Optical cable identity recognition and illegal access detection method and system for smart grid

Abstract

The application is suitable for the technical field of communication and power safety monitoring, and provides a kind of optical cable identity recognition and illegal access detection method and system for smart grid, the method comprises the following steps: real-time acquisition vibration signal, when the amplitude of vibration signal is greater than the preset amplitude threshold, and its energy is greater than the energy threshold calculated based on dynamic baseline energy, it is determined as effective disturbance event;In response to effective disturbance event or according to the preset inspection plan, visible light image and infrared thermal image of target optical cable node are synchronously collected;Visible light image and infrared thermal image are spliced to form a fusion image;The fusion image is input into the lightweight target detection network, and the target detection result is output;According to the target detection result, the actual identity information of the current optical cable is extracted, and whether the actual identity information of the optical cable is legal and whether there is illegal access behavior is determined.The application provides a new generation of verifiable, traceable and closed-loop protection capability for the physical layer security of smart grid communication.

Description

Technical Field

[0001] This invention belongs to the field of communication and power safety monitoring technology, and in particular relates to a method and system for optical cable identification and illegal access detection for smart grids. Background Technology

[0002] The reliable operation of smart grids heavily relies on the physical security and reliable identity of communication optical cables. In urban underground utility tunnels, cable tunnels, and substation access lines, dedicated power optical cables (such as GYTA and GYTS) are densely laid in pipes or brackets, constantly exposed to high humidity, high dust, and strong electromagnetic interference environments. In such scenarios, the cable identification information is easily degraded due to label detachment, color fading, and corrosion of junction boxes, and there is a lack of effective dynamic verification methods. Maintenance relies on static ledgers and manual inspections, making it difficult to detect covert activities such as illegal cable replacement, unauthorized branching, or splicing for eavesdropping, leading to widespread discrepancies between records and actual cables.

[0003] Existing identification methods mainly include QR code scanning, visual verification of color codes, and RFID tag reading. QR codes need to be clearly visible and unobstructed; however, they are often unreadable in underground manholes or corners of utility tunnels due to dirt, glare, or angle issues. While color code sequences use standard encoding (e.g., ABCD corresponding to different lines), they are susceptible to aging and paint coverage, and manual interpretation is highly subjective. Embedded RFID or NB-IoT tags can provide 128-bit encrypted IDs and are resistant to obstruction, but they only indicate the tag's existence and are not bound to the fiber optic cable itself. Attackers can transfer legitimate tags to illegitimate fiber optic cables, and the system cannot determine that the entity has been replaced when reading the "correct ID," creating an identity forgery vulnerability.

[0004] Multimodal sensing technology has been initially applied in some intelligent utility tunnel systems, but visible light and infrared images are still processed in isolation. Visible light is used to identify outer casing features, and infrared is used to detect abnormal heating of junction boxes. Neither is structurally aligned with wireless IDs and prior record information. When the visual system identifies a "blue-white-red" color code, but the corresponding record for the wireless ID is "yellow-blue-white," the system cannot automatically determine whether the fiber optic cable has been replaced, the tag has been stolen, or it is an identification error. The lack of a three-way consistency check between "visual features—wireless ID—record expectation" renders identity verification results unreliable.

[0005] Underground power fiber optic cables are subject to strict engineering specifications, including color coding, junction box types, branch spacing, and bending radii. For example, a provincial power grid stipulates that the color coding sequence for trunk fiber optic cables is yellow-blue-red-white, and branch cables must use dedicated splitter junction boxes. General target detection models (such as YOLOv8) do not incorporate such prior knowledge. In complex utility tunnel backgrounds (such as multiple cable crossings and reflective metal supports), they are prone to misidentifying illegally connected splitters as compliant joints or failing to detect concealed eavesdropping devices. The identification results lack constraints on engineering rationality.

[0006] Unreliable identity verification directly leads to the failure of unauthorized access detection. When the system cannot confirm whether the current optical cable is the entity recorded in the ledger, any abnormal alarm may be a false alarm. Therefore, there is an urgent need for a multimodal, prior-guided, end-to-end identity recognition method for underground scenarios to provide high-confidence identity anchors for optical cables, and on this basis, to accurately identify risks such as unauthorized access, identity drift, and tag forgery. Summary of the Invention

[0007] The purpose of this invention is to provide a method for optical cable identification and illegal access detection for smart grids, aiming to solve the above-mentioned technical problems.

[0008] To achieve the above objectives, the present invention provides a method for optical cable identification and unauthorized access detection in smart grids, comprising the following steps:

[0009] Vibration signals are acquired in real time, the amplitude and energy of the current vibration signal are calculated, and compared with the dynamic baseline energy; when the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event.

[0010] In response to the effective disturbance event or according to the preset inspection plan, the inspection platform is controlled to reach the target optical cable node, and the visible light image and infrared thermal image of the target optical cable node are collected simultaneously, and the electronic code of the wireless ID tag attached to the optical cable is read.

[0011] The visible light image and the upsampled infrared thermal image are stitched together along the channel dimension to form a fused image; the fused image is input into a preset lightweight target detection network, and the target detection results of preset categories in the fused image are output;

[0012] Based on the target detection results, the actual identity information of the current optical cable is extracted; the expected identity information corresponding to the target optical cable node is queried from the preset optical cable ledger prior database; the actual identity information and the expected identity information are compared item by item, and the electronic code is verified to be consistent with the code bound in the ledger; based on the comparison and verification results, it is determined whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior.

[0013] Furthermore, the steps of real-time acquisition of vibration signals, calculation of the amplitude and energy of the current vibration signal, and comparison with the dynamic baseline energy specifically include:

[0014] Vibration signal S vib Defined as:

[0015] ;

[0016] In the formula, Indicates that the fiber optic vibration sensor is in The analog voltage signal collected within a time period, measured in volts;

[0017] The expression for the amplitude of the vibration signal and the triggering condition are as follows:

[0018] ;

[0019] The expression for the energy of the vibration signal and the triggering condition are as follows:

[0020] ;

[0021] In the formula, the amplitude threshold is 0.5V, E th The energy threshold is 1.8 times the dynamic baseline energy. The dynamic baseline energy is obtained by: taking the current time as the endpoint, tracing back a preset time window, removing the time periods marked as valid disturbance events, and calculating the average vibration energy of the remaining silent periods. and standard deviation , obtain dynamic baseline energy .

