Tandem ultrahigh frequency sensing GIS three-dimensional positioning and monitoring system and method

Abstract

The invention discloses a series ultrahigh frequency sensing GIS three-dimensional positioning and monitoring system and method, and belongs to the technical field of high-voltage equipment state monitoring. The system comprises a sensing layer, a processing layer, a transmission layer and an application layer, wherein the sensing layer synchronously collects ultrahigh-frequency electromagnetic signals, sound wave vibration signals and SF6 gas state parameters through a plurality of serial multi-physical-quantity sensing nodes; the processing layer eliminates interference based on signal homology verification, calls an intelligent response model to dynamically compensate propagation path errors in combination with a GIS three-dimensional structure model and real-time environment data, outputs accurate three-dimensional coordinates of the partial discharge source by adopting a progressive time difference of arrival positioning algorithm, and identifies fault types and risk levels by fusing multi-dimensional features; the transmission layer encrypts and uploads the diagnosis report; and the application layer maps defect positions in the three-dimensional visual model and automatically generates an operation and maintenance work order according to risk levels. According to the invention, high-precision positioning, multi-source fusion diagnosis and closed-loop operation and maintenance are realized, and the accuracy and intelligent level of GIS equipment monitoring are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of high-voltage equipment monitoring technology, and specifically relates to a GIS three-dimensional positioning and monitoring system and method using a series ultra-high frequency sensor. Background Technology

[0002] In existing technologies, partial discharge monitoring in GIS (Gas Insulated Switchgear) mainly relies on ultra-high frequency (UHF) electromagnetic wave methods or ultrasonic methods for detection. A typical approach involves installing several UHF sensors on the outer wall or inside the GIS cavity. These sensors detect electromagnetic signals generated by partial discharge in the 300MHz–1.5GHz frequency band and use a time difference of arrival (TDOA) algorithm to locate the discharge source in two dimensions or a rough three-dimensional position. Some systems supplement this with ultrasonic sensors to collect vibration signals for cross-validation. Additionally, some approaches incorporate SF6 gas state parameters (such as temperature, humidity, and pressure) as auxiliary criteria. However, these methods generally employ static placement strategies, with sensor layout largely based on engineering experience, failing to fully consider the impact of the complex geometry of the GIS cavity on signal propagation paths and coverage blind spots.

[0003] The main technical problem with the above approach is that, in complex cavity structures and dynamic operating environments, existing positioning methods suffer from a significant decrease in 3D positioning accuracy because they ignore the non-ideal characteristics of electromagnetic wave propagation paths (such as reflection, refraction, insulator obstruction, and additional time delay caused by cavity deformation). Traditional TDOA algorithms typically assume that signals propagate in a straight line in a uniform medium, which cannot dynamically compensate for propagation errors caused by structural changes in actual operating conditions. This results in positioning results that deviate from the true position, making it difficult to meet the requirements of high-reliability operation and maintenance. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a GIS three-dimensional positioning and monitoring system and method with series ultra-high frequency sensing. By constructing a multi-physical quantity synchronous sensing network, introducing a sensor layout strategy optimized by swarm intelligence, designing an anti-interference mechanism based on homology verification, developing a progressive positioning algorithm that integrates cavity deformation compensation, and combining it with a deep reinforcement learning-driven intelligent response model, this invention achieves high-precision three-dimensional positioning of GIS local discharge sources, intelligent identification of fault types, dynamic assessment of risk levels, and automatic triggering of maintenance work orders, significantly improving the accuracy, real-time performance, and intelligence level of GIS equipment status perception.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A series-connected ultra-high frequency sensor GIS three-dimensional positioning and monitoring system includes: The sensing layer is used to synchronously collect ultra-high frequency electromagnetic signals, acoustic vibration signals, and gas state parameters within the GIS cavity through multiple serially connected multi-physical quantity sensing nodes. The multiple serially connected multi-physical quantity sensing nodes are arranged in a spiral pattern along the axis of the GIS cavity, including at least a spiral arrangement path that advances spirally along the cavity axis and a zigzag arrangement path that advances along the generatrix. The arc distance between adjacent nodes is 100-200mm, and the normal distance between the node and the inner wall of the cavity is 15-25mm. The serial arrangement enables the signals collected by each node to form a time-series correlation link, providing continuous data support for propagation path error compensation and progressive positioning.

[0006] The processing layer, connected to the sensing layer, is used to verify the homology of UHF electromagnetic signals and acoustic vibration signals, eliminate non-homogeneous interference signals, and generate purified multi-dimensional feature vectors. Combining the GIS cavity 3D structural model and real-time environmental data, it calls the intelligent response model to dynamically compensate for signal propagation path errors and uses a progressive positioning algorithm based on time difference of arrival to output the precise 3D coordinates of the partial discharge source. It integrates electrical features, acoustic features, and gas state parameters from the multi-dimensional feature vectors to identify the fault type of partial discharge and assess its risk level. It also periodically injects standard test signals into multiple series-connected multi-physical quantity sensing nodes, calibrates the signal response characteristics of each node based on feedback results, and incrementally updates the intelligent response model using new data accumulated during operation. The transport layer, connecting the processing layer, is used to encrypt and transmit diagnostic reports containing precise three-dimensional coordinates, fault types, and risk levels to the application. The application layer connects to the transport layer and is used to map defect locations in the GIS 3D visualization model, and trigger maintenance work order generation and closed-loop feedback mechanisms based on risk levels.

[0007] Preferably, the sensing layer comprises multiple series-connected multi-physical quantity sensing nodes, each integrating an ultra-high frequency sensor, an acoustic vibration sensor, and a gas state sensor. The spatial positions of these multiple series-connected multi-physical quantity sensing nodes are determined by a final position coordinate matrix generated through digital modeling and swarm intelligence optimization algorithms based on the geometric structure information of the monitored GIS cavity. Optionally, the node spacing is determined using a particle swarm optimization algorithm to ensure that the volume ratio of the monitoring blind zone is less than 2%. The series-connected multi-physical quantity sensing nodes are connected in series via shielded coaxial cables to form a unified, time-synchronized sensing link with a synchronization error better than ±1 ns. Based on the time-series correlation data of the series-connected nodes, a continuous model of the signal propagation path is constructed to improve the accuracy of path error compensation.

[0008] Preferably, the processing layer includes: The anti-interference processing module is used to verify the homology between ultra-high frequency electromagnetic signals and acoustic vibration signals, eliminate non-homogeneous interference signals, and generate purified multi-dimensional feature vectors. The three-dimensional positioning algorithm module is used to combine the GIS cavity three-dimensional structure model with real-time environmental data, call the intelligent response model to dynamically compensate for signal propagation path errors, and use a progressive positioning algorithm based on time difference of arrival to output the accurate three-dimensional coordinates of the local discharge source. The fault diagnosis module is used to fuse electrical features, acoustic features, and gas state parameters from a multi-dimensional feature vector to identify the fault type of partial discharge and assess the risk level. The self-calibration module is used to periodically inject standard test signals into each serial multi-physical quantity sensing node, calibrate the signal response characteristics of the node based on the feedback results, and incrementally update the intelligent response model using the operating data.

[0009] Preferably, the transport layer is configured to encrypt the diagnostic report before transmitting it to the application, and the transmission strategy is dynamically adjusted according to the risk level.

[0010] Preferably, the application layer includes: The GIS 3D visualization model module is used to load high-precision 3D models of GIS equipment and render defect identifiers based on the received precise 3D coordinates, and display fault types, risk levels and gas state information in association. The operation and maintenance linkage unit is used to automatically generate electronic maintenance work orders when the risk level is "abnormal" or "critical", push them to authorized operation and maintenance terminals, and receive on-site feedback to complete closed-loop operation and maintenance.

[0011] Preferably, it also includes a layout optimization subsystem, which is used to acquire GIS cavity geometry data, construct cavity topology model, generate candidate sensing paths, simulate and calculate monitoring coverage, use swarm intelligence optimization algorithm to determine the optimal spatial layout, and output the final position coordinate matrix to characterize the spatial position of multiple serial multi-physical quantity sensing nodes.

[0012] Preferably, the layout optimization subsystem is also equipped with a dynamic compensation mechanism to continuously monitor cavity deformation during system operation and trigger online updates of sensor layout parameters when the deformation exceeds a preset threshold.

[0013] Preferably, the progressive positioning process executed by the 3D positioning algorithm module includes: The preliminary three-dimensional coordinates of the local discharge source are output using the Chan algorithm; The intelligent response model is invoked to determine, based on the current environmental data, whether there is any deformation of the cavity structure that could cause the electromagnetic wave propagation path to deviate. If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each series multi-physical quantity sensing node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator blocking is quantified. By using additional propagation delay to correct the original arrival time, an equivalent arrival time that better reflects the actual working conditions is obtained; The arrival time difference measurement vector is recalculated based on the corrected equivalent arrival time, and the positioning algorithm is executed again. The iteration continues until the spatial coordinate change between two consecutive positioning results is less than the preset convergence threshold, and the accurate three-dimensional coordinates of the local discharge power source are output.

[0014] Preferably, the intelligent response model integrated in the processing layer is trained through a deep reinforcement learning framework. Its state inputs include cavity deformation, temperature gradient and vibration spectrum, and its action outputs are the propagation path correction coefficient and time delay compensation parameter in the positioning algorithm. The reward function is defined as the negative value of the positioning error, and the training objective is to minimize the three-dimensional positioning deviation of the local discharge source in a dynamic environment.

