Substation high-voltage chamber power equipment overheating fault identification system based on supercomputing and identification method thereof

Abstract

The invention discloses a transformer substation high-voltage chamber power equipment overheating fault identification system and method based on supercomputing, and relates to the technical field of transformer substation equipment fault identification, and the method comprises the following steps: S1, power equipment high-voltage contact classification; s2, acquiring the temperature of the contact on line; s3, PCA analysis and judgment are carried out; s4, judging temperature change; s5, positioning a fault contact; s6, fault trend prediction based on heat transfer theory and particle filtering; and S7, carrying out supercomputing parallel acceleration processing. According to the invention, through contact classification, on-line temperature acquisition, wavelet denoising processing, PCA analysis, a K-means algorithm, a temperature model constructed by heat transfer theory and parameter correction of a particle filter algorithm, the problems of data processing lag, untimely fault early warning and insufficient positioning precision of a traditional monitoring mode are effectively solved; and a powerful guarantee is provided for stable operation of power equipment in a high-voltage chamber of a transformer substation.

Description

Technical Field

[0001] This invention relates to the field of substation equipment fault identification technology, specifically to a supercomputing-based substation high-voltage room power equipment overheating fault identification system and method. Background Technology

[0002] The stable operation of high-voltage indoor power equipment in substations is the core guarantee for the safe power supply of the power system. Among them, overheating faults are one of the most common types of faults in high-voltage equipment. The operating temperature of power equipment is affected by multiple factors such as its own condition, ambient temperature, load current, and three-phase balance, which directly reflects the operating status of the equipment.

[0003] In existing technologies, temperature monitoring of high-voltage equipment mostly relies on manual inspections or data collection by a single sensor, which has the following drawbacks: low data processing efficiency, unable to meet the real-time analysis needs of massive monitoring data; delayed fault warning, lack of effective early warning mechanisms, making it difficult to identify potential faults in advance; and insufficient fault location accuracy, making it impossible to quickly locate fault contacts. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a supercomputing-based overheating fault identification system and method for power equipment in high-voltage rooms of substations, solving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying overheating faults in high-voltage electrical equipment in substations based on supercomputing, comprising the following steps: S1. Classification of high-voltage contacts in power equipment: High-voltage contacts of power equipment are classified according to their working environment. The classification dimensions include temperature and humidity of the installation area, load level and current fluctuation, insulation environment and corrosion risk, and three-phase balance condition. S2. Online acquisition of contact temperature: Temperature sensors are used to collect the temperature of key components of various contacts. The temperature is monitored online by a temperature monitoring system, and the temperature data is updated after each sampling cycle. Temperature datasets for various contacts are then constructed in chronological order. and remove the temperature dataset. The dataset containing invalid data is then subjected to wavelet denoising. S3, PCA analysis and judgment: Using PCA The analysis yields a dynamic principal component eigenvalue curve. The PCA window length is 10, and the principal component eigenvalue threshold is 20. The principal component eigenvalue curve is monitored in real time. When the principal component eigenvalue exceeds the threshold, proceed to step S4; otherwise, return to step S1. S4. Temperature Change Judgment: Determine the direction of temperature change. If the temperature rises, proceed to step S5; otherwise, return to step S1. S5. Locate the faulty contact: Based on K-means, faulty contacts can be located accurately, and faulty or potentially faulty contacts can be triggered to trigger system alarms and record faulty contact information. S6. Fault trend prediction based on heat transfer and particle filtering: The power equipment contacts are treated as a lumped heat capacity system. A heat conduction equation is established and solved to obtain an overheating fault temperature model. The LS-DM method is used to estimate the initial values ​​of the temperature model parameters using the initial temperature sequence. The initial values ​​of the parameters are substituted into the temperature model to predict the temperature of the contacts at future times and to grasp the fault development trend. When new temperature data is updated, the particle filter algorithm is used to correct the model parameters and return to step S5 to repeat the execution, so as to realize dynamic optimization of the model and rolling prediction of fault trends. S7, Supercomputing Parallel Acceleration Processing: The feature extraction algorithm, PCA early warning algorithm, K-means localization algorithm, and particle filter optimization algorithm in the above steps are deployed on the supercomputing platform. The parallel computing capabilities of the supercomputing platform are used to optimize and accelerate these algorithms, thereby improving the efficiency of data processing and algorithm execution.

