTS temperature, and GOOSE / SV message error rate are bound to specific links, avoiding the problem of isolated and unrelated multi-source data in traditional methods. Second, multi-source data can also correct the topology in reverse. When data and topology mismatches, the mapping relationship can be automatically verified and updated, solving the problem of static topology relying on manual configuration and prone to failure. The combination of these two technologies makes data acquisition more targeted and the topology more dynamically adaptable. Third, DS evidence fusion further integrates BiLSTM time-series analysis results, OTDR reflection peak characteristics, and multi-source data. It utilizes BiLSTM to capture the temporal patterns of latent faults, provides direct physical layer evidence through OTDR, and verifies fault consistency based on multi-source data. Compared to location methods that rely solely on a single type of data, this significantly improves the accuracy of fault identification. Meanwhile, the DCAE denoising module preserves the effective timing characteristics of BiLSTM under strong electromagnetic interference. BiLSTM, in turn, accurately triggers OTDR through output confidence (test time < 10ms). Combined with wavelength division multiplexing, it enables the transmission of service and monitoring signals on the same fiber. The three work together to solve the problem of poor data quality under strong interference and avoid frequent OTDR start-up interrupting services. Ultimately, it achieves a multimode fiber positioning error of ≤ 5m within 100m, an identification accuracy of ≥ 99% under 60dB interference, and zero service interruption, which far exceeds the performance of each technology when applied alone.

[0021] Beneficial effects of this invention:

[0022] (1) High positioning accuracy solves the pain point of short-distance multimode fiber.

[0023] Through dynamic virtual-real fusion and DS evidence theory verification, the multimode fiber loop positioning error within 100m is ≤5m, which is significantly reduced compared to traditional OTDRs. It can accurately identify the latent loss problem of mode dispersion and the problem of increased connector loss in multimode fibers, reducing the missed detection rate and avoiding protection malfunctions caused by latent faults.

[0024] (2) Zero business interruption, adaptable to uninterrupted operation and maintenance.

[0025] The invention employs wavelength division multiplexing (WDM) technology to enable the transmission of service signals and monitoring signals on the same fiber. The OTDR is activated only during fault warnings (test time < 10ms), eliminating the risk of service interruption. Even during peak periods of protection signal transmission, the invention can still stably collect data without affecting the real-time performance of GOOSE / SV messages.

[0026] (3) It has strong anti-interference ability and can adapt to complex field environments.

[0027] The DCAE module can improve the signal-to-noise ratio of -5dB low data to over 16dB, and maintain a fault identification accuracy of ≥99% even under 60dB strong electromagnetic interference. Compared with existing single-source data models, it improves accuracy and anti-interference capabilities, and can operate stably in strong interference areas near substation GIS equipment and high-voltage circuit breakers.

[0028] (4) The project is highly practical and easy to promote and apply.

[0029] This invention features low sample requirements. By combining simulation samples with transfer learning, the sample size requirement is reduced, allowing for rapid model deployment in newly built substations. It boasts good interface compatibility, supporting standard protocols such as IEC61850 and IEC61970, and can be directly integrated into existing SCADA systems with data interaction latency ≤20ms. Cost is relatively controllable; the core hardware (OMU, DTS) uses domestically produced equipment, reducing the single-station deployment cost by 30% compared to imported solutions, making it suitable for large-scale deployment. Attached Figure Description

[0030] The present invention will be further described below with reference to the accompanying drawings.

[0031] Figure 1 This is the offline topology initialization flowchart of the present invention;

[0032] Figure 2 This is a flowchart of the daily monitoring process of this invention;

[0033] Figure 3 This is a flowchart of the fault location and alarm process of the present invention;

[0034] Figure 4 Flowchart of a substation fiber optic fault location method based on virtual-real fusion and deep learning;

[0035] Figure 5 A schematic diagram of a substation fiber optic fault location system based on virtual-real fusion and deep learning. Detailed Implementation

[0036] See attached document Figure 1-5 The substation fiber optic fault location system based on virtual-real fusion and deep learning includes an SCD resolver, a physical link resolution unit, a topology mapping engine, a database, a wavelength division multiplexer, an optical monitoring unit (OMU), a distributed temperature sensor (DTS), a bit error rate detector, and an Ethernet transmission unit.

