An abnormality early warning and terminal temperature correction method for an electric energy metering box

Abstract

This invention relates to an anomaly warning and terminal temperature correction method for power metering boxes, aiming to solve the problems of terminal thermal response anomaly detection and adaptive model optimization. It includes: acquiring multi-dimensional time-series data through standardized thermal excitation calibration; extracting the characteristic inflection point time, slope change rate, platform maintenance duration, and steady-state temperature rise difference of the temperature dynamic trajectory; generating a thermal inertia fingerprint vector; and using this vector to initialize the simulator's equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient to construct a heat conduction differential model; inputting the current parameter stream into the model under real-time operating conditions to obtain the theoretical temperature response curve; and generating an anomaly trigger signal or normal operation indicator by combining the dynamic time warping similarity score and the real-time fingerprint deviation as dual criteria. This invention transforms anomaly judgment from purely data-driven to physical behavior consistency verification, reducing false alarm and false negative rates and enhancing the model's robustness to environmental and aging drift.

Description

Technical Field

[0001] This invention relates to the field of abnormal monitoring and early warning technology for electricity metering boxes, and in particular to an abnormal early warning and terminal temperature correction method for electricity metering boxes. Background Technology

[0002] Currently, anomaly monitoring and early warning systems for electricity metering boxes mainly rely on multi-source parameter acquisition and data-driven models, including real-time monitoring of multi-dimensional parameters such as terminal temperature, current, voltage, internal temperature and humidity, and ambient airflow velocity. Existing solutions generally employ statistical threshold matching, machine learning, or deep neural networks for anomaly pattern recognition. While these technologies can identify terminal anomalies to some extent, they generally suffer from limited accuracy, high false alarm rates, and insufficient model generalization ability. In actual operation, sudden environmental changes, load transients, or terminal aging can easily lead to misjudgments and missed alarms in data-driven models, and deep learning solutions struggle to achieve effective generalization in the absence of sufficient fault samples.

[0003] Current typical power metering box anomaly monitoring systems primarily improve identification capabilities through statistical feature thresholding, black-box supervised learning models, or federated learning and model distillation. These methods are suitable for scenarios with abundant data and relatively stable environmental factors. However, terminal temperature is affected by various factors such as load current, ambient temperature, and humidity, resulting in complex dynamic coupling relationships between parameters that are not simply linear correlations. In highly dynamic environments with strong coupling of multidimensional terminal parameters, uncontrollable factors such as the drift of the equipment's own physical properties and sensor offset severely restrict identification accuracy. The high computing power, data security, and edge deployment complexity issues brought about by massive historical data and frequent network communication have also become bottlenecks in typical technology applications.

[0004] Current technologies cannot fully exploit the inherent physical characteristics of terminals as thermodynamic entities. Their identification logic relies excessively on data correlation and lacks interpretable physical constraint mechanisms, failing to effectively reduce the risk of false alarms caused by strong coupling of multiple parameters and environmental transients. Especially when fault samples are missing, or when equipment aging or environmental cumulative effects cause baseline drift, traditional anomaly identification models often produce unpredictable erroneous judgments, affecting system robustness and early warning capabilities. Therefore, there is an urgent need for a method that can model physical properties based on the inherent thermal inertial fingerprint of terminals. Through offline calibration and online adaptive correction mechanisms, this method should transform the anomaly judgment logic from statistical thresholds and data fitting paradigms to physical behavior consistency checks, thereby improving the accuracy of terminal anomaly pattern recognition and the reliability of early warnings. Summary of the Invention

[0005] This application provides an abnormal early warning and terminal temperature correction method for electricity metering boxes, aiming to solve the problem that existing electricity metering box terminal abnormality detection methods rely too much on data statistics and ignore physical laws, resulting in high false alarm rates under complex operating conditions and reference failure after equipment aging.