[0015] Preferably, the gas state sensor includes a temperature sensor, a humidity sensor, and a pressure sensor for monitoring the SF6 gas medium.

[0016] A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor includes the following steps: Based on the geometric structure information of the monitored GIS cavity, the optimal spatial layout of multiple serial multi-physical quantity sensing nodes in the GIS cavity is determined through digital modeling and swarm intelligence optimization algorithms, and the final position coordinate matrix is ​​generated to characterize the spatial position of these nodes in the global coordinate system. Based on the final location coordinate matrix, deploy multiple serial multi-physical quantity sensing nodes and construct a physical sensing network consistent with the optimal spatial layout; Multiple serial multi-physical quantity sensing nodes are controlled to synchronously acquire ultra-high frequency electromagnetic signals, acoustic vibration signals and gas state parameters under a unified time sequence, and the acquired data are aggregated after being assigned a unified time stamp. The original signal after aggregation is verified based on the homology between the UHF signal and the acoustic signal. Non-homogeneous interference signals are eliminated, and a purified multidimensional feature vector is generated. By combining the GIS cavity 3D structural model with real-time environmental data, a pre-trained intelligent response model is invoked to dynamically compensate for signal propagation path errors. A progressive positioning algorithm based on time difference of arrival is adopted to output the precise 3D coordinates of the local discharge source. By integrating electrical features, acoustic features, and gas state parameters from multidimensional feature vectors, the fault types of partial discharge can be identified and their risk levels can be assessed. The precise 3D coordinates, fault type, and risk level are packaged into a diagnostic report, which is then encrypted and transmitted to the application. The defect location is mapped in the GIS 3D visualization model, and the operation and maintenance work order generation and closed-loop feedback mechanism are triggered according to the risk level. Standard test signals are periodically injected into multiple serially connected multi-physical quantity sensing nodes. The signal response characteristics of each node are calibrated based on the feedback results, and the intelligent response model is incrementally updated using new data accumulated during operation.

[0017] Preferably, based on the geometric structure information of the monitored GIS cavity, the optimal spatial layout of multiple serially connected multi-physical quantity sensing nodes within the GIS cavity is determined through digital modeling and swarm intelligence optimization algorithms, and a final position coordinate matrix is ​​generated to characterize the spatial positions of these nodes in the global coordinate system. The specific method is as follows: Acquire the internal geometry data of the GIS cavity, including the length, diameter, bending radius of each air chamber, curvature characteristics of flange connections and basin insulator areas, and obtain the initial empirical parameters for sensor layout; A three-dimensional unstructured mesh generation method is used to discretize the internal space of the GIS cavity and construct a high-fidelity cavity topology model; Generate at least two candidate linear topological paths in the cavity topology model, including a spiral path that advances spirally along the cavity axis and a polygonal path that advances along the generatrix. The monitoring coverage of UHF signals to the cavity interior space under spiral and piecewise linear arrangement paths was simulated and calculated using the finite-time integral method. If the monitoring coverage of the spiral arrangement path is more than 5% higher than that of the piecewise linear arrangement path, the spiral arrangement path is selected as the basic topology. Extract the set of key geometric parameters in the spiral arrangement path, including the arc distance between adjacent nodes, the tangent bending angle at the node of the curved segment, and the normal distance between the node and the inner wall of the cavity; With the goal of maximizing monitoring coverage and minimizing average positioning error, the particle swarm optimization algorithm is used to perform multi-objective iterative optimization on the set of key geometric parameters to obtain a preliminary optimized set of geometric parameters. Acquire data on cavity deformation caused by thermal expansion and contraction, internal pressure fluctuations, or external mechanical vibrations during the operation of GIS equipment; Based on deformation data, the Kalman filter method is used to estimate and predict the state of the initially optimized set of geometric parameters; if the predicted deformation amplitude exceeds the preset deformation threshold, the set of geometric parameters is updated to obtain the corrected geometric parameters. Based on the corrected geometric parameters, a final position coordinate matrix is ​​generated to characterize the spatial positions of multiple serial multi-physical quantity sensing nodes in the global coordinate system. Based on the final position coordinate matrix, another electromagnetic field numerical simulation is performed to generate a signal intensity distribution map inside the cavity. Image recognition and region growing algorithms are used to extract the monitoring blind zone from the signal intensity distribution map, and the proportion of the total volume of the blind zone to the total volume inside the cavity is calculated. If the proportion is less than 2%, the current layout scheme is confirmed to be effective; otherwise, the process returns to the particle swarm optimization algorithm step and iterates again until the coverage requirement is met.

[0018] Preferably, multiple series-connected multi-physical quantity sensing nodes are controlled to synchronously acquire ultra-high frequency electromagnetic signals, acoustic vibration signals, and gas state parameters under a unified time sequence, and the acquired data are aggregated after being assigned a unified time stamp. The specific method is as follows: Send unified synchronization timing commands to multiple serial multi-physical quantity sensing nodes; Each node synchronously starts its built-in UHF sensor to collect electromagnetic transient signals inside the GIS cavity at a sampling rate of no less than 1GS / s, and adds a uniform time stamp to the collected UHF raw waveform data; The acoustic wave sensors at each node are activated synchronously to collect acoustic wave vibration signals within the same time window as the UHF signal, and the original acoustic wave waveform data is time-aligned with the corresponding UHF signal. The temperature, humidity and pressure parameters output by the built-in gas state sensors of each node are read synchronously to form equipment status data that is synchronized with the waveform data in time. Each node will upload UHF raw waveform data, acoustic raw waveform data and gas state parameters with uniform time stamps to the data aggregation node through its respective sensor network; The data aggregation node packages, verifies, and standardizes the format of all uploaded data before sending it to the subsequent processing module.

[0019] Preferably, the converged original signal is filtered and denoised, and based on the homology verification between the UHF signal and the acoustic signal, non-homogeneous interference signals are eliminated to generate a purified multidimensional feature vector. The specific method is as follows: The aggregated mixed data packets are parsed to separate the ultra-high frequency waveform data, acoustic waveform data and gas state parameters corresponding to each node; A bandpass filter with a center frequency of 300MHz–1.5GHz is applied to the UHF waveform data, and wavelet threshold noise reduction is applied to the acoustic waveform data. Extracting the precise start time of the denoised UHF signal pulse It searches for the start time of the corresponding acoustic signal pulse within its preset time window. ; Calculate the time difference If the time difference is less than the set time tolerance threshold, and the cross-correlation coefficient of the two signals in the time-frequency domain is greater than the preset correlation threshold, then the signals are determined to be real partial discharge signals from the same source. Retain signals of the same origin and eliminate UHF signals that are identified as external electromagnetic interference; The amplitude, phase, frequency band energy distribution, acoustic wave dominant frequency and energy integral of the retained effective signal, along with the synchronously acquired gas temperature, humidity and pressure parameters, are combined into a multi-dimensional feature vector, which serves as the effective data output after purification.

[0020] Preferably, by combining the GIS cavity 3D structural model with real-time environmental data, a pre-trained intelligent response model is invoked to dynamically compensate for signal propagation path errors. A progressive positioning algorithm based on time difference of arrival is adopted to output the precise 3D coordinates of the local discharge source. The specific method is as follows: From the purified valid data, select at least four nodes whose amplitude exceeds the trigger threshold for UHF signals, and extract their precise arrival time when they receive the same partial discharge pulse. Using one node as a reference, calculate the signal arrival time difference of the other nodes relative to the reference node to form an arrival time difference measurement vector; Under the assumption of a homogeneous medium, the Chan algorithm is used to solve for the time difference measurement vector to obtain the preliminary three-dimensional coordinates of the local discharge source; The initial three-dimensional coordinates are associated with the GIS cavity three-dimensional structural model, and the temperature field and vibration data monitored in real time by the sensor network are input. The pre-trained intelligent response model is invoked to determine, based on the current environmental data, whether there is any deformation of the cavity structure that would cause the electromagnetic wave propagation path to deviate. If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator blocking is quantified. By using additional propagation delay to correct the original arrival time, an equivalent arrival time that better reflects the actual working conditions is obtained; The time difference vector is recalculated based on the corrected equivalent time of arrival, and the positioning algorithm is executed again. The iteration continues until the change in spatial coordinates between two consecutive positioning results is less than the preset convergence threshold, and the final accurate three-dimensional coordinates of the local discharge source are output.

[0021] Preferably, the electrical features, acoustic features, and gas state parameters from the multi-dimensional feature vector are integrated to identify the fault type of partial discharge and assess its risk level. The specific method is as follows: Extract the PRPD spectrum, center frequency, and bandwidth characteristics of the UHF signal corresponding to the precise three-dimensional coordinates from the purified effective data, as well as the energy integral and main frequency characteristics of the acoustic signal. Electrical and acoustic characteristics are fused with synchronously acquired gas state parameters to form a comprehensive feature vector; The comprehensive feature vector is input into the pre-trained fault diagnosis classification model, and the fault type with the highest probability is output. The fault types include tip discharge, floating potential body discharge, free metal particle discharge or internal air gap discharge of insulation. Based on the diagnosed fault type, discharge signal amplitude, pulse repetition rate, and deterioration trend of gas state parameters, combined with the preset risk assessment rules, the current defect risk level is calculated and classified as "normal", "caution", "abnormal" or "critical".

