Temperature online monitoring method and system for intelligent power distribution cabinet

Abstract

The invention belongs to the technical field of intelligent power distribution, and particularly relates to a temperature online monitoring method and system for an intelligent power distribution cabinet, and the method comprises the steps: collecting a three-dimensional temperature field, vibration and environment parameters, constructing a monitoring data set, and transmitting the monitoring data set to an edge computing platform; the cabinet body is divided into spatial infinitesimal elements, a heat flow graph model is constructed, a graph attention neural network is adopted to perform modeling on heat conduction association, and thermal fault probability distribution prediction is output; based on a prediction result, dynamically adjusting calculation power by utilizing a reconfigurable edge calculation platform, and determining hot spot positioning by combining infrared sensor data and a heat flow diagram model when a thermal fault risk is predicted to exist; and according to the hot spot information and historical fault data, an early warning threshold is self-calibrated through Bayesian optimization, and when the thermal fault probability exceeds the early warning threshold, active fault-tolerant control of the intelligent heat dissipation and contact adjusting mechanism is linked. Therefore, the problems of limited monitoring precision and range, lack of adaptability of an early warning mechanism, insufficient data utilization and the like in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent power distribution, and particularly relates to a temperature online monitoring method and system for an intelligent power distribution cabinet. BACKGROUND

[0002] In modern power systems, intelligent power distribution cabinets, as the key hub of power distribution and control, their stable operation is crucial to ensure the reliability and safety of power supply. With the continuous growth of power demand and the increasing degree of industrial automation, the power load carried by intelligent power distribution cabinets continues to increase, and the operating environment of internal electrical equipment becomes increasingly complex. Therefore, it is urgent to develop a high-precision, self-adaptive, actively controlled temperature online monitoring method and system that can fully utilize data, which has important practical significance for improving the operation stability of intelligent power distribution cabinets, prolonging the service life of equipment, and ensuring the safe and reliable operation of power systems.

[0003] However, the traditional intelligent power distribution cabinet temperature monitoring has many problems: the monitoring accuracy and range are limited, single-point or a few-point measurement is difficult to grasp the internal complex temperature distribution, and hot spot hidden dangers are easily missed; the early warning threshold is fixed and cannot be adjusted with the working condition, equipment aging and environmental changes, leading to false positives or false negatives; passive heat dissipation and fault handling, fixed power dissipation lacks pertinence, lacks active fault-tolerant means for contact failure, relies on manual inspection, and has low efficiency and high risk; data utilization is insufficient, only used for simple display and threshold judgment, not in-depth analysis, and cannot provide comprehensive decision support and predictive maintenance for operation and maintenance. SUMMARY

[0004] The application provides a temperature online monitoring method and system for an intelligent power distribution cabinet to solve the problems of limited monitoring accuracy and range, lack of adaptability of early warning mechanism, and insufficient data utilization in the prior art.

[0005] The first aspect embodiment of the application provides a temperature online monitoring method for an intelligent power distribution cabinet, comprising the following steps: collecting a three-dimensional temperature field, vibration and environmental parameters; constructing a multi-modal monitoring data set according to the three-dimensional temperature field, vibration and environmental parameters, and transmitting original data to an edge computing platform; after receiving the original data, dividing the cabinet body into spatial microelements, constructing a heat flow graph model, modeling the heat conduction correlation between nodes by using a graph attention neural network, outputting a thermal fault probability distribution prediction by analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow and load current; based on the thermal fault probability distribution prediction result, dynamically adjusting the computing power by using a reconfigurable edge computing platform, starting a decision engine when a thermal fault risk is predicted, analyzing in combination with infrared sensor data and the heat flow graph model to determine the thermal spot positioning; according to the information of the thermal spot positioning and historical fault data, self-calibrating the warning threshold through Bayesian optimization, when the thermal fault probability exceeds the warning threshold, linking intelligent heat dissipation and contact adjustment mechanisms for active fault tolerance control, and feeding back the control result to the monitoring center.

[0006] Preferably, the graph attention neural network is used to model the heat conduction correlation between nodes, comprising: constructing a heterogeneous graph structure, wherein the nodes include temperature monitoring points, vibration sensors and environmental parameter monitoring points, and the edges include heat conduction paths, vibration propagation paths and environmental influence relationships; dynamically assigning differentiated heat conduction weights to the nodes according to the heterogeneous graph structure; updating the edge weight values in combination with real-time temperature gradients according to the heat conduction weights, capturing neighborhood heat flow correlation and time sequence load influence through a spatiotemporal double attention mechanism; based on the neighborhood heat flow correlation and time sequence load influence, enhancing the rare fault recognition ability of the adversarial regularization training model, dynamically updating the graph node attributes and edge connection relationships in combination with real-time data, and forming a self-adaptive modeling framework.

[0007] Preferably, the formula of the adversarial regularization training model is: ; wherein, is an optimization model parameter; is an optimization generator parameter; is a task loss function; is a hyperparameter; ; is an objective function; is a parameter of a task model; is a parameter of an adversarial or regularization part.

[0008] Preferably, based on the thermal failure probability distribution prediction result, the reconfigurable edge computing platform is used to dynamically adjust the computing power, including: constructing a heterogeneous computing architecture; based on the heterogeneous computing architecture and the thermal failure probability distribution prediction result, determining the risk level; based on the thermal failure risk level, performing dynamic resource allocation, wherein when the probability distribution is concentrated and the peak value is far below the threshold value, it is classified as low risk; when the probability distribution has a certain dispersion and the peak value is close to the threshold value, it is classified as medium risk; when the probability distribution is dispersed and the peak value exceeds the threshold value, it is classified as high risk.

