GIS disconnecting switch multi-state intelligent sensing system

Abstract

The invention discloses a GIS disconnecting switch multi-state intelligent sensing system, and relates to the field of GIS disconnecting switch state monitoring. A self-calibration multi-mode sensor is deployed for multi-source data acquisition, and novel sensors including terahertz imaging and the like are included; in data preprocessing, deep learning noise reduction is applied, fuzzy entropy is used for dynamic weighted fusion, and abnormal values are processed by an improved algorithm; feature extraction is combined with a plurality of frontier algorithms to process vibration signals, and CRNN is used to analyze acoustic signals; the QPSO evidence theory is adopted for state fusion perception, the node relation is learned by means of GNN, and the weight is adjusted according to the information gain rate; state assessment and early warning are based on transfer learning, GAN and LSTM-attention mechanisms, and early warning priorities are ranked by FAHP. According to the invention, multi-mode accurate acquisition, intelligent data processing, deep feature mining, innovative fusion perception, accurate evaluation and early warning and efficient fault diagnosis and positioning are realized, the state of the GIS isolation switch can be comprehensively and accurately perceived, the system is self-learned and optimized, the equipment safety is guaranteed, the risk of a power system is reduced, and the power operation and maintenance benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of GIS isolating switch state monitoring, and in particular to a GIS isolating switch multi-state intelligent sensing system. Background Art

[0002] In power systems, GIS (Gas Insulated Metal-Enclosed Switchgear) disconnectors are key equipment for ensuring safe and stable grid operation. They are primarily used to isolate power sources, perform switching operations, and open and close low-current circuits. Their operational status directly impacts the reliability and economic efficiency of power systems. However, current state monitoring and awareness of GIS disconnectors face numerous challenges.

[0003] Existing monitoring methods mostly rely on single sensors, which can only capture a specific aspect of the equipment's status. For example, a vibration sensor can only monitor vibration, and a temperature sensor can only reflect contact heating. This single-source monitoring approach has significant limitations and cannot fully and accurately reflect the overall operating status of the GIS disconnector. Because GIS disconnector failures are often the result of multiple factors, data from a single sensor cannot capture this complex fault information, resulting in potential failures not being discovered in a timely manner, potentially leading to serious power accidents.

[0004] Traditional methods for data processing and analysis are relatively simple and crude. They typically use threshold-based assessments to assess equipment status. This approach lacks early warning capabilities for faults, often only sounding alarms after a significant failure has occurred, potentially causing significant damage to the power system. Furthermore, traditional methods lack the ability to deeply mine and analyze data, unable to extract valuable information from massive amounts of monitoring data, or effectively predict the future operating status of equipment, making them difficult to meet the demands of modern power systems for intelligent equipment management.

[0005] Furthermore, with the continuous development and upgrading of power systems, the structure and operating environment of GIS disconnectors are becoming increasingly complex, and the requirements for their status monitoring and perception are becoming increasingly stringent. Existing monitoring and perception methods have poor adaptability to complex operating conditions and environmental changes, and cannot promptly adjust monitoring strategies and analysis methods, resulting in reduced accuracy and reliability of monitoring results. Therefore, developing a method that can comprehensively, accurately, and intelligently perceive the multiple states of GIS disconnectors is of great practical significance. Summary of the Invention

[0006] The present invention proposes a multi-state intelligent sensing system for GIS disconnectors to solve the problems mentioned in the above-mentioned prior art.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: a GIS disconnector multi-state intelligent sensing system, comprising:

[0008] Multi-source data acquisition module: Multi-modal sensors are deployed at key locations of GIS disconnectors. Vibration sensors adaptively adjust sensitivity and sampling frequency, temperature sensors sense temperature changes, current sensors measure current contactlessly, acoustic sensors collect sound signals to locate sound sources, stress sensors monitor stress field changes, and a new terahertz imaging microstructure sensor is added to obtain microstructure image information of key internal components.

[0009] Data preprocessing module: This module uses a multi-scale residual noise reduction network to denoise vibration signals. It captures and removes noise features by constructing a convolution kernel. It also uses an adaptive weighted fusion algorithm to fuse temperature, current, and stress data. It also introduces fuzzy entropy theory, dynamically assigns weights based on data uncertainty through a formula, and uses a LOF-DBSCAN hybrid algorithm based on density and outlier detection to identify and process outliers.

[0010] Feature Extraction Module: This module uses EMD combined with variational mode decomposition (VMD) to extract vibration features. It also introduces an inverse heat conduction problem model, calculates the temperature gradient rate of change through finite element iteration, and uses a recurrent convolutional neural network (CRNN) to extract sound temporal and spatial features. It also extracts stress tensor features based on tensor decomposition to comprehensively describe the stress state.

[0011] State Fusion Perception Module: This module uses an evidence-based fusion method based on quantum-behavior particle swarm optimization (QPSO). The QPSO algorithm optimizes parameters, combines it with graph neural networks (GNNs) to learn node relationships, introduces information gain rate theory to adjust feature weights, and dynamically adjusts weights by calculating information gain rates.

