Cable temperature real-time monitoring and early warning method based on Internet of Things

Abstract

The invention relates to the technical field of power equipment state monitoring and signal devices, and particularly discloses a cable temperature real-time monitoring and early warning method based on the Internet of Things, which comprises the following steps of: deploying a temperature sensing node with edge computing capability at a key position of a cable, constructing a dynamic early warning model in which a resource scheduling engine driven by a shuffled frog leaping algorithm and an extreme gradient lifting classifier cooperate, and dynamically adjusting data acquisition frequency, transmitting power and reasoning depth to balance power consumption and precision; early warning information is uploaded to a cloud platform through a low-power wide area network, a global risk situation map is generated, and a three-level early warning mechanism is triggered. According to the technical scheme, self-adaptive, low-power-consumption and high-reliability cable temperature monitoring and early thermal fault early warning are achieved, and the intelligent level and emergency response efficiency of operation and maintenance of a power system are improved.

Description

A method for real-time monitoring and early warning of cable temperature based on the Internet of Things Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring and signal device application technology, specifically relating to a method for real-time monitoring and early warning of cable temperature based on the Internet of Things. Background Technology

[0002] With the rapid evolution of power Internet of Things (IoT) technology, real-time cable temperature monitoring systems have become a crucial infrastructure for ensuring the security and stability of energy transmission networks. By widely deploying temperature sensing nodes at cable joints and critical locations, the system can continuously monitor the power grid's operational status, providing key data for early detection of electrical faults and fire prevention. This monitoring model utilizes distributed sensing terminals to acquire high-frequency temperature rise data and relies on a back-end analysis platform for risk assessment, making it a core component of modern smart grid digital operation and maintenance systems.

[0003] Adaptive early warning technology combining edge computing and swarm intelligence optimization algorithms is a current research focus in the field of cable monitoring. This technology aims to combine machine learning models with heuristic optimization strategies to accurately extract and identify fault characteristics by analyzing multi-dimensional status data in real time at the monitoring terminal. By dynamically scheduling computing resources, the system can collaboratively configure the data acquisition frequency and algorithm inference depth based on the cable's operating environment, load status, and remaining power at the nodes, aiming to achieve low-power operation while ensuring real-time early warning.

[0004] Existing technologies often face limitations in handling cable monitoring tasks, struggling to balance energy efficiency control and early warning performance. Due to the limited power supply capabilities of monitoring nodes, especially passive RF terminals, high-intensity computing tasks lead to a surge in power consumption, hindering the continuous deployment of complex algorithms at the edge. Furthermore, traditional early warning schemes often rely on fixed-parameter models, failing to effectively compensate for physical characteristic deviations caused by cable aging and seasonal changes, resulting in insufficient early warning accuracy under complex operating conditions. In addition, existing systems lack deep coordination between computing resources and environmental characteristics, making it difficult to flexibly adjust monitoring strategies based on real-time load fluctuations, leading to performance issues when handling nonlinear state changes. Summary of the Invention

[0005] The purpose of this invention is to provide a method for real-time monitoring and early warning of cable temperature based on the Internet of Things, which can solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the technical solution adopted by this invention is as follows: a method for real-time monitoring and early warning of cable temperature based on the Internet of Things, comprising the following specific steps: Step 1: Deploying temperature sensing nodes with edge computing capabilities at key locations along the cable. Each temperature sensing node integrates an environmental sensing module, a wireless communication module, and a local computing unit to collect cable surface temperature data and perform local inference tasks; Step 2: Constructing a dynamic early warning model that integrates swarm intelligence optimization and ensemble learning. This dynamic early warning model is composed of a resource scheduling engine driven by a hybrid frog-leap algorithm and an extreme gradient boosting classifier, used to adaptively adjust monitoring strategies according to operating conditions. (Omitted); Step 3: The resource scheduling engine evaluates the current environmental status and node energy consumption level in real time, and dynamically configures the data acquisition frequency, wireless transmission power and classifier inference depth to reduce the overall power consumption of the system while ensuring the accuracy of the early warning; Step 4: The locally processed early warning information is uploaded to the cloud monitoring platform through a low-power wide area network. The cloud monitoring platform aggregates and analyzes the received data and generates a global risk situation map; Step 5: A multi-level early warning mechanism is triggered based on the global risk situation map. When the local temperature rise rate or absolute temperature value exceeds the preset threshold, the system automatically pushes an alarm command to the operation and maintenance terminal and activates the emergency plan.

[0007] Preferably, in step 1, the temperature sensing node adopts a passive radio frequency power supply architecture, and its energy acquisition depends on the electromagnetic field induction around the cable. It is also equipped with a micro energy storage unit to maintain short-term high-load operation, ensuring that critical data processing tasks can still be completed in a weak current environment.

[0008] Preferably, in step 2, the hybrid frog jumping algorithm simulates the foraging behavior of frog groups, dividing the entire optimization process into multiple subgroups. Each subgroup performs a local search, and the subgroups periodically exchange optimal solutions to achieve global exploration, quickly approximating the Pareto optimal solution set under limited computing resources.

[0009] Preferably, in step 2, the extreme gradient boosting classifier is trained based on historical fault samples. Its input features include temperature change rate, load current fluctuation amplitude, ambient humidity and cable service life. The output is a fault probability score, which is used to determine whether there is a potential thermal fault risk.