[0009] Preferably, when the thermal failure probability exceeds the warning threshold, the linkage intelligent cooling and contact adjustment mechanism performs active fault tolerance control, including: constructing a Bayesian variational autoencoder; according to the Bayesian variational autoencoder, combining positioning data, real-time output of the posterior probability of the thermal failure probability distribution; when the thermal failure posterior probability breaks through the warning threshold, triggering active fault tolerance control, wherein the intelligent cooling system performs directional cooling power optimization based on the thermal flow field simulation model, and the magnetostrictive actuator of the contact adjustment mechanism performs nanoscale dynamic compensation of the contact resistance.

[0010] Preferably, the formula of the thermal flow field simulation model is: ; Wherein, is the heat conduction term; is the material thermal conductivity; is the temperature gradient; Rc is the contact resistance of the contact; is the current; is the contact area volume.

[0011] Preferably, according to the three-dimensional temperature field, vibration and environmental parameters, a multi-modal monitoring data set is constructed, including: constructing a synchronous clock module; according to the synchronous clock module and the sampling time stamp, generating a synchronous sample sequence; according to the synchronous sample sequence, combining a data quality evaluation mechanism, performing feature transformation, and constructing a multi-modal feature data set, wherein the feature transformation includes converting the three-dimensional temperature field into a graph structure data with spatial microelement as node and heat conduction path as edge, converting the vibration signal into mel frequency cepstral coefficient, and embedding the environmental parameters as global features into the graph node attribute.

[0012] The second aspect embodiment of the application provides a temperature online monitoring system for a smart power distribution cabinet, comprising: an acquisition module, configured to acquire a three-dimensional temperature field, vibration and environmental parameters; a transmission module, configured to construct a multi-modal monitoring data set according to the three-dimensional temperature field, vibration and environmental parameters, and transmit original data to an edge computing platform; a construction module, configured to divide a cabinet body into spatial microelements after receiving the original data, construct a heat flow graph model, model heat conduction correlation between nodes by using a graph attention neural network, and output a thermal fault probability distribution prediction by analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow and load current; a prediction module, configured to dynamically adjust computing power by using a reconfigurable edge computing platform based on the thermal fault probability distribution prediction result, start a decision engine when a thermal fault risk is predicted, analyze infrared sensor data and the heat flow graph model, and determine thermal spot positioning; and a control module, configured to calibrate a warning threshold by Bayes optimization according to information of the thermal spot positioning and historical fault data, perform active fault tolerance control by linking intelligent heat dissipation and contact adjustment mechanisms when the thermal fault probability exceeds the warning threshold, and feed back control results to a monitoring hub.

[0013] The third aspect embodiment of the application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor executes the program to implement the temperature online monitoring method for a smart power distribution cabinet as described in the above embodiments.

[0014] The fourth aspect embodiment of the application provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the temperature online monitoring method for a smart power distribution cabinet as described in the above embodiments.

[0015] Therefore, the application has the following beneficial effects: This application's embodiments construct a comprehensive monitoring system covering the equipment's operating status through multimodal data acquisition of three-dimensional temperature fields, vibration, and environmental parameters, providing multi-source heterogeneous data support for thermal fault analysis. Based on a thermal flow graph model using spatial micro-element partitioning and graph attention neural networks, it accurately captures the spatiotemporal correlation of heat conduction between cabinet nodes, breaking through the limitations of traditional single-point monitoring and achieving dynamic prediction and risk quantification assessment of thermal fault probability distribution. A reconfigurable edge computing platform, combined with infrared sensor data and a thermal flow model, dynamically adjusts computing power in real time and accurately locates hot spots, significantly improving the timeliness and accuracy of fault location. A Bayesian-optimized self-calibrated early warning threshold mechanism adaptively optimizes the early warning strategy based on historical fault data, avoiding the lag defects of fixed thresholds. This, combined with the active fault-tolerant control of intelligent heat dissipation and contact adjustment mechanisms, comprehensively improves the safety and reliability of intelligent distribution cabinet operation, reduces maintenance costs and accident risks, and ensures the stable operation of the power system. Thus, it solves the problems of limited monitoring accuracy and range, lack of adaptability in early warning mechanisms, and insufficient data utilization in existing technologies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for online temperature monitoring of an intelligent power distribution cabinet according to an embodiment of this application; Figure 2 This is an example diagram of an edge computing system for automotive parts manufacturing provided according to an embodiment of this application; Figure 3 This is an example diagram of a multi-sensor synchronous monitoring system for substations provided according to an embodiment of this application; Figure 4 This is a diagram illustrating an example of power transformer fault diagnosis according to an embodiment of this application. Figure 5 This is an example diagram of a long-term operation monitoring system for high-voltage circuit breakers provided according to an embodiment of this application; Figure 6 This is a flowchart of a method for online temperature monitoring of an intelligent power distribution cabinet according to an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an online temperature monitoring system for an intelligent power distribution cabinet according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0018] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0019] The following description, with reference to the accompanying drawings, illustrates an online temperature monitoring method and system for intelligent distribution cabinets according to embodiments of this application. Addressing the lack of adaptability in early warning mechanisms mentioned in the background section, this application provides an online temperature monitoring method for intelligent distribution cabinets. This method constructs a comprehensive monitoring system covering the equipment's operating status through multimodal data acquisition of three-dimensional temperature fields, vibration, and environmental parameters, providing multi-source heterogeneous data support for thermal fault analysis. Based on a thermal flow graph model using spatial micro-element partitioning and graph attention neural networks, it accurately captures the spatiotemporal correlation of heat conduction between cabinet nodes, breaking through the limitations of traditional single-point monitoring and achieving dynamic prediction and risk quantification assessment of thermal fault probability distribution. A reconfigurable edge computing platform, combined with infrared sensor data and a thermal flow model, dynamically adjusts computing power in real time and accurately locates hot spots, significantly improving the timeliness and accuracy of fault location. A Bayesian-optimized self-calibrated early warning threshold mechanism adaptively optimizes the early warning strategy based on historical fault data, avoiding the lag defects of fixed thresholds. This mechanism, linked to intelligent heat dissipation and the active fault-tolerant control of the contact adjustment mechanism, comprehensively improves the safety and reliability of intelligent distribution cabinet operation, reduces maintenance costs and accident risks, and ensures the stable operation of the power system. This solves the problems of limited monitoring accuracy and range, lack of adaptability of early warning mechanisms, and insufficient data utilization in existing technologies.