[0012] Status assessment and early warning module: Build a status assessment model based on transfer learning and generative adversarial network (GAN), use historical data for pre-training, fine-tune in the target field using transfer learning technology, use long short-term memory network (LSTM) combined with attention mechanism to predict early warnings, and use fuzzy hierarchical analysis method (FAHP) to sort the warning priorities.

[0013] Furthermore, it also includes:

[0014] Real-time data transmission module: Adopts an intelligent transmission architecture based on software-defined networking (SDN) and network function virtualization (NFV), dynamically adjusts transmission paths and bandwidth allocation according to data traffic, priority, and network conditions. Homomorphic encryption technology is used in the transmission process to route and forward encrypted data. Blockchain-based distributed ledger technology is introduced to record data transmission logs. In the event of data transmission failure, a data retransmission mechanism is automatically triggered through blockchain smart contracts.

[0015] Furthermore, it also includes:

[0016] Fault diagnosis and location module: When the status assessment module determines that the GIS disconnector is in a faulty state, a joint fault diagnosis method based on Bayesian networks and deep learning is adopted. The deep learning model is used to extract sensor data features as Bayesian network input nodes. Faults are diagnosed by learning the conditional probability distribution between nodes. The Bayesian network structure is optimized by combining the particle swarm optimization algorithm (PSO). Using IoT and augmented reality (AR) positioning technology, IoT tags and AR devices are deployed on the GIS disconnector to obtain the location information of the faulty equipment in real time, and the fault location is intuitively displayed in the form of augmented reality.

[0017] Furthermore, the feature extraction module calculates the temperature gradient change rate through finite element iteration, and the formula is: T i is the calculated temperature, is the observed temperature, R(T) is the regularization term, and λ is the regularization parameter;

[0018] The generative adversarial network (GAN) variant Cycle-GAN is introduced to enhance features. Cycle-GAN performs unsupervised conversion between different types of features to generate diverse feature data. Deep belief network (DBN) is used to mine multi-source features. DBN stacks multiple layers of restricted Boltzmann machines (RBMs) to unsupervisedly learn high-order relationships between features.

[0019] Furthermore, the state fusion perception module introduces the information gain rate theory to adjust the feature weights during the fusion process. The formula is: Where IG(X|Y) is the information gain and IV(X) is the intrinsic value of feature X;

[0020] A tensor model is constructed using the tensor neural network (TNN) fusion method. Multi-source sensor data is represented in tensor form. TNN fuses features by learning the interaction relationship between tensors. The fusion process uses the quantum genetic algorithm (QGA) to optimize TNN parameters. Combined with the concept of information-physical fusion system (CPS), it deeply integrates physical world sensor data and virtual world model data.

[0021] Furthermore, the state assessment and warning module introduces the proximal policy optimization (PPO) algorithm in deep reinforcement learning to evaluate the state. The state assessment problem is modeled as a Markov decision process. The intelligent agent optimizes the evaluation strategy according to the reward function by interacting with the environment, and uses Monte Carlo tree search (MCTS) combined with neural networks to predict the state. MCTS simulates different state transition paths and combines neural networks to evaluate future states.

[0022] Furthermore, the multi-source data acquisition module adds an elemental analysis sensor based on laser-induced breakdown spectroscopy (LIBS) to analyze the elemental composition of the internal metal components of the GIS isolation switch, monitor the corrosion and wear of the components by analyzing the changes in elemental composition, use wireless energy harvesting technology to power some sensors, and deploy electromagnetic induction energy harvesters near the operating mechanism to collect the electromagnetic energy generated during operation to provide power for the sensors.

[0023] Furthermore, the data preprocessing module uses an adaptive weighted fusion algorithm for temperature, current, and stress data, and introduces fuzzy entropy theory into the weight calculation. The formula is: Where FEntropy(x i ) is the fuzzy entropy of the i-th sensor data;

[0024] An image super-resolution reconstruction algorithm is used to process terahertz microstructure images. The low-resolution image is reconstructed into a high-resolution image by combining the generative adversarial network (GAN) with the convolutional neural network (CNN). An adaptive Kalman filter algorithm is used to process the displacement sensor data, and the Kalman filter parameters are adjusted in real time according to the sensor measurement noise and changes in the system dynamic characteristics.

[0025] Furthermore, the feature extraction module extracts current data features based on wavelet packet transform and singular value decomposition (SVD), decomposes the current signal into different frequency bands using wavelet packet transform, decomposes the singular values of each frequency band signal, extracts the singular values as features, and uses deep autoencoder (DAE) to reduce the dimension of multi-source features. By learning data reconstruction errors, the key features of the data are automatically extracted.

[0026] Furthermore, it also includes:

[0027] Self-learning and optimization module: Utilizes online learning algorithms to update the parameters of the status assessment model, fault diagnosis model, and feature extraction model in real time based on newly collected data. A parameter optimization method based on a simulated annealing algorithm is used to optimize key system parameters. The optimization process is combined with IoT big data analysis technology to explore the patterns of equipment operation data.