[0037] The system includes a 3σ outlier removal module, a Min-Max normalization module, an early warning triggering unit, a denoising convolutional autoencoder (DCAE), a bidirectional long short-term memory network (BiLSTM), an optical time domain reflectometer (OTDR), a DS evidence theory fusion unit, a visualization topology display unit, an IEC61970 protocol interface, an audible and visual alarm unit, an SMS alarm unit, and a fault handling report generation unit. Their functions are as follows:

[0038] The SCD parser's core function is to import the substation SCD (Substation Configuration Description) file and extract information related to "virtual loops," such as the IED (Intelligent Electronic Device) list, secondary equipment virtual terminal link table, and network communication configuration (e.g., GOOSE / SV message transmission paths). This clarifies the logical connection relationships of secondary equipment within the substation and forms the logical foundation for constructing a "virtual-physical" mapping. The SCD parser utilizes existing digital substation SCD parsing tools.

[0039] The Physical Link Resolution Unit is responsible for importing the substation's physical link configuration file and extracting the core optical cable parameters recorded in the file. These parameters include the optical cable type (e.g., multimode / single-mode), laying length, joint coordinates (accurate to the specific location), and laying path (e.g., cable trench / aerial / underground), providing physical-dimensional data support for subsequent association of virtual loops with physical optical cables. The Physical Link Resolution Unit is developed using Python / C++ and is compatible with the substation's configuration management system.

[0040] Topology Mapping Engine: The core execution unit of the virtual-physical fusion system, its function is to associate and match the "virtual loop information" output by the SCD parser with the "physical loop information" output by the physical link parser, constructing a one-to-one mapping table of "virtual loop ID - physical optical cable ID - connector location". Simultaneously, it automatically generates a visual topology diagram (intuitively displaying the binding relationship between virtual loops and physical optical cables), solving the problem of isolated and unrelated virtual and physical loops in traditional technologies. The topology mapping engine is a tool for visualizing and analyzing network or system structures, helping to understand the layout and interactions of complex systems by drawing nodes and connections.

[0041] Database: Serving as the storage medium for topology data, its function is to store the "virtual-physical mapping table" and visualized topology map generated by the topology mapping engine. This ensures that the system can quickly retrieve topology data during subsequent daily monitoring and fault location, avoiding the need to re-parse the SCD and physical link files for each location, thus improving location efficiency. Configuration Database: Stores network topology, RTU device information, fiber optic attributes, test task configurations, etc. Commonly used are MySQL / PostgreSQL.

[0042] Wavelength division multiplexer (WDM): The core unit for solving the problem of "uninterrupted monitoring services." Its function is to use WDM technology to transmit system monitoring signals (such as OTDR test signals and OMU optical power monitoring signals) and normal substation service signals (such as GOOSE / SV control messages) in the same optical fiber, achieving zero-interruption monitoring and avoiding the drawback of traditional OTDR testing requiring interruption of fiber optic services. The WDM multiplexer used is an existing WDM multiplexer.

[0043] Optical Monitoring Unit (OMU): The core unit for optical power data acquisition, its function is to collect the optical power values ​​at both ends of the optical cable in real time and automatically calculate the optical power difference ΔP between the two ends. This ΔP is a key indicator for determining whether there is a fault in the optical fiber, and it is also one of the core data for accurate fault location. The Optical Monitoring Unit (OMU) uses an existing optical multiplexer (OMU).