[0020] Specifically, Figure 1 This is a schematic flowchart illustrating the online temperature monitoring method for intelligent power distribution cabinets provided in an embodiment of this application.

[0021] like Figure 1 As shown, the online temperature monitoring method for intelligent power distribution cabinets includes the following steps: In step S101, three-dimensional temperature field, vibration and environmental parameters are collected.

[0022] Among them, the three-dimensional temperature field describes the temperature distribution at various points inside the intelligent power distribution cabinet. Through the collection and analysis of temperature data in three-dimensional space, it presents the temperature changes at different locations and heights.

[0023] It is understood that the embodiments of this application accurately locate abnormal temperature areas, analyze heat propagation and heat exchange, and combine other parameters to assess equipment operating status, predict thermal failure risks, perform preventive maintenance, and ensure stable power distribution.

[0024] In step S102, a multimodal monitoring dataset is constructed based on the three-dimensional temperature field, vibration, and environmental parameters, and the raw data is transmitted to the edge computing platform.

[0025] Among them, edge computing platforms are distributed computing platforms located close to data sources or users.

[0026] It is understood that the embodiments of this application receive raw data from multimodal monitoring of intelligent power distribution cabinets, are close to the data source, process the data quickly, reduce transmission delays, detect anomalies in a timely manner, output thermal fault probability predictions, support hot spot location, and dynamically adjust computing power to improve monitoring real-time performance, reduce network dependence, and ensure data security and privacy.

[0027] For example, such as Figure 2 As shown, in automotive parts manufacturing, edge computing nodes are deployed on the production line to collect real-time data on equipment parameters such as vibration, temperature, and current. By running lightweight AI models (such as vibration detection algorithms based on ONNXRuntime) at the edge, the system can identify abnormal trends in equipment within 500 milliseconds, triggering early warnings and automatically adjusting the production process. This significantly improves efficiency compared to the 5-minute response time of traditional cloud processing. Simultaneously, edge nodes use a data filtering mechanism to upload only abnormal events to the cloud, reducing data transmission volume by 70%, lowering bandwidth costs, and ensuring the privacy and security of production data. Furthermore, long-term trend analysis is achieved through cloud-edge collaboration, reducing unplanned downtime by 40% annually and saving over 5 million yuan in maintenance costs.

[0028] In this embodiment, a multimodal monitoring dataset is constructed based on the three-dimensional temperature field, vibration, and environmental parameters. This includes: constructing a synchronization clock module; generating a synchronization sample sequence based on the synchronization clock module and the sampling timestamp; and performing feature transformation based on the synchronization sample sequence and a data quality assessment mechanism to construct a multimodal feature dataset. The feature transformation includes converting the three-dimensional temperature field into graph structure data with spatial micro-elements as nodes and heat conduction paths as edges, converting vibration signals into Mel frequency cepstral coefficients, and embedding environmental parameters as global features into the graph node attributes.

[0029] Among them, the synchronization clock module is a hardware or software functional unit that provides a unified time reference for distributed systems, network devices or IoT terminals, ensures accurate time synchronization of each device or module, and avoids data processing deviations or coordination failures caused by time differences.

[0030] It is understood that the embodiments of this application provide a unified time reference for distributed sensors to ensure that multi-source data such as three-dimensional temperature field, vibration and environmental parameters are accurately aligned in time, avoid data misalignment or analysis deviation caused by device time asynchrony, perform feature transformation based on synchronized data, ensure the time consistency of multimodal data, provide time calibration for graph neural network modeling, thermal fault prediction, etc., improve the accuracy of data fusion analysis, ensure the timeliness of hot spot location, early warning and active control, and avoid monitoring misjudgment or control lag caused by time difference.

[0031] For example, such as Figure 3 As shown, when temperature sensors, vibration sensors, and environmental monitoring modules are distributed and deployed inside the distribution cabinet of a substation, a synchronization clock module based on the IEEE 1588 precision clock protocol is integrated to ensure that all sensors synchronously collect data with microsecond-level accuracy (e.g., synchronizing timestamps every 100ms). When an abnormal rise in contact temperature is detected, the synchronization clock module precisely aligns the temperature change data (e.g., an 85°C temperature rise at 14:23:15.001s) with the synchronous vibration signal (abnormal high-frequency vibration at 14:23:15.001s), avoiding misjudgments of "temperature change and vibration asynchrony" caused by sensor clock deviation. Based on the synchronized multimodal data, the system can accurately construct a contact thermal vibration coupling model, predicting thermal faults caused by contact resistance degradation 4 hours in advance. Compared with traditional asynchronous monitoring schemes, the fault warning accuracy is improved by 35%.

[0032] In step S103, after receiving the raw data, the cabinet is divided into spatial micro-elements, a heat flow graph model is constructed, and a graph attention neural network is used to model the heat conduction correlation between nodes. By analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow and load current, a thermal fault probability distribution prediction is output.