[0028] Compared with the existing technology, the beneficial effects of the present invention are:

[0029] In terms of data acquisition, the use of multimodal, self-calibrating sensors enables comprehensive and accurate acquisition of multifaceted status information on GIS disconnectors. The addition of terahertz imaging microstructure sensors and laser-induced breakdown spectroscopy elemental analysis sensors provides more dimensional data for status perception, significantly improving the ability to perceive equipment status.

[0030] During the data preprocessing stage, the use of deep learning and improved algorithms can effectively remove noise, process outliers, and dynamically allocate weights for fusion based on data uncertainty, thereby improving data quality and reliability and providing a solid foundation for subsequent analysis.

[0031] The feature extraction module combines multiple advanced algorithms to deeply mine the features of different data types and accurately reflect the operating status of the device. Generative adversarial networks enhance feature diversity, while deep belief networks mine high-order relationships, making feature extraction more comprehensive and in-depth.

[0032] State fusion perception utilizes a variety of innovative methods, taking into account the complex relationships between multi-source data. By optimizing algorithms and adjusting feature weights, the fusion results more accurately represent device states. Combined with the concept of cyber-physical fusion systems, this further enhances the comprehensiveness and accuracy of state perception.

[0033] The status assessment and early warning module uses technologies such as transfer learning, generative adversarial networks, and deep reinforcement learning to accurately assess equipment status, predict faults in advance, and reasonably prioritize early warnings, helping operation and maintenance personnel to handle high-risk faults in a timely manner.

[0034] The fault diagnosis and positioning module combines Bayesian networks, deep learning, and augmented reality technologies to improve the accuracy of fault diagnosis and the speed of positioning, assisting operation and maintenance personnel in quickly repairing faults.

[0035] The self-learning and optimization module of the intelligent sensing system can update model parameters in real time according to new data, optimize key system parameters, improve system performance and adaptability, ensure the safe and stable operation of GIS disconnectors, reduce the operating risks of the power system, and have significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic block diagram of a multi-state intelligent perception method for GIS disconnect switches proposed by the present invention;

[0037] Figure 2 This is a schematic diagram of fault warning time comparison;

[0038] Figure 3 This is a schematic diagram for comparing data processing efficiency;

[0039] Figure 4 A schematic diagram showing the comparison of state assessment accuracy. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0042] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below with reference to the accompanying drawings.

[0043] Reference Figures 1 to 4 : A GIS disconnector multi-state intelligent perception system, comprising:

[0044] Multi-source data acquisition module: Multimodal sensors with self-calibration capabilities are deployed at key locations on GIS disconnectors. The vibration sensor utilizes a new sensor based on microelectromechanical systems (MEMS) that integrates piezoelectric and piezoresistive technologies. It can adaptively adjust its sensitivity and automatically switch the sampling frequency between 10kHz and 100kHz depending on the disconnector's operating conditions. The temperature sensor utilizes ultra-sensitive thermocouples made from nanomaterials, which can accurately sense minute temperature changes and collect data every two minutes. The current sensor utilizes the magneto-optical effect to achieve contactless, high-precision measurement of operating current. The acoustic sensor utilizes an array design and combines beamforming technology to not only collect sound signals but also locate the source of the sound. The stress sensor, based on fiber Bragg grating (FBG) technology, can monitor changes in complex stress fields in real time, collecting data every 5ms. Furthermore, a new microstructure sensor based on terahertz imaging has been added, which penetrates the GIS casing to obtain microstructural image information of key internal components, enriching the data dimension.

[0045] Data preprocessing module: A multi-scale residual noise reduction network based on deep learning is used to denoise the vibration signal. This network captures and removes noise characteristics in different frequency ranges by constructing multiple convolution kernels of different scales. Compared with traditional noise reduction methods, it can better preserve signal details. For temperature, current, stress and other data, an improved adaptive weighted fusion algorithm is used. The weight calculation introduces fuzzy entropy theory. The formula is: Where FEntropy(x i ) is the fuzzy entropy of the i-th sensor data, reflecting the uncertainty of the data, and n is the number of sensors. This formula dynamically assigns weights based on data uncertainty, improving fusion accuracy. Using an improved hybrid LOF-DBSCAN algorithm based on density and outlier detection, outliers can be identified and processed more accurately, distinguishing normal data clusters from outliers.

[0046] Feature extraction module: For vibration signals, empirical mode decomposition (EMD) combined with variational mode decomposition (VMD) is used to decompose the signal into multiple intrinsic mode functions (IMFs) through EMD. VMD is used to further subdivide the IMFs and accurately extract vibration features of different scales. The inverse heat conduction problem model is introduced into the temperature data, and the temperature gradient change rate is calculated through finite element iteration. The formula is: Where T i is the calculated temperature, is the observed temperature, R(T) is the regularization term, and λ is the regularization parameter to more accurately reflect temperature trends. From the acoustic signal, a recurrent convolutional neural network (CRNN) in deep learning is used to extract the temporal and spatial characteristics of the sound, accurately analyzing the frequency, intensity, and propagation direction of the sound. For stress data, stress tensor features are extracted based on tensor decomposition to comprehensively describe the stress state.