[0022] Furthermore, the optical cable ledger prior library is represented as a structured knowledge base, containing the following three types of core verification information:

[0023] Identity identification relationship: describes the binding relationship between the unique ID of the optical cable and the ID of the wireless tag;

[0024] Visual feature specification: describes the color code sequence encoding rules for the outer sheath of optical cables;

[0025] Component combination logic: describes the allowed attachment types and branch constraints.

[0026] Furthermore, the preset lightweight target detection network is a lightweight YOLOv8n multimodal fusion recognition network based on the prior guidance of the optical cable ledger prior library. It encodes the spatial constraints in the optical cable ledger prior library into a multi-channel prior heatmap and embeds Prior modules in front of the three-level detection heads respectively, injecting feature streams in an additive bias manner to achieve multi-scale spatial attention guidance.

[0027] Furthermore, the step of stitching the visible light image and the upsampled infrared thermal image together along the channel dimension to form a fused image specifically includes:

[0028] The mathematical expression for a visible light image is: ;

[0029] The mathematical expression for an infrared thermal image is: ;

[0030] The infrared thermal image is upsampled to have the same spatial dimension as the visible light image: bilinear interpolation is used for upsampling, and the upsampled infrared thermal image is used as the fourth channel. The channel dimension is then stitched together with the visible light image to form a four-channel fused image.

[0031] Furthermore, the optical cable ledger prior library contains three types of formalizable spatial constraints derived from power industry standards, engineering design drawings, and operation and maintenance ledgers: spatial location relationships, geometric structural characteristics, and component combination logic; the steps of inputting the fused image into a preset lightweight target detection network and outputting the target detection results for preset categories in the fused image specifically include:

[0032] A multi-channel prior heatmap based on the spatial constraints of the optical cable ledger prior library is generated, with each channel corresponding to a type of detection target. The types of detection targets include identity feature type and abnormal device type. The identity feature type includes color mark area, QR code, and nameplate. The abnormal device type includes optical splitter, eavesdropping box, and illegal junction box.

[0033] Before embedding the multi-channel prior heatmap into the three-stage detection head of the lightweight target detection network with additive bias; specifically, the multi-channel prior heatmap is first... Enhanced features are obtained by downsampling to three different scales using bilinear interpolation and then performing channel broadcasting and element-wise addition with the corresponding feature maps output from the Neck layer. :

[0034] ;

[0035] Among them, F i This is the original feature map at the i-th scale. ; The scale-adaptive guiding coefficients are then used; the enhanced features are then directly fed into the native detection head of the lightweight object detection network to output multi-class object detection results; Repeat(·) indicates that the tensor is repeatedly copied along the channel dimension to adapt the original feature map;

[0036] Specifically, the original detection head structure is preserved at the three scales of the lightweight target detection network, and the three-level detection results are integrated into multi-class structured detection outputs through a unified post-processing fusion mechanism. Specifically, each scale detection head outputs class confidence, bounding box coordinates and objectivity score, and the class set includes the classes corresponding to the multi-channel prior heatmap.

[0037] After the lightweight object detection network completes inference, the following fusion logic is executed:

[0038] Cross-scale nonmaximum suppression: merging three-level outputs at different scales;

[0039] Prior-guided confidence correction: The final confidence level is dynamically adjusted based on the response value of the detection box area to the multi-channel prior heatmap.

[0040] Structured mapping: The retained detection boxes are merged by category to generate a multi-dimensional identity description vector; each component contains the detection existence, location, confidence level and associated attributes, which are output as a unique detection result; the associated attributes include color code sequence and device temperature.

[0041] Further, based on the target detection results, the actual identity information of the current optical cable is extracted; the expected identity information corresponding to the target optical cable node is queried from the preset optical cable ledger prior database; the actual identity information and the expected identity information are compared item by item, and the electronic code is verified to be consistent with the code bound in the ledger; based on the comparison and verification results, the steps of determining whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior are specifically included:

[0042] The detection results of various targets are classified and summarized to form a structured actual identity information; the actual identity information includes the actual identified color mark sequence, the actual detected accessory type, the actual identified number of branches, and the electronic code read from the wireless ID tag;

[0043] The expected identity information corresponding to the target optical cable node is retrieved from the preset optical cable ledger database; the expected identity information includes: the expected color code sequence, the types of accessories allowed to be installed, the maximum number of branches allowed, and the electronic code to be bound.

[0044] The actual identity information is compared with the expected identity information item by item to perform a consistency check:

[0045] If the actual identified color mark sequence is inconsistent with the expected color mark sequence, it is judged as "identity mismatch";

[0046] If the actual detected accessory type is not within the range of allowed accessory types, it is determined to be an "illegal device";

[0047] If the number of branches actually identified exceeds the maximum number of branches allowed, it is judged as an "illegal branch";

[0048] If the electronic code read from the wireless ID tag does not match the electronic code that should be bound, it is determined to be "tag forgery";

[0049] If any of the above consistency checks fails, the current optical cable is deemed to have a security risk, triggering the corresponding alarm type and entering the alarm generation process; if all consistency checks pass, the current optical cable is deemed legal and recorded as normal.

[0050] Furthermore, the alarm generation process specifically includes:

[0051] Determine the alarm category based on the type of consistency check failure;

[0052] For each type of alarm, it is automatically packaged into multimodal evidence information;

[0053] Multimodal evidence information is organized into standardized, structured alarm data packets, and these alarm data packets are uploaded in real time.

[0054] Another objective of this invention is to provide a fiber optic cable identification and unauthorized access detection system for smart grids, used to implement the aforementioned fiber optic cable identification and unauthorized access detection method, comprising:

[0055] The event-driven triggering module is used to collect vibration signals in real time, calculate the amplitude and energy of the current vibration signal, and compare it with the dynamic baseline energy. When the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event.