[0044] Distributed Temperature Sensing (DTS): This optical cable environment and condition monitoring unit collects temperature data along the entire optical cable laying path (covering the entire cable length with no monitoring blind spots). It helps identify latent fiber optic faults caused by abnormal temperatures (such as high-temperature aging or localized overheating) and provides temperature-related information for fault type determination (e.g., a sudden temperature rise may correspond to cable burn-out). Existing continuous distributed Brillouin fiber optic temperature and strain sensors can be used for DTS.

[0045] Bit Error Rate (BER) Detector: This service signal quality monitoring unit monitors the BER of GOOSE / SV messages transmitted in optical fiber in real time. It determines the communication quality of the optical fiber link by analyzing BER changes. A sudden increase in BER may correspond to faults such as fiber micro-bending or loose connectors causing signal attenuation, providing supplementary data on communication quality for fault location. The BER detector uses an existing BER detector.

[0046] Ethernet Transmission Unit: This data transmission hub synchronously transmits the optical power difference data collected by the OMU, the temperature data collected by the DTS, and the bit error rate data collected by the bit error rate detector to the data preprocessing unit at a fixed period of 100ms. This ensures that multi-source data is delivered to subsequent processing stages in real time without delay. A communication link is established between the monitoring center and the remote testing unit using existing IP data networks or wireless networks (4G / 5G) to transmit control commands and test data.

[0047] 3σ outlier removal module: The core unit of data cleaning, its function is to use the 3σ criterion (i.e., to select data that meet (μ-3σ≤x≤μ+3σ), where μ is the data mean and σ is the standard deviation) to remove outliers caused by strong electromagnetic interference from substations in multi-source data (such as instantaneous jumps in optical power and meaningless extreme temperature values), so as to avoid outlier data affecting the accuracy of subsequent model calculations.

[0048] Min-Max Standardization Module: This data standardization unit uses the Min-Max standardization formula x'=(x-x_min) / (x_max-x_min) to uniformly transform the cleaned optical power difference, temperature, and bit error rate data to the [0,1] interval, eliminating the dimensional differences between different data (such as optical power difference in dB and temperature in °C), and providing standardized input data for subsequent deep learning models.

[0049] Early warning triggering unit: Fault early warning triggering node, whose function is to monitor the standardized optical power difference ΔP in real time and determine whether it exceeds the preset threshold range (-10 to -28dB): If ΔP exceeds the threshold, it indicates that the optical power attenuation of the optical fiber link is abnormal, and a fault early warning signal is immediately triggered to start the subsequent deep learning localization and precise localization process. If it does not exceed the threshold, it maintains the daily monitoring status.

[0050] Denoising Convolutional Autoencoder (DCAE): The core anti-interference unit, its function is to process noise data in a substation environment with strong electromagnetic interference of 60dB. Through a network structure of 5 convolutional layers and 6 transposed convolutional layers, it denoises the standardized multi-source data, improving the signal-to-noise ratio (SNR) of low data (e.g., -5dB) to above 16dB, while retaining effective fault-related features (such as the gradual trend of optical power difference and the slow increase in temperature), providing high-quality data for subsequent time-series feature extraction.

[0051] Bidirectional Long Short-Term Memory (BiLSTM) network: The initial fault location unit receives the data after DCAE denoising and captures the temporal correlation features of the data (such as the decay law of optical power difference over time and the gradual increase trend of bit error rate) through two hidden layers (each containing 128 neurons, with the Dropout parameter set to 0.2 to suppress overfitting). Finally, it outputs the initial fault location (such as "200m away from end A") and the corresponding confidence level (such as 90%), providing a preliminary reference for subsequent accurate fault location.

[0052] Optical Time Domain Reflectometer (OTDR): A physical layer fault feature acquisition unit. Its function is to be activated only once (test time < 10ms, to avoid service interruption) after the BiLSTM outputs preliminary location results. It emits optical pulses and receives backscattered / reflected signals from the optical cable, extracting reflection peak features to provide direct physical layer evidence for fault location, compensating for the lack of physical mechanism support in purely data-driven models. One OTDR test module is permanently and directly connected to a single fiber under test via a fixed fiber optic patch cord. This is a "one-to-one" physical connection mode. If monitoring multiple fibers is required, a corresponding number of OTDR modules and RTUs must be configured, or a simple optical distribution frame can be used for manual switching. The OTDR utilizes an existing optical time domain reflectometer.