[0047] State Fusion Perception Module: This module uses an evidence theory fusion method based on quantum-behavior particle swarm optimization (QPSO). The QPSO algorithm optimizes the parameters of the basic probability distribution function in evidence theory to improve fusion accuracy. Furthermore, it combines the graph neural network (GNN) in deep learning, uses each sensor feature as a node in the graph, and uses GNN to learn the complex relationships between nodes to perform multi-source feature fusion. During the fusion process, the information gain rate theory is introduced to adjust the feature weights. The formula is: Where IG(X|Y) is the information gain, and IV(X) is the intrinsic value of feature X. By calculating the information gain rate of each feature, its weight in the fusion is dynamically adjusted so that the fusion result can better reflect the actual status of the disconnector.

[0048] Status assessment and early warning module: Construct a status assessment model based on transfer learning and generative adversarial networks (GANs). A large amount of historical data on GIS disconnectors with known status is used to pre-train the model in the source domain. In the target domain (i.e., the currently monitored GIS disconnectors), transfer learning technology is used to fine-tune the model in combination with a small amount of current data. At the same time, GAN is introduced to enhance the generalization ability of the model. The generator in the generative adversarial network generates simulated disconnector status data, and the discriminator determines the authenticity of the data. Through adversarial training, the evaluation model's ability to recognize complex states is improved. The long short-term memory network (LSTM) is combined with the attention mechanism to predict future states and issue early warnings. When issuing warnings, the fuzzy analytic hierarchy process (FAHP) is used to sort the warning priorities of different fault types, giving priority to high-risk faults.

[0049] The present invention also includes the following modules:

[0050] The real-time data transmission module utilizes advanced technical architecture and encryption methods to ensure efficient and secure data transmission. Regarding the transmission architecture, an intelligent transmission system is built based on software-defined networking (SDN) and network function virtualization (NFV). SDN separates the network control plane from the data plane, enabling centralized network management and flexible configuration. Network administrators can formulate policies in the control plane based on real-time data traffic, service priorities, and current network conditions, dynamically adjusting data transmission paths to ensure data bypasses congested nodes and selects the optimal transmission path. NFV decouples traditional network functions, such as routers and firewalls, from dedicated hardware devices, running them as software on general-purpose servers. This enables rapid deployment and flexible expansion of network functions, providing greater flexibility in bandwidth allocation and ensuring that different types of data receive appropriate transmission bandwidth. Homomorphic encryption technology is utilized to ensure data security. This technology allows various operations, such as routing and forwarding, to be performed on data in an encrypted state. Even if data is intercepted during transmission, attackers cannot directly obtain the plaintext information because the homomorphic encryption algorithm ensures that the data can only be restored using a specific decryption key, thus greatly ensuring the security and privacy of data transmission. Furthermore, blockchain-based distributed ledger technology has been introduced. This technology records the time, source address, destination address, and data volume of each data transmission on each blockchain node, creating an immutable transmission log. In the event of a data transmission failure, the blockchain smart contract automatically detects and triggers a data retransmission mechanism. The smart contract pre-sets retransmission conditions and policies, enabling rapid retransmission without manual intervention, improving data transmission success rates and ensuring traceability throughout the data transmission process, facilitating troubleshooting and performance optimization for operations and maintenance personnel.

[0051] The present invention also includes the following modules:

[0052] Fault Diagnosis and Location Module: When the condition assessment module determines that the GIS disconnector is in a faulty state, fault diagnosis is immediately initiated. A joint fault diagnosis approach based on Bayesian networks and deep learning is first employed. Deep learning models, such as convolutional neural networks (CNNs), leverage their powerful feature extraction capabilities to perform in-depth analysis of data from various sensors, including vibration, temperature, and stress. Through multi-layer convolution and pooling operations, CNNs automatically learn complex patterns and features within the data. For example, they can extract specific frequency components from vibration signals and detect abnormal temperature rise trends from temperature data. The features extracted by the CNN serve as input nodes for the Bayesian network. A Bayesian network is a graphical network model based on probabilistic reasoning. It infers the probability of a fault by learning conditional probability distributions between nodes. For example, when a component's temperature rises abnormally and its vibration frequency changes, the Bayesian network calculates the probability of the corresponding fault type based on pre-learned probability relationships, thereby enabling fault diagnosis. To further improve diagnostic accuracy, the Bayesian network structure is optimized using an improved particle swarm optimization (PSO) algorithm. The PSO algorithm simulates the foraging behavior of bird flocks and searches for the optimal network structure parameters in the solution space, allowing the Bayesian network to more accurately reflect the intrinsic connection between faults and features. In terms of fault location, positioning technology based on the Internet of Things (IoT) and augmented reality (AR) is utilized. IoT tags are deployed on GIS disconnector equipment. These tags have built-in high-precision sensors and communication modules, which can collect equipment location information in real time and upload it to the network. When a fault is detected, operation and maintenance personnel use AR devices, such as smart glasses or tablets, to scan the fault area. By interacting with the IoT tags, the AR device intuitively displays the precise location information of the faulty equipment in the form of a virtual image in the field of view of the operation and maintenance personnel, clearly marking the fault location, greatly assisting the operation and maintenance personnel in quickly locating the fault point, and then efficiently carrying out repair work, effectively shortening the equipment fault repair time, and improving the reliability of power system operation.