[0006] Furthermore, in step S2, invalid data includes missing values, duplicate values, physical anomalies, logical anomalies, and fixed values.

[0007] Furthermore, in step S5, the specific process for locating the faulty contact using K-means is as follows: Principal component eigenvalues ​​and temperature change rate were selected as clustering features to construct a sample vector for each contact point. To form a clustered dataset ; Determining the optimal number of clusters using the elbow rule Initial settings That is, normal contact clusters and faulty or potentially hazardous contact clusters; For dataset Perform Z-score standardization to eliminate the influence of dimensions, initialize cluster centers, calculate the Euclidean distance between the sample and the cluster center, assign cluster labels, update cluster centers, until the change in cluster center is ≤0.001 or the number of iterations reaches 50. Clusters with average principal component eigenvalues ​​> 20 and average temperature change rate > 0 are defined as fault or potential fault clusters, and the contact points corresponding to the samples in these clusters are the fault or potential fault contact points.

[0008] Furthermore, in step S6, the heat conduction equation is as follows: Formula 1 Formula 2 Formula 3 Formula 4 In the formula, For contact volume; The contact surface area; For density, For specific heat capacity, For current, For resistance, The average thermal conductivity represents the entire boundary surface and is the total thermal conductivity including convection, radiation, and conduction. Represents ambient temperature; Represents time; Represents contact temperature. At the initial moment, for The contact temperature at any given moment.

[0009] Furthermore, the process for solving the overheating fault temperature model is as follows: Make ,but: Formula 5 Formula 6 Formula 7 Formula 8 Formula 9 Formula 10 Formula 11 Formula 12 Formula 13 Formula 14 By solving the above differential equations, we obtain the temperature model for overheating faults of power equipment contacts, namely Equation 14.

[0010] Furthermore, in step S6, the specific process for estimating the initial values ​​of the temperature model parameters using the LS-DM method is as follows: Extraction after pretreatment and continuous Temperature data from one sampling period is used as the initial sequence. Matching the PCA window length, denoted as The corresponding time series is Therefore, the LS-DM observation equation is constructed based on Equation 14, and the specific observation equation is as follows: in, For the parameters to be estimated, It is the product of current and resistance; To measure noise, a sample follows a mean of 0 and a variance of 1. The normal distribution; Then solve the least squares problem by minimizing the objective function: Get the initial values ​​of the parameters The solution formula is: in, Jacobian matrix (dimensions) ), Let be the initial temperature sequence vector, then:

[0011] Furthermore, the specific process for correcting the model parameters using the particle filter algorithm is as follows: generate Particles The particles obey a parameter with the current estimated value as mean and variance as... The normal distribution ; Based on newly acquired temperature data Calculate the weight of each particle: Where the probability density function : And normalize the weights: When the number of effective particles At that time, a new set of particles is generated by resampling according to the normalized weights. ; Calculate the mean of the new particle set: This serves as the corrected model parameter.

[0012] Furthermore, the following steps are also included: S8, Contact-related coupling correction: Using all monitoring contacts in the high-voltage chamber as nodes, with node characteristics including current temperature, principal component eigenvalues, temperature change rate, load current, and ambient temperature, and the physical connections between contacts as edges, with initial edge weights set based on the degree of physical influence, a contact relationship graph is formed. ,in For a set of nodes, For edge set, This is the initial edge weight matrix; The initial values ​​of the edge weights are set based on the degree of physical influence as follows: direct circuit connection weight = 1.0, adjacent installation weight = 0.6, and non-direct connection weight within the same loop = 0.3. Historical temperature data containing temperature sequences of related contact points during the same period are input into a graph neural network (GNN) model. The error between the actual temperature of a single contact point and the temperature of related contact points is used as the loss function to train the model to learn edge weight correction values, and the optimized correlation weight matrix is ​​output. Its dimensions , The total number of touch points, the number of model training iterations ≤ 100, and the convergence threshold of the loss function is 0.005; For the contact to be predicted Extract its associated node set Its satisfaction nodes Calculate the associated node pairs Temperature effect value ; The predicted temperature of a single contact point in step S6 is called... Make it related to the influence value By merging the data, the final corrected fault trend prediction value is obtained. ; If the associated node set Corrected predicted values ​​with ≥2 nodes If all exceed 80% of the equipment's allowable temperature threshold, it is determined to be a risk of collaborative failure, and an escalation alarm is triggered.