[0010] Preferably, the adjustment range of the data acquisition frequency in step 3 covers the range from once per second to once per minute. The specific value is determined by the resource scheduling engine based on the current load status and remaining power. During low-load periods at night, the low-frequency acquisition mode is preferred to extend the node's battery life.

[0011] Preferably, in step 3, the wireless transmission power is dynamically adjusted according to the transmission distance and channel quality to minimize the power consumption of the radio frequency module while ensuring communication reliability, and to avoid energy waste caused by frequent retransmissions.

[0012] Preferably, in step 3, the classifier inference depth refers to the number of decision tree layers in which the model participates in the calculation. The lightweight mode only uses the first few layers for rough judgment, while the high-precision mode activates all layers to improve the recognition accuracy. The switching between the two is uniformly controlled by the resource scheduling engine.

[0013] Preferably, in step 4, the low-power wide-area network adopts a narrowband IoT protocol stack, which supports massive terminal access and long-distance transmission, and has good penetration capability, making it suitable for stable communication in complex electromagnetic environments such as underground pipe corridors and tunnels.

[0014] Preferably, the multi-level early warning mechanism in step 5 is divided into three response levels. The first-level early warning corresponds to a minor anomaly, and only logs are recorded for subsequent analysis; the second-level early warning indicates that there is a significant temperature rise trend, which requires manual verification; and the third-level early warning indicates that the danger threshold has been reached, and power must be cut off for maintenance immediately.

[0015] Preferably, the cloud monitoring platform has a built-in aging compensation module, which can model and correct the drift of physical characteristics by combining the cable commissioning time and cumulative operating hours, thereby eliminating false alarms caused by material deterioration.

[0016] Preferably, the method further includes a seasonal adaptive calibration function, in which the system periodically compares the temperature distribution curves of the same period in history and automatically updates the benchmark parameters, so that the early warning logic can match the normal temperature change pattern under different climatic conditions.

[0017] Preferably, the temperature sensing node also integrates a self-diagnostic unit, which can periodically detect the sensor sensitivity decay and proactively report a maintenance request before the performance drops to a critical level, preventing monitoring blind spots due to hardware failure.

[0018] Preferably, the dynamic early warning model supports online incremental learning. Whenever a new real-world fault case is received, the system can fine-tune the classification boundary without affecting real-time services, thereby continuously improving its ability to generalize to unknown fault types in the future.

[0019] Preferably, the method further introduces a spatiotemporal correlation analysis mechanism, which not only focuses on single-point temperature abrupt changes, but also combines historical data of adjacent nodes for horizontal comparison to identify chain overheating events with propagation characteristics, thereby improving the reliability of early warning.

[0020] Compared with existing technologies, this invention has the following advantages: 1. By deeply integrating swarm intelligence optimization algorithms with ensemble learning models, a dynamic balance between computing resources and monitoring needs is achieved, reducing the energy consumption burden of edge nodes while ensuring the accuracy of early warnings; 2. The traditional fixed threshold alarm mechanism is abandoned in favor of an adaptive early warning logic based on environmental perception and equipment status, overcoming the impact of physical characteristic changes caused by cable aging and seasonal changes on system stability; 3. A complete closed-loop system covering local inference, cloud collaboration, and multi-level response is constructed, improving the intelligence level and emergency response efficiency of the entire monitoring network; The proposed hybrid leapfrog scheduling strategy can maximize system availability under limited energy constraints, and is particularly suitable for long-term unattended operation in large-scale deployment scenarios. Attached Figure Description

[0021] Figure 1 is a flowchart of the overall technical solution according to the present invention; Figure 2 is a schematic diagram of data flow according to the present invention; Figure 3 is a flowchart of dynamic early warning in the present invention, which is based on the resource scheduling engine driven by the hybrid frog leaping algorithm and the extreme gradient boosting classifier; Figure 4 is a flowchart of dynamically configuring the data acquisition frequency, wireless transmission power and classifier inference depth according to the environmental state and node energy consumption according to the present invention; Figure 5 is a flowchart of generating a global risk situation map and triggering a multi-level early warning mechanism by combining aging compensation and seasonal adaptive calibration according to the present invention. Detailed Implementation

[0022] Example 1: Please refer to Figures 1 to 5. To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0023] In the IoT-based real-time monitoring and early warning method for cable temperature, step 1 involves deploying temperature sensing nodes with edge computing capabilities at key locations along the cable. Specifically, the deployment density of the temperature sensing nodes is finely designed based on the cable's voltage level, laying environment, and historical fault frequency points. Each temperature sensing node integrates an environmental sensing module, a wireless communication module, and a local computing unit. The environmental sensing module not only includes a high-precision contact temperature sensor for real-time acquisition of the absolute temperature of the cable surface, but also integrates an environmental humidity sensor, an environmental temperature sensor, and a high-sensitivity vibration sensor to construct a multi-dimensional operating environment feature vector. The local computing unit uses a high-performance, low-power embedded microprocessor, which internally divides the space into a dedicated data processing area, an algorithm execution area, and a temporary storage area to handle complex local inference tasks and preliminary data cleaning logic.

