A cloud-based intelligent diagnostic system for switchgear status

Abstract

This invention relates to the field of intelligent monitoring and fault diagnosis technology for power equipment, specifically a cloud-based intelligent diagnostic system for switchgear status. It includes units for multi-source topology mapping, physical constraint embedding, global state reconstruction, residual decoupling diagnosis, and adaptive parameter correction. The system maps sensor data into a dynamic graph structure and generates constraint vectors using physical conservation logic. Its core is to correct the state based on physical constraints to reconstruct the state graph, locate faults by calculating the residual sequence between measured data and the reconstructed graph, and establish feedback closed-loop correction parameters. This invention achieves zero-sample diagnosis without relying on a large number of samples, solving the problem of pure data-driven solutions failing due to a lack of fault samples.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and fault diagnosis technology for power equipment, specifically to an intelligent diagnostic system for switchgear status based on a cloud platform. Background Technology

[0002] In the current power system switchgear operation and maintenance environment, the equipment involves key heat points such as plum blossom contacts and busbar lap surfaces. The operation process is accompanied by complex electromagnetic thermodynamic coupling effects and generates a large amount of multi-source heterogeneous sensor data. To monitor the equipment status, existing solutions generally adopt pure data-driven deep learning models, that is, directly using historical monitoring data to train neural networks to fit fault characteristics. Although this solution has a certain recognition ability when the sample is sufficient, the scarcity of fault samples in actual operation makes it difficult for the model to diagnose unknown long-tail faults. Moreover, due to the lack of constraints of physical conservation laws, the model is prone to output false states that violate physical common sense under non-steady-state conditions such as drastic load changes. In addition, the fixed parameter model cannot perceive the physical parameter drift caused by equipment aging, resulting in weak anti-interference ability, high false alarm rate, and significant degradation of diagnostic accuracy as the equipment operation cycle extends.

[0003] Therefore, how to establish effective supervision of data models by physical mechanisms and achieve highly robust and accurate diagnosis and full life cycle adaptation under zero-sample conditions has become an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a cloud-based intelligent diagnostic system for switchgear status. Specifically, the technical solution of this invention includes:

[0005] The multi-source topology mapping unit is configured to receive heterogeneous sensor data from the edge end and static design parameters of the switch cabinet; map key heat points and sensor locations as graph nodes; map heat conduction paths and electromagnetic coupling paths as graph edges; and generate dynamic graph structure data containing real-time node features and edge connection relationships.

[0006] The physical constraint embedding unit is configured to monitor the dynamic graph structure data in real time; calculate the gradient of each graph node change based on the current load input using preset physical conservation logic; and generate a physical mechanism constraint vector representing the strength of physical rules.

[0007] The global state reconstruction unit is configured to receive the dynamic graph structure data and the physical mechanism constraint vector; process the dynamic graph structure data using an optimization solution layer embedded with physical constraints; construct a penalty function based on the physical mechanism constraint vector; dynamically correct the state values ​​output by the graph neural network; and generate a reconstructed state map.

[0008] The residual decoupling diagnostic unit is configured to receive the measured heterogeneous sensor data and the reconstructed state map; calculate the difference between the two to generate a state residual sequence; use dynamic threshold judgment logic to distinguish between noise and fault; and output the fault location result.

[0009] An adaptive parameter correction unit is configured to establish a feedback closed loop; track the evolution trend of the state residual sequence through a time-series historical database; generate a parameter correction signal when a systematic deviation is detected; and feed the parameter correction signal back to the physical constraint embedding unit to update the key coefficients in the physical conservation logic.

[0010] Preferably, the configuration rules for the multi-source topology mapping unit are as follows:

[0011] The heterogeneous sensor data includes temperature, current, and partial discharge signals;

[0012] The static design parameters include the busbar layout structure and contact positions;

[0013] The key heating points include the plum blossom contact and the busbar lap surface.

[0014] Preferably, the configuration rules for the physical constraint embedding unit are as follows:

[0015] The physical conservation logic is set so that the Joule heat generated by the current is equal to the sum of heat absorbed by temperature rise, heat dissipated by convection, heat dissipated by radiation, and heat dissipated by conduction.

[0016] The physical mechanism constraint vector is transmitted to the global state reconstruction unit as a forced boundary condition in the computation process.

