Temperature Rise Monitoring Method and System for High-Voltage Fully Enclosed Vacuum Environmental Protection GIS Equipment

Abstract

The present application relates to the technical field of temperature monitoring, and discloses a temperature rise monitoring method and system for a high-voltage fully enclosed vacuum environmental protection type GIS device. The method includes: collecting temperature data through temperature sensors, and constructing a node feature matrix and an adjacency matrix according to the spatial position relationship of the SF6 gas chamber layer, the high-voltage conductor layer, and the outer shell enclosure layer; performing intra-layer graph convolution operations and inter-layer graph convolution operations to obtain a temperature spatial distribution feature vector, and performing temporal fusion to obtain a temperature dynamic change temporal feature vector; partitioning the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change temporal feature vector to obtain adaptive control parameters for each partition; inputting the adaptive control parameters into a multi-level compensation controller to calculate temperature compensation coefficients, and obtaining temperature correction parameters for each level, thereby improving the integrity and accuracy of temperature field distribution feature extraction, and realizing precise evaluation and hierarchical early warning of temperature rise risks at different levels.

Description

Technical Field

[0001] This application relates to the technical field of temperature monitoring, and particularly to a temperature rise monitoring method and system for a high-voltage fully enclosed vacuum environmental protection type GIS device. Background Art

[0002] The high-voltage fully enclosed vacuum environmental protection type GIS device is a key device in the power system, and its safe and stable operation has an important impact on the reliability of the power system. During actual operation, the temperature distribution characteristics inside the GIS device are complex, involving temperature coupling effects at multiple levels such as the SF6 gas chamber, high-voltage conductor, and outer shell. Traditional single-point temperature measurement methods are difficult to comprehensively reflect the temperature field distribution characteristics of the device.

[0003] Existing temperature rise monitoring methods for GIS devices mainly rely on empirical knowledge to judge temperature thresholds, lacking in-depth analysis of the dynamic correlation between temperatures at different levels of the device, resulting in low early warning accuracy. At the same time, due to the dynamic changes in the operating conditions of the device, the temperature field distribution shows obvious spatio-temporal correlation characteristics, and traditional static monitoring methods are difficult to accurately grasp the temperature change trend. In addition, multiple factors such as the SF6 gas pressure, load current, and ambient humidity in the GIS device will affect the distribution of the temperature field, and there are complex coupling relationships between these factors. Existing monitoring methods often consider these factors separately, lacking systematic multi-parameter fusion analysis, and it is difficult to achieve accurate early warning of temperature rise risks, affecting the operating reliability and safety of the device. Summary of the Invention

[0004] This application provides a temperature rise monitoring method and system for a high-voltage fully enclosed vacuum environmental protection type GIS device, thereby improving the integrity and accuracy of temperature field distribution feature extraction, and achieving precise assessment and hierarchical early warning of temperature rise risks at different levels.

[0005] In the first aspect of this application, a temperature rise monitoring method for a high-voltage fully enclosed vacuum environmental protection type GIS device is provided. The temperature rise monitoring method for the high-voltage fully enclosed vacuum environmental protection type GIS device includes:

[0006] Collect temperature data through temperature sensors, and construct a node feature matrix and an adjacency matrix according to the spatial position relationship of the SF6 gas chamber layer, high-voltage conductor layer, and outer shell closed layer;

[0007] Perform intra-layer graph convolution operations and inter-layer graph convolution operations on the node feature matrix and the adjacency matrix to obtain a temperature spatial distribution feature vector, and perform temporal fusion to obtain a temperature dynamic change temporal feature vector;

[0008] Partition the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change temporal feature vector to obtain adaptive control parameters for each partition;

[0009] Input the adaptive control parameters into a multi-level compensation controller to calculate the temperature compensation coefficient, and obtain the temperature correction parameters for each level.

[0010] The second aspect of the present application provides a temperature rise monitoring system for a high-voltage fully enclosed vacuum environmental protection type GIS device. The temperature rise monitoring system for the high-voltage fully enclosed vacuum environmental protection type GIS device includes:

[0011] An acquisition module, configured to collect temperature data through a temperature sensor, and construct a node feature matrix and an adjacency matrix according to the spatial position relationship of the SF6 gas chamber layer, the high-voltage conductor layer, and the enclosure layer;

[0012] An operation module, configured to perform intra-layer graph convolution operation and inter-layer graph convolution operation on the node feature matrix and the adjacency matrix to obtain a temperature spatial distribution feature vector, and perform temporal fusion to obtain a temperature dynamic change temporal feature vector;

[0013] A partitioning module, configured to partition the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change temporal feature vector to obtain the adaptive control parameters for each partition;

[0014] A compensation module, configured to input the adaptive control parameters into a multi-level compensation controller to calculate the temperature compensation coefficient, and obtain the temperature correction parameters for each level.

[0015] Compared with the prior art, the present application has the following beneficial effects: By constructing a three-layer network topology structure, systematic management of the temperature data of the SF6 gas chamber layer, the high-voltage conductor layer, and the enclosure layer is realized, improving the integrity and accuracy of the extraction of the temperature field distribution characteristics; By adopting a combination of intra-layer graph convolution and inter-layer graph convolution, the temperature correlation between different levels is effectively captured, enhancing the ability to extract the spatial distribution characteristics of the temperature field; By combining the temporal fusion mechanism of the state update gate and the parameter correlation gate, dynamic correlation analysis of temperature data with multiple parameters such as SF6 gas pressure and environmental humidity is realized; Through the partition reinforcement learning control strategy, different control parameters are adopted for different regions, improving the accuracy of temperature control; By adopting a multi-level compensation control mechanism, the coupling effects of gas pressure, load current, and environmental humidity are considered, improving the accuracy of temperature correction; Based on the risk warning mechanism of a multi-layer fully connected neural network, accurate assessment and classification warning of the temperature rise risks at different levels are realized. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] The structures, proportions, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention. Therefore, they do not have substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope that can be covered by the technical content disclosed in the present invention.

[0018] Figure 1 It is a schematic flowchart of the temperature rise monitoring method for the high-voltage fully enclosed vacuum environmental protection type GIS equipment provided by the embodiment of the present invention;

[0019] Figure 2 It is a schematic block diagram of the structure of the temperature rise monitoring system for the high-voltage fully enclosed vacuum environmental protection type GIS equipment provided by the embodiment of the present invention. Detailed implementation manners

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present invention.

[0021] The flowchart shown in the drawings is only an example, and does not necessarily include all the content and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0022] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0023] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the temperature rise monitoring method for the high-voltage fully enclosed vacuum environmental protection type GIS device in the embodiment of this application includes:

[0024] Step 100, collect temperature data through a temperature sensor, and construct a node feature matrix and an adjacency matrix according to the spatial position relationship of the SF6 gas chamber layer, the high-voltage conductor layer, and the housing enclosure layer;

[0025] It can be understood that the execution subject of this application can be the temperature rise monitoring system of the high-voltage fully enclosed vacuum environmental protection type GIS device, or it can also be a terminal or a server, and specific limitations are not made here. In the embodiment of this application, the server is used as the execution subject for illustration.