[0024] The temperature sensing node employs a passive radio frequency power supply architecture. Its energy acquisition process is highly dependent on the alternating electromagnetic field generated around the cable. The node integrates a high-efficiency electromagnetic induction coil. When alternating current flows through the cable, the coil generates an induced electromotive force (EMF) based on the principle of electromagnetic induction. This EMF is converted to direct current by a rectifier circuit and then powers the system through a voltage regulation module. To address energy supply interruptions under no-load or low-current conditions, the node is equipped with a micro-energy storage unit. This micro-energy storage unit uses long-life supercapacitors or high-energy-density thin-film batteries, capable of storing excess energy when electromagnetic energy is abundant. When energy acquisition is insufficient or the system is performing high-load computational tasks (such as complex AI inference), the power management logic automatically switches to energy storage power supply mode, ensuring the continuity of monitoring tasks and the real-time issuance of critical early warning information.

[0025] In the above method, step 2 involves constructing a dynamic early warning model that integrates swarm intelligence optimization and ensemble learning. This model logically consists of a two-layer collaborative architecture: the bottom layer is a fault classifier based on the Extreme Gradient Boosting (XGBoost) algorithm, and the top layer is a resource scheduling engine based on the Hybrid Frog Leaping Algorithm (SFLA). During the construction of the Extreme Gradient Boosting classifier, it is trained offline using a large number of historical fault samples. Its input features include, but are not limited to: the real-time temperature change rate of the cable (i.e., the time derivative of the difference between the current temperature and the previous temperature), the fluctuation amplitude of the load current (reflecting the dynamic characteristics of the heat source input), changes in ambient humidity (reflecting the potential changes in thermal stability caused by moisture in the insulation layer), and the cumulative service life of the cable (a physical benchmark characterizing material aging). The output of the classifier is a continuous value between 0 and 1, representing a probability score of fault occurrence, used to quantify the existence of potential thermal fault risks.

[0026] The hybrid frog-jumping algorithm drives the resource scheduling engine by simulating the foraging behavior of a frog swarm. The specific operation includes: First, the system initializes a frog swarm containing multiple individuals, each representing a set of system resource configuration parameters (such as acquisition frequency, transmission power, and model depth). Next, the entire swarm is divided into multiple subgroups, and a local search process is performed within each subgroup. Within each subgroup, the individual with the worst fitness is identified and guided to update its position towards the optimal individual within the subgroup or the globally optimal individual. This position update is described logically as: progressively adjusting the parameter configuration of the worst individual towards the parameter configuration of the target individual, with the adjustment amount determined by a step size factor that dynamically decays with the number of iterations. Subgroups periodically exchange information, that is, summarizing and re-dividing the optimal solutions of each subgroup to achieve path exploration globally. Through this multi-level collaborative evolution, the system can quickly find the Pareto optimal solution between energy consumption and early warning accuracy under limited edge computing resource constraints.

[0027] In the above method, step 3 involves dynamically configuring system parameters through a resource scheduling engine. The resource scheduling engine monitors two key dimensions in real time: environmental conditions (including cable load strength and extreme ambient temperature) and node energy efficiency levels (including current energy acquisition rate and remaining power percentage of energy storage units). Based on real-time feedback from these two dimensions, the engine dynamically adjusts the data acquisition frequency, wireless transmission power, and classifier inference depth at the millisecond level.

[0028] The data acquisition frequency is set between once per second and once per minute. When the cable is operating under peak load or when an abnormally large increase in the slope of temperature change is detected, the resource scheduling engine switches the acquisition frequency to a high-frequency mode (e.g., once per second) to capture minute transient temperature rises. During low-load periods at night or when the ambient temperature is stable, the system automatically reduces the acquisition frequency to a low-frequency mode (e.g., once per minute), significantly reducing the operating time of the sensing module and local computing unit.

[0029] Regarding the adjustment of wireless transmission power, the system has established a communication quality evaluation mechanism. This mechanism dynamically calculates the minimum power required to maintain reliable communication by monitoring the Received Signal Strength Indication (RSSI) and historical packet loss rate. When the channel quality between the node and the base station is excellent, the transmission current of the RF module is automatically reduced; when increased electromagnetic interference or signal attenuation is detected, the transmission power is appropriately increased to ensure that early warning information is not lost due to signal interruption.

[0030] The classifier inference depth refers to the number of decision tree layers involved in the computation of the extreme gradient boosting model. In normal monitoring mode, the resource scheduling engine control system only calls the first few layers of the basic decision trees (i.e., lightweight mode) to perform basic anomaly screening. At this time, the computational logic is relatively simple and power consumption is low. Once the lightweight inference result shows that the fault probability score exceeds the preset preliminary warning line, the system immediately activates all decision tree layers of the model (i.e., high-precision mode) to perform deep feature fusion and judgment to obtain the fault classification result with the highest confidence. This shallow-to-deep inference strategy effectively balances the energy-saving requirements of routine monitoring with the need for accurate identification during sudden faults.

[0031] In the above method, step 4 involves uploading the early warning information to the cloud monitoring platform via a Low Power Wide Area Network (LPWAN). The LPWAN uses a Narrow Band Internet of Things (NB-IoT) protocol stack, which supports extremely high-density connection access and deep coverage. During data upload, edge nodes perform feature compression on the original temperature sequence, transmitting only key statistics (such as mean, maximum, and rate of change) and the risk level derived locally. The cloud monitoring platform performs spatiotemporal aggregation analysis on the received multi-node data. By establishing a digital twin mapping of the entire cable line, the temperature data of each discrete node is mapped to a geographic coordinate system, generating a global risk situation map. This situation map displays the thermal pressure distribution of different sections in the form of a heat map and supports the retrospective analysis of historical trajectories.