[0013] Furthermore, in step S8, the optimization formula for the correlation weight matrix is ​​as follows: in, For the Sigmoid activation function, For contact At any moment temperature, Commonly associated nodes temperature, For a moment The ambient temperature is used as the denominator, and +20 is added to the denominator to avoid the denominator approaching 0 due to excessively low ambient temperatures. The GNN features used to learn node association features are calculated based on the temperature effect value. The formula is as follows: in, For contact Historical average temperature For contact Physical installation distance, For contact At any moment Temperature; The coupling correction formula is: in, This is a correction factor, ranging from 0.3 to 0.8, which is adaptively adjusted according to the contact type; copper contacts. =0.6, aluminum contacts =0.5, SF6 insulated environmental contacts =0.8, to ensure that the correction range conforms to the physical laws of heat transfer.

[0014] A supercomputing-based overheating fault identification system for power equipment in the high-voltage room of a substation is provided, wherein the system applies the aforementioned supercomputing-based overheating fault identification system method for power equipment in the high-voltage room of a substation.

[0015] This invention provides a supercomputing-based system and method for identifying overheating faults in high-voltage electrical equipment in substations, which has the following advantages: 1. This supercomputing-based overheating fault identification system and method for substation high-voltage room power equipment ensures data accuracy and effectiveness through contact point classification combined with online temperature acquisition and wavelet denoising. PCA analysis enables early warning of overheating faults, and temperature change analysis filters out truly risky situations. The K-means algorithm accurately locates faulty or potentially hazardous contacts. Subsequent parameter correction using a temperature model based on heat transfer and a particle filter algorithm reliably predicts fault development trends. Finally, leveraging the parallel computing capabilities of the supercomputing platform significantly improves the efficiency of data processing and algorithm execution, effectively solving the problems of data processing lag, untimely fault warnings, and insufficient positioning accuracy in traditional monitoring methods, thus providing strong support for the stable operation of substation high-voltage room power equipment.