[0022] Preferably, the precise 3D coordinates, fault type, and risk level are encapsulated into a diagnostic report, encrypted, and transmitted to the application. The defect location is mapped in the GIS 3D visualization model, and a maintenance work order generation and closed-loop feedback mechanism are triggered based on the risk level. The specific method is as follows: The precise three-dimensional coordinates, fault type, and risk level are packaged into a structured diagnostic report; The diagnostic report is encrypted; The encrypted diagnostic report is transmitted to the application. In the 3D visualization model of the GIS equipment loaded on the application side, defect identifiers are rendered according to precise 3D coordinates, and fault type, risk level and gas state information are displayed in association. When the risk level is "abnormal" or "critical", an electronic maintenance work order containing location information, diagnostic conclusions and handling suggestions is automatically generated and pushed to the authorized operation and maintenance terminal. Receive on-site processing feedback from the operation and maintenance terminal, update the work order status and archive the entire process data to form a closed-loop operation and maintenance record of "monitoring-diagnosis-decision-execution-feedback".

[0023] Preferably, standard test signals are periodically injected into multiple serially connected multi-physical quantity sensing nodes. The signal response characteristics of each node are calibrated based on the feedback results, and the intelligent response model is incrementally updated using new data accumulated during operation. The specific method is as follows: Generate standard electrical pulse signals and standard acoustic pulse signals with known amplitudes and waveforms according to a preset period; The standard signal is synchronously injected into the signal input terminal of each node; Each node collects and reports back the measured signal amplitude, waveform characteristics, and response time; The feedback signal is compared with the standard signal to calculate the channel attenuation coefficient and sensitivity deviation factor for each node. Based on the deviation factor, the compensation coefficients of each node in subsequent signal processing are dynamically updated to complete online calibration; The optimal compensation parameters, corresponding environmental data, and positioning error data obtained during this calibration process will be used as new training samples. By using newly added training samples, the intelligent response model is incrementally learned, its internal parameters are updated, and its adaptability to dynamic operating environments is improved.

[0024] Preferably, the intelligent response model is trained through a deep reinforcement learning framework. Its state inputs include cavity deformation, temperature gradient and vibration spectrum. The action outputs are the propagation path correction coefficient and time delay compensation parameter in the positioning algorithm. The reward function is defined as the negative value of the positioning error. The training objective is to minimize the three-dimensional positioning deviation of the local discharge source in a dynamic environment.

[0025] A storage medium storing an executable program, which, when executed, is used to implement the aforementioned GIS three-dimensional positioning and monitoring method using a series ultra-high frequency sensor.

[0026] The present invention can achieve the following beneficial effects: 1. This invention introduces an intelligent response model based on a three-dimensional cavity structure model and real-time environmental data to dynamically compensate for propagation path errors caused by cavity deformation, insulator obstruction, and electromagnetic wave reflection / refraction. It also employs a progressive time difference of arrival (TDOA) positioning algorithm to iteratively correct coordinates, effectively overcoming the problem of large positioning deviations in complex GIS structures using traditional methods, and achieving millimeter-level high-precision three-dimensional positioning.

[0027] 2. This invention utilizes the homology verification mechanism of ultra-high frequency signals and acoustic vibration signals to accurately eliminate non-homogeneous signals such as external electromagnetic interference. It combines electrical characteristics, acoustic characteristics and gas state parameters for multi-source fusion analysis, which significantly improves the accuracy of partial discharge identification and the reliability of fault type judgment.

[0028] 3. This invention automatically generates the optimal spatial layout of sensors through digital modeling and swarm intelligence optimization algorithms (such as particle swarm optimization), and introduces a dynamic compensation mechanism to update node position parameters online, ensuring that the volume ratio of monitoring blind spots is less than 2%, which greatly improves the ability to perceive the entire area of ​​complex GIS cavities.

[0029] 4. This invention periodically injects standard test signals to achieve online calibration of node sensitivity, and uses the operating data as new samples to incrementally learn the intelligent response model, enabling the system to have long-term stability and the ability to adapt to dynamic operating conditions.

[0030] 5. This invention automatically maps the precise positioning results and risk levels to the GIS 3D visualization model, triggers electronic work orders as needed and links with on-site feedback, forming an integrated closed-loop management of "monitoring-diagnosis-decision-execution-feedback", improving operation and maintenance efficiency and power grid safety.

[0031] 6. Compared to the distributed, discrete layout of existing technologies, the serial layout of this invention can form a continuous signal sensing link, reducing monitoring blind spots. Simultaneously, by utilizing the temporal correlation between nodes, the computational complexity of propagation path errors is reduced. Multiple serial multi-physical quantity sensing nodes are connected in series via shielded coaxial cables and sequentially connected to the data aggregation node, achieving cascaded signal transmission and unified synchronous control. Attached Figure Description

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a system structure diagram of the present invention; Figure 2 This is a diagram of the perception layer architecture of the present invention; Figure 3 This is a diagram of the application layer architecture of the present invention; Figure 4 This is a diagram of the processing layer architecture of the present invention; Figure 5 This is a diagram of the layout optimization subsystem architecture of the present invention; Figure 6 This is a flowchart of the method of the present invention; Figure 7 This is a diagram showing the signal intensity distribution inside the cavity of the present invention; Figure 8 This is a statistical chart showing the monitoring coverage and iteration count of this invention; Figure 9 This is a statistical chart showing the fault types and their accurate identification in this invention. Detailed Implementation

[0033] Example 1: Preferred solutions include Figures 1 to 5 As shown, a GIS three-dimensional positioning and monitoring system with series ultra-high frequency sensing includes: The sensing layer is used to synchronously collect ultra-high frequency electromagnetic signals, acoustic vibration signals and gas state parameters within the GIS cavity through multiple serial multi-physical quantity sensing nodes. The processing layer, connected to the sensing layer, is used to verify the homology of UHF electromagnetic signals and acoustic vibration signals, eliminate non-homogeneous interference signals, and generate purified multi-dimensional feature vectors. Combining the GIS cavity 3D structural model and real-time environmental data, it calls the intelligent response model to dynamically compensate for signal propagation path errors and uses a progressive positioning algorithm based on time difference of arrival to output the precise 3D coordinates of the partial discharge source. It integrates electrical features, acoustic features, and gas state parameters from the multi-dimensional feature vectors to identify the fault type of partial discharge and assess its risk level. It also periodically injects standard test signals into multiple series-connected multi-physical quantity sensing nodes, calibrates the signal response characteristics of each node based on feedback results, and incrementally updates the intelligent response model using new data accumulated during operation. The transport layer, connecting the processing layer, is used to encrypt and transmit diagnostic reports containing precise three-dimensional coordinates, fault types, and risk levels to the application. The application layer connects to the transport layer and is used to map defect locations in the GIS 3D visualization model, and trigger maintenance work order generation and closed-loop feedback mechanisms based on risk levels.

[0034] Furthermore, the sensing layer includes multiple serial multi-physical quantity sensing nodes, each node integrating an ultra-high frequency sensor, an acoustic vibration sensor, and a gas state sensor; the spatial position of the multiple serial multi-physical quantity sensing nodes is determined by the final position coordinate matrix generated based on the geometric structure information of the monitored GIS cavity through digital modeling and swarm intelligence optimization algorithms. Furthermore, in a 220kV GIS substation, a typical GIS cavity containing straight sections, L-shaped bends, and basin insulators was selected as the monitoring object. Based on the CAD model of this cavity (including geometric parameters such as flange position, bending radius, and chamber length), the optimal spatial layout scheme was generated using the layout optimization subsystem, and the final position coordinate matrix was output. Subsequently, according to the coordinate matrix, eight series-connected multi-physical quantity sensing nodes (numbered N1–N8) were installed along the axial direction of the inner wall of the GIS cavity in a spiral path. Each node integrates an ultra-high frequency (UHF) sensor (center frequency 900MHz, bandwidth 300MHz–1.5GHz), a piezoelectric acoustic vibration sensor (frequency response range 20kHz–200kHz), and an SF6 gas state sensor (including a three-in-one module for temperature, humidity, and pressure). All nodes are connected in series to the data aggregation unit via shielded coaxial cables and connected to a unified timing system (such as the IEEE 1588PTP protocol) to ensure that the sampling synchronization accuracy is better than ±1ns. Optionally, in this embodiment, the eight serial nodes are connected in a spiral along the GIS cavity, with an arc distance of 150mm between adjacent nodes and a normal distance of 20mm. The signal transmission delay of the serial link is controlled within 50ns.