[0053] In the present invention, the feature extraction module introduces a variant of the generative adversarial network (GAN), Cycle-GAN, for feature enhancement. Cycle-GAN consists of two generators and two discriminators, forming a double-loop structure. When processing sensor data of GIS disconnect switches, it can perform unsupervised conversion between different types of features. For example, when vibration feature data is input, one of the generators will try to convert it into acoustic feature data. The discriminator will determine whether the generated data belongs to the acoustic feature domain. Through continuous adversarial training, the conversion ability of the generator is optimized, thereby generating feature data with diversity. These newly generated data are not simply copied, but cover a wider feature space, greatly expanding the feature library and providing richer information for subsequent state analysis. At the same time, a deep belief network (DBN) is used to deeply mine multi-source features. DBN is stacked by multiple layers of restricted Boltzmann machines (RBM). Each RBM layer can learn different levels of features of the data. When processing multi-source data for GIS disconnectors, the underlying RBM can capture fundamental features such as vibration frequency and temperature variations. As the number of layers increases, the higher-level RBMs can learn the higher-order relationships between these fundamental features. For example, it can discover how the associated vibration and stress characteristics change synergistically when the contact temperature rises. Through this unsupervised learning approach, DBN can deeply explore the complex intrinsic connections between features, further improving the depth and breadth of feature extraction. This allows the features extracted from multi-source data to more accurately reflect the actual operating status of the GIS disconnector, providing a more valuable information foundation for subsequent status assessment and fault diagnosis, thereby improving the accuracy and reliability of the entire multi-state intelligent perception method.

[0054] In the present invention, the state fusion perception module adopts a fusion method based on tensor neural network (TNN). First, a tensor model is constructed to represent the data from multiple source sensors such as vibration, temperature, and stress in the form of tensors. Tensors can effectively integrate data information of different dimensions and different modes, and comprehensively present the complex structure of the data. TNN uses a unique neural network architecture to learn the complex interactive relationship between tensors and perform feature fusion on multi-source data. It not only considers the linear relationship of the data, but also captures the high-order correlation between different modal data. For example, when analyzing the operating status of the GIS disconnector, TNN can discover the deep correlation between the contact temperature change and the vibration characteristics of the transmission connecting rod, which may be ignored in traditional methods. During the fusion process, the quantum genetic algorithm (QGA) is used to optimize the parameters of TNN. QGA draws on the idea of quantum computing and introduces quantum bits to represent chromosomes, so that the algorithm has a wider search capability in the solution space. By simulating the evolution process of the quantum state, QGA can quickly find the optimal parameter combination of TNN, so that the fused features can more accurately characterize the state of the disconnector. For example, when adjusting the weight parameters of a neural network, QGA can avoid falling into local optimal solutions and find the globally optimal weight configuration, thereby improving the quality of feature fusion. At the same time, it combines the concept of cyber-physical fusion systems (CPS). The real operating data collected in real time by sensors in the physical world is deeply integrated with the model data obtained through modeling and simulation in the virtual world. The virtual model can predict and simulate the operating status of the disconnector, and verify and complement the actual sensor data. For example, when the actual sensor detects an abnormal temperature in a component, the virtual model can analyze the possible causes and subsequent development trends from a theoretical perspective. The combination of the two enables operation and maintenance personnel to perceive the operating status of the disconnector more comprehensively and accurately, providing strong support for the reliable operation of the equipment and fault prevention.

[0055] In this invention, the state assessment and early warning module incorporates the Proximal Policy Optimization (PPO) algorithm from deep reinforcement learning for state assessment. Modeling the state assessment problem as a Markov decision process means treating the operating state of the GIS disconnector as a series of discrete state nodes, each representing the device's operating status at a specific moment. The intelligent agent acts as an evaluator in this process, interacting with the environment (i.e., the operating state of the GIS disconnector). The agent adopts an evaluation strategy based on the current state and then adjusts this strategy based on a reward function fed back by the environment. The reward function is based on key metrics such as the accuracy of the state assessment and the timeliness of the early warning. For example, if the agent accurately assesses the device's fault state and issues a timely early warning, it receives a higher reward; conversely, if the assessment is incorrect or the early warning is delayed, the reward is lower. By continuously interacting with the environment and adjusting its strategy based on the rewards, the agent can gradually optimize its evaluation strategy and more accurately determine the disconnector's operating state. Simultaneously, Monte Carlo Tree Search (MCTS) is combined with a neural network for state prediction. MCTS explores the space of possible future states by simulating different state transition paths. It starts from the current state node and generates multiple possible state transition paths through random sampling and simulation. In each simulation process, the neural network is combined to evaluate the future state. With its powerful learning ability, the neural network can conduct in-depth analysis of the collected multi-source data (such as vibration, temperature, stress, etc.) and explore the potential patterns and trends behind the data. Based on the evaluation results of the neural network, MCTS selects the most likely state transition path and predicts possible failures in advance. This combination method allows the state prediction to not only take into account historical data and current status, but also conducts a forward-looking assessment of the future state through simulation and neural network analysis, greatly improving the accuracy and timeliness of the early warning, and providing strong support for operation and maintenance personnel to take measures in advance and prevent failures.