[0026] Specifically, temperature sensors installed in the SF6 gas chamber layer, high-voltage conductor layer, and outer shell enclosure layer of the GIS device are used to collect temperature data for the corresponding areas. In the SF6 gas chamber layer, multiple sensors are arranged to collect temperature data at each monitoring point inside the gas chamber. At the same time, the specific spatial position coordinates of these sensors are recorded. After pairing the temperature values and coordinate information, they are sorted into a set of sensing nodes for the gas chamber layer. In the high-voltage conductor layer, temperature sensors are installed at key parts such as circuit breaker contacts, disconnector switches, and bushing joints to collect temperature data at these positions. After pairing the data with their spatial position coordinates, a set of sensing nodes for the conductor layer is generated. In the outer shell enclosure layer, sensors are arranged to monitor the temperature changes of the device outer shell and support structure. After pairing these temperature data with the corresponding spatial positions, a set of sensing nodes for the outer shell layer is formed. The spatial coordinates of the nodes in each layer are standardized, and the spatial positions of the nodes at different levels are mapped to a unified coordinate system to ensure the consistency of all node data in the same coordinate system. At the same time, the collected temperature data is normalized, and the temperature data of different nodes is adjusted to a unified numerical range to eliminate the interference of different measurement ranges on subsequent calculations, generating a standardized sensing node matrix. Based on the standardized matrix, a node feature matrix is constructed. The node feature matrix is a key data structure that comprehensively describes the attributes of each node. It contains the normalized temperature value, spatial position, and layer information to which each node belongs. The layer attribute of each node is clearly marked as the SF6 gas chamber layer, high-voltage conductor layer, or outer shell enclosure layer. According to the spatial position information in the standardized sensing node matrix, the distance between any two sensing nodes is calculated to establish the spatial correlation between the nodes. By comparing these distance values with a preset threshold, it is determined whether two nodes have a direct connection relationship. If the distance between two nodes is within the in-layer connection threshold range and they belong to the same layer, they are defined as in-layer connected; if the distance between two nodes is within the inter-layer connection threshold range and they belong to different layers, they are defined as inter-layer connected. Through this judgment process, the relationship between the nodes is transformed into a binary adjacency matrix, where each element represents whether there is a connection between the corresponding two nodes. The binary adjacency matrix is associated with the node layer information. In the binary adjacency matrix, the in-layer connection edges and inter-layer connection edges are clearly marked. For example, the nature of the two types of connections is reflected through different identifiers or weights, comprehensively representing the connection relationship between the nodes and demonstrating the correlation characteristics within and between each layer. Through the above steps, a node feature matrix containing node temperature, spatial position, and layer information, as well as an adjacency matrix reflecting the connection relationship and hierarchical structure between the nodes, is constructed.

[0027] Step 200: Perform intra-layer and inter-layer graph convolution operations on the node feature matrix and the adjacency matrix to obtain the temperature spatial distribution feature vector, and perform temporal fusion to obtain the temperature dynamic change temporal feature vector;

[0028] Specifically, normalize the temperature data of each sensor node in the node feature matrix, and mark it in combination with the hierarchical information of each node. Divide all sensor nodes into three different hierarchical node groups, namely the gas chamber layer node group, the conductor layer node group, and the outer shell layer node group, to form a hierarchical grouping matrix. At the same time, based on the adjacency matrix, distinguish the connection relationships between nodes, and divide all connection edges into an intra-layer connection edge set and an inter-layer connection edge set according to the hierarchical labels of the sensor nodes to obtain a connection edge classification matrix. Perform matrix multiplication on the hierarchical grouping matrix and the first convolution kernel matrix to obtain an initial feature transformation result, and apply the ReLU activation function to this operation result to introduce non-linearity and generate an intra-layer initial feature matrix. Each node feature vector in the intra-layer initial feature matrix will contain an important representation of its local features. Calculate the attention coefficient for each node feature vector in the intra-layer initial feature matrix. By comprehensively considering the temperature gradient and topological distance between nodes, generate the attention weight coefficient for each node and construct an attention weight matrix. Perform weighted combination on the attention weight matrix and the intra-layer initial feature matrix, and combine it with the intra-layer connection edge set to perform graph convolution operation to extract the local feature interaction relationship between nodes and obtain an intra-layer spatial feature matrix, which reflects the temperature distribution characteristics within each layer. In order to capture the feature correlations between nodes at different levels, perform convolution operation on the node feature vectors in the intra-layer spatial feature matrix and the second convolution kernel matrix, and complete cross-layer feature transfer in combination with the inter-layer connection edge set to generate an inter-layer correlation feature matrix, which reflects the heat conduction relationship between different levels and reveals the cooperative effect of each layer in the overall temperature distribution. Perform feature splicing on the intra-layer spatial feature matrix and the inter-layer correlation feature matrix to obtain a spatial feature expression structure. Input the spliced feature vector into a fully connected layer for feature compression to obtain the temperature spatial distribution feature vector. Perform temporal fusion on the temperature spatial distribution feature vector, SF6 gas pressure data, and environmental humidity data. Capture the dynamic correlation between temperature features and external environmental parameters through a time series model to generate a temporal feature vector of temperature dynamic change, which comprehensively reflects the time variation law of the internal temperature rise of the GIS device and its response ability to environmental conditions.

[0029] Normalize the temperature spatial distribution feature vector and SF6 gas pressure data to eliminate the inconsistencies caused by different data types and numerical ranges, ensure the comparability of different parameters within the same numerical interval, and generate a temperature-pressure time series matrix. Align the temperature-pressure time series matrix with the environmental humidity data in the time dimension to ensure the consistency of sampling points of all parameter data on the time axis, and form a multi-parameter time series data matrix containing temperature, pressure, and humidity information. Based on the multi-parameter time series data matrix, calculate the proportion of historical information in the matrix through the state update gate to determine the comprehensive weight relationship between the data at each moment and its historical information, and generate a state update matrix. The state update gate analyzes the dynamic change law of the data at each time step, judges the importance of historical information at the current moment, and thus effectively captures the cumulative effect of historical information during the temperature rise process. At the same time, calculate the influence degree of each parameter in the multi-parameter time series data matrix through the parameter correlation gate. Based on the thermodynamic coupling relationship between SF6 gas pressure and temperature, and the correlation between environmental humidity and temperature change, generate a parameter correlation matrix, quantify the influence degree between different parameters, and dynamically adjust the weights of each parameter in the overall feature extraction process. Perform matrix multiplication on the state update matrix and the parameter correlation matrix, and comprehensively extract the key features of the data through cross-dimensional fusion calculation. The fused result is subjected to feature selection through the output gate, and the parameter combination most representative of the dynamic change of temperature rise is selected to obtain a fused feature matrix. Perform dynamic time attention calculation on the time series data in the fused feature matrix. Combine the operating state of the GIS device (such as device start-stop state) and the influence of load change on temperature characteristics to generate a time series weighted matrix. The dynamic time attention mechanism can assign different weights to the feature data at different moments, making the contribution of the data in the key operating stages of the device to the overall features more significant. Perform weighted superposition operation on the fused feature matrix and the time series weighted matrix to obtain an initial dynamic feature vector. Extract the time series features of the initial dynamic feature vector through a gated recurrent unit (GRU). Through its gating mechanism, the GRU can effectively capture the long-term dependence relationship of the feature vector in the time series and avoid the problem of gradient disappearance in traditional recurrent networks. Under the processing of the GRU, the initial dynamic feature vector is transformed into a temperature dynamic change time series feature vector, which reflects the temperature rise law of the GIS device under different environmental conditions and operating states.

[0030] Step 300: Divide the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector to obtain the adaptive control parameters for each partition;