[0033] Among them, the heat flow graph model is a graph structure model that divides the internal space of an intelligent power distribution cabinet into spatial micro-elements as nodes and heat conduction paths as edges.

[0034] It is understood that the embodiments of this application analyze the spatiotemporal relationship between node temperature, neighborhood heat flow and load current through graph neural network analysis, accurately capture temperature changes and heat propagation laws, output more accurate thermal fault probability prediction, and perform global correlation analysis on the cabinet temperature field to improve the comprehensiveness and accuracy of thermal fault monitoring.

[0035] In this embodiment, a graph attention neural network is used to model the heat conduction correlation between nodes, including: constructing a heterogeneous graph structure, wherein nodes include temperature monitoring points, vibration sensors, and environmental parameter monitoring points, and edges include heat conduction paths, vibration propagation paths, and environmental influence relationships; dynamically assigning differentiated heat conduction weights to nodes according to the heterogeneous graph structure; updating edge weights according to the heat conduction weights and in combination with real-time temperature gradients, and capturing neighborhood heat flow correlations and temporal load influences respectively through a spatiotemporal dual attention mechanism; enhancing the rare fault identification capability by training the model through adversarial regularization based on neighborhood heat flow correlations and temporal load influences, and dynamically updating graph node attributes and edge connection relationships in combination with real-time data to form an adaptive modeling framework.

[0036] Heterogeneous graph structures are graph structures that contain multiple types of nodes and multiple types of edges.

[0037] It is understood that the embodiments of this application comprehensively model the multiple entities and their complex relationships in the intelligent power distribution cabinet through a graph structure containing multiple types of nodes and multiple types of edges. It supports dynamically assigning differentiated heat conduction weights to nodes, effectively capturing multi-dimensional spatiotemporal coupling relationships such as neighborhood heat flow, vibration propagation, and environmental influences. It combines real-time data to dynamically update the model and enhance the ability to identify rare faults to build an adaptive modeling framework, accurately analyzing the physical relationships and fault evolution laws in complex systems.

[0038] For example, such as Figure 4 As shown, in power transformer fault diagnosis, a heterogeneous graph structure is used to construct a graph model that includes equipment component nodes such as windings, cores, and bushings; monitoring nodes such as temperature sensors, vibration sensors, and oil chromatography sensors; and environmental parameter nodes such as ambient temperature and humidity. The edges cover various types, including heat conduction paths between components, mechanical connections for vibration transmission, gas diffusion relationships in oil, and the influence of environmental parameters on insulation materials. Through this heterogeneous graph structure, the internal multi-entities of the transformer (such as physical components, monitoring devices, and environmental factors) and their complex physical relationships (such as thermal coupling, mechanical vibration transmission, and the influence of chemical changes) can be comprehensively characterized. It supports dynamically assigning differentiated influence weights to different component nodes in the development of thermal faults. Combined with a spatiotemporal attention mechanism, it captures the spatiotemporal coupling characteristics of multi-dimensional data such as oil temperature changes, abnormal vibration signals, and gas concentration fluctuations. At the same time, it utilizes the neighborhood information aggregation capability of graph neural networks to analyze weak correlation signals in the early stages of faults. Based on this, the model can dynamically update node attributes and edge connection strength by integrating multi-source monitoring data in real time, effectively identifying early signs of various types of faults such as winding overheating, core loosening, and bushing insulation deterioration, thereby improving the comprehensiveness of fault diagnosis and early warning capabilities of complex electrical equipment.

[0039] In this embodiment of the application, the formula for the adversarial regularization training model is: ; in, To optimize model parameters; To optimize generator parameters; The task loss function; For hyperparameters; ; The objective function is... These are the parameters of the task model; Parameters for the adversarial or regularization part.

[0040] It is understood that the embodiments of this application apply perturbations to the model by generating adversarial examples, forcing the model to learn more robust feature representations, solving the model bias problem caused by insufficient rare fault samples, enhancing the model's sensitivity to low-probability, high-risk abnormal states, and prompting the model to capture subtle nonlinear correlation differences in the fault evolution process by aligning the feature space of adversarial examples with real samples, avoiding the model from overfitting to common working condition data, improving the generalization ability to small sample fault modes, and combining the dynamic updates of node attributes and edge connections in the graph structure to accurately extract and stably identify rare fault features in complex electrical systems.

[0041] For example, in insulator fault detection for transmission lines, adversarial regularization training models can effectively address the scarcity of rare fault samples such as insulator cracks and damage. Adversarial samples (e.g., infrared thermographic images and vibration signal disturbances simulating minor electrolytic corrosion on the insulator surface) that closely approximate normal operating conditions are created through a generator network. These samples are then input into a graph neural network model, forcing the model to learn robust distinguishing features in the feature spaces of adversarial and real fault samples. During training, the discriminator and generator networks engage in a game-like interaction, prompting the model to capture subtle differences between normal and fault states (e.g., abnormal local temperature rise gradients and vibration frequency mode shifts), avoiding missed faults due to over-reliance on normal samples. This model can accurately identify early signs of insulator degradation that are difficult to detect using traditional methods (e.g., abnormal heat conduction caused by nanoscale cracks), improving the accuracy of rare fault identification by more than 30% and enhancing the fault early warning capabilities of transmission equipment in complex environments.

[0042] In step S104, based on the predicted thermal failure probability distribution, the computing power is dynamically adjusted using the reconfigurable edge computing platform. When a thermal failure risk is predicted, the decision engine is activated, and analysis is performed by combining infrared sensor data and heat flow map model to determine the location of the hot spot.