[0035] Furthermore, the processing layer includes: The anti-interference processing module is used to filter and reduce noise in the converged original signal, and to verify and eliminate non-originating interference signals based on the homology between ultra-high frequency electromagnetic signals and acoustic vibration signals, and to generate purified multi-dimensional feature vectors. The three-dimensional positioning algorithm module is used to select at least four UHF electromagnetic signals from series-connected multi-physical quantity sensing nodes whose amplitudes exceed the trigger threshold from the purified multi-dimensional feature vector, and extract the precise arrival time of the received partial discharge pulse. Taking one node as a reference, the signal arrival time difference of the other nodes relative to the reference node is calculated to form the arrival time difference measurement vector. Under the assumption of a homogeneous medium, the Chan algorithm is used for preliminary positioning to obtain the preliminary three-dimensional coordinates of the partial discharge source. Then, combined with the GIS cavity three-dimensional structure model and real-time environmental data, the intelligent response model is called to determine whether there is cavity structure deformation. If so, the electromagnetic wave propagation path is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator obstruction is quantified. The original arrival time is corrected using this additional propagation delay to obtain the equivalent arrival time that reflects the actual situation. The arrival time difference measurement vector is recalculated based on the corrected equivalent arrival time, and the positioning algorithm is executed again. The iteration continues until the spatial coordinate change of two consecutive positioning results is less than the preset convergence threshold, and the precise three-dimensional coordinates of the partial discharge source are output. The fault diagnosis module is used to extract the PRPD spectrum, center frequency, and bandwidth features of the ultra-high frequency electromagnetic signal corresponding to the precise three-dimensional coordinates from the purified multi-dimensional feature vector, as well as the energy integral and main frequency features of the acoustic vibration signal; it fuses the electrical features, acoustic features, and synchronously acquired gas state parameters to form a comprehensive feature vector; it inputs the comprehensive feature vector into a pre-trained fault diagnosis classification model and outputs the fault type with the highest probability, which includes tip discharge, floating potential body discharge, free metal particle discharge, or internal air gap discharge of insulation; and it calculates the risk level of the current defect based on the fault type, discharge signal amplitude, pulse repetition rate, and deterioration trend of gas state parameters, combined with preset risk assessment rules, which are divided into "normal", "caution", "abnormal", or "critical". The self-calibration module generates standard electrical pulse signals and standard acoustic pulse signals with known amplitudes and waveforms according to a preset period, and synchronously injects the standard signals into the signal input terminals of each series-connected multi-physical quantity sensing node; it receives the signal amplitude, waveform characteristics, and response time fed back by each node; it compares the feedback signals with the standard signals and calculates the channel attenuation coefficient and sensitivity deviation factor of each node; it dynamically updates the compensation coefficients of each node in subsequent signal processing based on the deviation factor; and it uses the optimal compensation parameters, corresponding environmental data, and positioning error data obtained during this calibration process as new training samples for incremental learning of the intelligent response model.

[0036] Furthermore, after system startup, all sensor nodes operate synchronously under unified timing commands: the UHF sensor continuously acquires electromagnetic transient signals within the cavity at a sampling rate of 1.25 GS / s; the acoustic sensor synchronously records mechanical vibration signals within the same time window; the gas state sensor outputs temperature, humidity, and pressure data every 100 ms, and timestamps the waveform data accordingly; all raw data are preprocessed locally and then uploaded to the data aggregation node of the processing layer. The anti-interference processing module first applies a 300 MHz–1.5 GHz bandpass filter to the UHF signal, performs 5-level decomposition of the acoustic signal using the db4 wavelet basis, and performs soft thresholding denoising; then it performs homology verification: for each UHF pulse, it searches for the corresponding acoustic pulse within a ±50 μs time window; if the time difference Δt < 20 μs and the cross-correlation coefficient ρ > 0.7, it is determined to be a homologous signal; otherwise, it is discarded; the retained effective signals are combined with their amplitude, phase, spectral energy, acoustic frequency, energy integral, and gas parameters to form a multi-dimensional feature vector. The 3D positioning algorithm module executes a progressive process: it filters at least four nodes with UHF amplitudes exceeding -65dBm (e.g., N2, N4, N6, N7) and extracts their precise times of arrival (TDOA); it calculates the TDOA vector using N2 as a reference; and it operates in a homogeneous SF6 medium (ε... r The initial coordinates P0 are obtained using the Chan algorithm at approximately 1.002. The intelligent response model is then invoked, and the temperature gradient, vibration spectrum, and cavity model are used to determine if significant deformation (e.g., thermal expansion offset > 2 mm) exists. If so, the propagation path is simulated using CSTStudioSuite based on the updated model, and the additional time delay τ is quantified. i ; Correct the arrival time to t′ i =t i -τ iReconstruct TDOA and reposition it, iterating until the coordinate change is <5mm, and output precise coordinates P; the intelligent response model is trained using deep reinforcement learning (such as PPO), the state input is [cavity deformation, temperature gradient, vibration frequency], the action output is [path correction coefficient, time delay compensation parameter], and the reward function is the negative value of the positioning error. The fault diagnosis module extracts the PRPD spectrum, center frequency (e.g., 850MHz), and 3dB bandwidth of the UHF signal, the energy integral and dominant frequency of the sound wave (e.g., 45kHz), and SF6 gas parameters (temperature 25℃, humidity 150ppm, pressure 0.5MPa), forming a 9-dimensional comprehensive feature vector which is input into the pre-trained XGBoost model. The output fault type probability is then calculated (e.g., "free metal particle discharge" probability 89%). The risk level is then assessed according to rules. For example, if free metal particle discharge is detected, and the signal amplitude is greater than -50dBm, the pulse repetition rate is greater than 500p / s, and humidity continues to rise, the system classifies the defect as "critical." If it is only a slight tip discharge and all parameters are stable, it is classified as "caution." The self-calibration module injects a standard UHF pulse (Gaussian pulse, FWHM=1ns, amplitude -40dBm) and a standard acoustic pulse (50kHz sine burst, duration 100μs) every 24 hours. Each node feeds back measured values, and the attenuation coefficient α is calculated. i With sensitivity deviation δ i Dynamically update the gain compensation coefficient G i =1 / (α i ·(1+δ i The calibration data is used as a new sample to trigger incremental learning of the intelligent response model.

[0037] Furthermore, the transport layer is configured to encrypt the diagnostic report before transmitting it to the application, and the transmission strategy is dynamically adjusted according to the risk level. The diagnostic report (including precise 3D coordinates, fault type, and risk level) is encrypted with AES-256 and transmitted to the application via fiber optic or 5G private network. When the risk level is "abnormal" or "critical," the system prioritizes transmission bandwidth and real-time performance to ensure that critical alarm information is delivered to maintenance personnel in a timely manner.

[0038] Furthermore, the application layer includes: The GIS 3D visualization model module is used to load high-precision 3D models of GIS equipment and render defect identifiers based on the received precise 3D coordinates, and display fault types, risk levels and gas state information in association. The operation and maintenance linkage unit is used to automatically generate an electronic maintenance work order containing location information, diagnostic conclusions and handling suggestions when the risk level is "abnormal" or "critical", and push it to the authorized operation and maintenance terminal; receive on-site handling feedback from the operation and maintenance terminal, update the work order status and archive the whole process data, forming a closed-loop operation and maintenance record of "monitoring-diagnosis-decision-execution-feedback".

[0039] Furthermore, the GIS 3D visualization model is developed based on the Unity3D engine, which can intuitively mark the location of defects (such as red icons) and pop-up prompts; the electronic work order includes the 3D coordinates of the defect (supports importing AR inspection equipment), fault mechanism analysis (such as "suspected free metal particles jumping in the tank") and handling suggestions (such as "arrange a power outage inspection, focusing on the N5-N7 section"); the work order is pushed to the substation operation and maintenance APP, and after the operation and maintenance personnel handle it on-site, they upload photos and handling results. The system automatically archives and closes the work order, forming a complete closed loop.

[0040] Furthermore, it also includes a layout optimization subsystem, which is configured as follows: Acquire the internal geometric shape data of the GIS cavity, including the length, diameter, bending radius, and curvature characteristics of each air chamber, flange connection, and basin insulator area, and obtain the initial empirical parameters for sensor layout; Based on geometric shape data, a three-dimensional unstructured mesh generation method is used to discretize the internal space of the GIS cavity and construct a high-fidelity cavity topology model. Generate at least two candidate linear topological paths in the cavity topology model, including a spiral path that advances spirally along the cavity axis and a polygonal path that advances along the generatrix. The monitoring coverage efficiency of UHF electromagnetic signals to the cavity interior space was simulated and calculated using the finite-time integral method under spiral and polygonal arrangement paths, respectively. If the monitoring coverage efficiency of the spiral arrangement path is more than 5% higher than that of the polygonal arrangement path, the spiral arrangement path is selected as the basic topology. Extract a set of key geometric parameters from the spiral arrangement path. The set of key geometric parameters includes the arc distance between adjacent serial multi-physical quantity sensing nodes, the tangent bending angle at the node located in the curved section, and the normal distance between the node and the inner wall of the cavity. With the optimization objectives of maximizing monitoring coverage and minimizing average positioning error, the particle swarm optimization algorithm is used to perform multi-objective iterative optimization on the set of key geometric parameters to obtain a preliminary optimized set of geometric parameters. Acquire data on cavity deformation caused by thermal expansion and contraction, internal pressure fluctuations, or external mechanical vibrations during the operation of GIS equipment; Based on deformation data, the Kalman filter method is used to estimate and predict the state of the initially optimized set of geometric parameters; if the predicted deformation amplitude exceeds the preset deformation threshold, the set of geometric parameters is updated to obtain the corrected geometric parameters. Based on the corrected geometric parameters, a final position coordinate matrix is ​​generated to characterize the spatial positions of multiple serial multi-physical quantity sensing nodes in the global coordinate system. Based on the final position coordinate matrix, another electromagnetic field numerical simulation is performed to generate a signal intensity distribution map inside the cavity. Image recognition and region growing algorithms are used to extract the monitoring blind zone from the signal intensity distribution map, and the proportion of the total volume of the blind zone to the total volume of the cavity is calculated. If the proportion is less than 2%, the current layout scheme is confirmed to be effective; otherwise, the process returns to the step of using the particle swarm optimization algorithm and iterative optimization is performed again until the coverage requirement is met.