[0053] DS Evidence Theory Fusion Unit: The core fusion unit for precise positioning, its function is to fuse and calculate four sets of "evidence" using DS evidence theory: the preliminary fault location and confidence level output by BiLSTM, the reflection peak features extracted by OTDR (corresponding to the fault location), the optical power difference data from the OMU, and the bit error rate data from the bit error rate detector. By calculating the confidence assignment function of each set of evidence, the location corresponding to the highest confidence level is selected as the final fault location, ensuring that the positioning error is ≤5m, thus achieving precise positioning.

[0054] Visualized Topology Display Unit: This fault information visualization unit overlays the final fault location, fault type (e.g., fiber breakage, loose connector), and affected virtual circuit IDs onto the virtual-physical fusion topology map generated by the offline topology construction module. This visually displays the specific location of the fault within the substation topology, allowing maintenance personnel to quickly understand the fault's impact range. Modern systems typically employ web front-end architectures such as Vue.js, React, or Angular, but traditional desktop clients (such as WPF and Qt) can also be used.

[0055] IEC61970 Protocol Interface: A standardized interface unit for system-to-external interaction, conforming to the IEC61970-301 (CIM / XML) standard. It has no model number and its function is to push the topology map with superimposed fault information and the final fault location data to the existing SCADA (Supervisory and Data Acquisition) system of the substation with a delay of ≤20ms, in accordance with the IEC61970 substation communication standard protocol. This ensures that the SCADA system can obtain fiber optic fault information in real time and achieves interconnection with the overall operation and maintenance system of the substation.

[0056] Audible and visual alarm unit: The on-site alarm unit is used to trigger optical alarms (such as flashing red warning lights) and acoustic alarms (such as buzzer sounds) in the substation operation and maintenance duty room and in the vicinity of the faulty optical cable, so as to remind on-site operation and maintenance personnel to pay attention to the fault immediately and quickly go to the fault area to handle it. The audible and visual alarm unit uses existing audible and visual alarm devices.

[0057] SMS Alarm Unit: The remote alarm unit automatically sends fault alarm SMS messages to the mobile phones of maintenance personnel. The content includes the fault location, such as the optical cable from the No. 3 main transformer to the protection panel, 150m away from end A, the fault type, and the suggested handling priority. This ensures that maintenance personnel can receive fault information in real time even if they are not on-site, thus avoiding missing the window of opportunity for emergency repair.

[0058] Fault Handling Report Generation Unit: Operation and Maintenance Guidance Unit. Its function is to automatically generate fault handling reports, including basic fault information (location, type, and time of occurrence), the impact of the fault on substation services (such as whether it affects the GOOSE trip signal), and specific handling suggestions (such as whether to activate the backup optical cable and carry out fault repair during off-peak hours). It provides standardized handling solutions for operation and maintenance personnel and improves the efficiency of fault repair.

[0059] A fault location method for substation fiber optic fault location system based on virtual-real fusion and deep learning is proposed. The deep collaborative design of virtual-real fusion technology and deep learning technology breaks through the bottleneck of traditional location technology through the bidirectional empowerment of the two.

[0060] The specific work process is as follows:

[0061] First, during the equipment installation and commissioning phase, offline topology initialization is performed. The SCD file and physical link configuration file of the target substation are imported. The SCD parser extracts the IED device list and virtual terminal link table, and parses the virtual terminal links and network communication configurations of the secondary equipment. The physical link configuration file extracts the optical cable type, length, connector coordinates, and laying path. The topology mapping engine establishes a mapping table of "virtual loop ID - physical optical cable ID - connector location" and generates a visual topology map, which is then stored in the database. For detailed procedures, please refer to [reference needed]. Figure 1 .