[0043] Thermal failure risk refers to the possibility that electrical equipment or systems may experience functional abnormalities, performance degradation, or even failure due to localized overheating, and the degree of safety and economic losses that may result.

[0044] It should be noted that the thermal failure risk levels include low risk (thermal failure probability P<0.3): prediction period ≥10 seconds, system runs smoothly, only CPU needs to execute lightweight graph neural network; medium risk (0.3≤P<0.7): prediction period 3-10 seconds, CPU multi-core parallel computing is activated; high risk (P≥0.7): prediction period ≤2 seconds, switch to GPU to execute high-performance graph neural network, accelerate the iteration of heat flow map model.

[0045] It is understood that the embodiments of this application, by fusing infrared sensor data with heat flow model analysis, accurately pinpoint the location of localized overheating in the equipment, providing precise targets for early intervention in thermal faults. Early identification of high-risk areas avoids the limitations of single-point monitoring, and relies on multi-source data fusion and model analysis to improve positioning accuracy. This provides data for subsequent operations such as dynamic calibration of hotspot warning thresholds and proactive fault-tolerant control, reducing the risk of equipment malfunction, performance degradation, and failure due to localized overheating.

[0046] In this embodiment of the application, based on the thermal failure probability distribution prediction results, the computing power is dynamically adjusted using a reconfigurable edge computing platform, including: constructing a heterogeneous computing architecture; determining the risk level based on the heterogeneous computing architecture and the thermal failure probability distribution prediction results; and performing dynamic resource allocation based on the thermal failure risk level. Specifically, when the probability distribution is concentrated and the peak value is much lower than the threshold, it is classified as low risk; when the probability distribution has a certain degree of dispersion and the peak value is close to the threshold, it is classified as medium risk; and when the probability distribution is dispersed and the peak value exceeds the threshold, it is classified as high risk.

[0047] Among them, heterogeneous computing architecture is a hybrid computing system architecture that integrates different types of computing units such as CPU, GPU, FPGA, and ASIC, and improves the overall computing efficiency and energy efficiency of the system by coordinating the processing of diverse tasks.

[0048] It is understood that the embodiments of this application reduce system energy consumption while ensuring the real-time analysis efficiency of the heat flow model by dynamically matching risk levels with computing resources (allocating low-power CPUs for low-risk scenarios and scheduling high-performance GPUs / ASICs for high-risk scenarios). Leveraging the complementary advantages of hybrid computing units, the computing power of the edge computing platform is elastically expanded and finely allocated, avoiding computing bottlenecks in high-risk scenarios, reducing resource idleness under low load, and improving the response speed, energy efficiency, and complex task processing capabilities of the thermal fault monitoring system, thus providing computing power support for real-time risk assessment and dynamic decision-making.

[0049] In step S105, based on the hot spot location information and historical fault data, the self-calibration warning threshold is optimized through Bayesian method. When the probability of thermal fault exceeds the warning threshold, the intelligent heat dissipation and contact adjustment mechanism are linked to perform active fault-tolerant control, and the control results are fed back to the monitoring center.

[0050] The contact adjustment mechanism is a mechanical device used to adjust parameters such as the contact position, pressure, and stroke of the contacts to ensure reliable electrical connection, stable contact resistance, and optimized conductivity.

[0051] It is understood that the embodiments of this application ensure the reliability of electrical connections and the stability of contact resistance by precisely adjusting parameters such as the contact position, pressure, and stroke of the contacts, thereby optimizing conductivity from the root and reducing heat loss and failure risk caused by poor contact. In active fault-tolerant control of thermal faults, it is linked with the intelligent heat dissipation system to dynamically respond to abnormal operating conditions, correct the contact state in real time, improve the stability and safety of equipment operation, extend the service life of electrical equipment, and enhance the overall fault tolerance of the system.

[0052] For example, such as Figure 5 As shown, during the long-term operation of a high-voltage circuit breaker, if the monitoring system detects insufficient contact pressure and abnormally high contact resistance (e.g., a 15% increase in heat generation power compared to normal conditions) due to mechanical wear of the contacts via hot spot location, the contact adjustment mechanism can automatically fine-tune the contact closing position and increase the contact pressure to the rated parameter range (e.g., from 0.8 N / mm² to 1.2 N / mm²) through a servo motor-driven screw transmission component. Simultaneously, the intelligent cooling system is activated to reduce local temperature rise. This process can complete contact state correction within 100ms, reducing contact resistance by 30%, effectively avoiding the risk of arc discharge or equipment burnout caused by poor contact. It is typically used for preventative maintenance of power distribution equipment in intelligent substations, reducing the incidence of contact-related faults by more than 60%.

[0053] In this embodiment, when the probability of a thermal failure exceeds a warning threshold, the intelligent heat dissipation system and the contact adjustment mechanism are linked to perform active fault-tolerant control, including: constructing a Bayesian variational autoencoder; based on the Bayesian variational autoencoder and combined with positioning data, outputting the posterior probability of the thermal failure probability distribution in real time; when the posterior probability of a thermal failure exceeds the warning threshold, active fault-tolerant control is triggered, wherein the intelligent heat dissipation system performs directional cooling power optimization based on a thermal flow field simulation model, and the magnetostrictive actuator of the contact adjustment mechanism performs nanometer-level dynamic compensation of contact resistance.

[0054] Among them, the Bayesian variational autoencoder is a generative model that combines Bayesian inference with variational methods.

[0055] It is understood that the embodiments of this application provide a probabilistic decision-making basis for active fault-tolerant control by quantifying uncertainty, capturing the potential random characteristics of thermal faults, processing complex data distribution to reduce the risk of misjudgment, and enabling the intelligent heat dissipation and contact adjustment mechanism to act precisely according to probability thresholds (such as directional optimization of cooling power and dynamic compensation of contact resistance), thereby improving the robustness and reliability of active control.