[0006] This application provides a method for abnormal early warning and terminal temperature correction in an electricity metering box, specifically including: Acquire multi-dimensional time-series data of the power metering box, which includes the dynamic trajectory of terminal temperature, current amplitude, temperature and humidity inside the box and ambient airflow speed corresponding to the stepped load sequence, to form the original calibration dataset. Based on the terminal temperature dynamic trajectory in the original calibration dataset, the characteristic inflection point time, slope change rate, platform maintenance duration and steady-state temperature rise difference of the temperature rise and fall segments are extracted to generate the terminal thermal inertia fingerprint vector. Using the prior constraints of physical properties in the terminal thermal inertial fingerprint vector, the three adjustable physical parameters of the thermal response simulator, namely the equivalent heat capacity, contact thermal resistance and convective heat transfer coefficient, are initialized to construct a differential model of heat conduction. The real-time collected multi-dimensional parameter stream of the current operating condition is input into the heat conduction differential model to calculate and output the theoretical temperature response curve reflecting the current load and environmental coupling state. A dynamic time warping comparison operation is performed on the real-time measured temperature curve and the theoretical temperature response curve to calculate the dynamic time warping similarity score, and the thermal inertia fingerprint vector corresponding to the real-time measured temperature curve is extracted simultaneously. Determine whether the dynamic time warping similarity score is lower than a preset morphological threshold and whether the deviation between the thermal inertial fingerprint vector and the terminal thermal inertial fingerprint vector exceeds a preset tolerance band. If both conditions are met, generate a terminal abnormal mode trigger signal; otherwise, generate a normal operation status identifier. In response to the terminal abnormal mode trigger signal, or in response to the preset time of periodic stable low-load maintenance, the terminal thermal inertial fingerprint vector is updated or the adjustable physical parameters are adjusted by comparing the current fingerprint with the aging model prediction range, thereby completing the adaptive closed-loop optimization of the abnormal identification model.

[0007] The present application provides a method for abnormal early warning and terminal temperature correction in an electricity metering box, which has the following beneficial effects: By extracting dynamic morphological features such as inflection point time, slope change rate, platform maintenance duration, and steady-state temperature rise difference throughout the temperature response, a thermal inertial fingerprint is constructed that reflects the inherent physical properties of terminals, such as equivalent heat capacity, contact thermal resistance, and heat dissipation capacity. This avoids the shortcomings of traditional absolute temperature threshold judgment, which is susceptible to environmental drift and sensor offset. Simultaneously, a lightweight physical simulator is built-in, using the thermal conduction differential equation as a priori constraint to generate theoretical temperature response curves. Combined with a dynamic time warping algorithm for morphological similarity comparison, anomaly detection based on physical consistency verification can be achieved even without a large number of historical fault samples, enhancing the interpretability and anti-interference capability of the diagnostic process. Furthermore, the system uses periodic low-load recalibration and fingerprint drift compensation mechanisms to compare the current fingerprint with the aging model's prediction interval and update model parameters differentially, forming a closed-loop adaptive long-term operation guarantee system. All calculations are completed on the local terminal, without relying on external data or complex network architectures, effectively balancing model accuracy with the real-time and security requirements of edge deployment. Attached Figure Description

[0008] Figure 1 This is a main flowchart of an abnormal early warning and terminal temperature correction method for an electricity metering box.

[0009] Figure 2 This is a sub-flowchart of a method for abnormal early warning and terminal temperature correction in an electricity metering box.

[0010] Figure 3 This is another sub-flowchart of a method for abnormal early warning and terminal temperature correction for electricity metering boxes. Detailed Implementation

[0011] Embodiments of the present invention are described in detail below. Examples of these 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 are only used to explain the present invention, and should not be construed as limiting the present invention.

[0012] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0013] like Figure 1 As shown, this application provides a method for abnormal early warning and terminal temperature correction of an electricity metering box, specifically including: S1: Obtain multi-dimensional time-series data of the power metering box under the standardized thermal excitation calibration process. The multi-dimensional time-series data includes the dynamic trajectory of terminal temperature, current amplitude, temperature and humidity inside the box and ambient airflow speed corresponding to the stepped load sequence, so as to form the original calibration dataset for constructing the physical constraint benchmark.

[0014] S2: Based on the terminal temperature dynamic trajectory in the original calibration dataset, extract the characteristic inflection point time, slope change rate, platform maintenance duration and steady-state temperature rise difference of the temperature rise and fall segments to generate a terminal thermal inertia fingerprint vector characterizing the inherent thermal response delay characteristics of the terminal.

[0015] S3: Using the prior constraints of physical properties in the terminal thermal inertia fingerprint vector, initialize the three adjustable physical parameters of the thermal response simulator: equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient, and construct a thermal conduction differential model that can reproduce the theoretical temperature response curve in real time.