[0017] Preferably, the processing logic of the global state reconstruction unit is as follows:

[0018] Utilizing graph neural network information aggregation mechanisms;

[0019] Deducing the state of adjacent unknown nodes from known sensor nodes;

[0020] When the initial state values ​​generated by data fitting violate the physical rules in the physical mechanism constraint vector, the inherent weights of the neural network remain unchanged;

[0021] Initiate the iterative optimization algorithm;

[0022] The output state value is corrected based on the penalty function;

[0023] The process continues until the output conforms to the physical conservation law within the allowable error range, thus obtaining the reconstructed state map representing the theoretical physical field distribution inside the health switch cabinet.

[0024] Preferably, the dynamic threshold determination logic in the residual decoupling diagnostic unit is as follows:

[0025] The dynamic threshold is calculated based on the uncertainty range of the physical model under the current operating conditions.

[0026] When the value in the state residual sequence exceeds the dynamic threshold and exhibits specific spatiotemporal distribution characteristics, a fault is determined to exist;

[0027] The specific spatiotemporal distribution characteristics include the continuous accumulation of heat at a node that deviates from the theoretical values ​​of the physical model.

[0028] The dynamic threshold automatically expands the allowable error range when the load fluctuates drastically.

[0029] Preferably, the adaptive parameter correction unit generates the parameter correction signal as follows:

[0030] Identify whether there are long-term, small, and stable systematic biases in the state residual sequence;

[0031] When the systematic deviation is detected, the parameter correction signal is generated and fed back.

[0032] The physical constraint embedding unit updates the heat dissipation coefficient or basic resistance value according to the parameter correction signal to adapt to the physical parameter drift caused by equipment aging.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. This system enforces physical constraints on the data-driven model and uses the reverse logic of reconstructing the normal state and detecting residuals. It does not require a large number of fault samples for training, thus achieving zero-sample diagnosis of unknown long-tail faults and solving the defect of existing pure data-driven solutions that fail due to the lack of fault samples.

[0035] 2. By making the thermal balance equation a hard constraint in advance, this system gives the black box model physical interpretability, which can effectively distinguish between normal temperature rise caused by load fluctuations that conform to physical laws and abnormal temperature rise caused by contact resistance deterioration. It eliminates the physical illusion caused by overfitting in pure data models and ensures the authenticity and reliability of the output results under non-steady-state conditions such as drastic load changes.

[0036] 3. This system has constructed a long-cycle parameter evolution closed-loop mechanism, which can identify systematic deviations in the state residual sequence that represent the natural aging trajectory of the physical attributes of the equipment, and dynamically correct the heat dissipation coefficient or basic resistance value in the physical conservation logic accordingly. This realizes the automatic adaptation of the model to the aging state of the equipment and solves the technical problem of the accuracy of fixed parameter models declining as the equipment operating cycle extends.

[0037] 4. This system utilizes a dynamic threshold calculated based on the uncertainty range of a physical model and a joint judgment logic of spatiotemporal features to automatically expand the allowable error range to shield false signals when the load fluctuates drastically. At the same time, it captures early and weak faults by tracking heat accumulation characteristics that cannot be explained by physical formulas, thus achieving an optimal balance between high-sensitivity diagnosis and low false alarm rate. Attached Figure Description

[0038] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0039] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0041] Example 1:

[0042] Please see Figure 1 A cloud-based intelligent diagnostic system for switchgear status includes:

[0043] The multi-source topology mapping unit is configured to receive heterogeneous sensor data from the edge end and static design parameters of the switch cabinet; map key heat points and sensor locations as graph nodes; map heat conduction paths and electromagnetic coupling paths as graph edges; and generate dynamic graph structure data containing real-time node features and edge connection relationships.

[0044] The physical constraint embedding unit is configured to monitor dynamic graph structure data in real time; calculate the gradient of each graph node change based on the current load input using preset physical conservation logic; and generate physical mechanism constraint vectors representing the strength of physical rules.

[0045] The global state reconstruction unit is configured to receive dynamic graph structure data and physical mechanism constraint vectors; process the dynamic graph structure data using an optimization solution layer embedded with physical constraints; construct a penalty function based on the physical mechanism constraint vectors; dynamically correct the state values ​​output by the graph neural network; and generate a reconstructed state map.