[0056] The multimodal data acquisition module is used to respond to the effective disturbance event or according to the preset inspection plan, control the inspection platform to arrive at the target optical cable node, and simultaneously acquire the visible light image and infrared thermal image of the target optical cable node, and read the electronic code of the wireless ID tag attached to the optical cable.

[0057] The target detection module is used to stitch the visible light image and the upsampled infrared thermal image in the channel dimension to form a fused image; input the fused image into a preset lightweight target detection network, and output the target detection results of preset categories in the fused image;

[0058] The consistency verification module is used to extract the actual identity information of the current optical cable based on the target detection result; query the expected identity information corresponding to the target optical cable node from the preset optical cable ledger prior library; compare the actual identity information with the expected identity information item by item; and verify whether the electronic code is consistent with the code bound in the ledger; and determine whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior based on the comparison and verification results.

[0059] The optical cable identity recognition and illegal access detection method for smart grids provided by this invention is based on event-driven, multimodal fusion, prior guidance, and end-to-end decision-making. With the core objective of building a trusted identity foundation, it achieves automatic identification and structured alarms for high-risk behaviors such as illegal splicing, identity forgery, and unauthorized branching. It provides a new generation of verifiable, traceable, and closed-loop protection capabilities for the physical layer security of smart grid communication. It is applicable to scenarios such as urban underground pipe corridors, cable tunnels, and substation entry and exit sections for high-confidence identity verification and end-to-end identification of illegal behaviors in power communication optical cables. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the structure of the optical cable identification and illegal access detection method for smart grids provided in an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of the lightweight target detection network provided in an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0063] The embodiments of this invention aim to systematically solve the three core bottlenecks existing in the safety monitoring of underground optical cables:

[0064] I. Weak identity recognition foundation: Relying on a single visual tag or isolated wireless ID, it is easy for the "ledger ID ≠ actual optical cable" identity to drift due to dirt, obstruction, or tag forgery;

[0065] Second, source information is fragmented: Modal data such as visible light, infrared, and wireless ID are processed in isolation, lacking a three-dimensional consistency verification mechanism of "visual features - wireless ID - ledger expectation", making it difficult to distinguish between optical cable replacement, tag theft, or identification errors.

[0066] Third, lack of prior knowledge: The general target detection model does not embed the engineering specifications of power optical cables (such as color code sequence, accessory type, laying constraints), which leads to the lack of physical rationality in the recognition results and makes it easy to miss hidden illegal devices.

[0067] Therefore, such as Figure 1As shown, in one embodiment of the present invention, a method for optical cable identification and illegal access detection for smart grids is provided, which is particularly suitable for situations where automatic identification verification and security status assessment of optical cables are required in electromagnetically confined spaces such as underground utility tunnels and cable tunnels due to unclear optical cable identification information, missing tags, or the risk of illegal splicing; the method specifically includes the following steps:

[0068] S1. Real-time acquisition of vibration signals, calculation of the amplitude and energy of the current vibration signal, and comparison with the dynamic baseline energy; when the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event;

[0069] S2. In response to the effective disturbance event or according to the preset inspection plan, control the inspection platform to reach the target optical cable node, and simultaneously collect the visible light image and infrared thermal image of the target optical cable node, and read the electronic code of the wireless ID tag attached to the optical cable.

[0070] S3. The visible light image and the upsampled infrared thermal image are stitched together in the channel dimension to form a fused image; the fused image is input into a preset lightweight target detection network, and the target detection results of preset categories in the fused image are output.

[0071] S4. Based on the target detection results, extract the actual identity information of the current optical cable; query the expected identity information corresponding to the target optical cable node from the preset optical cable ledger prior database, compare the actual identity information with the expected identity information item by item, and verify whether the electronic code is consistent with the code bound in the ledger; based on the comparison and verification results, determine whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior.

[0072] Correspondingly, in another embodiment of the present invention, a fiber optic cable identification and unauthorized access detection system for smart grids is also provided to implement the above-mentioned fiber optic cable identification and unauthorized access detection method. This system includes vibration sensing units (such as fiber optic vibration sensors), wireless identification (ID) tags deployed at key nodes of the fiber optic cable, a mobile inspection platform (such as an inspection robot, handheld terminal, or edge computing node), and an edge server. The inspection platform integrates an infrared thermal imager for acquiring infrared thermal images, a visible light camera for acquiring visible light images, a wireless ID reader for reading the wireless identification tag (electronic code), and an embedded AI computing unit. The embedded AI computing unit includes an event-driven triggering module, a multimodal data acquisition module, and a target detection module.

[0073] The event-driven triggering module is used to collect vibration signals in real time, calculate the amplitude and energy of the current vibration signal, and compare it with the dynamic baseline energy. When the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event.

[0074] The multimodal data acquisition module is used to respond to the effective disturbance event or according to the preset inspection plan, control the inspection platform to arrive at the target optical cable node, and simultaneously acquire the visible light image and infrared thermal image of the target optical cable node, and read the electronic code of the wireless ID tag attached to the optical cable.

[0075] The target detection module is used to stitch the visible light image and the upsampled infrared thermal image in the channel dimension to form a fused image; input the fused image into a preset lightweight target detection network, and output the target detection results of preset categories in the fused image;

[0076] The edge server includes a consistency verification module and an alarm generation module;

[0077] The consistency verification module is used to extract the actual identity information of the current optical cable based on the target detection result; query the expected identity information corresponding to the target optical cable node from the preset optical cable ledger prior database; compare the actual identity information with the expected identity information item by item; and verify whether the electronic code is consistent with the code bound in the ledger; and determine whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior based on the comparison and verification results.

[0078] The alarm generation module is used to initiate the alarm generation process based on the corresponding alarm type.