[0031] It should be noted that the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector are deeply fused in the spatial and temporal dimensions to generate a spatio-temporal feature matrix. The spatio-temporal feature matrix contains the temperature distribution characteristics of each monitoring area inside the GIS device and its dynamic law of change over time. By performing hierarchical dimensionality reduction on this matrix, redundant information is removed while key features are retained, obtaining a region classification feature vector. The region classification feature vector is input into the classification network, and region division is carried out according to the thermal characteristics and temperature dynamic laws of different parts inside the GIS device. The core heating parts such as circuit breaker contacts and disconnector switches are classified as key areas; the positions such as conductor joints and bushing joints are classified as transition areas; the areas such as the shell and support structure are classified as ordinary areas. Through this step, a region category label matrix is generated. A reinforcement learning training environment is constructed for the region category label matrix. The reinforcement learning environment defines the control objectives and the operable range of the system by setting the state space and the action space. The state space defines the temperature states of each partition, including temperature distribution characteristics, dynamic change trends, etc., while the action space covers the possible temperature control actions that each partition can take, such as adjusting the operation mode of the heat dissipation system, optimizing the gas flow path, etc. By setting the temperature control requirements for different regions, the reinforcement learning environment parameters are generated. The reinforcement learning environment parameters are input into the reinforcement learning agent, and the intelligent agent explores and samples the temperature control actions for each partition to generate an action sample set. During this process, the reinforcement learning agent tries different control actions in a trial-and-error manner and evaluates each action through the constructed reward function. The design of the reward function comprehensively considers factors such as temperature control accuracy, response speed, temperature uniformity, and long-term stability, ensuring that the control strategy for each partition can not only achieve precise temperature rise regulation, but also quickly respond to environmental changes and maintain stability during long-term operation. By combining the action sample set and the partition reward values, the reinforcement learning agent can continuously optimize the control strategy. To improve the quality of the control strategy, the action sample set and the partition reward values are used to train the policy network. The training process uses the policy gradient algorithm, and by optimizing the parameters of the policy network, a policy network model suitable for each partition is gradually generated. These models can dynamically generate highly adaptable temperature control strategies according to the characteristics of different partitions. To ensure that the performance of the policy network model reaches the optimal state, its performance is evaluated and the parameters are fine-tuned. The trained policy network models are respectively input into the temperature control accuracy evaluator, the response speed evaluator, and the stability evaluator to evaluate their control performance in the actual environment. The evaluation results will be used as feedback to fine-tune the policy network parameters, obtaining an optimized policy network model. Based on the optimized policy network model, a temperature control action sequence is generated. These action sequences represent the specific control measures that each partition should take under different states. By converting the temperature control action sequence into actual control parameters, the adaptive control parameters for each partition are generated.The adaptive control parameters dynamically adjust the temperature rise control strategy according to the partition characteristics, ensuring uniform temperature distribution and good thermal stability during the operation of the GIS equipment, thereby improving the safety and reliability of the equipment.

[0032] Step 400: Input the adaptive control parameters into the multi-level compensation controller to calculate the temperature compensation coefficient, and obtain the temperature correction parameters at each level.

[0033] Specifically, the adaptive control parameters are input into the multi-level compensation controller. The compensation controller performs hierarchical processing on the control parameters according to the structural characteristics of the GIS device and the thermodynamic characteristics of different levels to generate a set of hierarchical control parameters. The control parameters of the gas chamber layer mainly focus on the relationship between the pressure and temperature of the SF6 gas. The parameters of the high-voltage conductor layer mainly analyze the thermal effect of the load current, while the outer shell layer conducts condensation analysis based on the relationship between the environmental humidity and the surface temperature. The hierarchical processing ensures that the control parameters of each layer can accurately match their unique thermodynamic and operating characteristics. After the hierarchical processing is completed, for different levels, the compensation controller uses corresponding physical models and calculation methods to calculate the temperature compensation coefficients. For the gas chamber layer, the SF6 gas pressure-temperature coupling model is used for analysis. The pressure and temperature data of the gas are input into the model, and correlation calculations are performed based on the coupling relationship between the gas pressure and temperature to obtain the temperature compensation coefficient of the gas chamber layer. This coefficient can effectively reflect the thermodynamic change characteristics inside the SF6 gas and ensure the accuracy of the temperature rise monitoring and control of the gas chamber layer. At the same time, in the high-voltage conductor layer, the load current data is input into the load current heat accumulation model, and calculations are performed based on the heat accumulation relationship between the current intensity and the conductor temperature rise to obtain the temperature compensation coefficient of the conductor layer. Through calculation, the influence of the load current on the conductor temperature rise is quantified, so as to accurately compensate and control the thermal effect of the high-voltage conductor layer. For the outer shell layer, the influence of environmental humidity on the temperature rise and condensation risk is mainly analyzed. By inputting the data collected by the environmental humidity sensor into the humidity distribution calculation module, the humidity distribution calculation result is obtained, and it is input into the environmental humidity condensation model for condensation risk analysis. Based on the relationship between the humidity distribution and the surface temperature, this model can accurately evaluate the condensation risk of the outer shell layer and generate the temperature compensation coefficient of the outer shell layer to ensure that the outer shell layer can still maintain temperature uniformity in an environment with large humidity changes and avoid the influence of condensation phenomena on the operation of the equipment. After calculating the temperature compensation coefficients of each layer, according to the corresponding relationship between the hierarchical control parameter set and the compensation coefficients of each layer, a compensation control matrix is generated. This matrix organically combines the control parameters of each layer with their corresponding compensation coefficients to form a unified framework that comprehensively reflects the temperature rise compensation requirements of each layer. To ensure the rationality and stability of the compensation parameters, stability constraint processing is performed on the parameters in the compensation control matrix. By analyzing the temperature change limits of each layer, the correction range of each compensation parameter is determined, and a constraint correction matrix is generated to avoid monitoring deviations or system instabilities caused by over-compensation or under-compensation. The constraint correction matrix is input into the temperature compensation calculation module, and the temperature correction amount of each layer is calculated according to the temperature compensation coefficients of each level to generate the final temperature correction parameters.

[0034] Deeply fuse the temperature correction parameter, the temperature spatial distribution feature vector, and the temperature dynamic change time series feature vector. Integrate these data by level and characteristic dimension through feature combination to ensure the spatio-temporal consistency of all parameters. Standardize the fused data to unify the numerical range of the data and eliminate the influence of dimensional differences on the calculation results, generating a standardized feature matrix. Based on the standardized feature matrix, perform temperature rise risk analysis and risk level division for the SF6 gas chamber layer, the high-voltage conductor layer, and the outer shell closed layer respectively. For the SF6 gas chamber layer, use the gas temperature feature in the standardized feature matrix and the corrected parameter values to calculate the temperature deviation value of the gas chamber layer through the temperature anomaly detection module, and compare these deviation values with the preset insulation breakdown threshold for comparative analysis. If the deviation value exceeds the threshold range, it is determined that there is an insulation breakdown risk, and a risk level matrix of the gas chamber layer is generated in combination with the detection results. The value of each node in the matrix reflects the risk level distribution of different regions of the gas chamber. For the high-voltage conductor layer, extract the temperature gradient features of the circuit breaker contacts, disconnector switches, and bushing joints from the standardized feature matrix and input them into the temperature rise detection module. By calculating the temperature change rate of these key parts and combining their heat conduction characteristics, evaluate the temperature rise risk of each region within the conductor layer and generate a risk level matrix of the conductor layer to reflect the thermal safety of each key part within the high-voltage conductor layer. At the same time, the temperature rise risk analysis of the outer shell closed layer is mainly based on the temperature uniformity detection, and the surface temperature distribution feature in the standardized feature matrix is associated with the environmental temperature difference for calculation. By analyzing the deviation between the surface temperature change of different regions of the equipment shell and the environmental temperature, evaluate the temperature rise uniformity and condensation risk of the outer shell layer and generate a risk level matrix of the outer shell layer. Through the independent analysis of the gas chamber layer, the conductor layer, and the outer shell layer, risk level matrices corresponding to the three layers are formed respectively, and each matrix records the temperature rise risk information of each node in that layer. Input the risk level matrix of the gas chamber layer, the risk level matrix of the conductor layer, and the risk level matrix of the outer shell layer into a multi-layer fully connected neural network. In the neural network, through the layer-by-layer processing of linear transformation and non-linear activation functions, extract the potential correlations in the three-layer risk matrices and generate a comprehensive hierarchical risk feature vector containing risk information of different levels. Input the hierarchical risk feature vector into the Softmax classification module, calculate the probability distribution of various risks, and combine the preset temperature rise risk thresholds of the gas chamber layer, the conductor layer, and the outer shell layer to complete multi-level risk assessment and generate a warning probability matrix. Each element of the warning probability matrix represents the warning probability corresponding to the level and risk level. Perform correlation analysis on the warning probability matrix and the real-time operating state data of the GIS equipment, and combine the temperature rise rate and the temperature rise duration of the equipment to comprehensively judge the actual risk level of each region and output the corresponding classification warning signal. In this way, the warning signal can dynamically reflect the operating state and potential risks of the equipment.