[0032] To eliminate the interference of physical aging on the accuracy of early warnings, the cloud monitoring platform has a built-in aging compensation module. This module constructs a material performance degradation model based on the commissioning date, cumulative operating hours, and number of long-term thermal cycles experienced by each cable. The material performance degradation model is described logically using text: the system periodically calculates an aging correction coefficient, which exhibits a non-linear growth trend with increasing cable service time. This aging correction coefficient is applied to the original temperature threshold uploaded by local nodes, achieving dynamic calibration of physical characteristic drift and avoiding false alarms caused by natural aging of the cable insulation layer.

[0033] In the above method, step 5 involves triggering a multi-level early warning mechanism. The system has a strict preset response hierarchy: Level 1 Early Warning (Blue Level): Triggered when the system detects a slight fluctuation in local temperature, but before a significant temperature rise trend has formed. At this time, the system only records the event in the local log and synchronizes a low-frequency heartbeat packet to the cloud for subsequent operational analysis, without triggering manual intervention. Level 2 Early Warning (Orange Level): Triggered when the temperature rise rate exceeds the preset unit time change threshold, or when the absolute temperature value enters the warning range. The system immediately pushes a detailed anomaly report to the mobile terminal of the maintenance personnel, including the specific geographical location, current load current, and predicted temperature rise curve, requiring manual remote video verification or on-site inspection. Level 3 Early Warning (Red Level): Triggered when the local temperature rise rate shows an exponential increase, or when the absolute temperature value exceeds the limit tolerance temperature of the insulation material. At this time, the system automatically activates the emergency plan, issues an emergency trip command to the circuit breaker control circuit (in scenarios with automatic power-off conditions), and simultaneously activates all surrounding sensing nodes to enter the highest frequency acquisition mode to track the thermal runaway process in real time and prevent the accident from spreading.

[0034] Furthermore, this embodiment also introduces a seasonal adaptive calibration function. The system maintains a feature library containing environmental parameters for all four seasons and periodically compares the current temperature distribution curve with the historical baseline curve for the same period. If the current temperature rise is a natural mapping caused by a seasonal increase in ambient temperature, the system will automatically shift the warning baseline to ensure that the warning logic can accurately distinguish between normal climate fluctuations and actual cable overheating faults.

[0035] The temperature sensing node also integrates a self-diagnostic unit. This unit periodically performs hardware self-tests, including detecting the resistance drift of the thermistor, the VSWR of the RF module, and the internal resistance of the micro-energy storage unit. Once the hardware performance is detected to have degraded to a preset critical level (e.g., sensor sensitivity attenuation exceeding 15%), the node will use its remaining power to send a maintenance request packet, informing the maintenance platform of the specific hardware failure risk and preventing monitoring blind spots caused by the hardware operating in harsh environments for extended periods.

[0036] The dynamic early warning model also possesses online incremental learning capabilities. Whenever maintenance personnel report a real fault case or rule out a false alarm, the feature data of that case is labeled and fed back to the training engine. The system uses these new samples to fine-tune the decision boundary of the extreme gradient boosting classifier, updating the knowledge base without retraining the entire model. This continuous learning mechanism enables the system to constantly evolve, possessing a stronger generalization ability to recognize complex fault modes that may appear in the future and have never been seen before.

[0037] Furthermore, to enhance the reliability of early warnings, the system introduces a spatiotemporal correlation analysis mechanism. The cloud-based monitoring platform not only analyzes the numerical changes of individual sensing nodes in isolation but also extracts data from adjacent nodes for horizontal comparison. If multiple adjacent nodes simultaneously detect a temperature rise trend, and the starting points of the temperature rise exhibit a logical temporal sequence, the system will identify this as a cascading overheating event with propagation characteristics (such as a pipe gallery fire or a through-type cable fault), thus improving the authority and accuracy of early warnings.

[0038] Example 2: Based on Example 1, this example further refines the communication mechanism and data aggregation logic for the complex electromagnetic environment of large-scale urban underground utility tunnels.

[0039] In the IoT-based real-time cable temperature monitoring and early warning method, the temperature sensing node in step 1 further enhances its anti-interference capability. Considering the strong alternating electromagnetic interference generated by high-voltage cables inside the utility tunnel, the wireless communication module adopts frequency-hopping spread spectrum technology. When performing wireless transmission tasks, the system rapidly switches between multiple sub-channels according to a pre-agreed pseudo-random sequence. This switching process can be described logically as follows: the system evaluates the signal-to-noise ratio of each sub-channel in real time, prioritizes frequency bands with interference levels lower than the set value for data carrier transmission, and ensures that early warning information can be transmitted with zero bit error rate in extremely complex electromagnetic pulse environments.

[0040] In step 2, this embodiment introduces a cost-sensitive learning mechanism for training the extreme gradient boosting classifier. Since cable faults are low-probability events, the historical data of normally operating cables far outweighs the fault data. To address the imbalanced sample problem, the system assigns higher weight coefficients to negative samples representing faults in the loss function construction. When the classifier misclassifies true fault samples, the resulting loss value is amplified several times, while the misclassification of normal samples retains its normal weight. This weight allocation strategy forces the model to be more inclined to capture fault features during iterative optimization, improving the system's sensitivity to sudden thermal accidents.