[0016] S4: Input the real-time collected multi-dimensional parameter flow of the current operating condition into the heat conduction differential model, calculate and output the theoretical temperature response curve reflecting the current load and environmental coupling state, and use it as a benchmark reference signal for subsequent morphological consistency comparison.

[0017] S5: Perform dynamic time warping comparison operation on the real-time measured temperature curve and the theoretical temperature response curve, calculate the dynamic time warping similarity score that characterizes the morphological similarity between the two, and simultaneously extract the thermal inertia fingerprint vector corresponding to the real-time measured temperature curve.

[0018] S6: Determine whether the dynamic time warping similarity score is lower than the preset morphological threshold and whether the deviation between the thermal inertial fingerprint vector and the terminal thermal inertial fingerprint vector exceeds the preset tolerance band. If both conditions are met, generate a terminal abnormal mode trigger signal; otherwise, generate a normal operation status indicator.

[0019] S7: In response to the terminal abnormal mode trigger signal, or in response to the preset periodic stable low-load maintenance arrival time, the terminal thermal inertial fingerprint vector is updated or the adjustable physical parameters are adjusted by comparing the current fingerprint with the aging model prediction range, thereby completing the adaptive closed-loop optimization of the abnormal identification model.

[0020] Step S1: Obtain multi-dimensional time-series data of the power metering box under a standardized thermal excitation calibration process. This multi-dimensional time-series data includes the dynamic trajectory of terminal temperatures, current amplitude, internal temperature and humidity, and ambient airflow velocity corresponding to the stepped load sequence, forming the original calibration dataset used to construct a physical constraint benchmark. Specifically, this includes: S1.1: The programmable current source control unit in the power metering box is processed to issue instructions. Based on the preset standardized thermal excitation calibration protocol, a stepped load sequence control signal including zero load, half load, full load and fall-off stages is generated to drive the terminal block into the controlled thermal response excitation state and output standardized load excitation timing data.

[0021] The S1.1 sub-step aims to stimulate the thermal response characteristics of the terminals by precisely controlling the load excitation. It reads the preset standardized thermal excitation calibration protocol configuration parameters and analyzes the target current amplitude and duration thresholds for four stages: zero load, half load, full load, and fallback. The programmable current source control unit receives the analyzed instruction sequence, generates corresponding digital control signals, and drives the power amplifier circuit to output a stepped load current. In the zero load stage, the control loop current is maintained near 0A for a preset duration to establish a thermal balance reference state. Then, it switches to the half load stage, linearly ramping up the current to 50% of its rated value and maintaining this constant until the terminal temperature change rate falls below the preset steady-state criterion. Next, it enters the full load stage, increasing the current to 100% of its rated value to simulate the heating limit under extreme conditions and record the temperature rise dynamics. Finally, it executes the fallback stage, rapidly cutting off or reducing the current to the zero load level to capture the heat dissipation trajectory during natural cooling. Throughout the process, the control unit monitors the stability of the output current in real time, ensuring that the rise and fall steepness of the load step meets the microsecond-level response requirements specified in the protocol, and eliminating thermal excitation noise introduced by power supply fluctuations. Through the controlled, stepped load sequence distribution process described above, the abstract calibration protocol is transformed into concrete physical electrical excitation, driving the terminals into a reproducible thermal response excitation state. Standardized load excitation timing data is output, providing a precise time reference and operating condition identifier for subsequent multi-dimensional parameter synchronous acquisition. For example, with a rated current of 60A, the system maintains 0A for 30 seconds during the zero-load phase, 30A for 60 seconds during the half-load phase, 60A for 60 seconds during the full-load phase, and then drops to 0A for 30 seconds during the fallback phase. The control unit refreshes the PWM duty cycle at a 1ms cycle, ensuring that the current step transition time is less than 10ms. During the full-load phase, the measured current fluctuation range is controlled within ±0.5A, effectively stimulating the complete thermal inertia characteristics of the terminals. The output load excitation timing data contains precise current jump time markers, significantly improving the accuracy of subsequent thermal response curve segment alignment.

[0022] S1.2: Based on standardized load excitation timing data, the high-precision temperature sensing array, Hall current sensor and multi-parameter environmental monitoring module are synchronously triggered for acquisition and processing. Multi-channel high-speed analog-to-digital conversion technology is used to acquire the transient values ​​of terminal surface temperature, loop current amplitude, air temperature and humidity inside the box and the original sampling points of ambient airflow velocity in parallel, so as to form a multi-source heterogeneous real-time parameter stream data set.