[0016] 2. This supercomputing-based substation high-voltage room power equipment overheating fault identification system and its identification method constructs a contact association diagram that closely matches the actual physical connection relationship. It uses a graph neural network model to learn and optimize the correlation between contacts, fully considering the physical correlations such as circuit topology connections and mutual heat dissipation effects between contacts in the high-voltage room equipment. By calculating the temperature influence value of the associated nodes on the target contact and correcting the prediction results, it breaks through the limitations of traditional isolated prediction of a single contact. It can not only effectively identify multi-contact collaborative faults and associated conduction faults, avoiding the omission of such faults, but also reduce the impact of associated contact interference on the temperature prediction of a single contact, significantly improving the accuracy of temperature prediction in complex scenarios. At the same time, this step can be directly integrated into the process of steps S1-S7, further improving the reliability and applicability of the fault identification system without affecting the overall real-time performance. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the power equipment overheating fault identification system of the present invention; Figure 2 This is a schematic diagram of the PCA-based early warning algorithm for overheating faults according to the present invention. Figure 3 This is a schematic diagram of the model optimization and fault trend prediction algorithm based on particle filtering of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0019] like Figures 1-3 As shown, the present invention provides a technical solution: a method for identifying overheating faults in power equipment in the high-voltage room of a substation based on supercomputing, comprising the following steps: S1. Classification of high-voltage contacts in power equipment: High-voltage contacts of power equipment are classified according to their working environment. The classification dimensions include temperature and humidity of the installation area, load level and current fluctuation, insulation environment and corrosion risk, and three-phase balance condition. S2. Online acquisition of contact temperature: Temperature sensors are used to collect the temperature of key components of various contacts. The temperature is monitored online by a temperature monitoring system, and the temperature data is updated after each sampling cycle. Temperature datasets for various contacts are then constructed in chronological order. and remove the temperature dataset. The dataset containing invalid data is then subjected to wavelet denoising. Invalid data includes missing values, duplicate values, physical outliers, logical outliers, and fixed values; S3, PCA analysis and judgment: Using PCA The analysis yields a dynamic principal component eigenvalue curve. The PCA window length is 10, and the principal component eigenvalue threshold is 20. The principal component eigenvalue curve is monitored in real time. When the principal component eigenvalue exceeds the threshold, proceed to step S4; otherwise, return to step S1. S4. Temperature Change Judgment: Determine the direction of temperature change. If the temperature rises, proceed to step S5; otherwise, return to step S1. S5. Locate the faulty contact: Based on K-means, faulty contacts can be located accurately, and faulty or potentially faulty contacts can be triggered to trigger system alarms and record faulty contact information. The specific process for locating faulty contacts using K-means is as follows: Principal component eigenvalues ​​and temperature change rate were selected as clustering features to construct a sample vector for each contact point. To form a clustered dataset ; Determining the optimal number of clusters using the elbow rule Initial settings That is, normal contact clusters and faulty or potentially hazardous contact clusters; For dataset Perform Z-score standardization to eliminate the influence of dimensions, initialize cluster centers, calculate the Euclidean distance between the sample and the cluster center, assign cluster labels, update cluster centers, until the change in cluster center is ≤0.001 or the number of iterations reaches 50. Clusters with average principal component eigenvalues ​​> 20 and average temperature change rate > 0 are defined as fault or potential fault clusters, and the contact points corresponding to the samples in these clusters are fault or potential fault contact points. S6. Fault trend prediction based on heat transfer and particle filtering: The power equipment contacts are treated as a lumped heat capacity system. A heat conduction equation is established and solved to obtain an overheating fault temperature model. The LS-DM method is used to estimate the initial values ​​of the temperature model parameters using the initial temperature sequence. The initial values ​​of the parameters are substituted into the temperature model to predict the temperature of the contacts at future times and to grasp the fault development trend. When new temperature data is updated, the particle filter algorithm is used to correct the model parameters and return to step S5 to repeat the execution, so as to realize dynamic optimization of the model and rolling prediction of fault trends. The heat conduction equation is as follows: Formula 1 Formula 2 Formula 3 Formula 4 In the formula, For contact volume; The contact surface area; For density, For specific heat capacity, For current, For resistance, The average thermal conductivity represents the entire boundary surface and is the total thermal conductivity including convection, radiation, and conduction. Represents ambient temperature; Represents time; Represents contact temperature. At the initial moment, for The contact temperature at any given moment; The process for solving the overheating fault temperature model is as follows: Make ,but: Formula 5 Formula 6 Formula 7 Formula 8 Formula 9 Formula 10 Formula 11 Formula 12 Formula 13 Formula 14 By solving the above differential equations, we obtain the temperature model for overheating faults of power equipment contacts, namely Equation 14. The specific procedure for estimating the initial values ​​of the parameters of the temperature model using the LS-DM method is as follows: Extraction after pretreatment and continuous Temperature data from one sampling period is used as the initial sequence. Matching the PCA window length, denoted as The corresponding time series is