[0056] For example, in the prediction of contact faults in high-voltage switchgear, a probabilistic analysis model is constructed using a Bayesian variational autoencoder. Data from contact temperature and contact pressure sensors, as well as vibration signals from the operating mechanism, are input into the model. Bayesian inference and variational methods are used to model the latent variable posterior distribution of contact resistance degradation, and the posterior distribution of the contact fault probability (e.g., the probability confidence level of oxide film thickening and mechanical deformation leading to poor contact) is output in real time. When the posterior probability of the contact fault output by the model exceeds the warning threshold, the system can precisely trigger the contact adaptive adjustment mechanism based on the probability confidence level (e.g., dynamically adjusting the contact spring pressure or triggering a surface coating repair procedure). Simultaneously, the maintenance strategy is optimized by combining the uncertainty quantification results of the probability distribution (e.g., immediate repair for high-confidence faults, and extended monitoring cycles for low-confidence faults). This method captures the potential random characteristics of the contact state through probabilistic learning (e.g., the random influence of environmental humidity on the oxidation rate), effectively handling the noise and distribution uncertainty of multi-source monitoring data, and reducing the contact fault misjudgment rate by 40% compared to traditional deterministic models.

[0057] In this embodiment of the application, the formula for the thermal flow field simulation model is: ; in, This is the term for heat conduction; The thermal conductivity of the material; Rc is the temperature gradient; Rc is the contact resistance of the contact. For current; Let V be the volume of the contact area.

[0058] It is understood that the embodiments of this application analyze the temperature gradient, heat flux density and heat dissipation bottleneck of hot spots to guide the dynamic optimization of directional cooling power (such as accurately allocating cooling resources to high-risk hot spot areas), avoid the energy waste and inefficiency of traditional uniform heat dissipation, combine the fault prediction results of the heat flow map model, and intelligently target the configuration of heat dissipation resources. By simulating the effects of different heat dissipation strategies, the design of contact adjustment, ventilation structure and other aspects is optimized, thereby improving the accuracy and energy efficiency ratio of thermal management in active fault-tolerant control and suppressing the risk of faults caused by local overheating.

[0059] The online temperature monitoring method for intelligent distribution cabinets proposed in this application constructs a comprehensive monitoring system covering the equipment's operating status through multimodal data acquisition of three-dimensional temperature fields, vibration, and environmental parameters, providing multi-source heterogeneous data support for thermal fault analysis. Based on a thermal flow graph model using spatial micro-element partitioning and graph attention neural networks, it accurately captures the spatiotemporal correlation of heat conduction between cabinet nodes, breaking through the limitations of traditional single-point monitoring and achieving dynamic prediction and risk quantification assessment of thermal fault probability distribution. A reconfigurable edge computing platform, combined with infrared sensor data and a thermal flow model, dynamically adjusts computing power in real time and accurately locates hot spots, significantly improving the timeliness and accuracy of fault location. A Bayesian-optimized self-calibrated early warning threshold mechanism adaptively optimizes the early warning strategy based on historical fault data, avoiding the lag defects of fixed thresholds. This, combined with the active fault-tolerant control of intelligent heat dissipation and contact adjustment mechanisms, comprehensively improves the safety and reliability of intelligent distribution cabinet operation, reduces maintenance costs and accident risks, and ensures stable operation of the power system. Therefore, it solves the problems of limited monitoring accuracy and range, lack of adaptability in early warning mechanisms, and insufficient data utilization in existing technologies.

[0060] The following will illustrate a specific embodiment of a method for online temperature monitoring in intelligent power distribution cabinets, such as... Figure 6 As shown, it includes: Multiple types of industrial-grade sensors, including Omega K thermocouples, PCBPiezotronics vibration sensors, and Vaisala temperature and humidity sensors, are deployed in key areas inside the intelligent power distribution cabinet to collect data on temperature, vibration, and environmental conditions. All sensors support the standard Modbus protocol. Simultaneously, a reconfigurable heterogeneous computing module, composed of an NXP X8M CPU, an NVIDIA Jetson Xavier NX GPU, and a Xilinx Zynq UltraScale+ MPSoC FPGA, is configured to balance performance, power consumption, and cost, handling data acquisition and preprocessing.

[0061] A heterogeneous graph structure was constructed, defining 50 nodes containing temperature, vibration, and environmental parameters. Sixty-six edges (60+20+10) were determined through SolidWorks modeling, ANSYS Fluent thermal simulation, and COMSOL Multiphysics mechanical dynamics simulation to account for heat conduction, vibration propagation, and environmental influences. The PyTorch Geometric library was used to encode node features and edge weights. An 8-head multi-head attention mechanism was employed to calculate neighborhood heat flow correlations. Adversarial regularization training was performed based on the WGAN-GP architecture and the TensorFlow Probability library. Using one year's worth of data from 100 power distribution cabinets (including 200 thermal failure cases), the model was trained for 72 hours using stochastic gradient descent to optimize its performance.

[0062] Based on the thermal failure probability distribution, the system is divided into low, medium, and high risk levels. Low-risk cases are handled solely by the CPU, medium-risk cases utilize GPU inference, and high-risk cases combine GPU and FPGA and initiate thermal flow field simulation to achieve dynamic allocation of computing power. When the thermal failure probability exceeds 0.6, infrared thermal imager data is fused, and hot spots are located using a thermal flow map model. Axial flow fans are driven for directional cooling based on thermal flow field simulation, and a PID algorithm is used to adjust the airflow. When the posterior probability of a contact failure output by the Bayesian variational autoencoder exceeds 0.8, a magnetostrictive actuator is triggered to adjust the contact pressure, and a maintenance work order is sent.