[0079] In this embodiment of the invention, an event-driven trigger-priority response mechanism is used to integrate multimodal perception data and structured engineering priors to achieve end-to-end authentication and unauthorized access detection from pixel-level observation to operation-level semantics; the complete process is as follows:

[0080] I. Event-driven triggering and priority response mechanism:

[0081] Vibration signal S vib Defined as:

[0082] ;

[0083] In the formula, Indicates that the fiber optic vibration sensor is in The analog voltage signal acquired within a time period is measured in volts (V); R represents the set of real numbers; this vibration signal is output by a phase-sensitive optical time-domain reflectometer or an embedded piezoelectric sensor deployed along the optical cable, used to sense external disturbances such as knocking, pulling, and well opening, and the following dual-threshold triggering conditions are set:

[0084] The expression for the amplitude of the vibration signal and the triggering condition are as follows:

[0085] ;

[0086] The expression for the energy of the vibration signal and the triggering condition are as follows:

[0087] ;

[0088] In the formula, the amplitude threshold is 0.5V, E th The energy threshold is 1.8 times the dynamic baseline energy. The dynamic baseline energy is obtained by: taking the current time as the endpoint, backtracking through a preset time window (e.g., 10 minutes), removing the time periods marked as valid disturbance events, and calculating the average vibration energy of the remaining silent periods. and standard deviation , obtain dynamic baseline energy .

[0089] It should be noted that the system determines a disturbance event as valid only when both of the above triggering conditions are met simultaneously, triggering subsequent multimodal data synchronous acquisition; this mechanism enables the system to adapt to changes in environmental noise and reduce the false trigger rate.

[0090] II. Synchronous Acquisition of Multimodal Data:

[0091] This invention requires the simultaneous acquisition of four types of heterogeneous sensing data as the input basis for optical cable identification and unauthorized access detection. To improve system energy efficiency and data relevance, this invention employs an event-driven dual-mode acquisition mechanism: on one hand, the inspection platform periodically arrives at the optical cable node according to plan to perform routine multimodal data acquisition; on the other hand, fiber optic vibration sensors are deployed at key locations on the optical cable. When a suspected human disturbance event meeting both amplitude and energy thresholds is detected, the inspection platform is proactively triggered to proceed to the optical cable node, activating the visible light camera, infrared thermal imager, and wireless ID reader / writer to achieve on-demand acquisition of high-value data and avoid ineffective continuous observation. The four types of heterogeneous sensing data are described in Table 1. The above four types of data are aligned using hardware-level timestamps to form a unique event quadruple.

[0092] Table 1. Data Description for Four Types of Heterogeneous Sensing

[0093]

[0094] III. Construction of the Optical Cable Ledger Priority Database and Priority Guidance Mechanism:

[0095] To improve the physical rationality and engineering credibility of identity recognition results, this embodiment of the invention requires the construction of a priori optical cable ledger and the formalization of the domain knowledge therein into computable structured rules, which serve as logical guidance signals in the network reasoning process.

[0096] (1) Construction of the optical cable ledger prior library: The optical cable ledger prior library is represented as a structured knowledge base, containing the following three types of core verification information:

[0097] Identification relationship: describes the binding relationship between the unique ID of the optical cable and the RFID tag ID; for example: "Optical cable GD-2025-0876 is bound to RFID ID a3f8c2e1";

[0098] Visual feature specifications: describe the color code sequence encoding rules for the outer sheath of optical cables; for example: "The color code sequence for 500kV trunk optical cables is yellow-blue-red-white";

[0099] Component combination logic: describes the allowed accessory types and branch constraints; for example: "The maximum number of branches for the trunk optical cable is 0, and only OPGW dedicated junction boxes are allowed."

[0100] The aforementioned prior knowledge is derived from power industry standards (such as DL / T 5344), engineering design drawings, and operation and maintenance ledgers, and is common engineering knowledge known to those skilled in the art. An example of the optical cable ledger prior knowledge library is shown in Table 2.

[0101] Table 2. Structure of the Optical Cable Ledger Priority Storage

[0102]

[0103] (2) Application of the prior library guidance mechanism: The optical cable ledger prior library does not directly participate in the loss calculation of the YOLOv8n network, but is used as a structured verification rule to filter and correct the rationality of the detection results after the network inference is completed. Its application runs through the post-processing stage of the detection results, and the specific implementation is as follows:

[0104] a. Verification location: YOLO output post-processing stage; b. Mechanism of action: logic attention guidance;

[0105] The optical cable ledger prior library is essentially a logical attention mask. Its rules (such as "color mark must be four segments" and "attachments must not contain splitters") will suppress untrusted engineering outputs generated by general target detection models in the context of complex utility tunnels (such as misjudging illegal splitters as compliant joints).

[0106] IV. Construction and Application of Lightweight Target Detection Networks:

[0107] To efficiently and robustly extract optical cable identification features and anomalous device information from multimodal inputs, this invention constructs a priori-guided lightweight YOLOv8n multimodal fusion recognition network as a lightweight target detection network. While maintaining the high detection accuracy and low computational cost of YOLOv8n, this lightweight target detection network innovatively encodes the spatial constraints in the optical cable ledger prior library into a multi-channel prior heatmap, and embeds Prior modules before each of the three-level detection heads, injecting feature streams with additive bias to achieve multi-scale spatial attention guidance. Please refer to the appendix for a detailed structural diagram. Figure 2 .

[0108] (1) Improved multimodal input layer: In this embodiment of the invention, two heterogeneous image data types, visible light and infrared, are acquired simultaneously and unified as network input. Specifically:

[0109] Visible light images are captured by a visible light camera mounted on the inspection platform, and their mathematical expression is: ;

[0110] The visible light image provides detailed information on the color code of the fiber optic cable sheath, the QR code label, the appearance of the junction box, and the texture and structural details of abnormal devices (such as beam splitters and eavesdropping boxes), which is the main basis for identifying the characteristics of the fiber optic cable.