Therefore, the LS-DM observation equation is constructed based on Equation 14, and the specific observation equation is as follows: in, For the parameters to be estimated, It is the product of current and resistance; To measure noise, a sample follows a mean of 0 and a variance of 1. The normal distribution; Then solve the least squares problem by minimizing the objective function: Get the initial values ​​of the parameters The solution formula is: in, Jacobian matrix (dimensions) ), Let be the initial temperature sequence vector, then: The specific process of correcting model parameters using the particle filter algorithm is as follows: generate Particles The particles obey a parameter with the current estimated value as mean and variance as... The normal distribution ; Based on newly acquired temperature data Calculate the weight of each particle: Where the probability density function : And normalize the weights: When the number of effective particles At that time, a new set of particles is generated by resampling according to the normalized weights. ; Calculate the mean of the new particle set: This serves as the corrected model parameter; S7, Supercomputing Parallel Acceleration Processing: The feature extraction algorithm, PCA early warning algorithm, K-means localization algorithm and particle filter optimization algorithm in the above steps are deployed on the supercomputing platform. The parallel computing capabilities of the supercomputing platform are used to optimize and accelerate the process, thereby improving the efficiency of data processing and algorithm execution. Based on the above description, this invention ensures the accuracy and effectiveness of data by combining contact point classification with online temperature acquisition and wavelet denoising. It achieves early warning of overheating faults through PCA analysis, and then filters out situations with real risks by judging temperature changes. It uses the K-means algorithm to accurately locate faulty or potential contact points. Subsequently, the temperature model built based on heat transfer and the parameter correction of the particle filter algorithm can reliably predict the fault development trend. Finally, relying on the parallel computing capabilities of the supercomputing platform, the efficiency of the entire data processing and algorithm execution is greatly improved. It effectively solves the problems of data processing lag, untimely fault warning and insufficient positioning accuracy of traditional monitoring methods, and provides a strong guarantee for the stable operation of power equipment in the high-voltage room of substations. S8, Contact-related coupling correction: Using all monitoring contacts in the high-voltage chamber as nodes, with node characteristics including current temperature, principal component eigenvalues, temperature change rate, load current, and ambient temperature, and the physical connections between contacts as edges, with initial edge weights set based on the degree of physical influence, a contact relationship graph is formed. ,in For a set of nodes, For edge set, This is the initial edge weight matrix; The initial values ​​of the edge weights are set based on the degree of physical influence as follows: direct circuit connection weight = 1.0, adjacent installation weight = 0.6, and non-direct connection weight within the same loop = 0.3. Historical temperature data containing temperature sequences of related contact points during the same period are input into a graph neural network (GNN) model. The error between the actual temperature of a single contact point and the temperature of related contact points is used as the loss function to train the model to learn edge weight correction values, and the optimized correlation weight matrix is ​​output. Its dimensions , The total number of touch points, the number of model training iterations ≤ 100, and the convergence threshold of the loss function is 0.005; For the contact to be predicted Extract its associated node set Its satisfaction nodes Calculate the associated node pairs Temperature effect value ; The predicted temperature of a single contact point in step S6 is called... Make it related to the influence value By merging the data, the final corrected fault trend prediction value is obtained. ; If the associated node set Corrected predicted values ​​with ≥2 nodes If all exceed 80% of the equipment's allowable temperature threshold, it is determined to be a risk of collaborative failure, and an escalation alarm is triggered. The formula for optimizing the correlation weight matrix is ​​as follows: in, For the Sigmoid activation function, For contact At any moment temperature, Commonly associated nodes temperature, For a moment The ambient temperature is used as the denominator, and +20 is added to the denominator to avoid the denominator approaching 0 due to excessively low ambient temperatures. The GNN features used to learn node association features are calculated based on the temperature effect value. The formula is as follows: in, For contact Historical average temperature For contact Physical installation distance, For contact At any moment Temperature; The coupling correction formula is: in, This is a correction factor, ranging from 0.3 to 0.8, which is adaptively adjusted according to the contact type; copper contacts. =0.6, aluminum contacts =0.5, SF6 insulated environmental contacts =0.8, to ensure that the correction range conforms to the physical laws of heat transfer; Based on the above description, this invention constructs a contact association diagram that closely matches the actual physical connection relationship, uses a graph neural network model to learn and optimize the association between contacts, and fully considers the physical associations such as the circuit topology connection and heat dissipation mutual influence between contacts in high-voltage room equipment. By calculating the temperature influence value of the associated nodes on the target contact and correcting the prediction results, it breaks through the limitations of traditional isolated prediction of a single contact. It can not only effectively identify multi-contact collaborative faults and associated conduction faults, avoiding the omission of such faults, but also reduce the impact of associated contact interference on the temperature prediction of a single contact, significantly improving the accuracy of temperature prediction in complex scenarios. At the same time, this step can be directly integrated into the process of steps S1-S7, further improving the reliability and applicability of the fault identification system without affecting the overall real-time performance.