[0063] Data is uploaded to the cloud via TCP using a Huawei 5G DTU module, and real-time data is stored in InfluxDB for 30 days. Abnormal data is compressed before transmission. The heat flow model is retrained weekly in the cloud and updated differentially to edge nodes, with versions managed by Git. A web-based monitoring platform based on Vue.js was developed, utilizing ECharts, Three.js, and D3.js to provide real-time visualization of the distribution cabinet's heat flow map, fault probability trend curves, and active control status.

[0064] In summary, this invention achieves real-time acquisition and efficient processing of multi-dimensional data from power distribution cabinets by deploying multiple types of sensors and heterogeneous computing modules; it constructs a heterogeneous graph structure and combines it with graph neural networks and adversarial training to accurately capture complex device relationships and improve thermal fault identification capabilities; it dynamically allocates computing power based on risk levels, locates hot spots, and triggers intelligent heat dissipation and contact adjustment to achieve targeted resource configuration and rapid response; and through edge-cloud collaboration and a visualization platform, it ensures continuous model optimization and intuitive operation and maintenance monitoring, improving fault warning accuracy and equipment operational reliability, while reducing energy consumption and operation and maintenance costs.

[0065] Next, referring to the accompanying drawings, a temperature online monitoring system for intelligent power distribution cabinets according to embodiments of this application is described.

[0066] Figure 7 This is a block diagram of an online temperature monitoring system for an intelligent power distribution cabinet according to an embodiment of this application.

[0067] like Figure 7 As shown, the online temperature monitoring system 10 for intelligent power distribution cabinets includes: a data acquisition module 100, a transmission module 200, a data acquisition module 300, a prediction module 400, and a control module 500.

[0068] The system comprises the following modules: Acquisition module 100, which acquires three-dimensional temperature field, vibration, and environmental parameters; Transmission module 200, which constructs a multimodal monitoring dataset based on the three-dimensional temperature field, vibration, and environmental parameters, and transmits the raw data to the edge computing platform; Construction module 300, which, upon receiving the raw data, divides the cabinet into spatial micro-elements, constructs a heat flow map model, and uses a graph attention neural network to model the heat conduction correlation between nodes. By analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow, and load current, it outputs a predicted thermal fault probability distribution; Prediction module 400, which, based on the predicted thermal fault probability distribution, dynamically adjusts the computing power using the reconfigurable edge computing platform. When a thermal fault risk is predicted, it activates the decision engine, combines infrared sensor data with the heat flow map model, and analyzes the data to determine the hot spot location; Control module 500, based on the hot spot location information and historical fault data, optimizes and self-calibrates the warning threshold using Bayesian optimization. When the thermal fault probability exceeds the warning threshold, it activates the intelligent heat dissipation and contact adjustment mechanism for active fault-tolerant control, and simultaneously feeds the control results back to the monitoring center.

[0069] It should be noted that the foregoing explanation of the embodiment of the online temperature monitoring method for intelligent distribution cabinets also applies to the online temperature monitoring system for intelligent distribution cabinets in this embodiment, and will not be repeated here.

[0070] The online temperature monitoring system for intelligent distribution cabinets proposed in this application constructs a comprehensive monitoring system covering the equipment's operating status through multimodal data acquisition of three-dimensional temperature fields, vibration, and environmental parameters, providing multi-source heterogeneous data support for thermal fault analysis. Based on a thermal flow graph model using spatial micro-element partitioning and graph attention neural networks, it accurately captures the spatiotemporal correlation of heat conduction between cabinet nodes, breaking through the limitations of traditional single-point monitoring and achieving dynamic prediction and risk quantification assessment of thermal fault probability distribution. A reconfigurable edge computing platform, combined with infrared sensor data and a thermal flow model, dynamically adjusts computing power in real time and accurately locates hot spots, significantly improving the timeliness and accuracy of fault location. A Bayesian-optimized self-calibrated early warning threshold mechanism adaptively optimizes the early warning strategy based on historical fault data, avoiding the lag defects of fixed thresholds. This, combined with the active fault-tolerant control of intelligent heat dissipation and contact adjustment mechanisms, comprehensively improves the safety and reliability of intelligent distribution cabinet operation, reduces maintenance costs and accident risks, and ensures stable operation of the power system. Thus, it solves the problems of limited monitoring accuracy and range, lack of adaptability in early warning mechanisms, and insufficient data utilization in existing technologies.

[0071] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0072] When the processor 802 executes the program, it implements the online temperature monitoring method for intelligent power distribution cabinets provided in the above embodiments.

[0073] Furthermore, electronic devices also include: Communication interface 803 is used for communication between memory 801 and processor 802.

[0074] The memory 801 is used to store computer programs that can run on the processor 802.

[0075] The memory 801 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0076] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0077] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0078] The processor 802 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0079] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for online temperature monitoring of an intelligent power distribution cabinet.

[0080] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0081] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0082] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0083] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0084] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0085] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for online temperature monitoring in intelligent power distribution cabinets, characterized in that, include: Collect three-dimensional temperature field, vibration, and environmental parameters; Based on the three-dimensional temperature field, vibration, and environmental parameters, a multimodal monitoring dataset is constructed, and the raw data is transmitted to an edge computing platform. After receiving the raw data, the cabinet is divided into spatial micro-elements, a heat flow map model is constructed, and a graph attention neural network is used to model the heat conduction correlation between nodes. By analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow and load current, a thermal failure probability distribution prediction is output. Based on the predicted thermal failure probability distribution, the computing power is dynamically adjusted using a reconfigurable edge computing platform. When a thermal failure risk is predicted, the decision engine is activated, and the infrared sensor data is combined with the heat flow map model to perform analysis and determine the location of the hot spot. Based on the hot spot location information and historical fault data, the self-calibration warning threshold is optimized through Bayesian methods. When the probability of a thermal fault exceeds the warning threshold, the intelligent heat dissipation and contact adjustment mechanism are linked to perform active fault-tolerant control, and the control results are fed back to the monitoring center.