[0023] It receives the standardized load excitation timing data generated by S1.1 and distributes it as a synchronization trigger signal to the high-precision temperature sensing array, Hall current sensor and multi-parameter environmental monitoring module.

[0024] Configure the sampling clock phase of the multi-channel high-speed analog-to-digital converter (ADC) to ensure that each sensor channel starts data acquisition simultaneously within a microsecond time window, eliminating timing misalignment errors caused by serial acquisition.

[0025] A high-precision temperature sensing array captures changes in the heat distribution on the terminal surface in real time and outputs a temperature-voltage analog signal that reflects the transient thermal response. This signal directly maps the heat accumulation process of the terminal material.

[0026] The electromagnetic field strength in the synchronous sensing circuit of the Hall current sensor is linearly converted into a voltage signal that is proportional to the instantaneous current amplitude, accurately recording the dynamic current trajectory during the load step process.

[0027] The multi-parameter environmental monitoring module collects output data from the air temperature and humidity sensors and the miniature anemometer inside the chamber in parallel to obtain the original values ​​of the environmental boundary conditions that affect the convective heat transfer efficiency.

[0028] Each channel ADC discretizes the analog signal at a preset high sampling rate, converting continuous physical quantities into digital quantized values, forming a multi-source heterogeneous original sampling point sequence containing timestamps.

[0029] By using a hardware interrupt mechanism, the raw sampling points of all channels are packaged and stored in a high-speed cache queue to ensure that data is not lost and that a strict timing correspondence is maintained during rapid load switching.

[0030] By using the parallel synchronous acquisition and processing method described above, the load excitation signal from the previous step is transformed into a multi-source heterogeneous real-time parameter stream data set containing temperature, current and environmental parameters, thereby achieving spatiotemporal consistency acquisition of multi-dimensional physical field data and providing a phase-bias-free data foundation for the subsequent construction of high-fidelity thermal inertial fingerprints.

[0031] For example, the ADC sampling rate is set to 10kHz, the temperature sensor uses a PT100 platinum resistance thermometer paired with a 24-bit ADC, and the current sensor has a range of 0-100A and an accuracy of 0.5%. At the instant a 30A step load is applied, the system simultaneously captures the temperature rise slope of 0.5℃ / s, the current stability of 30.02A, the ambient temperature of 25.3℃, and the wind speed of 0.2m / s. All data points are aligned within a 100μs time window, eliminating the lag of more than 5ms caused by traditional polling acquisition and significantly improving the temporal resolution of thermal response feature extraction.

[0032] S1.3: Perform timestamp alignment and noise filtering on the multi-source heterogeneous real-time parameter stream data set, use the sliding window mean filtering algorithm to remove glitch data caused by high-frequency electromagnetic interference, and resample the original sampling points of terminal temperature transient values, loop current amplitude and ambient airflow velocity at different sampling frequencies to a unified time base to generate time-synchronized multi-dimensional time-series data frames.

[0033] Receive a multi-source heterogeneous real-time parameter stream data set from S1.2, which includes terminal temperature transient values, loop current amplitudes, and raw sampling points of ambient airflow velocity at different sampling frequencies.

[0034] Timestamp alignment is performed on multi-source heterogeneous data. Based on the system master clock, a linear interpolation algorithm is used to map the high-frequency sampling points of the Hall current sensor and the low-frequency sampling points of the environmental monitoring module to a unified time grid, eliminating timing misalignment caused by hardware response delay.

[0035] For the aligned data sequence, a sliding window mean filtering algorithm is applied to suppress noise. The window length is set and the data points within the window are arithmetically averaged to effectively remove glitch data caused by high-frequency electromagnetic interference.

[0036] The filtered terminal temperature transient values, loop current amplitude, and ambient airflow velocity data are resampled according to a unified time base to ensure that each physical quantity has a corresponding numerical record at each time step.

[0037] The resampled multidimensional data is encapsulated into structured data frames in chronological order. Each data frame contains temperature, current, humidity and wind speed fields at the same time, forming a time-synchronized multidimensional time-series data frame.