[0041] Furthermore, during the initial deployment phase, the layout optimization subsystem input GIS cavity geometry data (total length 8m, including 2 bends with R=300mm) to construct an unstructured grid model (number of cells ≈120,000), and simulated the coverage efficiency of spiral path (pitch 200mm) and polygonal path in the monitoring area. The results show that the coverage efficiency of the spiral path reaches 92.3%, while that of the broken line path is 86.1%. Therefore, the spiral path was selected as the basic topology. Key parameters were extracted: arc distance d = 150 mm, normal distance h = 20 mm, and maximum tangent angle θ = 15°. The PSO algorithm was used with the objective function of maximizing monitoring coverage and minimizing average positioning error. After 50 iterations, the optimal parameter set was obtained. Deformation data during operation was introduced (laser ranging showed an expansion of 1.8 mm at the bend). Kalman filtering predicted that the deformation would reach 2.3 mm in the next 24 hours, exceeding the preset deformation threshold of 2 mm, triggering parameter correction. After updating the coordinate matrix, electromagnetic field simulation was performed again. The blind zone volume ratio was reduced to 1.7%, which is lower than the acceptance standard of 2%, confirming the effectiveness of the current layout scheme.

[0042] Furthermore, the layout optimization subsystem is also equipped with a dynamic compensation mechanism to continuously monitor cavity deformation during system operation and trigger online updates of sensor layout parameters when the deformation exceeds a preset threshold. Moreover, during system operation, the dynamic compensation mechanism continuously monitors cavity strain gauge data; once the deformation exceeds the limit, it immediately triggers online updates of layout parameters without manual intervention.

[0043] Furthermore, the progressive positioning process executed by the 3D positioning algorithm module includes: The preliminary three-dimensional coordinates of the local discharge source are output using the Chan algorithm; The intelligent response model is invoked to determine, based on the current environmental data, whether there is any deformation of the cavity structure that could cause the electromagnetic wave propagation path to deviate. If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each series multi-physical quantity sensing node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator blocking is quantified. By using additional propagation delay to correct the original arrival time, an equivalent arrival time that reflects the actual situation is obtained; The arrival time difference measurement vector is recalculated based on the corrected equivalent arrival time, and the positioning algorithm is executed again. The iteration continues until the spatial coordinate change between two consecutive positioning results is less than the preset convergence threshold, and the accurate three-dimensional coordinates of the local discharge power source are output.

[0044] Furthermore, the intelligent response model integrated in the processing layer is trained through a deep reinforcement learning framework. Its state inputs include cavity deformation, temperature gradient, and vibration spectrum. The action outputs are the propagation path correction coefficient and time delay compensation parameter in the positioning algorithm. The reward function is defined as the negative value of the positioning error. The training objective is to minimize the three-dimensional positioning deviation of the local discharge source in a dynamic environment.

[0045] Furthermore, the gas state sensor includes a temperature sensor, a humidity sensor, and a pressure sensor for monitoring the SF6 gas medium. More specifically, the gas state sensor is a three-in-one integrated module that outputs the temperature (typically 25°C), humidity (typically 150 ppm), and pressure (typically 0.5 MPa) of the SF6 gas in real time. The data is strictly time-synchronized with UHF and acoustic signals to assist in fault diagnosis and risk assessment.

[0046] Example 2: like Figures 6 to 9 As shown, a GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor includes the following steps: S1. Based on the geometric structure information of the monitored GIS cavity, the optimal spatial layout of multiple serial multi-physical quantity sensing nodes in the GIS cavity is determined through digital modeling and swarm intelligence optimization algorithms, and the final position coordinate matrix is ​​generated to characterize the spatial position of these nodes in the global coordinate system. S2. Deploy multiple serial multi-physical quantity sensing nodes according to the final position coordinate matrix, and construct a physical sensing network consistent with the optimal spatial layout; S3. Control multiple series-connected multi-physical quantity sensing nodes to synchronously acquire ultra-high frequency electromagnetic signals, acoustic vibration signals and gas state parameters under a unified time sequence, and then aggregate the acquired data after assigning a unified time stamp to them. S4. Based on the homology verification between the UHF signal and the acoustic signal, the original signal after aggregation is processed, non-homogeneous interference signals are eliminated, and a purified multidimensional feature vector is generated. S5. Combining the GIS cavity 3D structural model with real-time environmental data, the pre-trained intelligent response model is called to dynamically compensate for signal propagation path errors. A progressive positioning algorithm based on time difference of arrival is adopted to output the accurate 3D coordinates of the local discharge source. S6. By integrating electrical features, acoustic features, and gas state parameters from multidimensional feature vectors, the fault types of partial discharge can be identified and their risk levels can be assessed. S7. The precise three-dimensional coordinates, fault type and risk level are packaged into a diagnostic report, which is then encrypted and transmitted to the application. The defect location is mapped in the GIS three-dimensional visualization model, and the operation and maintenance work order generation and closed-loop feedback mechanism are triggered according to the risk level. S8. Periodically inject standard test signals into multiple serial multi-physical quantity sensing nodes, calibrate the signal response characteristics of each node based on the feedback results, and incrementally update the intelligent response model using new data accumulated during operation.

[0047] Furthermore, the specific implementation methods for each step are as follows: S1 Specific Method: S1.1 Acquire the internal geometric data of the GIS cavity, including the length, diameter, bending radius of each air chamber, curvature characteristics of the flange connection and the basin insulator area, and obtain the initial empirical parameters for sensor layout; S1.2 uses a three-dimensional unstructured mesh generation method to discretize the internal space of the GIS cavity and construct a high-fidelity cavity topology model; S1.3 Generate at least two candidate linear topological paths in the cavity topology model, including a spiral path that advances spirally along the cavity axis and a polygonal path that advances along the generatrix. S1.4 The monitoring coverage of the UHF signal to the cavity interior space under the spiral arrangement path and the piecewise linear arrangement path is simulated and calculated respectively using the finite-integration time-domain (FIT) method. If the monitoring coverage of the spiral arrangement path is more than 5% higher than that of the piecewise linear arrangement path, the spiral arrangement path is selected as the basic topology. S1.5 Extracts the set of key geometric parameters in the spiral arrangement path, including the arc distance d between adjacent nodes (unit: mm) and the tangent bending angle at the node of the curved segment. (Unit: °), normal distance h between the node and the inner wall of the cavity (unit: mm); S1.6 aims to maximize monitoring coverage C and minimize average positioning error. To achieve the objective, the Particle Swarm Optimization (PSO) algorithm is used to perform multi-objective iterative optimization on the set of key geometric parameters. The objective function is:

[0048] in: The monitoring coverage rate (unit: %) is defined as the ratio of the volume of the effectively sensed area to the total volume of the cavity. The average positioning error (unit: mm) is defined as the average Euclidean distance between multiple simulated positioning results and the actual source location. , , The maximum permissible error (e.g., 100 mm) is preset for normalization to eliminate the influence of dimensions. For normalized weight coefficients, satisfying , ,and ; A preliminary optimized set of geometric parameters is obtained; S1.7 Acquire cavity deformation data caused by thermal expansion and contraction, internal pressure fluctuations, or external mechanical vibrations during the operation of GIS equipment; S1.8 Based on deformation data, the Kalman filter method is used to estimate and predict the state of the initially optimized set of geometric parameters; if the predicted deformation amplitude exceeds the preset deformation threshold (e.g., 2mm), the set of geometric parameters is updated to obtain the corrected geometric parameters; S1.9 Based on the corrected geometric parameters, generate the final position coordinate matrix to characterize the spatial positions of multiple cascaded multi-physical quantity sensing nodes in the global coordinate system. ,in: This represents the total number of sensor nodes. For the first The three-dimensional coordinates (in meters) of each node in the global coordinate system, i=1, 2... N .

[0049] S1.10 Performs another electromagnetic field numerical simulation based on the final position coordinate matrix (using CSTStudioSuite) to generate a signal intensity distribution map inside the cavity, as shown below. Figure 7 As shown; S1.11 uses image recognition and region growing algorithms to extract the monitoring blind zone region from the signal strength distribution map and calculates the proportion of the total volume of the blind zone to the total volume of the cavity. If the proportion is less than 2%, the current layout scheme is considered effective; otherwise, it returns to S1.6 and iterates again until the coverage requirement is met. A statistical graph of monitoring coverage and iteration count is shown below. Figure 8 As shown.

[0050] For example, in a typical cavity of 8m in length in a 220kV GIS substation, including two elbows with a radius of 300mm and pot-type insulators, an unstructured mesh of approximately 120,000 elements was constructed. Simulations showed that the coverage efficiency of the spiral path (pitch 200mm) was 92.3%, and that of the broken-line path was 86.1%. After 50 generations of PSO optimization, the optimal parameters were obtained. , , Laser ranging detected an actual expansion of 1.8mm at the bend. Kalman filtering predicted that the deformation would reach 2.3mm within 24 hours, which is greater than the 2mm threshold, triggering a correction. After resimulation, the blind zone ratio was 1.7% (less than 2%), confirming the layout's effectiveness.