[0035] In the embodiments of the present application, by constructing a three-layer network topology, systematic management of temperature data for the SF6 gas chamber layer, high-voltage conductor layer, and outer shell enclosure layer is achieved, improving the integrity and accuracy of the extraction of temperature field distribution characteristics; by combining in-layer graph convolution and inter-layer graph convolution, the temperature correlation between different levels is effectively captured, enhancing the ability to extract the spatial distribution characteristics of the temperature field; through a temporal fusion mechanism combining a state update gate and a parameter correlation gate, dynamic correlation analysis of temperature data with multiple parameters such as SF6 gas pressure and ambient humidity is realized; through a partitioned reinforcement learning control strategy, different control parameters are adopted for different regions, improving the accuracy of temperature control; a multi-level compensation control mechanism is adopted, considering the coupling effects of gas pressure, load current, and ambient humidity, improving the accuracy of temperature correction; a risk warning mechanism based on a multi-layer fully connected neural network realizes precise assessment and hierarchical warning of temperature rise risks at different levels.

[0036] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0037] Pair the temperature sensor data collected at the internal monitoring points in the gas chamber layer of the SF6 gas chamber with the spatial position coordinates of the sensors to obtain a set of gas chamber layer sensing nodes;

[0038] Pair the temperature sensor data collected at the circuit breaker contacts, disconnector switches, and bushing joints in the high-voltage conductor layer with the spatial position coordinates of the sensors to obtain a set of conductor layer sensing nodes;

[0039] Pair the temperature sensor data collected at the equipment outer shell and support structure in the outer shell enclosure layer with the spatial position coordinates of the sensors to obtain a set of outer shell layer sensing nodes;

[0040] Perform unified spatial coordinate system and temperature data normalization processing on the set of gas chamber layer sensing nodes, the set of conductor layer sensing nodes, and the set of outer shell layer sensing nodes to obtain a standardized sensing node matrix;

[0041] Based on the temperature data in the standardized sensing node matrix, construct a node feature matrix and mark the hierarchical attributes to obtain a node feature matrix containing hierarchical information;

[0042] Calculate the three-dimensional Euclidean distance between any two sensor nodes according to the spatial coordinate information in the standardized sensing node matrix to obtain a node distance matrix, and compare and judge the distance values in the node distance matrix with the preset in-layer connection threshold and inter-layer connection threshold to generate a binary adjacency matrix;

[0043] Perform an association operation on the binary adjacency matrix and the node hierarchical information, and mark the in-layer connection edges and inter-layer connection edges in the binary adjacency matrix to obtain an adjacency matrix.

[0044] Specifically, the monitoring points in the SF6 gas chamber layer consist of multiple temperature sensors arranged inside the gas chamber, and these sensors collect temperature data at different positions in the gas chamber. , where represents the th sensor. At the same time, the spatial position of each sensor is represented by three-dimensional coordinates , which is defined as the actual position of the sensor in three-dimensional space. By pairing the temperature data collected by each sensor with its corresponding spatial position coordinates, the following set is formed:

[0045] ;

[0046] where is the set of sensor nodes in the gas chamber layer, represents the total number of sensors in the gas chamber layer. Similarly, in the high-voltage conductor layer, temperature data at key parts such as circuit breaker contacts, disconnector switches, and bushing joints are collected by sensors and the corresponding spatial position coordinates . The set of sensing nodes in the conductor layer is defined as:

[0047] ;

[0048] where represents the set of sensor nodes in the conductor layer, is the number of sensors in the conductor layer. For the enclosure layer, the sensors arranged collect temperature data of the equipment enclosure and the support structure and the spatial position coordinates , and its node set is defined as:

[0049] ;

[0050] where represents the set of sensing nodes in the conductor layer, is the number of sensors in the enclosure layer. To unify the spatial coordinate systems of each layer and ensure that the positions of sensing nodes in different layers have a consistent reference system, , and are subjected to unified coordinate transformation, assuming the reference coordinate system . The transformed coordinates are expressed as:

[0051] ;

[0052] where are the coordinates in the unified coordinate system, is the rotation matrix, is the translation vector. Through the unification of the coordinate system, a standardized sensing node matrix is generated :

[0053] ;

[0054] Among them, each row represents a sensing node, and its characteristics include temperature data, spatial coordinates, and hierarchical attributes. Based on the standardized matrix , calculate the three-dimensional Euclidean distance between any two sensor nodes :

[0055] ;

[0056] wherein, and respectively represent the indices of two nodes. By performing the above calculations for all node pairs, a node distance matrix is generated, and its element represents the distance between the th and the nd nodes. According to the preset intra-layer connection threshold and the inter-layer connection threshold , judge the distance values in to generate a binary adjacency matrix

[0057] ;

[0058] wherein, the element in represents whether nodes and are connected. Perform an association operation on the binary adjacency matrix A and the hierarchical information in the node feature matrix to mark the intra-layer connection edges and the inter-layer connection edges, and generate an adjacency matrix containing hierarchical attributes.

[0059] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0060] Normalize the temperature data of each sensor node in the node feature matrix and perform hierarchical marking. According to the hierarchical information to which the sensor nodes belong, divide the sensor nodes into an air chamber layer node group, a conductor layer node group, and a housing layer node group to obtain a hierarchical grouping matrix;

[0061] Distinguish the connection relationships in the adjacency matrix, and divide the connection edges into an intra-layer connection edge set and an inter-layer connection edge set according to the hierarchical markings between the sensor nodes to obtain a connection edge classification matrix;

[0062] Perform matrix multiplication on the hierarchical grouping matrix and the first convolutional kernel matrix to obtain the operation result, and perform ReLU activation function processing on the operation result to obtain the initial intra-layer feature matrix;

[0063] Calculate the attention coefficient for each node feature vector in the initial intra-layer feature matrix, generate the node weight coefficient based on the node temperature gradient and topological distance, and obtain the attention weight matrix;

[0064] Perform weighted combination on the attention weight matrix and the initial intra-layer feature matrix, and perform graph convolution operation with the intra-layer connection edge set to obtain the intra-layer spatial feature matrix;

[0065] Perform convolution operation on the feature vectors between different hierarchical nodes in the intra-layer spatial feature matrix and the second convolutional kernel matrix, and perform feature transfer with the inter-layer connection edge set to obtain the inter-layer correlation feature matrix;

[0066] Perform feature splicing on the intra-layer spatial feature matrix and the inter-layer correlation feature matrix, and perform feature compression through a fully connected layer to obtain the temperature spatial distribution feature vector;

[0067] Perform temporal fusion on the temperature spatial distribution feature vector with SF6 gas pressure data and environmental humidity data to obtain the temperature dynamic change temporal feature vector.

[0068] Specifically, the node feature matrix represents the main features of each sensor node, including temperature data, spatial coordinates, and the hierarchical information it belongs to, in the form of:

[0069] ;

[0070] Among them, is the node feature matrix, represents the temperature data of the th node, is the three-dimensional spatial coordinate of the node, is the hierarchical label of the node (such as gas chamber layer, conductor layer, shell layer), represents the total number of nodes. Normalize the temperature data to eliminate the difference in numerical range. The normalization formula is:

[0071] ;

[0072] Among them, is the normalized temperature value, and are the minimum and maximum values of the temperatures of all nodes respectively. Map the temperature value to the interval [0, 1]. According to the hierarchical label , the nodes are divided into three groups, namely the air chamber layer node group, the conductor layer node group, and the housing layer node group, to form a hierarchical grouping matrix , in the form of:

[0073] ;

[0074] Among them, respectively contain the node features belonging to the corresponding levels. At the same time, for the adjacency matrix the connection relationships in are distinguished. For any two nodes and , if they belong to the same level and meet the connection conditions (such as the Euclidean distance is less than the intra-layer threshold), then their connection belongs to an intra-layer connection; otherwise, if they belong to different levels and meet the cross-layer connection conditions, then their connection belongs to an inter-layer connection. The intra-layer connection set and the inter-layer connection set are defined as:

[0075] ;

[0076] ;

[0077] Perform matrix multiplication on the hierarchical grouping matrix and the first convolution kernel matrix to extract the initial feature transformation results of each layer:

[0078] ;