[0062] After entering the daily monitoring phase, the multi-source data acquisition module achieves same-fiber transmission through a wavelength division multiplexer. The optical monitoring unit (OMU) acquires the optical power at both ends of the optical cable in real time and calculates the power difference ΔP. The distributed temperature sensing (DTS) acquires the temperature along the entire optical cable. The bit error rate detector monitors the GOOSE / SV message bit error rate with an acquisition cycle of 100ms. After the data is transmitted to the preprocessing unit via Ethernet, outliers caused by electromagnetic interference are removed using the 3σ criterion (μ-3σ≤x≤μ+3σ). Then, the data is processed to the [0,1] range using the Min-Max normalization formula x'=(x-x_min) / (x_max-x_min). If ΔP exceeds the preset threshold (-10~-28dB), a fault warning is triggered. The flowchart for this phase can be found in the diagram. Figure 2 .

[0063] Then, the initial fault localization stage begins. The standardized dataset is input into the denoising convolutional autoencoder (DCAE) module to remove noise caused by 60dB strong electromagnetic interference. The denoised data is then fed into the BiLSTM module, which captures the temporal features of the data through two hidden layers (128 units per layer, Dropout=0.2) and outputs the initial fault location and corresponding confidence level.

[0064] Finally, the process enters the precise fault location and alarm stage. The OTDR module is triggered (test time < 10ms) to acquire the reflection peak characteristics of the target optical cable. Using DS evidence theory, four sets of evidence are fused: BiLSTM output results, OTDR reflection peak characteristics, optical power difference, and bit error rate. The location corresponding to the highest confidence level is calculated and selected as the final fault location (error ≤ 5m). Virtual loop information regarding the fault type, location, and impact is overlaid onto the visualized topology map and pushed to the SCADA system via the IEC61970 protocol. Simultaneously, on-site audible and visual alarms and SMS alarms for maintenance personnel are triggered. A fault handling report is generated, including priority activation of backup optical cables and repair during off-peak hours. This forms a complete closed-loop process from topology construction, data acquisition, fault identification to result output and alarms. For details of this stage, please refer to [link / reference needed]. Figure 3 .

[0065] The overall process of the substation fiber optic fault location system based on virtual-real fusion and deep learning can be referenced. Figure 4 .

[0066] Example

[0067] 1) Multi-source data acquisition module

[0068] The DTS module was replaced with a Brillouin Optical Time-Domain Reflectometer (BOTDR) to simultaneously acquire temperature and strain data of the optical cable. This solution is suitable for high-altitude, high-vibration mountain substations and can simultaneously identify both temperature-induced damage and vibration-induced micro-bending faults in the optical cable. The BOTDR module costs 12,000 RMB more than the DTS module, but its positioning error is ≤5m, and its performance is comparable to the original solution.

[0069] 2) Deep learning models

[0070] The DCAE and BiLSTM combined model is replaced with an ensemble model of CNN and Extreme Gradient Boosting (XGBoost). The CNN module is used to extract the spatial features of OTDR reflection peaks, while the XGBoost module uses the spatial features output by the CNN, plus the optical power difference and temperature, as input to construct 100 decision trees and optimize the fault classification results. This model is suitable for newly built substations and can be trained without a large number of simulation samples. It offers high model accuracy, reduced training time, and meets the requirements for rapid deployment.

[0071] 3) Topology Construction

[0072] Based on SCD file parsing, GIS map data is integrated to construct a 3D virtual-real fusion topology. The latitude and longitude coordinates of the optical cable laying are located via GPS, and the cable burial depth and surrounding environment are marked on the GIS map. During topology updates, the mapping relationship is corrected by incorporating GIS environmental data. This method is suitable for long-distance outdoor optical cables, facilitating rapid fault location by maintenance personnel in the field.