[0051] The comparison results of the two paths are shown in Table 1: Table 1 Comparison of Results

[0052] S3 specific method: S3.1 sends a unified synchronization timing command to multiple serial multi-physical quantity sensing nodes; S3.2 Each node synchronously starts its built-in UHF sensor to collect electromagnetic transient signals in the GIS cavity at a sampling rate of no less than 1GS / s, and adds a uniform time stamp to the collected UHF raw waveform data; S3.3 Synchronously starts the acoustic wave sensors of each node, collects acoustic wave vibration signals within the same time window as the UHF signal, and aligns the original acoustic wave waveform data with the corresponding UHF signal in time. S3.4 synchronously reads the temperature, humidity and pressure parameters output by the built-in gas state sensors of each node to form equipment status data that is synchronized with the waveform data in time; S3.5 Each node will upload UHF raw waveform data, acoustic raw waveform data and gas state parameters with uniform time stamps to the data aggregation node through its respective sensor network; The S3.6 data aggregation node packages, verifies, and standardizes the format of all uploaded data before sending it to the subsequent processing module.

[0053] Hardware configuration: Each node integrates a UHF sensor (center frequency 900MHz, bandwidth 300MHz–1.5GHz), a piezoelectric acoustic wave sensor (frequency response 20kHz–200kHz), and an SF6 three-in-one gas sensor (accuracy: temperature ±0.5℃, humidity ±3%RH, pressure ±0.01MPa); the sampling rate is set to 1.25GS / s; gas parameters are output once every 100ms; all nodes are connected in series via shielded coaxial cable and connected to the IEEE1588PTP time synchronization system, with a synchronization error of <±1ns.

[0054] S4 specific method: S4.1 parses the aggregated mixed data packets and separates the ultra-high frequency waveform data, acoustic waveform data and gas state parameters corresponding to each node; S4.2 applies a bandpass filter with a center frequency of 300MHz–1.5GHz to the UHF waveform data and uses wavelet threshold noise reduction to process the acoustic waveform data. S4.3 Extracting the precise start time of the denoised UHF signal pulse (Unit: s), and searches for the start time of the corresponding acoustic signal pulse within its preset time window (±50μs). (Unit: s); S4.4 Calculate the time difference (Unit: s), if And the cross-correlation coefficients of the two signals in the time-frequency domain:

[0055] If the value is greater than 0.7, then the group of signals is determined to be a true partial discharge signal from the same source; where: The number of sampling points involved in the calculation; For the first The amplitude of the ultra-high frequency signal at each sampling point (unit: V); For the first The amplitude of the acoustic signal at each sampling point (unit: V); for Mean (unit: V); for Mean (unit: V); S4.5 retains signals of the same origin and eliminates UHF signals that are identified as external electromagnetic interference; S4.6 combines the amplitude, phase, frequency band energy distribution, acoustic wave dominant frequency and energy integral of the retained effective signal, as well as the synchronously acquired gas temperature, humidity and pressure parameters, into a 9-dimensional multidimensional feature vector, which is used as the effective data output after purification.

[0056] In this embodiment, wavelet denoising employs a db4 wavelet basis, 5-level decomposition, and soft thresholding. ,in The noise standard deviation is expressed in V. This is the signal length.

[0057] S5 specific methods: S5.1 Select at least four UHF signals from the purified valid data whose amplitudes exceed the trigger threshold (-65dBm), and extract their precise arrival times when they receive the same partial discharge pulse. (Unit: s), where ,and ; S5.2 uses one of the nodes (index is...) Using the reference node as an example, calculate the signal arrival time difference (TDOA) of the other nodes relative to the reference node to form the TDOA measurement vector. (Unit: s), where ; S5.3 Under the assumption of a homogeneous dielectric (SF6 relative permittivity) electromagnetic wave propagation speed ,in Using the speed of light in vacuum, the Chan algorithm is employed to solve for the TDOA measurement vector, yielding the preliminary three-dimensional coordinates of the partial discharge source. (Unit: m); The core formula of the Chan algorithm is:

[0058] in: Let be a geometric matrix, and let its first... Behavior ; Measure the covariance matrix for TDOA (unit: s²); Let be a nonlinear term vector, and its first term be... The elements are (Unit: m²) For the first Known locations of the sensor nodes (in meters); S5.4 associates the initial three-dimensional coordinates with the GIS cavity three-dimensional structural model and inputs the temperature field and vibration data monitored in real time by the sensor network; S5.5 invokes a pre-trained intelligent response model to determine, based on current environmental data, whether there is any cavity structure deformation that could cause the electromagnetic wave propagation path to deviate. S5.6 If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction, and insulator obstruction is quantified. (Unit: s); S5.7 corrects the original arrival time by adding a propagation delay to obtain an equivalent arrival time that better reflects actual operating conditions. (Unit: s); S5.8 Recalculate the TDOA vector based on the corrected equivalent time of arrival, and execute the positioning algorithm again, iterating until the spatial coordinate change between two consecutive positioning results is reached. Output the final precise three-dimensional coordinates of the partial discharge source. (Unit: m).

[0059] In this embodiment, the intelligent response model is trained using the Proximal Policy Optimization (PPO) algorithm. The state inputs are cavity deformation, temperature gradient, and vibration dominant frequency. The action outputs are path correction coefficients and time delay compensation parameters. The reward function is the negative value of the positioning error.

[0060] S6 specific method: S6.1 Extract the PRPD spectrum, center frequency, bandwidth characteristics of the UHF signal corresponding to the precise three-dimensional coordinates from the purified effective data, as well as the energy integral and main frequency characteristics of the acoustic signal; S6.2 integrates electrical and acoustic characteristics with synchronously acquired gas state parameters to form a comprehensive feature vector; S6.3 inputs the comprehensive feature vector into the pre-trained XGBoost fault diagnosis classification model and outputs the fault type with the highest probability, including tip discharge, floating potential body discharge, free metal particle discharge or internal air gap discharge of insulation. S6.4 Based on the diagnosed fault type, discharge signal amplitude, pulse repetition rate, and deterioration trend of gas state parameters, combined with the preset risk assessment rules, calculate the current defect risk level and classify it as "normal", "caution", "abnormal", or "critical".

[0061] The eigenvector definition (9 dimensions) includes: UHF center frequency (MHz); UHF 3dB bandwidth (MHz); pulse repetition rate (p / s); sound energy integral (V²·s); sound wave dominant frequency (kHz); gas temperature (°C); gas humidity (%RH); and gas pressure (MPa).

[0062] S7 specific method: S7.1 encapsulates precise three-dimensional coordinates, fault type, and risk level into a structured diagnostic report; S7.2 uses AES-256 encryption for diagnostic reports; S7.3 transmits encrypted reports to the application via fiber optic or 5G private networks; S7.4 renders defect identifiers based on precise 3D coordinates in the GIS 3D visualization model (based on the Unity3D engine) loaded on the application side, and displays the fault type, risk level and gas state information in association. S7.5 When the risk level is "abnormal" or "critical", an electronic maintenance work order containing location information, diagnostic conclusions and handling suggestions will be automatically generated and pushed to the authorized operation and maintenance terminal. S7.6 receives on-site processing feedback from the operation and maintenance terminal, updates the work order status, and archives the entire process data, forming a closed-loop operation and maintenance record of "monitoring-diagnosis-decision-execution-feedback".

[0063] Example of a work order in this embodiment: "Defect coordinates (2.31m, 0.45m, 1.12m), fault type: free metal particle discharge (probability 89%), risk level: critical, recommendation: arrange a power outage inspection within 72 hours, focusing on the N5–N7 section."

[0064] S8 specific method: S8.1 Generates standard electrical pulse signals and standard acoustic pulse signals with known amplitude and waveform according to a preset period (24 hours); S8.2 synchronously injects the standard signal into the signal input terminal of each node; S8.3 Each node collects and feeds back the measured signal amplitude, waveform characteristics, and response time; S8.4 compares the feedback signal with the standard signal and calculates the channel attenuation coefficient for each node:

[0065] in: The theoretical amplitude of the standard signal (unit: V or dBm); For the first Measured amplitude at each node (unit: V or dBm); S8.5 dynamically updates the compensation coefficients of each node in subsequent signal processing based on the attenuation coefficient.

[0066] Complete online calibration; in, This is the signal compensation coefficient, used to correct node acquisition deviations; Channel attenuation coefficient, used to quantify the degree of signal attenuation.

[0067] S8.6 The optimal compensation parameters, corresponding environmental data, and positioning error data obtained during this calibration process will be used as new training samples; S8.7 uses newly added training samples to incrementally learn the intelligent response model, and updates the weights using online gradient descent:

[0068] in: This is the model parameter vector; For the learning rate, a value of 0.001 is recommended. The loss function; This is to add a new training sample set.

[0069] A standard UHF pulse is a Gaussian pulse. ,in (FWHM=1ns), amplitude -40dBm; standard acoustic pulse is a 50kHz sine burst, lasting 100μs; the calibration process is automatically triggered without manual intervention.