[0111] Infrared thermal images are acquired synchronously by an infrared thermal imager, and their mathematical expression is: ;

[0112] This infrared thermal image reflects the temperature distribution on the surface of the fiber optic junction box and its accessories. It can effectively detect local overheating caused by continuous operation of illegal beam splitters, eavesdropping boxes, etc., and has the ability to penetrate some visual obstructions (such as stains and fog).

[0113] Since infrared thermal images typically have lower resolution than visible light images, they need to be upsampled to achieve the same spatial dimension. Specifically, this embodiment uses bilinear interpolation for upsampling and uses the upsampled infrared image as the fourth channel, concatenating it with the visible light image to form a four-channel fused input. To accommodate this four-channel input, this embodiment expands the input channels of the first convolutional (Conv) layer of YOLOv8n from the standard 3 to 4, while keeping the other parameters unchanged. The weights of this layer are initialized as follows: the first three channels are loaded with weights pre-trained on ImageNet by YOLOv8n, and the fourth channel (infrared) is randomly initialized using a Gaussian distribution.

[0114] (2) Generation of multi-channel prior heatmaps based on spatial constraints formalized from the optical cable ledger prior library: The optical cable ledger prior library contains three types of formalizable spatial constraints, derived from power industry standards, engineering design drawings, and operation and maintenance ledgers, specifically including:

[0115] Spatial positional relationships: such as "the color mark should be in the horizontal strip area in the center of the image" and "the junction box should be in the lower 1 / 3 area of ​​the image";

[0116] Geometric structural characteristics: such as "the width of the color mark should be 30-50 pixels" and "the aspect ratio of the junction box is ∈ [1.5, 2.5]";

[0117] Component combination logic: such as "the main optical cable must not have a splitter" and "the eavesdropping box is considered abnormal across the entire area";

[0118] A 6-channel prior heatmap is generated based on the above spatial constraints. Each channel corresponds to one type of detection target: identity feature type (channels 1-3: color mark area, QR code, nameplate); abnormal device type (channels 4-6: optical splitter, eavesdropping box, illegal junction box); assuming the size of the acquired visible light image is... The generation rules are as follows:

[0119] A. Detection target: Color mark area ;

[0120] Prior condition: Located in the horizontal band-shaped region in the center of the image;

[0121] Generation rules: ;

[0122] B. Detection target: QR code ;

[0123] Prior conditions: Located in the adjacent area to the right or below the color mark, and relatively small in size (approximately 30-60 pixels wide).

[0124] Generation rules: ;

[0125] C. Target of inspection: Nameplate ;

[0126] Prior conditions: Located in the upper left or upper right quadrant of the image, rectangular in shape, with an aspect ratio of 2:3;

[0127] Generation rules: ;

[0128] D. Target for detection: Optical splitter ;

[0129] Prerequisite: The main optical cable is located in the central longitudinal band of the image, and optical splitters are prohibited in this area;

[0130] Generation rules: ;

[0131] E. Detection Target: Eavesdropping Box ;

[0132] Prior condition: Any occurrence at any location is considered an anomaly; the model should suppress the response to prevent false alarms.

[0133] Generation rules: ;

[0134] F. Detection Target: Illegal Junction Boxes ;

[0135] Precondition: A valid junction box should be located in the lower 1 / 3 of the image; therefore, the invalid area is the upper 2 / 3.

[0136] Generation rules: ;

[0137] Before combining the aforementioned 6-channel feature maps into a prior heatmap and embedding it with additive bias into the third-level detector head of the YOLOv8n network, specifically, the global 6-channel prior heatmap is first... The enhanced features are obtained by downsampling to three scales (80×80, 40×40, and 20×20) using bilinear interpolation and then performing channel broadcasting and element-wise addition with the corresponding feature maps output from the Neck layer. :

[0138] ;

[0139] Among them, F i This is the original feature map at the i-th scale. ; The coefficients are scale-adaptive guiding coefficients; Repeat(·) represents repeatedly copying the tensor to adapt the original feature map along the channel dimension. This operation achieves dynamic modulation of the feature space by the ledger prior without changing the network structure and parameter quantity, enabling the network to actively enhance the response of legitimate regions (such as color marks and junction boxes) during the inference stage, while suppressing false detections of illegal devices (such as beam splitters and eavesdropping boxes). The enhanced features are then directly fed into the YOLOv8n native detection head, which can output detection results for 6 types of targets.

[0140] (3) Detection Head and Output: The original detection head structure is retained on the three scales (P3 / P4 / P5) of the YOLOv8n network, and the three-level detection results are integrated into 6 types of structured detection outputs through a unified post-processing fusion mechanism. Specifically, each scale detection head outputs category confidence, bounding box coordinates and objectivity score. The category set includes categories corresponding to the multi-channel prior heatmap, specifically including the following 6 items: identity feature category (3 categories): color mark area, QR code, nameplate; abnormal device category (3 categories): optical splitter, eavesdropping box, illegal junction box.

[0141] After the lightweight object detection network completes inference, the following fusion logic is executed:

[0142] Cross-scale nonmaximum suppression (NMS): Combines the three-level outputs of different scales (P3 / P4 / P5) to eliminate duplicate detections;

[0143] Prior-guided confidence correction: The final confidence level is dynamically adjusted based on the response values ​​of the detection box area to the 6-channel prior heatmap.

[0144] Structured mapping: The retained detection boxes are merged by category to generate a 6-dimensional identity description vector. Each component contains the detection existence, location, confidence level, and associated attributes (such as color code sequence and equipment temperature). As a unique detection result output, it is directly sent to the subsequent consistency verification module for multi-dimensional comparison with the optical cable ledger prior library, realizing end-to-end mapping from pixel-level detection to operation and maintenance-level semantics.

[0145] V. Structured Identity Consistency Verification:

[0146] After completing multimodal recognition, this embodiment of the invention performs structured identity consistency verification on the target detection results to determine whether the current optical cable matches the ledger record and whether there is any unauthorized access. This process specifically includes the following steps:

[0147] First, the detection results of various targets are classified and summarized to form a structured actual identity information; the actual identity information includes the actual identified color code sequence (such as "yellow, blue, red, white", etc.), the actual detected accessory type (such as "OPGW junction box", "optical splitter", etc.), the actual identified number of branches, and the electronic code read from the wireless ID tag.