[0046] The residual decoupling diagnostic unit is configured to receive measured heterogeneous sensor data and reconstructed state maps; calculate the difference between the two to generate a state residual sequence; use dynamic threshold judgment logic to distinguish noise from faults; and output fault location results.

[0047] The adaptive parameter correction unit is configured to establish a feedback closed loop; track the evolution trend of the state residual sequence through a time-series historical database; generate a parameter correction signal when a systematic deviation is detected; and feed the parameter correction signal back to the physical constraint embedding unit to update the key coefficients in the physical conservation logic.

[0048] The system constructs a computing architecture that deeply couples physical mechanisms with data-driven approaches;

[0049] The multi-source topology mapping unit serves as an interface for mapping physical entities to digital space, performing the task of structured data reorganization. This unit accesses heterogeneous sensor data from the edge acquisition terminal and the static design parameters of the equipment. Instead of directly performing numerical calculations, it executes graph construction logic: it instantiates the physical coordinates of key heat-generating entities and sensors within the switch cabinet as graph nodes, and instantiates the physical association paths determined by the thermodynamic conduction equation and the electromagnetic field coupling equation as graph edges. This process outputs dynamic graph structure data, which is a high-dimensional tensor that encapsulates the real-time state attributes of nodes and the topological connection relationships between nodes, providing various boundary conditions with spatial physical meaning for subsequent calculations.

[0050] The physical constraint embedding unit operates upstream of the data stream and is used to establish a regulatory mechanism for the evolution of data based on physical laws. This unit uses pre-set physical conservation logic to calculate the time-varying gradient that each node should follow in the theoretical physical field, based on the input current load data. By comparing this gradient with real-time data, the unit generates a physical mechanism constraint vector. This vector, as a quantified control signal, represents the severity of the physical laws' constraints on the changes in state variables at the current moment. Under steady-state load, this vector applies a highly weighted penalty coefficient to temperature abrupt changes, while dynamically reducing the constraint strength during drastic load changes.

[0051] The global state reconstruction unit receives dynamic graph structure data and physical mechanism constraint vectors, and performs state deduction using an optimization solution layer embedded with physical constraints. During the process of fitting the data by the graph neural network, once the initial state value output by the intermediate layer violates the boundary conditions set by the physical mechanism constraint vector, this unit immediately triggers the penalty function mechanism. At this time, the system locks the weight parameters of the neural network without adjustment, and instead starts an iterative optimization algorithm to directly fine-tune the global state value at the output end until the output result converges within the error band allowed by the physical conservation law, and finally generates a reconstructed state map. This map is not a simple interpolation result, but represents the theoretical physical field distribution that all nodes inside an ideal and healthy switchgear should present under the current working conditions.

[0052] The residual decoupling diagnostic unit performs differential analysis and fault determination tasks. This unit calculates the Euclidean distance or Manhattan distance between the measured sensor data and the reconstructed state map to generate a state residual sequence. The dynamic threshold determination logic integrated within this unit calculates the dynamic threshold in real time based on the uncertainty of the physical model under the current operating conditions. The system only identifies a fault when the value of the residual sequence exceeds the dynamic threshold and exhibits specific clustering characteristics in the spatiotemporal dimension.

[0053] The adaptive parameter correction unit constructs a long-term parameter evolution closed loop. This unit performs trend term analysis on the state residual sequence in the time-series historical database to separate the residuals of sudden failures from the residuals of trend drift. When the algorithm identifies a monotonic, small, and long-term systematic deviation in the residual sequence, it determines that it is a physical parameter drift caused by equipment aging, and then generates a parameter correction signal. This signal is fed back to the physical constraint embedding unit to fine-tune the key coefficients in the physical conservation logic, so as to realize the automatic adaptation of the model to the aging state of the equipment.

[0054] This embodiment addresses the shortcomings of existing technologies that fail when there are few fault samples by establishing physical constraints to enforce supervision of the data-driven model. The system utilizes the reverse logic of reconstructing the normal state and detecting residuals to achieve zero-sample diagnosis of unknown long-tail faults. The introduction of physical constraints enables the system to interpret environmental noise and real faults, improving anti-interference robustness while ensuring high sensitivity. The closed-loop parameter correction mechanism gives the system the ability to adaptively evolve throughout its entire life cycle, solving the technical problem of the accuracy of fixed parameter models declining with equipment aging.