[0041] Regarding the resource scheduling engine in step 3, this embodiment refines the fitness evaluation function of the hybrid leapfrog algorithm. This function not only considers the accuracy of fault detection and the energy consumption of nodes, but also introduces the dimension of information timeliness. The specific calculation logic is as follows: the system calculates the impact weight of early warning delay on potential economic losses. When the cable is under heavy load, this impact weight increases rapidly, guiding the hybrid leapfrog algorithm to allocate resources towards high-frequency acquisition and high-power transmission, ensuring absolute real-time information even at the cost of some energy efficiency.

[0042] In the data aggregation stage of step 4, the cloud monitoring platform introduces topology consistency verification. Due to the complex environment of underground utility tunnels, nodes may experience physical displacement or damage after deployment due to external factors. The platform automatically constructs a logical topology by analyzing the correlation coefficients of data reported by adjacent nodes. If the temperature change of a certain node consistently mismatches with its physical neighbors, the system will automatically trigger topology reconstruction logic to re-identify the actual spatial affiliation of the corresponding node, ensuring the spatial mapping accuracy of the global risk situation map.

[0043] For the multi-level early warning in step 5, this embodiment adds a group collaborative confirmation step. When a single node triggers a level 2 or 3 early warning, the system no longer immediately issues an alarm command, but first sends a collaborative observation request to other nodes within a 5-meter radius of the corresponding node. After receiving the request, the surrounding nodes immediately and temporarily increase the sensitivity of their sensing modules. Only when more than 2 / 3 of the collaborative nodes confirm the presence of abnormal temperature rise will the early warning command be finally confirmed as valid and pushed to the operation and maintenance terminal. This early warning confirmation mechanism based on spatial consensus reduces the false alarm rate caused by single-point sensor failure or localized sporadic thermal interference.

[0044] Example 3: This example focuses on describing the long-term operation guarantee mechanism of the present invention in extreme climatic environments, especially the application details of cable monitoring in frigid regions.

[0045] In step 1, the micro-energy storage unit of the temperature sensing node incorporates thermal protection logic. When the ambient temperature drops below zero, the decreased electrolyte activity leads to a sharp reduction in energy storage capacity. The node integrates tiny self-heating resistors, directly driven by the energy harvesting module. When the system senses that the ambient temperature is too low and the energy storage unit voltage is dropping too rapidly, it allocates a portion of the electromagnetic induction energy to the heating resistors, maintaining the energy storage unit within its optimal operating temperature range. This ensures that the edge computing unit still has sufficient peak power to perform complex hybrid leapfrog scheduling operations even in cold winter conditions.

[0046] In step 2, when building the model, the temperature history fluctuation variance was added to the feature input of the extreme gradient boosting classifier. During cold seasons, drastic fluctuations in ambient temperature (such as cold waves) can interfere with the surface temperature of the cable. By introducing variance features, the model can identify which temperature changes are due to compensatory heat conduction caused by a rapid drop in ambient temperature, and which are due to abnormal heat generation caused by internal defects in the cable.

[0047] Regarding the dynamic configuration logic in step 3, the resource scheduling engine automatically adjusts the search space of the hybrid frog-leaping algorithm based on seasonal characteristics. In winter, due to the better overall heat dissipation of the cables, the system relaxes the warning restrictions on absolute temperature but tightens the monitoring of the temperature rise slope. The scheduling engine allocates more computational weight to the slope analysis model while reducing the heartbeat packet frequency of the wireless transmission module to save energy for more refined feature extraction.

[0048] In step 4, the seasonal adaptive calibration function of the cloud monitoring platform uses a sliding window comparison method. The system maintains an environmental temperature benchmark window for 30 days, updating the statistical average value within the window daily. By calculating the deviation between the current real-time temperature and this dynamic benchmark, the system can accurately offset the slow temperature rise during seasonal changes, ensuring that the warning benchmark remains within a reasonable dynamic fluctuation range, rather than a rigid fixed value.

[0049] After the warning is triggered in step 5, the system also has an emergency load recommendation function. When a level 3 warning is triggered and the power cannot be cut off immediately, the cloud platform combines the current cable load curve and the entire network power dispatch data, and provides specific load reduction recommendations by executing simulation exercise calculation logic. This recommendation is described in textual logic as follows: calculate the power proportion of each branch circuit, prioritize cutting off the power load in non-core areas, until the estimated temperature rise trend of the target section tends to level off. This function expands the monitoring system from a simple sensor to an intelligent management terminal with decision-making support capabilities.

[0050] Example 4: This example details the data security processing logic and long-term meta-model self-evolution process involved in this invention.

[0051] In the data acquisition phase of step 1, the sensing node performs local desensitization and encryption on all raw temperature data. The local computing unit uses a lightweight encryption algorithm to obfuscate the voltage values ​​output by the sensors, generating an encrypted digital sequence. Only a cloud platform with a valid authorization key can restore this digital sequence. This process effectively prevents the illegal interception of monitoring data during transmission over a low-power wide-area network, ensuring the security of core business data related to cable operating status.