[0070] This method achieves high-precision three-dimensional positioning, multi-source fusion diagnosis, closed-loop operation and maintenance management, and adaptive calibration of partial discharge in GIS equipment, possessing strong robustness and engineering feasibility. For example... Figure 9 As shown, the present invention demonstrates good performance in fault prediction.

[0071] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A GIS three-dimensional positioning and monitoring system using a series-connected ultra-high frequency sensor, characterized in that, include: The sensing layer is used to synchronously collect ultra-high frequency electromagnetic signals, acoustic vibration signals and gas state parameters in the GIS cavity through multiple series-connected multi-physical quantity sensing nodes; the multiple series-connected multi-physical quantity sensing nodes are arranged in series along the axis of the GIS cavity. The processing layer, connected to the sensing layer, is used to verify the homology between ultra-high frequency electromagnetic signals and acoustic vibration signals, eliminate non-homogeneous interference signals, and generate purified multi-dimensional feature vectors. Combining the GIS cavity 3D structural model with real-time environmental data, it calls the intelligent response model to dynamically compensate for signal propagation path errors and uses a progressive positioning algorithm based on time difference of arrival to output the precise 3D coordinates of the partial discharge source. It integrates electrical features, acoustic features, and gas state parameters from the multi-dimensional feature vectors to identify the fault type of partial discharge and assess its risk level. Standard test signals are periodically injected into multiple serial multi-physical quantity sensing nodes, the signal response characteristics of each node are calibrated based on the feedback results, and the intelligent response model is incrementally updated using new data accumulated during operation. The transport layer, connecting the processing layer, is used to encrypt and transmit diagnostic reports containing precise three-dimensional coordinates, fault types, and risk levels to the application. The application layer connects to the transport layer and is used to map defect locations in the GIS 3D visualization model, and trigger maintenance work order generation and closed-loop feedback mechanisms based on risk levels.

2. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 1, characterized in that, The sensing layer includes multiple serial multi-physical quantity sensing nodes, each node integrating an ultra-high frequency sensor, an acoustic vibration sensor, and a gas state sensor; the spatial position of the multiple serial multi-physical quantity sensing nodes is determined by the final position coordinate matrix generated based on the geometric structure information of the monitored GIS cavity through digital modeling and swarm intelligence optimization algorithms.

3. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 1, characterized in that, The processing layer includes: The anti-interference processing module is used to verify the homology between ultra-high frequency electromagnetic signals and acoustic vibration signals, eliminate non-homogeneous interference signals, and generate purified multi-dimensional feature vectors. The three-dimensional positioning algorithm module is used to combine the GIS cavity three-dimensional structure model with real-time environmental data, call the intelligent response model to dynamically compensate for signal propagation path errors, and use a progressive positioning algorithm based on time difference of arrival to output the accurate three-dimensional coordinates of the local discharge source. The fault diagnosis module is used to fuse electrical features, acoustic features, and gas state parameters from a multi-dimensional feature vector to identify the fault type of partial discharge and assess the risk level. The self-calibration module is used to periodically inject standard test signals into each serial multi-physical quantity sensing node, calibrate the signal response characteristics of the node based on the feedback results, and incrementally update the intelligent response model using the operating data.

4. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 1, characterized in that, The transport layer is configured to encrypt the diagnostic report before transmitting it to the application, and the transmission strategy is dynamically adjusted according to the risk level.

5. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 1, characterized in that, The application layer includes: The GIS 3D visualization model module is used to load high-precision 3D models of GIS equipment and render defect identifiers based on the received precise 3D coordinates, and display fault types, risk levels and gas state information in association. The operation and maintenance linkage unit is used to automatically generate electronic maintenance work orders when the risk level is "abnormal" or "critical", push them to authorized operation and maintenance terminals, and receive on-site feedback to complete closed-loop operation and maintenance.

6. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 1, characterized in that, It also includes a layout optimization subsystem, which is used to acquire GIS cavity geometry data, construct cavity topology model, generate candidate sensing paths, simulate and calculate monitoring coverage, use swarm intelligence optimization algorithm to determine the optimal spatial layout, and output the final position coordinate matrix to characterize the spatial position of multiple serial multi-physical quantity sensing nodes.

7. A GIS three-dimensional positioning and monitoring system based on a series UHF sensor according to claim 6, characterized in that, The layout optimization subsystem is also equipped with a dynamic compensation mechanism to continuously monitor cavity deformation during system operation and trigger online updates of sensor layout parameters when the deformation exceeds a preset threshold.

8. A GIS three-dimensional positioning and monitoring system with series ultra-high frequency sensing according to claim 3, characterized in that, The progressive positioning process executed by the 3D positioning algorithm module includes: The preliminary three-dimensional coordinates of the local discharge source are output using the Chan algorithm; The intelligent response model is invoked to determine, based on the current environmental data, whether there is any deformation of the cavity structure that could cause the electromagnetic wave propagation path to deviate. If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each series multi-physical quantity sensing node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator blocking is quantified. By using additional propagation delay to correct the original arrival time, an equivalent arrival time that better reflects the actual working conditions is obtained; The arrival time difference measurement vector is recalculated based on the corrected equivalent arrival time, and the positioning algorithm is executed again. The iteration continues until the spatial coordinate change between two consecutive positioning results is less than the preset convergence threshold, and the accurate three-dimensional coordinates of the local discharge power source are output.

9. A GIS three-dimensional positioning and monitoring system based on a series ultra-high frequency sensor according to claim 1, characterized in that, The intelligent response model integrated in the processing layer is trained through a deep reinforcement learning framework. Its state inputs include cavity deformation, temperature gradient and vibration spectrum. The action outputs are the propagation path correction coefficient and time delay compensation parameter in the positioning algorithm. The reward function is defined as the negative value of the positioning error. The training objective is to minimize the three-dimensional positioning deviation of the local discharge source in a dynamic environment.

10. A GIS three-dimensional positioning and monitoring system based on a series ultra-high frequency sensor according to claim 2, characterized in that, Gas state sensors include temperature sensors, humidity sensors, and pressure sensors for monitoring SF6 gaseous media.

11. A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor, characterized in that, A GIS three-dimensional positioning and monitoring system using a series UHF sensor as described in any one of claims 1-10 includes the following steps: Based on the geometric structure information of the monitored GIS cavity, the optimal spatial layout of multiple serial multi-physical quantity sensing nodes in the GIS cavity is determined through digital modeling and swarm intelligence optimization algorithms, and the final position coordinate matrix is ​​generated to characterize the spatial position of these nodes in the global coordinate system. Based on the final location coordinate matrix, deploy multiple serial multi-physical quantity sensing nodes and construct a physical sensing network consistent with the optimal spatial layout; Multiple serial multi-physical quantity sensing nodes are controlled to synchronously acquire ultra-high frequency electromagnetic signals, acoustic vibration signals and gas state parameters under a unified time sequence, and the acquired data are aggregated after being assigned a unified time stamp. The original signal after aggregation is verified based on the homology between the UHF signal and the acoustic signal. Non-homogeneous interference signals are eliminated, and a purified multidimensional feature vector is generated. By combining the GIS cavity 3D structural model with real-time environmental data, a pre-trained intelligent response model is invoked to dynamically compensate for signal propagation path errors. A progressive positioning algorithm based on time difference of arrival is adopted to output the precise 3D coordinates of the local discharge source. By integrating electrical features, acoustic features, and gas state parameters from multidimensional feature vectors, the fault types of partial discharge can be identified and their risk levels can be assessed. The precise 3D coordinates, fault type, and risk level are packaged into a diagnostic report, which is then encrypted and transmitted to the application. The defect location is mapped in the GIS 3D visualization model, and the operation and maintenance work order generation and closed-loop feedback mechanism are triggered according to the risk level. Standard test signals are periodically injected into multiple serially connected multi-physical quantity sensing nodes. The signal response characteristics of each node are calibrated based on the feedback results, and the intelligent response model is incrementally updated using new data accumulated during operation.

12. A GIS three-dimensional positioning and monitoring method based on a series UHF sensor according to claim 11, characterized in that, Based on the geometric structure information of the monitored GIS cavity, the optimal spatial layout of multiple serially connected multi-physical quantity sensing nodes within the GIS cavity is determined through digital modeling and swarm intelligence optimization algorithms. A final position coordinate matrix is ​​then generated to characterize the spatial positions of these nodes in the global coordinate system. The specific method is as follows: Acquire the internal geometry data of the GIS cavity, including the length, diameter, bending radius of each air chamber, curvature characteristics of flange connections and basin insulator areas, and obtain the initial empirical parameters for sensor layout; A three-dimensional unstructured mesh generation method is used to discretize the internal space of the GIS cavity and construct a high-fidelity cavity topology model; Generate at least two candidate linear topological paths in the cavity topology model, including a spiral path that advances spirally along the cavity axis and a polygonal path that advances along the generatrix. The monitoring coverage of UHF signals to the cavity interior space under spiral and piecewise linear arrangement paths was simulated and calculated using the finite-time integral method. If the monitoring coverage of the spiral arrangement path is more than 5% higher than that of the piecewise linear arrangement path, the spiral arrangement path is selected as the basic topology. Extract the set of key geometric parameters in the spiral arrangement path, including the arc distance between adjacent nodes, the tangent bending angle at the node of the curved segment, and the normal distance between the node and the inner wall of the cavity; With the goal of maximizing monitoring coverage and minimizing average positioning error, the particle swarm optimization algorithm is used to perform multi-objective iterative optimization on the set of key geometric parameters to obtain a preliminary optimized set of geometric parameters. Acquire data on cavity deformation caused by thermal expansion and contraction, internal pressure fluctuations, or external mechanical vibrations during the operation of GIS equipment; Based on deformation data, the Kalman filter method is used to estimate and predict the state of the preliminarily optimized set of geometric parameters; If the predicted deformation exceeds the preset deformation threshold, the geometric parameter set is updated to obtain the corrected geometric parameters; Based on the corrected geometric parameters, a final position coordinate matrix is ​​generated to characterize the spatial positions of multiple serial multi-physical quantity sensing nodes in the global coordinate system. Based on the final position coordinate matrix, another electromagnetic field numerical simulation is performed to generate a signal intensity distribution map inside the cavity. Image recognition and region growing algorithms are used to extract the monitoring blind zone from the signal intensity distribution map, and the proportion of the total volume of the blind zone to the total volume of the cavity is calculated. If the proportion is less than 2%, the current layout scheme is confirmed to be effective; otherwise, return to the particle swarm optimization algorithm step and iterate again until the coverage requirement is met.