[0079] Among them, is the convolution kernel matrix for feature transformation, is the intra-layer initial feature matrix. For each node feature vector in , calculate its attention coefficient to highlight the importance of key nodes. The attention coefficient is calculated based on the node temperature gradient and topological distance:

[0080] ;

[0081] Among them, is the learned attention vector, and are the feature vectors of node and node , || represents the vector concatenation operation. Through all construct the attention weight matrix . Combine the attention weight matrix with the intra-layer initial feature matrix through weighted combination, and perform graph convolution operation in combination with the intra-layer connection edge set to generate the intra-layer spatial feature matrix:

[0082] ;

[0083] Among them, is the convolution kernel weight, is the activation function. Convolve the node features in the in-layer spatial feature matrix with the second convolution kernel matrix and perform feature transfer with the set of inter-layer connection edges to generate an inter-layer correlation feature matrix:

[0084] ;

[0085] Concatenate the features of the in-layer spatial feature matrix and the inter-layer correlation feature matrix:

[0086] ;

[0087] Compress the concatenated features through a fully connected layer to generate the final temperature spatial distribution feature vector:

[0088] ;

[0089] Among them, and are the weights and biases of the fully connected layer. Perform temporal fusion on the temperature spatial distribution feature vector with the SF6 gas pressure data and the environmental humidity data and adopt a time attention mechanism to extract dynamic features:

[0090] ;

[0091] Through this process, generate a temperature dynamic change temporal feature vector that comprehensively describes the temperature distribution and dynamic changes.

[0092] In a specific embodiment, the process of performing temporal fusion on the temperature spatial distribution feature vector with the SF6 gas pressure data and the environmental humidity data to obtain the temperature dynamic change temporal feature vector may specifically include the following steps:

[0093] Perform numerical normalization on the temperature spatial distribution feature vector and the SF6 gas pressure data to obtain a temperature-pressure temporal matrix, and align the temperature-pressure temporal matrix with the environmental humidity data to obtain a multi-parameter temporal data matrix;

[0094] Calculate the historical information ratio of the multi-parameter temporal data matrix through a state update gate to obtain a state update matrix;

[0095] Calculate the parameter influence degree of the multi-parameter temporal data matrix through a parameter correlation gate, and generate a parameter correlation matrix according to the coupling relationship between the SF6 gas pressure and temperature and the correlation degree between the environmental humidity and temperature;

[0096] Perform matrix multiplication on the state update matrix and the parameter correlation matrix, and perform feature selection through the output gate to obtain the fused feature matrix;

[0097] Perform dynamic time attention calculation on the time series data in the fused feature matrix, and generate a time series weighted matrix according to the device start-stop state and load changes;

[0098] Perform weighted superposition operation on the fused feature matrix and the time series weighted matrix to obtain the initial dynamic feature vector, and perform time series feature extraction on the initial dynamic feature vector through the gated recurrent unit to obtain the temperature dynamic change time series feature vector.

[0099] Specifically, the temperature spatial distribution feature vector and the SF6 gas pressure data respectively represent the spatial distribution characteristics of the internal temperature of the device and the change trend of the SF6 gas pressure. Assume that the temperature feature and the gas pressure data are vectors at corresponding moments, and record the parameter changes at time steps respectively. To eliminate the influence of different dimensions on data analysis, numerical normalization is performed on these two vectors. The normalization process uses the following formula:

[0100] ;

[0101] Among them, and are the normalized temperature value and pressure value at the th time step respectively, and respectively represent the minimum and maximum values of temperature and pressure. After normalization, the two normalized vectors are combined to form a temperature-pressure time series matrix:

[0102] ;

[0103] Align the environmental humidity data with the temperature-pressure time series matrix to form a multi-parameter time series data matrix. The data alignment is achieved by matching the sampling points at each time step to ensure and the humidity data are synchronized in the time dimension. The aligned multi-parameter time series data matrix is expressed as:

[0104] ;

[0105] Calculate the proportion of historical information for the multi-parameter time series data matrix through the state update gate to generate the state update matrix . The state update gate controls the influence of the historical state on the current time step through a non-linear transformation, and its calculation formula is:

[0106] ;

[0107] Among them, is the weight matrix of the state update gate, is the bias vector, is the activation function (such as Sigmoid). Each element of represents the influence ratio of historical information on the current moment parameters. At the same time, through the parameter correlation gate, the parameter influence degree calculation is carried out on the multi-parameter time series data matrix . According to the thermodynamic coupling relationship between SF6 gas pressure and temperature, and the correlation degree of condensation risk between environmental humidity and temperature, a parameter correlation matrix

[0108] ;

[0109] Among them, and are the weights and biases of the parameter correlation gate, and the softmax function is used to map the result to a probability distribution to quantify the contribution of different parameters to the temperature rise change. Multiply the state update matrix with the parameter correlation matrix by matrix multiplication, and perform feature selection on the result through the output gate to generate a fused feature matrix :

[0110] ;

[0111] Among them, · represents element-wise multiplication. Perform dynamic time attention calculation on the time series data in the fused feature matrix to capture the dynamic influence of device start-stop state and load change on temperature rise. The calculation formula of the dynamic time attention weight matrix is:

[0112] ;

[0113] Among them, and are the weights and biases of the time attention module. By performing weighted superposition operation on the dynamic time attention weight matrix and the fused feature matrix, the initial dynamic feature vector is obtained:

[0114] ;

[0115] Input the initial dynamic feature vector into the gated recurrent unit (GRU) to extract time series features, and obtain the final temperature dynamic change time series feature vector . The update formula of GRU is:

[0116] ;

[0117] Among them, is the update gate, is the candidate hidden state, is the hidden state at the previous moment, represents element-wise multiplication.

[0118] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0119] Perform feature fusion on the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector in the spatial and temporal dimensions to obtain a spatio-temporal feature matrix, and perform hierarchical dimensionality reduction processing on the spatio-temporal feature matrix to obtain a region classification feature vector;

[0120] Input the region classification feature vector into the classification network, divide the core heating parts of the circuit breaker contacts and disconnector into key areas, the positions of the conductor joints and bushing joints into transition areas, and the parts of the housing and support structure into ordinary areas to obtain a region category label matrix;

[0121] Construct a reinforcement learning training environment for the region category label matrix, set the state space and action space according to the temperature control requirements of each partition to obtain reinforcement learning environment parameters;

[0122] Input the reinforcement learning environment parameters into the reinforcement learning agent, explore and sample the temperature control actions for each partition to obtain an action sample set, and construct a reward function according to the temperature control accuracy, response speed, temperature uniformity, and long-term stability to obtain a partition reward value;

[0123] Perform policy network training on the action sample set and the partition reward value, update the network parameters using the policy gradient algorithm to obtain a policy network model for each partition;

[0124] Input the policy network model into the temperature control accuracy evaluator, response speed evaluator, and stability evaluator respectively to obtain control performance evaluation results, and fine-tune the policy network parameters according to the control performance evaluation results to obtain an optimized policy network model;

[0125] Generate a temperature control action sequence according to the optimized policy network model, and convert the temperature control action sequence into adaptive control parameters for each partition.