[0052] In the model architecture of step 2, the system introduces a meta-learning mechanism. As monitoring time accumulates, data from different geographical locations and cable types constitutes a massive task set. The cloud platform uses these task sets to train the model's model, i.e., the meta-learner. The goal of the meta-learner is to master how to quickly adjust extreme gradients to improve the hyperparameters of the classifier. When a new monitoring node is deployed to a completely new environment, the meta-learner can quickly generate a customized set of initial model weights for the corresponding node based on a very small amount of initial data (e.g., one day's worth of running data), shortening the system's online trial period.

[0053] In the scheduling engine of step 3, the iterative logic of the hybrid frog-leaping algorithm adds a strategy to avoid communication collisions. Since LPWAN networks face collision risks when nodes are densely deployed, the scheduling engine introduces a random offset based on the node's physical address hash value when determining the transmission time of each node. In this way, the upload times of warnings from a large number of nodes are staggered at a microscale, avoiding network congestion and ensuring deterministic delay of warning information in complex network environments.

[0054] In step 4, the aging compensation module of the cloud platform introduces cumulative damage calculation logic based on the theory of chemical reaction rates. The system integrates and accumulates the time periods during which the cable operates at high temperatures, and calculates the probability density of molecular chain breakage in the insulation material according to a verbalized variant of the Arrhenius law. When this probability density reaches a critical threshold, the system will automatically generate a health warning report even if no immediate high temperature is detected, prompting the management department that the insulation life of this section of the cable is nearing its end and a replacement plan needs to be arranged in advance.

[0055] In step 5, the system also integrates closed-loop feedback verification. Whenever the early warning mechanism is triggered and performs a corresponding action (such as pushing an alarm to the terminal), the system continuously monitors the temperature change curve over the next 30 minutes. If the temperature does not drop or stabilize as expected after the alarm is issued, the system will determine that the early warning intervention is ineffective and automatically escalate the response level, skipping the second-level intervention and directly entering the third-level emergency response process. This closed-loop logic based on intervention effectiveness evaluation improves the system's success rate in handling serious incidents.

[0056] Example 5: This example describes the distributed deployment and collaborative monitoring implementation of the present invention in a multi-level power distribution network in a large industrial park.

[0057] In step 1, the sensing nodes within the park are divided into several autonomous regions. Each region selects a node with stronger energy acquisition capabilities or more stable power supply conditions as its regional coordination center. The remaining nodes act as subordinate sensing terminals, prioritizing local inference. This hierarchical deployment model not only reduces the frequency of long-distance transmission from individual nodes to the cloud but also enables rapid inter-regional collaboration.

[0058] In the model collaboration step 2, the extreme gradient boosting classifier is deconstructed into a distributed feature extraction model. Each sensing node extracts high-order temperature features locally, while the complex nonlinear fusion logic is handled by a regional coordination center or cloud platform. This distributed computing architecture fully utilizes the computing power at the edge while maintaining the depth of global decision-making.

[0059] Regarding the dynamic scheduling in step 3, the hybrid frog-leap algorithm in this embodiment introduces regional energy consumption balance constraints. When performing resource allocation optimization, the algorithm not only pursues Pareto optimality for individual nodes but also requires that the average remaining power of all nodes in the region remain consistent. Specifically, when the power of a node's energy storage unit is low, adjacent nodes will automatically take over some of its monitoring tasks (e.g., by increasing the sensing range or increasing the inspection frequency to cover the blind spots of weak-current nodes), while the weak-current nodes enter deep sleep mode to accumulate energy.

[0060] In step 4, the cloud monitoring platform enhances its prediction of the overall risk situation by introducing long short-term memory (LSTM) logic. The system analyzes the sequential characteristics of temperature changes over the past 24 hours and predicts the temperature trend over the next 2 hours. If the prediction indicates that a certain area is about to reach a dangerous threshold, the system will issue a scheduling instruction in advance, guiding the resource scheduling engine to enter a high-frequency data collection state, thus advancing the warning trigger time and providing more buffer time for operation and maintenance.

[0061] In step 5, during the multi-level early warning execution, the system added a linkage interface with the park's fire protection system. When a level 3 early warning is triggered and accompanied by an ambient smoke detector alarm, the warning information will be simultaneously pushed to the park's automated fire control center. This cross-system business collaboration allows secondary disasters caused by cable faults to be nipped in the bud.

[0062] Example 6: This example details the self-healing and robustness enhancement technology of sensing nodes to ensure the system has an ultra-long service cycle in an unattended environment.

[0063] In step 1, the hardware design of the temperature sensing node employs a redundant sensing architecture. Each node integrates two independent temperature sensing units, using different operating principles (e.g., one is a thermistor, and the other is an infrared sensor). The local computing unit compares the readings of the two units in real time. When the difference exceeds a reasonable range, the self-diagnostic unit initiates logical arbitration: by combining the historical correlation between ambient humidity and current load, it identifies and removes the drifting unit, switches to the backup unit to continue operating, and marks the sensor as being in a sub-healthy state in the reported information.

[0064] In the model update phase of step 2, this embodiment describes a sample augmentation logic based on a generative adversarial network. To improve the ability of the extreme gradient boosting classifier to identify rare and extreme faults, the cloud platform uses existing fault samples as seeds and synthesizes a large amount of virtual feature data simulating extreme working conditions through a generative adversarial process described in textual logic. Adding this virtual data to the training set enables the model to maintain its sharp discriminative ability when facing faults that rarely occur in reality but have extremely serious consequences (such as instantaneous overheating caused by lightning strikes).