2. The online temperature monitoring method for intelligent power distribution cabinets according to claim 1, characterized in that, A graph attention neural network is used to model the heat conduction correlation between nodes, including: Construct a heterogeneous graph structure, where nodes include temperature monitoring points, vibration sensors, and environmental parameter monitoring points, and edges include heat conduction paths, vibration propagation paths, and environmental influence relationships; Based on the heterogeneous graph structure, nodes are dynamically assigned differentiated heat conduction weights; Based on the heat conduction weights and combined with the real-time temperature gradient, the edge weights are updated, and the spatiotemporal dual attention mechanism is used to capture the correlation of neighborhood heat flow and the influence of time-series loads respectively. Based on the correlation of neighborhood heat flow and the influence of time-series load, the ability to identify rare faults is enhanced by training the model through adversarial regularization. Combined with real-time data, the graph node attributes and edge connection relationships are dynamically updated to form an adaptive modeling framework.

3. The online temperature monitoring method for intelligent power distribution cabinets according to claim 2, characterized in that, The formula for the adversarial regularization training model is: ； in, To optimize model parameters; To optimize generator parameters; The task loss function; For hyperparameters; ; The objective function is... These are the parameters of the task model; Parameters for the adversarial or regularization part.

4. The online temperature monitoring method for intelligent power distribution cabinets according to claim 1, characterized in that, Based on the predicted thermal failure probability distribution, the computing power is dynamically adjusted using a reconfigurable edge computing platform, including: Build heterogeneous computing architecture; Based on the heterogeneous computing architecture and the predicted thermal failure probability distribution, the risk level is determined. Based on the risk level of thermal failure, dynamic resource allocation is carried out. When the probability distribution is concentrated and the peak value is much lower than the threshold, it is classified as low risk; when the probability distribution is somewhat dispersed and the peak value is close to the threshold, it is classified as medium risk; when the probability distribution is dispersed and the peak value exceeds the threshold, it is classified as high risk.

5. The online temperature monitoring method for intelligent power distribution cabinets according to claim 1, characterized in that, When the probability of a thermal failure exceeds the warning threshold, the intelligent heat dissipation and contact adjustment mechanism are activated for proactive fault-tolerant control, including: Construct a Bayesian variational autoencoder; Based on the Bayesian variational autoencoder and combined with the positioning data, the posterior probability of the thermal fault probability distribution is output in real time. When the posterior probability of a thermal fault exceeds the warning threshold, active fault-tolerant control is triggered. In this process, the intelligent heat dissipation system optimizes the cooling power based on the thermal flow field simulation model, and the magnetostrictive actuator of the contact adjustment mechanism performs nanometer-level dynamic compensation of the contact resistance.

6. The online temperature monitoring method for intelligent power distribution cabinets according to claim 5, characterized in that, The formula for the heat flow field simulation model is: ； in, This is the term for heat conduction; The thermal conductivity of the material; Rc is the temperature gradient; Rc is the contact resistance of the contact. For current; Let V be the volume of the contact area.

7. The online temperature monitoring method for intelligent power distribution cabinets according to claim 1, characterized in that, Based on the aforementioned three-dimensional temperature field, vibration, and environmental parameters, a multimodal monitoring dataset is constructed, including: Construct a synchronous clock module; Based on the synchronization clock module and the sampling timestamp, a synchronization sample sequence is generated; Based on the synchronized sample sequence and combined with the data quality assessment mechanism, feature transformation is performed to construct a multimodal feature dataset. The feature transformation includes converting the three-dimensional temperature field into graph structure data with spatial micro-elements as nodes and heat conduction paths as edges, converting vibration signals into Mel frequency cepstral coefficients, and embedding environmental parameters as global features into graph node attributes.

8. A temperature online monitoring system for intelligent power distribution cabinets, characterized in that, include: The data acquisition module is used to acquire three-dimensional temperature fields, vibration, and environmental parameters. The transmission module is used to construct a multimodal monitoring dataset based on the three-dimensional temperature field, vibration and environmental parameters, and transmit the raw data to the edge computing platform; The construction module is used to divide the cabinet into spatial micro-elements after receiving the raw data, construct a heat flow map model, use a graph attention neural network to model the heat conduction correlation between nodes, and output a thermal failure probability distribution prediction by analyzing the spatiotemporal correlation of node temperature, neighborhood heat flow and load current. The prediction module is used to dynamically adjust the computing power using the reconfigurable edge computing platform based on the predicted thermal failure probability distribution. When a thermal failure risk is predicted, the decision engine is activated to analyze the data by combining infrared sensor data with the heat flow map model to determine the location of the hot spot. The control module is used to optimize the self-calibration warning threshold through Bayesian optimization based on the hot spot location information and historical fault data. When the probability of thermal failure exceeds the warning threshold, it will link the intelligent heat dissipation and contact adjustment mechanism to perform active fault-tolerant control, and at the same time, feed the control results back to the monitoring center.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the online temperature monitoring method for an intelligent distribution cabinet as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the online temperature monitoring method for intelligent power distribution cabinets as described in any one of claims 1-7.

Images