13. A GIS three-dimensional positioning and monitoring method based on a series UHF sensor according to claim 11, characterized in that, Multiple serially connected multi-physical quantity sensing nodes are controlled to synchronously acquire ultra-high frequency electromagnetic signals, acoustic vibration signals, and gas state parameters under a unified time sequence. The acquired data are then aggregated after being assigned a unified time stamp. The specific method is as follows: Send unified synchronization timing commands to multiple serial multi-physical quantity sensing nodes; Each node synchronously starts its built-in UHF sensor to collect electromagnetic transient signals inside the GIS cavity at a sampling rate of no less than 1GS / s, and adds a uniform time stamp to the collected UHF raw waveform data; The acoustic wave sensors at each node are activated synchronously to collect acoustic wave vibration signals within the same time window as the UHF signal, and the original acoustic wave waveform data is time-aligned with the corresponding UHF signal. The temperature, humidity and pressure parameters output by the built-in gas state sensors of each node are read synchronously to form equipment status data that is synchronized with the waveform data in time. Each node will upload UHF raw waveform data, acoustic raw waveform data and gas state parameters with uniform time stamps to the data aggregation node through its respective sensor network; The data aggregation node packages, verifies, and standardizes the format of all uploaded data before sending it to the subsequent processing module.

14. A GIS three-dimensional positioning and monitoring method based on a series UHF sensor according to claim 11, characterized in that, The converged original signal is filtered and denoised. Based on the homology verification between the UHF signal and the acoustic signal, non-homogeneous interference signals are eliminated to generate a purified multidimensional feature vector. The specific method is as follows: The aggregated mixed data packets are parsed to separate the ultra-high frequency waveform data, acoustic waveform data and gas state parameters corresponding to each node; A bandpass filter with a center frequency of 300MHz–1.5GHz is applied to the UHF waveform data, and wavelet threshold noise reduction is applied to the acoustic waveform data. Extracting the precise start time of the denoised UHF signal pulse It searches for the start time of the corresponding acoustic signal pulse within its preset time window. ; Calculate the time difference If the time difference is less than the set time tolerance threshold, and the cross-correlation coefficient of the two signals in the time-frequency domain is greater than the preset correlation threshold, then the signals are determined to be real partial discharge signals from the same source. Retain signals of the same origin and eliminate UHF signals that are identified as external electromagnetic interference; The amplitude, phase, frequency band energy distribution, acoustic wave dominant frequency and energy integral of the retained effective signal, along with the synchronously acquired gas temperature, humidity and pressure parameters, are combined into a multi-dimensional feature vector, which serves as the effective data output after purification.

15. A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor according to claim 1, characterized in that, By combining the GIS cavity 3D structural model with real-time environmental data, a pre-trained intelligent response model is invoked to dynamically compensate for signal propagation path errors. A progressive positioning algorithm based on time difference of arrival is adopted to output the precise 3D coordinates of the local discharge source. The specific method is as follows: From the purified valid data, select at least four nodes whose amplitude exceeds the trigger threshold for UHF signals, and extract their precise arrival time when they receive the same partial discharge pulse. Using one node as a reference, calculate the signal arrival time difference of the other nodes relative to the reference node to form an arrival time difference measurement vector; Under the assumption of a homogeneous medium, the Chan algorithm is used to solve for the time difference measurement vector to obtain the preliminary three-dimensional coordinates of the local discharge source; The initial three-dimensional coordinates are associated with the GIS cavity three-dimensional structural model, and the temperature field and vibration data monitored in real time by the sensor network are input. The pre-trained intelligent response model is invoked to determine, based on the current environmental data, whether there is any deformation of the cavity structure that would cause the electromagnetic wave propagation path to deviate. If significant deformation exists, the electromagnetic wave propagation path from the initial coordinates to each node is simulated and calculated based on the updated cavity geometry model, and the additional propagation delay caused by cavity wall reflection, refraction and insulator blocking is quantified. By using additional propagation delay to correct the original arrival time, an equivalent arrival time that better reflects the actual working conditions is obtained; The time difference vector is recalculated based on the corrected equivalent time of arrival, and the positioning algorithm is executed again. The iteration continues until the change in spatial coordinates between two consecutive positioning results is less than the preset convergence threshold, and the final accurate three-dimensional coordinates of the local discharge source are output.

16. A GIS three-dimensional positioning and monitoring method using a series UHF sensor according to claim 1, characterized in that, By integrating electrical features, acoustic features, and gas state parameters from a multi-dimensional feature vector, the fault type of partial discharge is identified and its risk level is assessed. The specific method is as follows: Extract the PRPD spectrum, center frequency, and bandwidth characteristics of the UHF signal corresponding to the precise three-dimensional coordinates from the purified effective data, as well as the energy integral and main frequency characteristics of the acoustic signal. Electrical and acoustic characteristics are fused with synchronously acquired gas state parameters to form a comprehensive feature vector; The comprehensive feature vector is input into the pre-trained fault diagnosis classification model, and the fault type with the highest probability is output. The fault types include tip discharge, floating potential body discharge, free metal particle discharge or internal air gap discharge of insulation. Based on the diagnosed fault type, discharge signal amplitude, pulse repetition rate, and deterioration trend of gas state parameters, combined with the preset risk assessment rules, the current defect risk level is calculated and classified as "normal", "caution", "abnormal" or "critical".

17. A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor according to claim 1, characterized in that, The precise 3D coordinates, fault type, and risk level are encapsulated into a diagnostic report, which is then encrypted and transmitted to the application. The defect location is mapped in the GIS 3D visualization model, and a maintenance work order generation and closed-loop feedback mechanism are triggered based on the risk level. The specific method is as follows: The precise three-dimensional coordinates, fault type, and risk level are packaged into a structured diagnostic report; The diagnostic report is encrypted; The encrypted diagnostic report is transmitted to the application. In the 3D visualization model of the GIS equipment loaded on the application side, defect identifiers are rendered according to precise 3D coordinates, and fault type, risk level and gas state information are displayed in association. When the risk level is "abnormal" or "critical", an electronic maintenance work order containing location information, diagnostic conclusions and handling suggestions is automatically generated and pushed to the authorized operation and maintenance terminal. Receive on-site processing feedback from the operation and maintenance terminal, update the work order status and archive the entire process data to form a closed-loop operation and maintenance record of "monitoring-diagnosis-decision-execution-feedback".

18. A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor according to claim 1, characterized in that, Standard test signals are periodically injected into multiple serially connected multi-physical quantity sensing nodes. The signal response characteristics of each node are calibrated based on the feedback results. The intelligent response model is incrementally updated using new data accumulated during operation. The specific method is as follows: Generate standard electrical pulse signals and standard acoustic pulse signals with known amplitudes and waveforms according to a preset period; The standard signal is synchronously injected into the signal input terminal of each node; Each node collects and reports back the measured signal amplitude, waveform characteristics, and response time; The feedback signal is compared with the standard signal to calculate the channel attenuation coefficient and sensitivity deviation factor for each node. Based on the deviation factor, the compensation coefficients of each node in subsequent signal processing are dynamically updated to complete online calibration; The optimal compensation parameters, corresponding environmental data, and positioning error data obtained during this calibration process will be used as new training samples. By using newly added training samples, the intelligent response model is incrementally learned, its internal parameters are updated, and its adaptability to dynamic operating environments is improved.

19. A GIS three-dimensional positioning and monitoring method using a series-connected ultra-high frequency sensor according to claim 2, characterized in that, The intelligent response model is trained using a deep reinforcement learning framework. Its state inputs include cavity deformation, temperature gradient, and vibration spectrum. Its action outputs are the propagation path correction coefficient and time delay compensation parameter in the positioning algorithm. The reward function is defined as the negative value of the positioning error. The training objective is to minimize the three-dimensional positioning deviation of the local discharge source in a dynamic environment.

20. A storage medium, characterized in that, It stores an executable program, which, when executed, is used to implement a GIS three-dimensional positioning and monitoring method based on a series ultra-high frequency sensor as described in any one of claims 11-19.

Images