[0073] 4) Verification Module

[0074] The OTDR with DS evidence theory is replaced by an optical time-domain reflectometer (OTDR) with a threshold comparison method. The simplified OTDR has a 15m blind zone and a test range of 0-5km, collecting only reflection peak position data. The difference between the BiLSTM output position and the OTDR reflection peak position is compared. If the difference is <10m, the BiLSTM result is used directly; if the difference is ≥10m, the OTDR result is used. This method is suitable for small and medium-sized substations of 110kV and below, where positioning accuracy requirements are slightly lower (≤10m), but the cost is approximately 8,000 RMB lower than the original OTDR, meeting the requirements for the operation and maintenance of small and medium-sized substations.

Claims

1. A substation fiber optic fault location system based on virtual-real fusion and deep learning, characterized in that, Based on the deep mapping of "physical circuit-logical circuit" in intelligent substations, and integrating multi-source sensing data with deep learning algorithms, a full-process positioning system is constructed, encompassing "data acquisition - virtual-real modeling - intelligent diagnosis - visualization presentation," including the following modules: 1) Offline topology construction module: The offline topology construction module is the basic support module for the system to realize "virtual and real integration" positioning. It consists of SCD parser, physical link parsing unit, topology mapping engine and database, and is enabled during the equipment installation and debugging phase. 2) Multi-source data acquisition module: The multi-source data acquisition module is the core module for the system to realize real-time monitoring. It includes wavelength division multiplexer, optical monitoring unit (OMU), distributed temperature sensor (DTS), bit error rate detector and Ethernet transmission unit, which runs continuously during the daily monitoring phase. 3) Data preprocessing unit: The data preprocessing unit is a key unit connecting multi-source data acquisition and intelligent computing. It consists of a 3σ outlier removal module, a Min-Max standardization module, and an early warning triggering unit. 4) Deep learning processing module: The deep learning processing module is the intelligent core module for the system to achieve initial fault localization. It consists of a denoising convolutional autoencoder (DCAE) and a bidirectional long short-term memory network (BiLSTM), and is activated after the fault warning is triggered. 5) Fault Precision Location Module: The fault precision location module is the key module for the system to achieve high-precision location. It consists of an OTDR and DS evidence theory fusion unit and is activated after the deep learning processing module outputs the preliminary results. 6) Alarm and Interaction Module: The alarm and interaction module is the terminal module of the system for transmitting fault information and providing operation and maintenance guidance. It includes a visual topology display unit, an IEC61970 protocol interface, an audible and visual alarm unit, an SMS alarm unit, and a fault handling report generation unit. It is activated after the fault is accurately located.

2. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The offline topology construction module is used to import SCD and physical link files, parse device configuration and optical cable parameters, establish a mapping table of "virtual loop ID-physical optical cable ID-connector location", generate a visual topology map and store it, and provide virtual and physical anchor points for subsequent positioning.

3. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The multi-source data acquisition module is used to achieve the transmission of service and monitoring signals on the same fiber through wavelength division multiplexing. It collects optical power difference ΔP, optical cable temperature, and GOOSE / SV message error rate at a period of 100ms, and transmits them to the preprocessing unit via Ethernet.

4. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The data preprocessing unit is used to remove electromagnetic interference anomalies and standardize the data to the [0,1] range. When ΔP exceeds the -10 to -28 dB threshold, a fault warning is immediately triggered.

5. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The fault precise location module is used to obtain the optical cable reflection peak characteristics by OTDR, fuse BiLSTM results, reflection peak characteristics, ΔP and bit error rate, and output the final fault location with a location error ≤5m.

6. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The alarm and interaction module is used to overlay fault information on the topology map, push it to the SCADA system via protocol with a delay of ≤20ms, trigger on-site and SMS alarms, and generate an operation and maintenance suggestion report.

7. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The physical link parsing unit is responsible for importing the substation physical link configuration file and extracting the core optical cable parameters recorded in the file, including optical cable model, laying length, joint coordinates, laying path, and "physical loop" information, to provide physical dimension data support for subsequent association of virtual loops and physical optical cables.

8. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The topology mapping engine is the core execution unit of the virtual-physical fusion system. It associates and matches the "virtual loop information" output by the SCD parser with the "physical loop information" output by the physical link parser unit to construct a one-to-one mapping table of "virtual loop ID - physical optical cable ID - connector location". At the same time, it automatically generates a visual topology map to intuitively show the binding relationship between virtual loops and physical optical cables, solving the problem of isolated and unrelated virtual and physical loops in traditional technologies.

9. The substation fiber optic fault location system based on virtual-real fusion and deep learning according to claim 1, characterized in that, The visualization topology display unit is a fault information visualization unit. It overlays the final fault location, fault type, and affected virtual circuit ID information onto the virtual-real fusion topology map generated by the offline topology construction module, intuitively displaying the specific location of the fault in the substation topology, making it easier for operation and maintenance personnel to quickly understand the scope of the fault's impact.

10. The location method of the substation fiber optic fault location system based on virtual-real fusion and deep learning as described in claim 1, characterized in that, The deep collaborative design of virtual-real fusion technology and deep learning technology breaks through the bottleneck of traditional positioning technology through the mutual empowerment of the two. The specific working process is as follows: First, during the equipment installation and commissioning phase, offline topology initialization is performed. The SCD file and physical link configuration file of the target substation are imported. The IED device list and virtual terminal link table are extracted through the SCD parser, and the virtual terminal links and network communication configuration of the secondary equipment are parsed. The optical cable model, length, joint coordinates and laying path are extracted from the physical link configuration file. The topology mapping engine establishes a mapping table of "virtual loop ID-physical optical cable ID-joint location" and generates a visual topology map which is stored in the database. After entering the daily monitoring phase, the multi-source data acquisition module realizes same-fiber transmission through wavelength division multiplexing, the optical monitoring unit (OMU) collects the optical power at both ends of the optical cable in real time and calculates the power difference ΔP, the distributed temperature sensor (DTS) collects the temperature of the optical cable throughout the entire process, and the bit error rate detector monitors the bit error rate of GOOSE / SV messages with an acquisition cycle of 100ms. After the data is transmitted to the preprocessing unit via Ethernet, the abnormal values ​​caused by electromagnetic interference are removed by the 3σ criterion μ-3σ≤x≤μ+3σ. Then, the data is processed to the [0,1] interval by the Min-Max normalization formula x'=(x-x_min) / (x_max-x_min). If ΔP exceeds the preset threshold of -10 to -28dB, a fault warning is triggered. Then, the initial fault localization stage begins. The standardized dataset is input into the denoising convolutional autoencoder (DCAE) module to remove noise caused by 60dB strong electromagnetic interference. The denoised data is then fed into the BiLSTM module, which captures the temporal features of the data through two hidden layers, each with 128 units and Dropout=0.2, and outputs the initial fault location and corresponding confidence level. Finally, the system enters the precise fault location and alarm stage. The OTDR module is triggered, with a test time of <10ms, to obtain the reflection peak characteristics of the target optical cable. Using DS evidence theory, the system integrates four sets of evidence: BiLSTM output results, OTDR reflection peak characteristics, optical power difference, and bit error rate. The location corresponding to the highest confidence level is calculated and selected as the final fault location, with an error of ≤5m. The virtual loop information of fault type, location, and impact is superimposed onto the visualized topology map and pushed to the SCADA system via the IEC61970 protocol. At the same time, on-site audible and visual alarms and SMS alarms for maintenance personnel are triggered, and a fault handling report is generated, including priority activation of backup optical cables and repair during off-peak hours. This forms a complete closed-loop process from topology construction, data acquisition, fault identification to result output and alarm.

Images