[0148] Next, the expected identity information corresponding to the target optical cable node is queried from the preset optical cable ledger database; the expected identity information includes: the expected color code sequence, the types of accessories that can be installed, the maximum number of branches allowed, and the electronic code that should be bound.

[0149] Then, the actual identity information and the expected identity information are compared item by item to perform a consistency check:

[0150] If the actual identified color mark sequence is inconsistent with the expected color mark sequence (e.g., incorrect order or missing segment), it is judged as "identity mismatch";

[0151] If the actual detected accessory type is not within the range of allowed accessory types (e.g., the ledger only allows "OPGW junction box", but "optical splitter" is detected), it is determined to be an "illegal device";

[0152] If the number of branches actually identified exceeds the maximum allowed number of branches (e.g., the trunk optical cable is specified to have 0 branches, but branches are detected), it is judged as an "illegal branch";

[0153] If the electronic code read from the wireless ID tag does not match the electronic code that should be bound, it is determined to be "tag forgery";

[0154] If any of the above consistency checks fails, the current optical cable is deemed to have a security risk, triggering the corresponding alarm type and entering the alarm generation process; if all consistency checks pass, the current optical cable is deemed legal and recorded as normal.

[0155] This mechanism ensures that identity verification does not rely on a single modality or a single feature, but rather on multi-dimensional cross-validation, fundamentally solving the problem of "inconsistency between the ledger and the entity" and providing a reliable foundation for the detection of unauthorized access.

[0156] VI. Generation of end-to-end illegal access alarms:

[0157] After completing the structured identity consistency verification, this embodiment of the invention automatically generates a structured illegal access alarm based on the verification result and abnormal device detection information, realizing an end-to-end closed loop from pixel-level recognition to operation and maintenance-level semantics.

[0158] Specifically, the system determines the alarm category based on the failure type of the consistency check: if the color code sequence is inconsistent with the ledger, it is determined as "identity mismatch"; if an unauthorized accessory (such as an optical splitter or eavesdropping box) is detected, it is determined as "illegal device"; if the number of branches exceeds the ledger's allowed limit, it is determined as "unauthorized branch"; if the wireless tag number does not match the ledger binding number, it is determined as "tag forgery".

[0159] For each type of alarm, the system automatically encapsulates it into multimodal evidence information, including: the time and geographical location of the alarm occurrence; the unique identifier of the fiber optic cable involved; the type of violation; and supporting evidence (such as the type of device detected, the surface temperature of the device, the color code sequence, the wireless tag number, and screenshots of relevant areas).

[0160] Then, the system organizes the above multimodal evidence information into standardized structured alarm data packets, in the format of structured text (such as JSON format) that can be directly parsed by the operation and maintenance system, and uploads them to the utility tunnel safety master station or dispatch center in real time.

[0161] This alarm data packet can directly trigger subsequent handling procedures, such as: automatically dispatching a manual re-inspection work order; activating the on-site audible and visual alarm device; and dispatching an inspection robot to the node for secondary confirmation. The entire process requires no manual intervention, realizing automatic detection, automatic identification, automatic alarm, and automatic linkage of illegal fiber optic cable access, forming a complete end-to-end intelligent protection closed loop.

[0162] Example 1: To enable those skilled in the art to clearly understand and implement the present invention, the following describes the specific implementation process of the present invention in detail, taking into account the actual deployment scenario of the underground integrated pipe gallery in the core area of ​​a certain city.

[0163] I. System Deployment and Initialization:

[0164] Location: Section G of the underground utility tunnel, node number GZ-G34;

[0165] Deployment equipment:

[0166] Vibration sensing unit: Fiber Bragg grating sensor, attached to the surface of the optical cable junction box;

[0167] Wireless identification tag: 128-bit AES encrypted UHF RFID chip, bound to optical fiber cable GD-2025-0876;

[0168] Inspection platform: IR-2000 pipe gallery inspection robot, equipped with a visible light camera, infrared thermal imager, RFID reader and Jetson Xavier NX edge computing unit;

[0169] The fiber optic cable ledger prior library is preloaded in the edge server.

[0170] II. Specific Implementation Steps:

[0171] Step 1, Event Triggering and Reporting: At 14:30:05 on a certain day, external personnel attempted to open the manhole cover of manhole GZ-G34 and connect the optical splitter. The fiber Bragg grating sensor detected a vibration signal with an amplitude of 1.2 volts and an energy of 0.031 joules; the dynamic baseline energy was 0.010 joules (calculated from the previous 10-minute silent period); the dual threshold conditions were met (amplitude > 1.0 V and energy > 3 × 0.010 J); the sensor reported the event to the edge server via LoRa: "Node GZ-G34 experienced a valid disturbance at 14:30:05".

[0172] Step 2, Multimodal Data Acquisition: The inspection robot arrived at node GZ-G34 as planned at 14:30:30: Querying the edge server, it found unprocessed events at this node and started fine acquisition mode; Synchronously acquiring: Visible light image: The optical cable color code is displayed as "yellow-blue-red-white", and a small gray box has been added to the right side of the junction box; Infrared thermal image: After upsampling, the temperature of the gray box area is 36.5℃; Wireless tag data: The encrypted ID is read as a3f8c2e1...; Event flag: The value is 1 (indicating that there is a reported event).

[0173] Step 3, Lightweight YOLOv8n Multimodal Recognition: The inspection robot inputs the four-channel fused image into the lightweight YOLOv8n network: The network outputs 6 types of detection results, with key items including: "Color mark area": ​​confidence level 0.96, location covers the central strip area of ​​the image; "OPGW junction box": confidence level 0.92; "Optical splitter": confidence level 0.89, location is on the right side of the junction box, associated temperature 36.5℃.