[0065] For the dynamic adjustment of wireless transmission power in step 3, this embodiment introduces a cooperative relay mode. When a sensing node located deep underground detects extremely poor channel quality and excessively high single-transmission power consumption, the resource scheduling engine initiates a relay search. It sends short-range probe signals to surrounding nodes, searching for neighboring nodes with better channel conditions to serve as data relay points. This multi-hop transmission process can be described logically as follows: the original warning information is divided into multiple small data packets, relayed by neighboring nodes, and finally delivered to the gateway. This method avoids a single node exhausting its power by forcibly increasing transmission power.

[0066] In the cloud-based analysis in step 4, the aging compensation module further considers the thermal fatigue effect. Cables experience minute thermal expansion and contraction during alternating day and night loads, which, over time, can lead to microcracks in the metal sheath. By analyzing the peak-to-valley differences in historical load curves, the system establishes a thermal fatigue damage accumulation matrix, incorporating this dimension into the global risk situation map, thus achieving a leap from single-point overheating monitoring to system fatigue assessment.

[0067] In step 5, the multi-level early warning mechanism adds a geofencing function for maintenance personnel. When a level 2 or 3 early warning is triggered, the cloud platform automatically scans the real-time geographical location of all maintenance personnel. Early warning instructions are prioritized and pushed to the mobile devices of individuals closest to the fault location who possess the corresponding maintenance qualifications. Simultaneously, the system automatically plans the optimal inspection route and pushes it to the navigation interface, ensuring minimal response time.

[0068] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time monitoring and early warning of cable temperature based on the Internet of Things, characterized in that, The method includes the following steps: Step 1, deploying temperature sensing nodes with edge computing capabilities along the cable line. The temperature sensing nodes integrate an environmental sensing module, a wireless communication module, and a local computing unit to collect cable surface temperature data and perform local inference tasks; Step 2, constructing a dynamic early warning model composed of a resource scheduling engine driven by a hybrid frog-leap algorithm and an extreme gradient boosting classifier, used to adjust the monitoring strategy according to the operating conditions; Step 3, evaluating the environmental status and node energy consumption level through the resource scheduling engine, and dynamically configuring the data acquisition frequency, wireless transmission power, and classifier inference depth; Step 4, uploading the locally processed early warning information to the cloud monitoring platform through a low-power wide-area network. The cloud monitoring platform aggregates and analyzes the received data and generates a global risk situation map; Step 5, triggering a multi-level early warning mechanism based on the global risk situation map. When the local temperature rise rate or absolute temperature value exceeds a preset threshold, an alarm command is pushed to the operation and maintenance terminal and an emergency plan is activated.

2. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: In step 1, the deployment density of the temperature sensing nodes is set based on the cable's voltage level, laying environment, and historical fault frequency points. The environmental sensing module includes a contact temperature sensor, an ambient humidity sensor, an ambient temperature sensor, and a vibration sensor to construct a multi-dimensional operating environment feature vector. The local computing unit is internally divided into a data processing area, an algorithm execution area, and a temporary storage area. The temperature sensing node adopts a passive radio frequency power supply architecture, and its energy acquisition depends on the alternating electromagnetic field generated around the cable. The node integrates an induction coil to generate an induced electromotive force based on the principle of electromagnetic induction, and converts the induced electromotive force into direct current through a rectifier circuit. The temperature sensing node is equipped with a micro energy storage unit composed of a supercapacitor or thin-film battery, which is used to switch to energy storage power supply mode by power management logic when the electromagnetic energy acquisition is lower than a preset operating threshold, or when the system is performing a preset high-load computing task.

3. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: The execution process of the hybrid frog-jumping algorithm in step 2 is as follows: The system initializes a frog population containing multiple individuals, each representing a set of system resource configuration parameters, including acquisition frequency, transmission power, and model inference depth; the frog population is divided into multiple subgroups, and a local search is performed within each subgroup to identify individuals with fitness values ​​in the lowest preset range; the parameter configuration of the individuals with fitness values ​​in the subgroup is adjusted stepwise towards the optimal individual parameter configuration direction within the subgroup or the globally optimal individual parameter configuration direction, and the adjustment amount is determined by a step size factor that dynamically decays with the number of iterations; information is exchanged periodically between the subgroups, and the optimal solutions of each subgroup are summarized and re-divided to find the Pareto optimal solution set between energy consumption and early warning accuracy under the constraint of limited edge computing resources.

4. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: In step 2, the extreme gradient boosting classifier is trained based on historical fault samples. Its input features include the real-time temperature change rate of the cable, the fluctuation range of the load current, the change of ambient humidity, and the cumulative service life of the cable. The real-time temperature change rate of the cable is the result of taking the time derivative of the difference between the current temperature and the previous temperature. The output of the extreme gradient boosting classifier is a continuous value between 0 and 1, which is used to represent the probability score of the fault occurrence, so as to quantify whether there is a potential thermal fault risk. A cost-sensitive learning mechanism is introduced during the training process. By assigning a higher weight coefficient to the negative samples representing faults than to the positive samples in the loss function, the loss value generated by the classifier for negative samples is greater than the loss value generated for positive samples when discrimination bias occurs.

5. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: In step 3, the resource scheduling engine monitors the cable load intensity, extreme ambient temperature, current energy acquisition rate, and remaining power percentage of the energy storage unit in real time; the resource scheduling engine dynamically adjusts the data acquisition frequency based on the monitored data. When the cable is operating under peak load or when the temperature change slope increases to a preset rate of change threshold, the resource scheduling engine switches the data acquisition frequency to a preset high-frequency mode. When the cable is operating under low load at night or when the ambient temperature is in a stable range, the resource scheduling engine reduces the data acquisition frequency to a preset low-frequency mode to reduce the operating time of the sensing module and the local computing unit.

6. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: The wireless transmission power configuration process in step 3 is as follows: The system establishes a communication quality evaluation mechanism, calculates the minimum power value required to maintain reliable communication by monitoring the received signal strength indication and historical packet loss rate; when the channel quality between the node and the base station is higher than the preset quality threshold, the transmission current of the radio frequency module is reduced; when electromagnetic interference enhancement or signal attenuation is detected, the wireless transmission power is increased; the classifier inference depth refers to the number of decision tree layers in the extreme gradient boosting classifier participating in the calculation; in the normal monitoring mode, the system only calls the first few layers of the basic decision tree of the extreme gradient boosting classifier to perform lightweight inference; when the fault probability score obtained by lightweight inference exceeds the preset preliminary warning line, the system activates all decision tree layers of the extreme gradient boosting classifier to enter the high-precision mode and performs feature fusion and judgment.

7. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: In step 4, the low-power wide-area network adopts a narrowband IoT protocol stack; during the data upload process, the temperature sensing node performs feature compression on the original temperature sequence and transmits preset statistics and risk levels derived from local inference. The cloud-based monitoring platform establishes a digital twin mapping of the entire cable line, mapping the temperature data of each discrete node to a geographic coordinate system, and generating a global risk situation map in the form of a heat map to display the distribution of thermal pressure. The cloud-based monitoring platform has a built-in aging compensation module, which constructs a material performance degradation model based on the cable's commissioning date, cumulative operating hours, and number of thermal cycles. The system periodically calculates the aging correction coefficient, which shows a non-linear increasing trend with the increase of the cable's service time. The aging correction coefficient is applied to the original temperature threshold uploaded by the local node to achieve dynamic calibration of physical property drift.

8. The method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: The multi-level early warning mechanism in step 5 includes a three-level response hierarchy; Level 1 warning corresponds to the system detecting local temperature fluctuations that have not yet formed a preset temperature rise trend. The system records the event in the local log and synchronizes a heartbeat packet to the cloud. Level 2 warning corresponds to the temperature rise rate exceeding a preset unit time change threshold, or the absolute temperature value entering a preset warning range. The system pushes an anomaly report containing geographical location, load current, and temperature rise curve to the mobile terminals of maintenance personnel. Level 3 warning corresponds to the local temperature rise rate showing exponential growth, or the absolute temperature value exceeding the insulation material's limit tolerance temperature. The system issues a trip command to the circuit breaker control circuit and activates surrounding sensing nodes to enter the highest frequency acquisition mode. When a single node triggers a Level 2 or Level 3 warning, the system sends a collaborative observation request to other nodes within a preset range around the single node. When more than a preset proportion of collaborative nodes confirm the existence of abnormal temperature rise, the warning command is confirmed to be valid.

9. A method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 1, characterized in that: The method also includes a seasonal adaptive calibration step. The system maintains a feature library containing environmental parameters for all four seasons and periodically compares the current temperature distribution curve with the historical baseline curve for the same period. If the correlation between the current temperature rise and the seasonal environmental temperature rise is higher than a preset correlation threshold, the system automatically shifts the warning baseline. The temperature sensing node integrates a self-diagnostic unit for periodically performing hardware self-tests. The hardware self-tests include detecting the resistance drift of the thermistor, the VSWR of the radio frequency module, and the internal resistance of the micro energy storage unit. When the hardware performance drops to a preset critical level, the temperature sensing node sends a maintenance request packet to the cloud monitoring platform; the dynamic early warning model supports online incremental learning. When it receives manually labeled fault cases or false alarm events, the system uses the feature data of the fault cases or false alarm events to fine-tune the decision boundary of the extreme gradient boosting classifier.

10. A method for real-time monitoring and early warning of cable temperature based on the Internet of Things according to claim 7, characterized in that: The method also includes a spatiotemporal correlation analysis step. The cloud monitoring platform extracts historical data from adjacent nodes for horizontal comparison. If multiple adjacent nodes simultaneously detect a temperature rise trend and the temperature rise start points satisfy a logical sequence on the time axis, it is identified as a chain overheating event with propagation characteristics. In step 1, the wireless communication module of the temperature sensing node adopts frequency hopping spread spectrum technology. The system evaluates the signal-to-noise ratio of each sub-channel in real time and selects a frequency band with an interference level lower than a set value for data carrier. When determining the transmission time of each temperature sensing node, the resource scheduling engine introduces a random offset based on the node's physical address hash value to stagger the warning upload time of different nodes. The aging compensation module introduces cumulative damage calculation logic based on chemical reaction rate theory. It integrates and accumulates the time period during which the cable is in high-level operation to calculate the probability density of molecular chain breakage in the insulation material. When the probability density reaches a critical threshold, a health warning report is generated.

Images