[0126] Specifically, the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector respectively represent the static distribution characteristics and time variation laws of the temperature inside the device. Assume is a -dimensional vector matrix, where is the number of time steps, is the dimension of spatial features. To fuse these two types of features in the spatial and temporal dimensions, and are feature-aligned and concatenated to generate a spatio-temporal feature matrix :

[0127] ;

[0128] where concat represents the feature concatenation operation, and the result is a matrix. The first row is the spatial feature, and the remaining rows are the temporal features. Hierarchical dimensionality reduction is performed on the spatio-temporal feature matrix to reduce redundant information and extract key features. The dimensionality reduction is achieved through a linear transformation, and its formula is:

[0129] ;

[0130] where, is the dimensionality reduction weight matrix, is the bias vector, is the activation function (such as ReLU). The output is a low-dimensional matrix containing the key features after spatio-temporal fusion. The dimensionality-reduced feature matrix is input into a classification network, and the regions are classified according to the device structure characteristics and temperature distribution rules. Assume that the output of the classification network is the region classification feature vector , and its formula is:

[0131] ;

[0132] where, and are the weights and biases of the classification network, and the softmax function maps the output to a probability distribution. According to the classification results, the core heating parts such as circuit breaker contacts and disconnectors in the device are classified as key areas, the positions such as conductor connections and bushing joints are classified as transition areas, and the housing and support structures are classified as ordinary areas, forming a region category label matrix . After the region classification is completed, the region category label matrix is used to construct the reinforcement learning training environment. The core of reinforcement learning lies in setting a reasonable state space and action space . The state space contains parameters such as temperature, pressure, and humidity of each partition, and the action space is defined as a set of control actions such as adjusting radiator switches and gas flow rates. The reinforcement learning environment parameter is expressed as:

[0133] ;

[0134] Among them, is the reward function, which is used to evaluate the quality of actions. The reinforcement learning agent interacts with the environment to explore and sample the temperature control actions of each partition, generating an action sample set D. At the same time, a reward function is constructed based on temperature control accuracy, response speed, temperature uniformity, and long-term stability:

[0135] ;

[0136] Among them, is the reward value at the th step, respectively represent the control accuracy, response speed, temperature uniformity, and stability scores, are the weights. Based on the action sample set and the reward value , the policy network is trained using the policy gradient algorithm to update the network parameters :

[0137] ;

[0138] Among them, is the learning rate, is the probability that the policy network selects action under state . After training is completed, the policy network model is input into the temperature control accuracy evaluator, response speed evaluator, and stability evaluator respectively to evaluate the control performance. Assuming the evaluation result is , by further fine-tuning the policy network parameters, an optimized policy network model is generated. Based on the optimized policy network model, a temperature control action sequence is generated:

[0139] ;

[0140] Among them, is the optimized policy network model. The action sequence is converted into adaptive control parameters for each partition to precisely regulate the temperature rise of the GIS device.

[0141] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0142] Input the adaptive control parameters into the multi-level compensation controller, and perform hierarchical processing on the control parameters according to the hierarchical attributes of the controller to obtain a set of hierarchical control parameters;

[0143] Input gas pressure and temperature data into the SF6 gas pressure-temperature coupling model, and perform correlation analysis according to the coupling relationship between gas pressure and temperature to obtain the temperature compensation coefficient of the gas chamber layer;

[0144] Input the load current data of the high-voltage conductor layer into the load current heat accumulation model, and perform heat accumulation calculation according to the relationship between the current intensity and the conductor temperature rise to obtain the conductor layer temperature compensation coefficient;

[0145] Perform humidity distribution calculation on the data collected by the environmental humidity sensor to obtain the humidity distribution calculation result, and input the humidity distribution calculation result into the environmental humidity condensation model to perform condensation risk analysis according to the relationship between humidity and surface temperature to obtain the enclosure layer temperature compensation coefficient;

[0146] Perform parameter matching according to the hierarchical control parameter set and the corresponding relationship of the temperature compensation coefficients of each layer to obtain the compensation control matrix;

[0147] Perform stability constraint processing on the parameters in the compensation control matrix, determine the parameter correction range according to the temperature change limit values of each level to obtain the constraint correction matrix;

[0148] Input the constraint correction matrix into the temperature compensation calculation module, and perform temperature correction amount calculation according to the compensation coefficients of each level to obtain the temperature correction parameters of each level.

[0149] Specifically, the adaptive control parameter is a comprehensive adjustment parameter set obtained through previous analysis, including the regulation requirements of the gas chamber layer, the high-voltage conductor layer, and the enclosure layer, and its form is expressed as:

[0150] ;

[0151] Among them, respectively correspond to the control parameter sets of the gas chamber layer, the conductor layer, and the enclosure layer. In the multi-level compensation controller, according to the hierarchical attributes of the controller, the input control parameters are separated and grouped according to the hierarchical attributes to generate a hierarchical control parameter set. Define the control parameter vector of each level as:

[0152] ;

[0153] Each element in represents a control requirement for a specific level. For example, the parameters of the gas chamber layer include the gas flow adjustment coefficient and the valve adjustment coefficient, the parameters of the high-voltage conductor layer involve the current distribution ratio, and the enclosure layer includes the radiator opening parameter and the enclosure heat conduction efficiency adjustment coefficient. For the gas chamber layer, use the SF6 gas pressure-temperature coupling model to perform correlation analysis of gas pressure and temperature. The gas pressure data and the temperature data are input into the coupling model, and the basic formula of the model is:

[0154] ;

[0155] Among them, represents the temperature change of the gas chamber layer, is the sensitivity coefficient of gas pressure to temperature change, is the actual gas pressure, is the reference gas pressure value. Through analysis, is converted into the temperature compensation coefficient of the gas chamber layer , which is used to correct the temperature rise behavior of the gas chamber layer. For the high-voltage conductor layer, the load current data is input into the load current heat accumulation model to calculate the influence of current intensity on the temperature rise of the conductor layer. The heat accumulation formula of the model is:

[0156] ;

[0157] Among them, is the heat accumulation amount, is the resistance value of the conductor, is the load current, is the time. Based on the calculation result, the temperature change of the conductor layer is expressed as:

[0158] ;

[0159] Among them, is the heat capacity of the conductor. Through the result, the temperature compensation coefficient of the conductor layer is generated. For the outer shell layer, the environmental humidity data is input into the humidity distribution calculation module to calculate the distribution characteristics of humidity on the outer shell surface. The result of the humidity distribution calculation is combined with the environmental humidity condensation model. The basic formula for condensation risk analysis is:

[0160] ;

[0161] Among them, is the condensation risk factor, is the environmental humidity, is the dew point temperature, is the outer shell surface temperature. Through the calculation result of the condensation risk factor, the temperature compensation coefficient of the outer shell layer is generated. Combining the temperature compensation coefficients of the above three levels, according to the correspondence between the hierarchical control parameter set and the temperature compensation coefficient, a compensation control matrix is generated:

[0162] ;

[0163] Among them, the diagonal elements of the compensation control matrix represent the temperature compensation coefficients of each layer. To ensure the stability of the compensation parameters, for Perform stability constraint processing on the parameters in it. Based on the temperature change limit values at each level, determine the parameter correction range:

[0164] ;

[0165] Wherein, and respectively represent the minimum and maximum allowable values of the compensation coefficient, and is the constraint correction matrix. Input the constraint correction matrix into the temperature compensation calculation module, and calculate the temperature correction amount according to the temperature compensation coefficients at each level:

[0166] ;

[0167] Wherein, represents the final temperature correction parameter of each layer.

[0168] In a specific embodiment, the method for monitoring the temperature rise of a high-voltage fully enclosed vacuum environmental protection type GIS device further includes the following steps:

[0169] Perform feature combination and data standardization processing on the temperature correction parameter, the temperature spatial distribution feature vector, and the temperature dynamic change time series feature vector to obtain a standardized feature matrix;

[0170] Perform temperature anomaly detection on the standardized feature matrix for the SF6 gas chamber layer, and generate a gas chamber layer risk level matrix according to the comparison and analysis of the gas temperature deviation value and the insulation breakdown threshold;

[0171] Perform temperature rise detection on the standardized feature matrix for the high-voltage conductor layer, and generate a conductor layer risk level matrix according to the temperature gradient calculation of the circuit breaker contact, the disconnector, and the bushing joint;

[0172] Perform temperature uniformity detection on the standardized feature matrix for the outer shell enclosure layer, and generate an outer shell layer risk level matrix according to the calculation of the surface temperature distribution and the environmental temperature difference;

[0173] Input the gas chamber layer risk level matrix, the conductor layer risk level matrix, and the outer shell layer risk level matrix into a multi-layer fully connected neural network, and through linear transformation and non-linear activation function processing, obtain a hierarchical risk feature vector;

[0174] Perform Softmax classification on the hierarchical risk feature vector, and perform multi-level early warning calculation according to the preset temperature rise risk thresholds for the gas chamber layer, the conductor layer, and the outer shell layer to obtain an early warning probability matrix;

[0175] Perform correlation analysis on the early warning data in the early warning probability matrix and the device operation state data, and output corresponding hierarchical early warning signals according to the temperature rise rate and the duration.