[0055] Example 2:

[0056] The configuration rules for multi-source topology mapping units are as follows:

[0057] Heterogeneous sensor data includes temperature, current, and partial discharge signals;

[0058] Static design parameters include busbar layout structure and contact position;

[0059] Key hot spots include the plum blossom contacts and the busbar lap surfaces.

[0060] The multi-source topology mapping unit executes strict data type definitions and node mapping protocols; the data access module is configured to process three types of heterogeneous data streams in parallel: temperature scalar field data reflecting thermodynamic state, current vector data reflecting electromagnetic load level, and high-frequency pulse signals of partial discharge reflecting the state of insulating medium; during the graph construction phase, the static design parameters retrieved by the unit cover the three-dimensional busbar arrangement geometry and contact spatial coordinates inside the switchgear; based on the statistical distribution law of thermal faults, the system forcibly defines the plum blossom contacts and busbar lap surfaces as first-level critical heating points and maps them as core hub nodes in the graph structure;

[0061] This configuration rule clarifies the granularity and focus of mapping physical entities to digital twins; by explicitly mapping high-failure-rate components such as plum blossom contacts as graph nodes, it ensures that the diagnostic algorithm can concentrate computing resources on key physical components, significantly improving the spatial resolution and diagnostic efficiency for locating faults in the core conductive circuit.

[0062] Example 3:

[0063] The configuration rules for the physical constraint embedding unit are as follows:

[0064] The physical conservation logic is set so that the Joule heat generated by the electric current is equal to the sum of heat absorbed by the temperature rise, heat dissipated by convection, heat dissipated by radiation, and heat dissipated by conduction.

[0065] The physical mechanism constraint vector is used as a forced boundary condition in the computation process and is sent to the global state reconstruction unit.

[0066] The physical constraint embedded unit solidifies inviolable physical laws; the physical conservation logic is strictly defined as the thermal balance differential equation: that is, the Joule heat increment generated by the current flowing through the conductor per unit time must be strictly equal to the sum of the heat absorbed by the conductor material due to temperature rise, the heat dissipated to the environment by the conductor surface through convection and radiation, and the heat conducted by the node to adjacent potential nodes through the physical connection structure; the dimension of this physical mechanism constraint vector is consistent with the number of graph nodes, and each element in the vector corresponds to the residual value of the energy balance equation of a graph node; the physical mechanism constraint vector generated based on this logic is not used as a suggestion, but is defined as a strong constraint penalty term in the operation process of the global state reconstruction unit; this means that in the subsequent optimization solution process, any calculation result that violates the range of this vector will result in a large penalty value in the objective function, thereby driving the optimization algorithm to converge to the solution space that conforms to the physical laws;

[0067] This embodiment endows the black-box AI model with physical interpretability by making the thermal balance equation a hard constraint in advance. This design eliminates the physical illusion caused by overfitting in pure data models from a mechanistic perspective, and can distinguish between normal temperature rises that conform to Joule's law caused by load fluctuations and abnormal temperature rises that violate the standard parameters of Joule's law caused by deterioration of contact resistance.

[0068] Example 4:

[0069] The global state reconstruction unit processing logic is as follows:

[0070] Utilizing graph neural network information aggregation mechanisms;

[0071] Deducing the state of adjacent unknown nodes from known sensor nodes;

[0072] When the initial state values ​​generated by data fitting violate the physical rules in the physical mechanism constraint vector, the inherent weights of the neural network remain unchanged;

[0073] Initiate the iterative optimization algorithm;

[0074] The output state value is corrected based on the penalty function;

[0075] The process continues until the output conforms to the physical conservation law within the allowable error range, resulting in a reconstructed state map representing the theoretical physical field distribution inside the health switch cabinet.

[0076] The global state reconstruction unit employs a physics-guided inference strategy. This unit utilizes a graph neural network information aggregation algorithm, based on the weight matrix of the graph edges, to propagate and infer the state information of known sensor nodes to unknown nodes in the neighboring region that do not have sensors installed. During the inference loop, the system continuously monitors the fit between the output values ​​and the physical rules. When it detects that the initially fitted state values ​​violate the physical mechanism constraint vector, the system executes a non-standard backpropagation strategy: keeping the weight matrix inside the neural network frozen to prevent the model from remembering incorrect physical laws, and instead initiating a gradient-based iterative optimization algorithm. This algorithm calculates the gradient of the penalty function with respect to the output layer state values, rather than the gradient with respect to the neural network weights, and follows the gradient descent direction. The algorithm aims to minimize the penalty function, directly correcting the output layer state values ​​successively. The correction process iterates continuously until the state values ​​of all nodes fall within the allowable error domain defined by the physical conservation laws; at this point, the output dataset is confirmed as the reconstructed state map.