[0056] In the present invention, the multi-source data acquisition module adds an elemental analysis sensor based on laser-induced breakdown spectroscopy (LIBS), providing a new perspective for the status monitoring of metal components inside GIS disconnectors. LIBS technology is based on high-energy laser pulses focused on the surface of metal components, instantly evaporating and ionizing a small amount of material to form a plasma. When the plasma cools, the atoms and ions in it emit light with a specific wavelength. This light is analyzed by a spectrometer to accurately identify the type and content of elements in the metal components. By continuously monitoring changes in elemental composition, it is possible to effectively understand the corrosion, wear and other conditions of the components. For example, when the copper content in the contact area decreases due to long-term wear, or when impurity elements increase, the LIBS sensor can capture these changes in time, providing key information for status assessment, allowing operation and maintenance personnel to predict potential failure risks in advance. Wireless energy harvesting technology is used to power some sensors, further improving the practicality and stability of the system. An electromagnetic induction energy harvester is deployed near the operating mechanism, and its working principle is based on Faraday's law of electromagnetic induction. When the disconnector operating mechanism operates, it generates a changing magnetic field. This magnetic field shift creates an induced electromotive force in the induction coil within the energy harvester, converting the electromagnetic energy into electrical energy. Through rectification, filtering, and other circuit processing, the alternating current is converted into stable direct current (DC), which is stored in the energy storage device, providing a continuous and stable power supply for nearby sensors. This power supply method effectively reduces wiring costs and avoids the complexities of cabling and maintenance. It also increases the flexibility of sensor deployment, allowing sensors to be installed in difficult-to-wire locations. This ensures the stable and reliable operation of the multi-source data acquisition module, providing a solid data acquisition foundation for the multi-state intelligent sensing of GIS disconnectors.

[0057] In this invention, the data preprocessing module uses a deep learning-based image super-resolution reconstruction algorithm to process terahertz microstructure images. Terahertz waves can penetrate a variety of materials and have unique reflection properties for metals, among others. Therefore, terahertz microstructure images can reveal the microstructural information of GIS disconnector internal components. However, such images often have low resolution due to factors such as the acquisition equipment or the environment. This algorithm innovatively combines a generative adversarial network (GAN) with a convolutional neural network (CNN). The GAN consists of a generator and a discriminator. The generator learns the mapping relationship from low-resolution to high-resolution images and attempts to generate a high-resolution image; the discriminator determines whether the generated image is a true high-resolution image or generated by the generator. Through continuous adversarial training between the two, the generator's generation capability is gradually improved. The CNN, with its powerful feature extraction capabilities, deeply learns and extracts local features of the image, further optimizing image details. The two work together to reconstruct low-resolution terahertz microstructure images into high-resolution images, clearly revealing microstructural details of disconnector internal components, such as wear textures and microcracks on the contact surfaces. This facilitates subsequent accurate feature extraction and in-depth analysis of component conditions. An adaptive Kalman filter algorithm is also used to process displacement sensor data. Displacement sensors are used to monitor position changes in GIS disconnector components. However, their measurement data is susceptible to noise, and the system's dynamic characteristics vary with operating conditions. The adaptive Kalman filter algorithm monitors changes in sensor measurement noise and system dynamic characteristics in real time. By estimating the statistical characteristics of the noise and adaptively adjusting the system's state transition model, the Kalman filter parameters are dynamically updated. For example, when frequent disconnector operation causes increased system vibration and measurement noise, the algorithm can promptly adjust parameters to more accurately predict and estimate the displacement state, effectively filtering out noise and improving the accuracy of the displacement data, providing a reliable basis for displacement data-based state assessment and fault diagnosis.