[0020] A supercomputing-based overheating fault identification system for power equipment in the high-voltage room of a substation is provided, wherein the system applies the aforementioned supercomputing-based overheating fault identification system method for power equipment in the high-voltage room of a substation.

[0021] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for identifying overheating faults in high-voltage electrical equipment in substations based on supercomputing, characterized in that: Includes the following steps: S1. Classification of high-voltage contacts in power equipment: High-voltage contacts of power equipment are classified according to their working environment. The classification dimensions include temperature and humidity of the installation area, load level and current fluctuation, insulation environment and corrosion risk, and three-phase balance condition. S2. Online acquisition of contact temperature: Temperature sensors are used to collect the temperature of key components of various contact points. The temperature is monitored online by a temperature monitoring system, and the temperature data is updated after each sampling cycle. Temperature datasets for various contact points are then constructed in chronological order. and remove the temperature dataset. The dataset containing invalid data is then subjected to wavelet denoising. S3, PCA analysis and judgment: Using PCA The analysis yields a dynamic principal component eigenvalue curve. The PCA window length is 10, and the principal component eigenvalue threshold is 20. The principal component eigenvalue curve is monitored in real time. When the principal component eigenvalue exceeds the threshold, proceed to step S4; otherwise, return to step S1. S4. Temperature Change Judgment: Determine the direction of temperature change. If the temperature rises, proceed to step S5; otherwise, return to step S1. S5. Locate the faulty contact: Based on K-means, faulty contacts can be located accurately, and faulty or potentially faulty contacts can be triggered to trigger system alarms and record faulty contact information. S6. Fault trend prediction based on heat transfer and particle filtering: The power equipment contacts are treated as a lumped heat capacity system. A heat conduction equation is established and solved to obtain an overheating fault temperature model. The LS-DM method is used to estimate the initial values ​​of the temperature model parameters using the initial temperature sequence. The initial values ​​of the parameters are substituted into the temperature model to predict the temperature of the contacts at future times and to grasp the fault development trend. When new temperature data is updated, the particle filter algorithm is used to correct the model parameters and return to step S5 to repeat the execution, so as to realize dynamic optimization of the model and rolling prediction of fault trends. S7, Supercomputing Parallel Acceleration Processing: The feature extraction algorithm, PCA early warning algorithm, K-means localization algorithm, and particle filter optimization algorithm in the above steps are deployed on the supercomputing platform. The parallel computing capabilities of the supercomputing platform are used to optimize and accelerate these algorithms, thereby improving the efficiency of data processing and algorithm execution.

2. The method for identifying overheating faults in high-voltage substation power equipment based on supercomputing, as described in claim 1, is characterized in that: In step S2, invalid data includes missing values, duplicate values, physical anomalies, logical anomalies, and fixed values.

3. The method for identifying overheating faults in high-voltage substation power equipment based on supercomputing, as described in claim 1, is characterized in that: In step S5, the specific process for locating faulty contacts using K-means is as follows: Principal component eigenvalues ​​and temperature change rate were selected as clustering features to construct a sample vector for each contact point. To form a clustered dataset ; Determining the optimal number of clusters using the elbow rule Initial settings That is, normal contact clusters and faulty or potentially hazardous contact clusters; For dataset Perform Z-score standardization to eliminate the influence of dimensions, initialize cluster centers, calculate the Euclidean distance between the sample and the cluster center, assign cluster labels, update cluster centers, until the change in cluster center is ≤0.001 or the number of iterations reaches 50. Clusters with average principal component eigenvalues ​​> 20 and average temperature change rate > 0 are defined as fault or potential fault clusters, and the contact points corresponding to the samples in these clusters are the fault or potential fault contact points.

4. The method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing as described in claim 1, characterized in that: In step S6, the heat conduction equation is as follows: Formula 1 Formula 2 Formula 3 Formula 4 In the formula, For contact volume; The contact surface area; For density, For specific heat capacity, For current, For resistance, The average thermal conductivity represents the entire boundary surface and is the total thermal conductivity including convection, radiation, and conduction. Represents ambient temperature; Represents time; Represents contact temperature. At the initial moment, for The contact temperature at any given moment.