[0077] This processing logic innovatively adopts an optimization path that fixes model weights and corrects output values, effectively solving the problem that deep learning models are easily affected by noise and output results that violate physical common sense under non-steady-state conditions; it ensures that the final generated reconstructed state map strictly follows physical laws, providing a high-confidence and pure theoretical benchmark for subsequent residual diagnosis.

[0078] Example 5:

[0079] The dynamic threshold determination logic in the residual decoupling diagnostic unit is as follows:

[0080] The dynamic threshold is calculated based on the uncertainty range of the physical model under the current operating conditions.

[0081] When the values ​​in the state residual sequence exceed the dynamic threshold and exhibit specific spatiotemporal distribution characteristics, a fault is determined to exist;

[0082] Specific spatiotemporal distribution characteristics include the continuous accumulation of heat at a certain node that deviates from the theoretical values ​​of the physical model;

[0083] The dynamic threshold automatically expands the allowable error range when the load fluctuates drastically.

[0084] Retrieve historical state residual data under the same or similar load conditions from the time series historical database, calculate its statistical variance and probability distribution, and use this to construct the uncertainty range of the physical model under the current working condition and calculate the dynamic threshold.

[0085] The residual decoupling diagnostic unit deploys adaptive fault adjudication logic; the dynamic threshold is not determined by a fixed constant, but by real-time calculation of the confidence interval of the physical model under the current instantaneous operating conditions, derived from the statistical characteristics of historical data; when non-steady-state conditions such as drastic load fluctuations or sudden changes in ambient temperature are detected, the algorithm automatically expands the allowable error range of the dynamic threshold to compensate for the hysteresis error of the physical model in the transient response; the fault judgment follows the dual verification principle: an alarm is triggered only when the amplitude of the state residual sequence exceeds the dynamic threshold and exhibits specific distribution characteristics in the spatiotemporal dimension; the specific spatiotemporal distribution characteristics are defined as: on a single or adjacent topological node, a heat accumulation phenomenon that cannot be explained by physical formulas occurs for multiple consecutive time steps;

[0086] This embodiment solves the industry problem of false alarms in high-sensitivity diagnostic systems by combining dynamic thresholds and spatiotemporal features. The dynamic threshold mechanism effectively shields false signals under non-steady-state conditions, while the tracking of heat accumulation features ensures that the system can keenly capture weak fault signals such as early contact failures, achieving the optimal balance between robustness and sensitivity.

[0087] Example 6:

[0088] The process by which the adaptive parameter correction unit generates the parameter correction signal is as follows:

[0089] Identify whether there are long-term, small, and stable systematic biases in the state residual sequence;

[0090] When a systematic deviation is detected, a parameter correction signal is generated and fed back.

[0091] The physical constraint embedded unit updates the heat dissipation coefficient or basic resistance value based on the parameter correction signal to adapt to the drift of physical parameters caused by equipment aging.

[0092] The adaptive parameter correction unit runs a long-cycle drift detection algorithm. This unit performs sliding window statistical analysis on the state residual sequence to identify and separate systematic deviation components with long-term duration, small amplitude, and stable variance. Unlike abrupt fault signals, such deviations characterize the natural aging trajectory of the equipment's physical properties. After ruling out sensor zero-point drift or faults through multi-source sensor data consistency checks, once such deviations are confirmed, the unit immediately generates a parameter correction signal containing drift quantification information and feeds it back to the front end. The physical constraint embedding unit responds to this signal by fine-tuning and updating key coefficients in the physical conservation logic: specifically, increasing the base resistance value to reflect contact surface oxidation, or adjusting the heat dissipation coefficient to reflect the effect of dust accumulation.