[0176] Specifically, the temperature correction parameter, the temperature spatial distribution eigenvector, and the temperature dynamic change time series eigenvector are subjected to feature combination and data standardization processing to generate a standardized feature matrix. Assume that the temperature correction parameter is , the temperature spatial distribution eigenvector is , and the temperature dynamic change time series eigenvector is , which is a matrix, where is the number of time steps, is the feature dimension. Concatenate with and to form a combined feature matrix:

[0177] ;

[0178] where concat represents the feature concatenation operation. To ensure the comparability of different data sources, perform normalization on , map each eigenvalue to the [0, 1] interval, and the normalization formula is:

[0179] ;

[0180] where is the normalized eigenvalue, is the original value, and are the minimum and maximum values of the column features respectively. Input the standardized feature matrix into the temperature anomaly detection module to analyze the temperature of the SF6 gas chamber layer. Assume that the temperature data of the SF6 gas chamber layer is , and calculate the temperature deviation value of each monitoring point:

[0181] ;

[0182] where is the temperature deviation value of the th node, is the reference temperature threshold. Compare the deviation value with the insulation breakdown threshold :

[0183] ;

[0184] Generate the gas chamber layer risk level matrix , and each element reflects the risk level of that node. For the high-voltage conductor layer, analyze the temperature rise of the circuit breaker contacts, disconnector switches, and bushing joints. Assume that the temperature data of the conductor layer is , and calculate the temperature gradient:

[0185] ;

[0186] Among them, is the temperature gradient of the th node. The risk level is divided according to the gradient magnitude:

[0187] ;

[0188] Generate the risk level matrix of the conductor layer . The temperature uniformity of the enclosure layer is calculated and analyzed through the surface temperature distribution and the temperature difference from the environment. Assume that the surface temperature of the enclosure is , and the ambient temperature is . Define the temperature difference as:

[0189] ;

[0190] Calculate the standard deviation of the temperature difference to represent the uniformity, and divide the risk level according to :

[0191] ;

[0192] Generate the risk level matrix of the enclosure layer . Input the three risk level matrices , , into a multi-layer fully connected neural network, and use linear transformation and non-linear activation functions to extract hierarchical risk features:

[0193] ;

[0194] Among them, and are the weights and biases of the neural network, is the activation function. Perform Softmax classification on the hierarchical risk feature vector to calculate the probabilities of each risk level:

[0195] ;

[0196] Generate the early warning probability matrix . Correlate with the device operating state data (such as the temperature rise rate and duration), and output a graded early warning signal.

[0197] The above describes the temperature rise monitoring method for the high-voltage fully enclosed vacuum environmental protection type GIS device in the embodiment of the present application. Next, the temperature rise monitoring system 10 for the high-voltage fully enclosed vacuum environmental protection type GIS device in the embodiment of the present application will be described. Please refer to Figure 2, in one embodiment, the temperature rise monitoring system 10 of the high-voltage fully enclosed vacuum environmental protection type GIS device in the embodiment of the present application includes:

[0198] An acquisition module 11, configured to collect temperature data through a temperature sensor, and construct a node feature matrix and an adjacency matrix according to the spatial position relationship of the SF6 gas chamber layer, the high-voltage conductor layer, and the outer shell enclosure layer;

[0199] An operation module 12, configured to perform intra-layer graph convolution operation and inter-layer graph convolution operation on the node feature matrix and the adjacency matrix, obtain a temperature spatial distribution feature vector, and perform temporal fusion to obtain a temperature dynamic change temporal feature vector;

[0200] A partitioning module 13, configured to partition the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change temporal feature vector to obtain adaptive control parameters for each partition;

[0201] A compensation module 14, configured to input the adaptive control parameters into a multi-level compensation controller to calculate temperature compensation coefficients, and obtain temperature correction parameters for each level.

[0202] Through the collaborative cooperation of the above-mentioned various components, by constructing a three-layer network topology structure, the systematic management of the temperature data of the SF6 gas chamber layer, the high-voltage conductor layer, and the outer shell enclosure layer is realized, and the integrity and accuracy of the extraction of the temperature field distribution characteristics are improved; the combination of intra-layer graph convolution and inter-layer graph convolution effectively captures the temperature correlation between different levels and enhances the extraction ability of the temperature field spatial distribution characteristics; the temporal fusion mechanism combining the state update gate and the parameter correlation gate realizes the dynamic correlation analysis of temperature data with multiple parameters such as SF6 gas pressure and environmental humidity; through the partition reinforcement learning control strategy, different control parameters are adopted for different regions, improving the accuracy of temperature control; the multi-level compensation control mechanism is adopted, considering the coupling effects of gas pressure, load current, and environmental humidity, improving the accuracy of temperature correction; the risk warning mechanism based on the multi-layer fully connected neural network realizes the accurate assessment and hierarchical warning of the temperature rise risks at different levels.

[0203] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0204] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0205] As described above, the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A temperature rise monitoring method for high-voltage fully enclosed vacuum environmentally friendly GIS equipment, characterized in that: The method comprises: The temperature data is collected by temperature sensors, and a node feature matrix and an adjacency matrix are constructed according to the spatial position relationship among the SF6 gas chamber layer, the high-voltage conductor layer and the shell closed layer; specifically, the following steps are performed: the temperature sensor data collected at the monitoring point inside the gas chamber in the SF6 gas chamber layer are paired with the spatial position coordinates of the sensor to obtain a gas chamber layer sensor node set; the temperature sensor data collected at the circuit breaker contacts, disconnectors and bushing joints in the high-voltage conductor layer are paired with the spatial position coordinates of the sensor to obtain a conductor layer sensor node set; the temperature sensor data collected at the equipment casing and the supporting structure in the shell closed layer are paired with the spatial position coordinates of the sensor to obtain a shell layer sensor node set; the gas chamber layer sensor node set and the conductor layer sensor node set are paired. The set and the shell layer sensor node set are subjected to spatial coordinate system unification and temperature data normalization processing to obtain a standardized sensor node matrix; a node feature matrix is ​​constructed and hierarchical attribute marking is performed based on the temperature data in the standardized sensor node matrix to obtain a node feature matrix containing hierarchical information; the three-dimensional Euclidean distance between any two sensor nodes is calculated according to the spatial coordinate information in the standardized sensor node matrix to obtain a node distance matrix, and the distance values ​​in the node distance matrix are compared and judged with the preset intra-layer connection threshold and inter-layer connection threshold to generate a binary adjacency matrix; the binary adjacency matrix is ​​associated with the node hierarchical information, and the intra-layer connection edges and inter-layer connection edges are marked in the binary adjacency matrix to obtain an adjacency matrix; Performing intra-layer graph convolution operations and inter-layer graph convolution operations on the node feature matrix and the adjacency matrix to obtain a temperature spatial distribution feature vector, and performing time series fusion to obtain a temperature dynamic change time series feature vector; Partitioning the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector to obtain adaptive control parameters for each partition; The adaptive control parameters are input into the multi-level compensation controller to calculate the temperature compensation coefficient, so as to obtain the temperature correction parameters of each level.