[0174] Step 4, Structured Identity Consistency Verification (Executed on the Edge Server): After receiving the detection results, the edge server performs the following verifications: Color Code Consistency: The detected color code sequence is ["yellow", "blue", "red", "white"], which matches the ledger → Pass; Attachment Compliance: "Optical splitter" is detected, but the ledger only allows "OPGW junction box" → Fail; Number of Branches: 1 branch is detected, but the ledger specifies a maximum of 0 → Fail; RFID Binding: The decrypted ID matches the ledger → Pass.

[0175] Step 5, End-to-end illegal access alarm generation: Due to the failure of attachment and branch verification, the system determines it as "illegal connection" and generates the following structured alarm:

[0176] {

[0177] Time: "2025-11-19T14:30:32",

[0178] Location: "Pipe Gallery GZ-G34",

[0179] "Optical cable identification": "GD-2025-0876",

[0180] "Type of violation": "Illegal connection"

[0181] "evidence":{

[0182] "Detection device": "Optical splitter"

[0183] Equipment temperature: 36.5

[0184] "Color code sequence":["yellow","blue","red","white"],

[0185] "Wireless Tag Number": "a3f8c2e1d9b7..."

[0186] }

[0187] }

[0188] III. Technical Feasibility Description:

[0189] All modules (fiber optic grating sensor, RFID, YOLOv8n, prior verification engine) are existing technologies or open-source components; the optical cable ledger prior verification library can be exported from the power PMS system and structured; the inspection platform only performs lightweight detection, and complex verification is completed on the edge server to meet low power consumption requirements; this embodiment only needs to change the content of the ledger library to be extended to other underground utility tunnel nodes.

[0190] In summary, the solution provided by the embodiments of the present invention has the following significant advantages and beneficial effects compared with the prior art:

[0191] I. Significantly improve the reliability and anti-interference capability of optical cable identification: By integrating visible light images, infrared thermal imaging and encrypted wireless ID information, a multimodal identity evidence chain is constructed, which effectively overcomes the complex environmental interference commonly found in underground utility tunnels, such as dirt, obstruction, tag detachment and low illumination, and significantly improves the robustness and confidence of identification.

[0192] Second, achieve three-dimensional consistency verification of "visual-wireless-ledger" to solve the problem of identity drift: The engineering specifications of power optical cables (such as color code sequence, accessory type, branch constraint) are structured into a priori library of optical cable ledgers and aligned with multi-modal recognition results in multiple dimensions. This can automatically identify hidden risks such as label forgery, optical cable replacement, and illegal splicing, and ensure the authenticity of identity from the source.

[0193] Third, it supports low-cost, low-power edge deployment mode: it adopts vibration sensing units as remote event sentinels, which only trigger responses when there is a high confidence disturbance. The inspection robot will then prioritize handling the event in subsequent tasks, avoiding the need to deploy a full set of vision equipment for each node, thus significantly reducing system deployment and maintenance costs.

[0194] IV. Achieve end-to-end intelligent identification and structured alarm output for unauthorized access: Based on prior verification and abnormal device detection results, directly generate operation and maintenance level semantic alarms (such as "identity mismatch" and "unauthorized connection") without manual review, and support seamless integration with the main station system to form a closed-loop handling process.

[0195] It should be noted that each of the above modules can be implemented as a computer program, which can run on a computer device. The computer device's memory can store the computer program that makes up each module, enabling the processor to execute each step of the above method.

[0196] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0197] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for optical cable identification and unauthorized access detection in smart grids, characterized in that, Includes the following steps: Vibration signals are acquired in real time, the amplitude and energy of the current vibration signal are calculated, and compared with the dynamic baseline energy; when the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event. In response to the effective disturbance event or according to the preset inspection plan, the inspection platform is controlled to reach the target optical cable node, and the visible light image and infrared thermal image of the target optical cable node are collected simultaneously, and the electronic code of the wireless ID tag attached to the optical cable is read. The visible light image and the upsampled infrared thermal image are stitched together along the channel dimension to form a fused image; the fused image is input into a preset lightweight target detection network, and the target detection results of preset categories in the fused image are output; Based on the target detection results, the actual identity information of the current optical cable is extracted; the expected identity information corresponding to the target optical cable node is queried from the preset optical cable ledger prior database; the actual identity information and the expected identity information are compared item by item, and the electronic code is verified to be consistent with the code bound in the ledger; based on the comparison and verification results, it is determined whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior. The optical cable ledger prior library contains three types of formalizable spatial constraints derived from power industry standards, engineering design drawings, and operation and maintenance ledgers: spatial location relationships, geometric structural characteristics, and component combination logic. The steps of inputting the fused image into a preset lightweight target detection network and outputting the target detection results for preset categories in the fused image specifically include: A multi-channel prior heatmap based on the spatial constraints of the optical cable ledger prior library is generated, with each channel corresponding to a type of detection target. The types of detection targets include identity feature type and abnormal device type. The identity feature type includes color mark area, QR code, and nameplate. The abnormal device type includes optical splitter, eavesdropping box, and illegal junction box. Before embedding the multi-channel prior heatmap into the three-stage detection head of the lightweight target detection network with additive bias; specifically, the multi-channel prior heatmap is first... Enhanced features are obtained by downsampling to three different scales using bilinear interpolation and then performing channel broadcasting and element-wise addition with the corresponding feature maps output from the Neck layer. : ； Among them, F i Let i be the original feature map at the i-th scale, i=1,2,3; The scale-adaptive guiding coefficients are then used; the enhanced features are then directly fed into the native lightweight object detection network to output multi-class object detection results; Repeat(·) indicates that the tensor is repeatedly copied along the channel dimension to adapt the original feature map; Specifically, the original detection head structure is preserved at the three scales of the lightweight target detection network, and the three-level detection results are integrated into multi-class structured detection outputs through a unified post-processing fusion mechanism. Specifically, each scale detection head outputs class confidence, bounding box coordinates and objectivity score, and the class set includes the classes corresponding to the multi-channel prior heatmap. After the lightweight object detection network completes inference, the following fusion logic is executed: Cross-scale nonmaximum suppression: merging three-level outputs at different scales; Prior-guided confidence correction: The final confidence level is dynamically adjusted based on the response value of the detection box area to the multi-channel prior heatmap. Structured mapping: The retained detection boxes are merged by category to generate a multi-dimensional identity description vector; each component contains the detection existence, location, confidence level and associated attributes, which are output as a unique detection result; the associated attributes include color code sequence and device temperature.