[0093] The closed-loop correction mechanism constructed in this embodiment overcomes the defect that static physical models cannot adapt to equipment aging; by continuously absorbing aging information from operating data to update physical parameters, it ensures that the theoretical benchmark of the diagnostic system is always synchronized with the current actual life cycle state of the equipment, thereby ensuring that the system maintains high-precision diagnostic capabilities throughout the decades-long equipment operation cycle.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-based intelligent diagnostic system for switchgear status, characterized in that, include: The multi-source topology mapping unit is configured to receive heterogeneous sensor data at the edge and static design parameters of the switch cabinet; Key heat points and sensor locations are mapped as graph nodes; heat conduction paths and electromagnetic coupling paths are mapped as graph edges; dynamic graph structure data containing real-time node features and edge connection relationships is generated. The physical constraint embedding unit is configured to monitor the dynamic graph structure data in real time; and to calculate the gradient of change of each graph node based on the current load input using preset physical conservation logic. Generate a physical mechanism constraint vector representing the strength of physical rules; The global state reconstruction unit is configured to receive the dynamic graph structure data and the physical mechanism constraint vector; and to process the dynamic graph structure data using an optimization solution layer embedded with physical constraints. A penalty function is constructed based on the physical mechanism constraint vector; the state values ​​output by the graph neural network are dynamically corrected; and a reconstructed state graph is generated. The residual decoupling diagnostic unit is configured to receive the measured heterogeneous sensor data and the reconstructed state map; calculate the difference between the two to generate a state residual sequence; Utilize dynamic threshold judgment logic to distinguish between noise and faults; Output the fault location results; The adaptive parameter correction unit is configured to establish a feedback closed loop; The evolution trend of the state residual sequence is tracked through a time-series historical database; When a systematic deviation is detected, a parameter correction signal is generated; the parameter correction signal is fed back to the physical constraint embedding unit to update the key coefficients in the physical conservation logic.

2. The intelligent diagnostic system for switchgear status based on a cloud platform according to claim 1, characterized in that, The configuration rules for the multi-source topology mapping unit are as follows: The heterogeneous sensor data includes temperature, current, and partial discharge signals; The static design parameters include the busbar layout structure and contact positions; The key heating points include the plum blossom contact and the busbar lap surface.

3. The intelligent diagnostic system for switchgear status based on a cloud platform according to claim 1, characterized in that, The configuration rules for the physical constraint embedding unit are as follows: The physical conservation logic is set so that the Joule heat generated by the current is equal to the sum of heat absorbed by temperature rise, heat dissipated by convection, heat dissipated by radiation, and heat dissipated by conduction. The physical mechanism constraint vector is transmitted to the global state reconstruction unit as a forced boundary condition in the computation process.

4. The intelligent diagnostic system for switchgear status based on a cloud platform according to claim 1, characterized in that, The processing logic of the global state reconstruction unit is as follows: Utilizing graph neural network information aggregation mechanisms; Deducing the state of adjacent unknown nodes from known sensor nodes; When the initial state values ​​generated by data fitting violate the physical rules in the physical mechanism constraint vector, the inherent weights of the neural network remain unchanged; Initiate the iterative optimization algorithm; The output state value is corrected based on the penalty function; The process continues until the output conforms to the physical conservation law within the allowable error range, thus obtaining the reconstructed state map representing the theoretical physical field distribution inside the health switch cabinet.

5. The intelligent diagnostic system for switchgear status based on a cloud platform according to claim 1, characterized in that, The dynamic threshold determination logic in the residual decoupling diagnostic unit is as follows: The dynamic threshold is calculated based on the uncertainty range of the physical model under the current operating conditions. When the value in the state residual sequence exceeds the dynamic threshold and exhibits specific spatiotemporal distribution characteristics, a fault is determined to exist; The specific spatiotemporal distribution characteristics include the continuous accumulation of heat at a node that deviates from the theoretical values ​​of the physical model. The dynamic threshold automatically expands the allowable error range when the load fluctuates drastically.

6. The intelligent diagnostic system for switchgear status based on a cloud platform according to claim 1, characterized in that, The process by which the adaptive parameter correction unit generates the parameter correction signal is as follows: Identify whether there are long-term, small, and stable systematic biases in the state residual sequence; When the systematic deviation is detected, the parameter correction signal is generated and fed back. The physical constraint embedding unit updates the heat dissipation coefficient or basic resistance value according to the parameter correction signal to adapt to the physical parameter drift caused by equipment aging.

Images