[0058] In the present invention, the feature extraction module performs feature extraction on the current data based on wavelet packet transform and singular value decomposition (SVD). The current signal contains rich information about the contact state of the GIS disconnector contacts and the working state of the operating mechanism. Wavelet packet transform is a powerful time-frequency analysis tool that can decompose the current signal into multiple different frequency bands to achieve a more detailed characterization of the signal. Through this decomposition, the characteristics of the current signal under different frequency components can be analyzed separately. For example, the high-frequency part may reflect the transient changes at the moment of contact, while the low-frequency part is related to the smooth operation of the operating mechanism. After completing the wavelet packet transform, singular value decomposition is performed on each frequency band signal. Singular value decomposition is a matrix decomposition technology that can decompose the signal matrix into multiple singular values and corresponding vectors. These singular values reflect the energy distribution of the signal in different dimensions and have good stability and feature representativeness. Extracting singular values as features can more accurately capture the essential characteristics of the current signal. For example, when contacts are in poor contact, the singular values of the current signal will undergo specific changes. By monitoring these changes, the contact state and the operating status of the operating mechanism can be effectively determined, providing key evidence for fault diagnosis. Furthermore, a deep autoencoder (DAE) is used to reduce the dimensionality of multi-source features. In the multi-state intelligent perception of GIS disconnectors, multi-source sensor data such as vibration, temperature, and stress are collected, with a wide range of feature dimensions. A DAE is an unsupervised learning model consisting of an encoder and a decoder. The encoder maps high-dimensional input data to a low-dimensional feature space, while the decoder attempts to reconstruct the original data from the low-dimensional features. By learning the reconstruction error of the data, the DAE can automatically identify and extract the key features of the data, remove redundant information, and reduce the feature dimensionality. This not only reduces the computational complexity of subsequent data processing but also highlights the key features of the data, improving the efficiency and accuracy of state assessment and fault diagnosis, and enabling the entire multi-state intelligent perception system to operate more efficiently.

[0059] The present invention also includes the following modules:

[0060] Self-Learning and Optimization Module: Utilizing online learning algorithms, such as the gradient descent-based online random forest algorithm, real-time model updates are achieved. Online random forests are composed of multiple decision trees and can handle dynamic data streams. During GIS disconnector operation, new sensor data is continuously generated. Based on this new data, the gradient descent-based online random forest algorithm calculates the gradients of model parameters and gradually adjusts the parameters of the state assessment model, fault diagnosis model, and feature extraction model along the gradient direction. For example, when new vibration and temperature data are collected, the model can promptly learn new features and patterns in the data, enabling more accurate assessment of the disconnector's current state, diagnosis of potential faults, and extraction of effective features. Simultaneously, a parameter optimization method based on simulated annealing is used to fine-tune key system parameters. Drawing on the principles of solid-state annealing, the simulated annealing algorithm performs a random search within the solution space. The algorithm optimizes the sensor sampling frequency based on dynamic changes in equipment operation, ensuring that critical information is captured without excessively increasing the data volume. In the fusion algorithm, the weights of different data sources are optimized for more efficient fusion of multi-source data. For neural network hyperparameters, such as the number of layers and neurons, a simulated annealing algorithm continuously tries different parameter combinations to find the hyperparameter settings that optimize neural network performance, thereby improving the system's ability to perceive and analyze complex operating conditions. During the optimization process, IoT big data analysis technology is incorporated. By collecting and integrating large amounts of GIS disconnector operating data, big data analysis tools and algorithms are used to uncover potential patterns in the data. For example, operating data from equipment in different seasons and under different loads is analyzed to identify patterns in parameters such as temperature and vibration, as well as the correlation between these patterns and fault occurrence. These discovered patterns provide strong data support for the system's self-learning and optimization, enabling the intelligent perception system to continuously adapt to changes in the equipment's operating environment and operating conditions, continuously improving the system's performance and adaptability, and providing more effective guarantees for the reliable operation of GIS disconnectors.

[0061] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A GIS disconnector multi-state intelligent sensing system, characterized in that: Includes the following modules: Multi-source data acquisition module: Multi-modal sensors are deployed at key locations of GIS disconnectors. Vibration sensors adaptively adjust sensitivity and sampling frequency, temperature sensors sense temperature changes, current sensors measure current contactlessly, acoustic sensors collect sound signals to locate sound sources, stress sensors monitor stress field changes, and a new terahertz imaging microstructure sensor is added to obtain microstructure image information of key internal components. Data preprocessing module: This module uses a multi-scale residual noise reduction network to denoise vibration signals. It captures and removes noise features by constructing a convolution kernel. It also uses an adaptive weighted fusion algorithm to fuse temperature, current, and stress data. It also introduces fuzzy entropy theory, dynamically assigns weights based on data uncertainty through a formula, and uses a LOF-DBSCAN hybrid algorithm based on density and outlier detection to identify and process outliers. Feature Extraction Module: This module uses EMD combined with variational mode decomposition (VMD) to extract vibration features. It also introduces an inverse heat conduction problem model, calculates the temperature gradient rate of change through finite element iteration, and uses a recurrent convolutional neural network (CRNN) to extract sound temporal and spatial features. It also extracts stress tensor features based on tensor decomposition to comprehensively describe the stress state. State Fusion Perception Module: This module uses an evidence-based fusion method based on quantum-behavior particle swarm optimization (QPSO). The QPSO algorithm optimizes parameters, combines it with graph neural networks (GNNs) to learn node relationships, introduces information gain rate theory to adjust feature weights, and dynamically adjusts weights by calculating information gain rates. Status assessment and early warning module: Build a status assessment model based on transfer learning and generative adversarial network (GAN), use historical data for pre-training, fine-tune in the target field using transfer learning technology, use long short-term memory network (LSTM) combined with attention mechanism to predict early warnings, and use fuzzy hierarchical analysis method (FAHP) to sort the warning priorities.

2. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: Also includes: Real-time data transmission module: Adopts an intelligent transmission architecture based on software-defined networking (SDN) and network function virtualization (NFV), dynamically adjusts transmission paths and bandwidth allocation according to data traffic, priority, and network conditions. Homomorphic encryption technology is used in the transmission process to route and forward encrypted data. Blockchain-based distributed ledger technology is introduced to record data transmission logs. In the event of data transmission failure, a data retransmission mechanism is automatically triggered through blockchain smart contracts.

3. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: Also includes: Fault diagnosis and location module: When the status assessment module determines that the GIS disconnector is in a faulty state, a joint fault diagnosis method based on Bayesian networks and deep learning is adopted. The deep learning model is used to extract sensor data features as Bayesian network input nodes. Faults are diagnosed by learning the conditional probability distribution between nodes. The Bayesian network structure is optimized by combining the particle swarm optimization algorithm (PSO). Using IoT and augmented reality (AR) positioning technology, IoT tags and AR devices are deployed on the GIS disconnector to obtain the location information of the faulty equipment in real time, and the fault location is intuitively displayed in the form of augmented reality.

4. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The feature extraction module calculates the temperature gradient change rate through finite element iteration, and the formula is T i is the calculated temperature, is the observed temperature, R(T) is the regularization term, and λ is the regularization parameter; The generative adversarial network (GAN) variant Cycle-GAN is introduced to enhance features. Cycle-GAN performs unsupervised conversion between different types of features to generate diverse feature data. Deep belief network (DBN) is used to mine multi-source features. DBN stacks multiple layers of restricted Boltzmann machines (RBMs) to unsupervisedly learn high-order relationships between features.

5. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The state fusion perception module introduces the information gain rate theory to adjust the feature weights during the fusion process. The formula is: Where IG(X|Y) is the information gain and IV(X) is the intrinsic value of feature X; A tensor model is constructed using the tensor neural network (TNN) fusion method. Multi-source sensor data is represented in tensor form. TNN fuses features by learning the interaction relationship between tensors. The fusion process uses the quantum genetic algorithm (QGA) to optimize TNN parameters. Combined with the concept of information-physical fusion system (CPS), it deeply integrates physical world sensor data and virtual world model data.

6. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The state assessment and warning module introduces the proximal policy optimization (PPO) algorithm in deep reinforcement learning to evaluate the state. The state assessment problem is modeled as a Markov decision process. The intelligent agent optimizes the evaluation strategy according to the reward function by interacting with the environment, and uses Monte Carlo tree search (MCTS) combined with neural networks to predict the state. MCTS simulates different state transition paths and combines neural networks to evaluate the future state.

7. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The multi-source data acquisition module adds an elemental analysis sensor based on laser-induced breakdown spectroscopy (LIBS) to analyze the elemental composition of the metal components inside the GIS disconnector. By analyzing changes in elemental composition, it monitors the corrosion and wear of the components. Wireless energy harvesting technology is used to power some sensors, and an electromagnetic induction energy harvester is deployed near the operating mechanism to collect the electromagnetic energy generated during operation to provide power for the sensors.

8. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The data preprocessing module uses an adaptive weighted fusion algorithm for temperature, current, and stress data. The weight calculation introduces fuzzy entropy theory. The formula is: Where FEntropy(x i ) is the fuzzy entropy of the i-th sensor data; An image super-resolution reconstruction algorithm is used to process terahertz microstructure images. The low-resolution image is reconstructed into a high-resolution image by combining the generative adversarial network (GAN) with the convolutional neural network (CNN). An adaptive Kalman filter algorithm is used to process the displacement sensor data, and the Kalman filter parameters are adjusted in real time according to the sensor measurement noise and changes in the system dynamic characteristics.

9. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: The feature extraction module extracts current data features based on wavelet packet transform and singular value decomposition (SVD). It uses wavelet packet transform to decompose the current signal into different frequency bands, decomposes the singular values of each frequency band signal, extracts the singular values as features, and uses deep autoencoders (DAEs) to reduce the dimensionality of multi-source features. By learning data reconstruction errors, it automatically extracts key data features.

10. The GIS disconnector multi-state intelligent sensing system according to claim 1 is characterized in that: Also includes: Self-learning and optimization module: Utilizes online learning algorithms to update the parameters of the status assessment model, fault diagnosis model, and feature extraction model in real time based on newly collected data. A parameter optimization method based on a simulated annealing algorithm is used to optimize key system parameters. The optimization process is combined with IoT big data analysis technology to explore the patterns of equipment operation data.

Images