5. The method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing, as described in claim 4, is characterized in that: The overheating fault temperature model is obtained by solving the differential equation, resulting in Equation 14: Equation 14.

6. The method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing as described in claim 1, characterized in that: In step S6, the specific procedure for estimating the initial values ​​of the temperature model parameters using the LS-DM method is as follows: Extraction after pretreatment and continuous Temperature data from one sampling period is used as the initial sequence. Matching the PCA window length, denoted as The corresponding time series is Therefore, the LS-DM observation equation is constructed based on Equation 14, and the specific observation equation is as follows: in, For the parameters to be estimated, It is the product of current and resistance; To measure noise, a sample follows a mean of 0 and a variance of 1. The normal distribution; Then solve the least squares problem by minimizing the objective function to obtain the initial values ​​of the parameters. The solution formula is: in, Jacobian matrix (dimensions) ), Let be the initial temperature sequence vector, then: 。 7. The method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing, as described in claim 6, is characterized in that: The specific process of correcting model parameters using the particle filter algorithm is as follows: generate Particles The particles obey a parameter with the current estimated value as mean and variance as... The normal distribution ; Based on newly acquired temperature data Calculate the weight of each particle and normalize the weights; When the number of effective particles At that time, a new set of particles is generated by resampling according to the normalized weights. ; Calculate the mean of the new particle set: This serves as the corrected model parameter.

8. The method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing, as described in claim 1, is characterized in that: It also includes the following steps: S8, Contact-related coupling correction: Using all monitoring contacts in the high-voltage chamber as nodes, with node characteristics including current temperature, principal component eigenvalues, temperature change rate, load current, and ambient temperature, and the physical connections between contacts as edges, with initial edge weights set based on the degree of physical influence, a contact relationship graph is formed. ,in For a set of nodes, For edge set, This is the initial edge weight matrix; The initial values ​​of the edge weights are set based on the degree of physical influence as follows: direct circuit connection weight = 1.0, adjacent installation weight = 0.6, and non-direct connection weight within the same loop = 0.

3. Historical temperature data containing temperature sequences of related contact points during the same period are input into a graph neural network (GNN) model. The error between the actual temperature of a single contact point and the temperature of related contact points is used as the loss function to train the model to learn edge weight correction values, and the optimized correlation weight matrix is ​​output. Its dimensions , The total number of touch points, the number of model training iterations ≤ 100, and the convergence threshold of the loss function is 0.005; For the contact to be predicted Extract its associated node set Its satisfaction nodes Calculate the associated node pairs Temperature effect value ; The predicted temperature of a single contact point in step S6 is called... Make it related to the influence value By merging the data, the final corrected fault trend prediction value is obtained. ; If the associated node set Corrected predicted values ​​with ≥2 nodes If all exceed 80% of the equipment's allowable temperature threshold, it is determined to be a risk of collaborative failure, and an escalation alarm is triggered.

9. A method for identifying overheating faults in high-voltage substation power equipment based on supercomputing, as described in claim 8, characterized in that: In step S8, the formula for optimizing the correlation weight matrix is ​​as follows: in, For the Sigmoid activation function, For contact At any moment temperature, Commonly associated nodes temperature, For a moment The ambient temperature is used as the denominator, and +20 is added to the denominator to avoid the denominator approaching 0 due to excessively low ambient temperatures. The GNN features used to learn node association features are calculated based on the temperature effect value. The formula is as follows: in, For contact Historical average temperature For contact Physical installation distance, For contact At any moment Temperature; The coupling correction formula is: in, This is a correction factor, ranging from 0.3 to 0.8, which is adaptively adjusted according to the contact type; copper contacts. =0.6, aluminum contacts =0.5, SF6 insulated environmental contacts =0.8, to ensure that the correction range conforms to the physical laws of heat transfer.

10. A supercomputing-based overheating fault identification system for power equipment in a substation high-voltage room, as described in claim 1, is characterized in that: The system application includes a method for identifying overheating faults in substation high-voltage room power equipment based on supercomputing, as described in any one of claims 1-9.

Images