2. The optical cable identification and illegal access detection method for smart grids according to claim 1, characterized in that, The steps of real-time acquisition of vibration signals, calculation of the amplitude and energy of the current vibration signal, and comparison with the dynamic baseline energy specifically include: Vibration signal S vib Defined as: ； In the formula, Indicates that the fiber optic vibration sensor is in The analog voltage signal collected within a time period, measured in volts; The expression for the amplitude of the vibration signal and the triggering condition are as follows: ； The expression for the energy of the vibration signal and the triggering condition are as follows: ； In the formula, the amplitude threshold is 0.5V, E th The energy threshold is 1.8 times the dynamic baseline energy. The dynamic baseline energy is obtained by: taking the current time as the endpoint, tracing back a preset time window, removing the time periods marked as valid disturbance events, and calculating the average vibration energy of the remaining silent periods. and standard deviation , obtain dynamic baseline energy .

3. The optical cable identification and illegal access detection method for smart grids according to claim 1, characterized in that, The optical cable ledger prior knowledge base is represented as a structured knowledge base, containing the following three types of core verification information: Identity identification relationship: describes the binding relationship between the unique ID of the optical cable and the ID of the wireless tag; Visual feature specification: describes the color code sequence encoding rules for the outer sheath of optical cables; Component combination logic: describes the allowed attachment types and branch constraints.

4. The optical cable identification and illegal access detection method for smart grids according to claim 3, characterized in that, The preset lightweight target detection network is a lightweight YOLOv8n multimodal fusion recognition network based on the prior guidance of the optical cable ledger prior library. It encodes the spatial constraints in the optical cable ledger prior library into a multi-channel prior heatmap and embeds Prior modules in front of the three-level detection heads respectively, injecting feature streams in an additive bias manner to achieve multi-scale spatial attention guidance.

5. The optical cable identification and illegal access detection method for smart grids according to claim 4, characterized in that, The step of stitching the visible light image and the upsampled infrared thermal image together along the channel dimension to form a fused image specifically includes: The mathematical expression for a visible light image is: ; The mathematical expression for an infrared thermal image is: ; The infrared thermal image is upsampled to have the same spatial dimension as the visible light image: bilinear interpolation is used for upsampling, and the upsampled infrared thermal image is used as the fourth channel. The channel dimension is then stitched together with the visible light image to form a four-channel fused image.

6. The optical cable identification and illegal access detection method for smart grids according to claim 1, characterized in that, Based on the target detection results, extract the actual identity information of the current optical cable; query the expected identity information corresponding to the target optical cable node from the preset optical cable ledger prior library, compare the actual identity information with the expected identity information item by item, and verify whether the electronic code is consistent with the code bound in the ledger; Based on the comparison and verification results, the steps to determine whether the actual identity information of the optical cable is legitimate and whether there is any unauthorized access include: The detection results of various targets are classified and summarized to form a structured actual identity information; the actual identity information includes the actual identified color mark sequence, the actual detected accessory type, the actual identified number of branches, and the electronic code read from the wireless ID tag; The expected identity information corresponding to the target optical cable node is retrieved from the preset optical cable ledger database; the expected identity information includes: the expected color code sequence, the types of accessories allowed to be installed, the maximum number of branches allowed, and the electronic code to be bound. The actual identity information is compared with the expected identity information item by item to perform a consistency check: If the actual identified color mark sequence is inconsistent with the expected color mark sequence, it is judged as "identity mismatch"; If the actual detected accessory type is not within the range of allowed accessory types, it is determined to be an "illegal device"; If the number of branches actually identified exceeds the maximum number of branches allowed, it is judged as an "illegal branch"; If the electronic code read from the wireless ID tag does not match the electronic code that should be bound, it is determined to be "tag forgery"; If any of the above consistency checks fails, the current optical cable is deemed to have a security risk, triggering the corresponding alarm type and entering the alarm generation process; if all consistency checks pass, the current optical cable is deemed legal and recorded as normal.

7. The optical cable identification and illegal access detection method for smart grids according to claim 6, characterized in that, The alarm generation process specifically includes: Determine the alarm category based on the type of consistency check failure; For each type of alarm, it is automatically packaged into multimodal evidence information; Multimodal evidence information is organized into standardized, structured alarm data packets, and these alarm data packets are uploaded in real time.

8. A fiber optic cable identification and illegal access detection system for smart grids, used to implement the fiber optic cable identification and illegal access detection method according to any one of claims 1-7, characterized in that, include: The event-driven triggering module is used to acquire vibration signals in real time, calculate the amplitude and energy of the current vibration signal, and compare it with the dynamic baseline energy. When the amplitude of the vibration signal is greater than a preset amplitude threshold and its energy is greater than the energy threshold calculated based on the dynamic baseline energy, it is determined to be a valid disturbance event. The multimodal data acquisition module is used to respond to the effective disturbance event or according to the preset inspection plan, control the inspection platform to arrive at the target optical cable node, and simultaneously acquire the visible light image and infrared thermal image of the target optical cable node, and read the electronic code of the wireless ID tag attached to the optical cable. The target detection module is used to stitch the visible light image and the upsampled infrared thermal image in the channel dimension to form a fused image; input the fused image into a preset lightweight target detection network, and output the target detection results of preset categories in the fused image; The consistency verification module is used to extract the actual identity information of the current optical cable based on the target detection result; query the expected identity information corresponding to the target optical cable node from the preset optical cable ledger prior library; compare the actual identity information with the expected identity information item by item; and verify whether the electronic code is consistent with the code bound in the ledger; and determine whether the actual identity information of the optical cable is legal and whether there is any illegal access behavior based on the comparison and verification results.

Images