2. The temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The intra-layer graph convolution operation and the inter-layer graph convolution operation are performed on the node feature matrix and the adjacency matrix to obtain the temperature spatial distribution feature vector, and the time series fusion is performed to obtain the temperature dynamic change time series feature vector, including: Normalizing and hierarchically marking the temperature data of each sensor node in the node feature matrix, dividing the sensor nodes into an air chamber layer node group, a conductor layer node group and an outer shell layer node group according to the hierarchical information to which the sensor nodes belong, and obtaining a hierarchical grouping matrix; The connection relationship in the adjacency matrix is ​​distinguished, and the connection edges are divided into an intra-layer connection edge set and an inter-layer connection edge set according to the layer labels between the sensor nodes, so as to obtain a connection edge classification matrix; Performing a matrix multiplication operation on the hierarchical grouping matrix and the first convolution kernel matrix to obtain an operation result, and performing ReLU activation function processing on the operation result to obtain an initial feature matrix within the layer; Calculating the attention coefficient for each node feature vector in the initial feature matrix in the layer, generating a node weight coefficient according to the node temperature gradient and the topological distance, and obtaining an attention weight matrix; The attention weight matrix is ​​weightedly combined with the initial feature matrix in the layer, and a graph convolution operation is performed with the set of connection edges in the layer to obtain a spatial feature matrix in the layer; Performing convolution operation on feature vectors between nodes at different levels in the intra-layer spatial feature matrix and a second convolution kernel matrix, and performing feature transfer with the inter-layer connection edge set to obtain an inter-layer correlation feature matrix; Performing feature concatenation on the intra-layer spatial feature matrix and the inter-layer correlation feature matrix, and performing feature compression through a fully connected layer to obtain a temperature spatial distribution feature vector; The temperature spatial distribution characteristic vector is time-series fused with the SF6 gas pressure data and the ambient humidity data to obtain a temperature dynamic change time-series characteristic vector.

3. The temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment according to claim 2 is characterized in that: The time series fusion of the temperature spatial distribution feature vector with the SF6 gas pressure data and the ambient humidity data to obtain the temperature dynamic change time series feature vector includes: The temperature spatial distribution characteristic vector and the SF6 gas pressure data are numerically normalized to obtain a temperature and pressure time series matrix, and the temperature and pressure time series matrix is ​​data aligned with the ambient humidity data to obtain a multi-parameter time series data matrix; Perform historical information ratio calculation on the multi-parameter time series data matrix through a state update gate to obtain a state update matrix; Calculating the parameter influence of the multi-parameter time series data matrix through a parameter association gate, and generating a parameter association matrix according to the coupling relationship between SF6 gas pressure and temperature and the degree of association between ambient humidity and temperature; Performing matrix multiplication operation on the state update matrix and the parameter association matrix, and performing feature selection through an output gate to obtain a fused feature matrix; Performing dynamic time attention calculation on the time series data in the fusion feature matrix, and generating a time series weighted matrix according to the start / stop state of the device and the load change; The fusion feature matrix and the timing weighted matrix are weighted superimposed to obtain an initial dynamic feature vector, and the timing feature of the initial dynamic feature vector is extracted through a gated cycle unit to obtain a temperature dynamic change timing feature vector.

4. The temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The partitioning of the monitoring area according to the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector to obtain the adaptive control parameters of each partition includes: Performing feature fusion in spatial and temporal dimensions according to the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector to obtain a spatiotemporal feature matrix, and performing hierarchical dimensionality reduction processing on the spatiotemporal feature matrix to obtain a regional classification feature vector; The regional classification feature vector is input into the classification network, the core heating parts of the circuit breaker contacts and disconnectors are divided into key areas, the conductor connection and the bushing joint are divided into transition areas, and the housing and the supporting structure are divided into ordinary areas, so as to obtain a regional category label matrix; Constructing a reinforcement learning training environment for the regional category label matrix, setting a state space and an action space according to the temperature control requirements of each partition, and obtaining reinforcement learning environment parameters; Input the reinforcement learning environment parameters into the reinforcement learning agent, perform exploration sampling on the temperature control action of each partition to obtain an action sample set, and construct a reward function according to the temperature control accuracy, response speed, temperature uniformity and long-term stability to obtain a partition reward value; Performing policy network training on the action sample set and the partition reward value, and updating network parameters using a policy gradient algorithm to obtain a policy network model for each partition; The strategy network model is respectively input into a temperature control accuracy evaluator, a response speed evaluator and a stability evaluator to obtain a control performance evaluation result, and the strategy network parameters are fine-tuned according to the control performance evaluation result to obtain an optimized strategy network model; A temperature control action sequence is generated according to the optimized policy network model, and the temperature control action sequence is converted into adaptive control parameters of each partition.

5. The temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The step of inputting the adaptive control parameters into the multi-level compensation controller to calculate the temperature compensation coefficient to obtain the temperature correction parameters of each level includes: Inputting the adaptive control parameters into a multi-level compensation controller, performing hierarchical processing on the control parameters according to hierarchical attributes of the controller, and obtaining a hierarchical control parameter set; The gas pressure and temperature data are input into the SF6 gas pressure-temperature coupling model, and a correlation analysis is performed based on the coupling relationship between gas pressure and temperature to obtain the gas chamber layer temperature compensation coefficient; The load current data of the high-voltage conductor layer is input into the load current heat accumulation model, and the heat accumulation calculation is performed according to the relationship between the current intensity and the conductor temperature rise to obtain the conductor layer temperature compensation coefficient; Performing humidity distribution calculation on the data collected by the environmental humidity sensor to obtain a humidity distribution calculation result, and inputting the humidity distribution calculation result into the environmental humidity condensation model, performing condensation risk analysis based on the relationship between humidity and surface temperature, and obtaining an outer shell temperature compensation coefficient; Performing parameter matching according to the correspondence between the hierarchical control parameter set and the temperature compensation coefficients of each layer to obtain a compensation control matrix; Performing stability constraint processing on the parameters in the compensation control matrix, determining the parameter correction range according to the temperature change limit of each level, and obtaining a constraint correction matrix; The constraint correction matrix is ​​input into the temperature compensation calculation module, and the temperature correction amount is calculated according to the compensation coefficient of each level to obtain the temperature correction parameters of each level.

6. The temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The method of calculating the temperature rise risk probability at different levels according to the temperature correction parameter, the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector, and outputting a graded warning signal includes: Performing feature combination and data standardization processing on the temperature correction parameter, the temperature spatial distribution feature vector and the temperature dynamic change time series feature vector to obtain a standardized feature matrix; Performing temperature anomaly detection on the SF6 gas chamber layer on the standardized characteristic matrix, and generating a gas chamber layer risk level matrix based on comparative analysis of the gas temperature deviation value and the insulation breakdown threshold; Performing temperature rise detection on the high-voltage conductor layer of the standardized characteristic matrix, and generating a conductor layer risk level matrix according to temperature gradient calculation of the circuit breaker contacts, disconnectors and bushing joints; Performing temperature uniformity detection on the enclosure sealing layer of the standardized characteristic matrix, and generating an enclosure layer risk level matrix based on calculation of surface temperature distribution and ambient temperature difference; Inputting the air chamber layer risk level matrix, the conductor layer risk level matrix and the shell layer risk level matrix into a multi-layer fully connected neural network, and processing them through linear transformation and nonlinear activation function to obtain a hierarchical risk feature vector; Performing Softmax classification on the hierarchical risk feature vector, performing multi-level warning calculation according to preset temperature rise risk thresholds of the air chamber layer, the conductor layer and the outer shell layer, and obtaining a warning probability matrix; The warning data in the warning probability matrix are correlated with the equipment operation status data and analyzed, and corresponding graded warning signals are output according to the temperature rise rate and duration.

7. A temperature rise monitoring system for high-voltage, fully enclosed, vacuum, environmentally friendly GIS equipment, characterized in that: Used to implement the temperature rise monitoring method of high-voltage fully enclosed vacuum environmentally friendly GIS equipment as claimed in claim 1, the temperature rise monitoring system of the high-voltage fully enclosed vacuum environmentally friendly GIS equipment comprises: The acquisition module is used to collect temperature data through a temperature sensor and construct a node feature matrix and an adjacency matrix according to the spatial position relationship between the SF6 gas chamber layer, the high-voltage conductor layer and the shell sealing layer; An operation module, used for performing intra-layer graph convolution operation and inter-layer graph convolution operation on the node feature matrix and the adjacency matrix to obtain a temperature spatial distribution feature vector, and performing time series fusion to obtain a temperature dynamic change time series feature vector; A partitioning module, used to partition the monitoring area according to the temperature spatial distribution characteristic vector and the temperature dynamic change time series characteristic vector, and obtain adaptive control parameters of each partition; The compensation module is used to input the adaptive control parameters into the multi-level compensation controller to calculate the temperature compensation coefficient and obtain the temperature correction parameters of each level.

Images