[0038] Multidimensional time-series data refers to a structured data frame sequence that contains multiple physical quantity fields such as terminal temperature transient value, loop current amplitude, box temperature and humidity and ambient airflow velocity under the same time base after being processed by timestamp alignment, noise filtering and unified time base resampling.

[0039] By employing the aforementioned timestamp alignment, sliding window filtering, and unified time base resampling processing methods, the multi-source heterogeneous original sampling points from the previous step are transformed into multi-dimensional time-series data frames with high signal-to-noise ratio and strict time synchronization, thereby achieving the technical effect of providing a high-quality data foundation for subsequent thermal response segmentation and feature extraction.

[0040] For example, the sliding window length N is set to 5, and the sampling frequency is uniformly resampled to 10Hz. The collected temperature data containing ±2℃ high-frequency noise is filtered. Calculations show that the standard deviation of the temperature curve after filtering is reduced from 1.2℃ to 0.3℃, significantly improving data smoothness. Simultaneously, the 1kHz sampling data from the current sensor and the 1Hz sampling data from the environmental sensor are aligned to a 100ms time grid through linear interpolation, eliminating a maximum timing deviation of 50ms and ensuring millisecond-level accuracy in the inflection point time during subsequent thermal inertial fingerprint extraction.

[0041] S1.4: Based on time-synchronized multi-dimensional time-series data frames, perform data truncation and segmentation marking processing for the complete thermal response process. Based on the transition edges of the stepped load sequence control signal, identify the start and end boundaries of the temperature rise segment, the plateau maintenance segment, and the temperature fall segment. Cut the continuous time-synchronized multi-dimensional time-series data frames into independent thermal response data segments corresponding to each load condition, so as to construct structured thermal response data units with operating conditions related.

[0042] For time-synchronized multi-dimensional time-series data frames, the current transition edges in the stepped load sequence control signal are extracted as reference time anchors for dividing the thermal response stages. An edge detection algorithm is used to identify the rising edge of the current amplitude transitioning from low to high level, marking it as the heating start time; and to identify the falling edge of the current amplitude returning from high to low level, marking it as the cooling start time. Based on the reference time anchors, the corresponding temperature response inflection points are located in the temperature dynamic trajectory, dividing the continuous temperature data stream into three independent intervals: a temperature rise segment, a plateau maintenance segment, and a temperature fall segment. For the temperature rise segment data, the rate of change of temperature difference between adjacent sampling points is calculated, and intervals with a rate of change exceeding a preset threshold are selected as effective thermal excitation response areas, eliminating initial noise interference segments. For the plateau maintenance segment data, the duration for which the temperature fluctuation amplitude is less than the set tolerance band is statistically analyzed to determine whether a thermal equilibrium state has been achieved; if not, the truncation window is extended until the steady-state condition is met. For the temperature fall segment data, a temperature decay curve of a fixed duration is truncated based on the cooling start time to ensure that the complete natural cooling process characteristics are included. Each segmented independent thermal response data fragment is associated and bound with its corresponding load condition identifier to construct a structured thermal response data unit containing timestamps, current levels, temperature ranges, and environmental parameters. Through the above segmentation, marking, and truncation processes, continuous unstructured time-series data is transformed into discrete, physically meaningful load condition-associated data units, achieving precise isolation of thermal response features and providing a high signal-to-noise ratio data foundation for subsequent fingerprint feature extraction.

[0043] For example, the current transition edge detection threshold is set to 5% of the rated current, and the sampling frequency is 10Hz. In a step excitation range of 0A to 60A, a current rising edge is detected at t=10s, marked as the heating start point; a current falling edge is detected at t=120s, marked as the cooling start point. The temperature rise segment from t=10s to t=40s is selected, and the temperature slope within this interval is calculated, discarding the sensor response delay data from the first 2 seconds. The plateau maintenance segment from t=40s to t=120s is selected, and the temperature fluctuation range is monitored to be within ±0.5℃, confirming thermal equilibrium. The temperature fall segment from t=120s to t=180s is selected, and the natural cooling curve is recorded. Finally, a structured data unit containing three intervals is generated, each unit with operating condition labels "full load heating," "full load steady state," and "zero load cooling," significantly improving data integrity and effectively supporting the accurate calculation of subsequent thermal inertial fingerprints.

[0044] S1.5: Perform metadata encapsulation and persistent storage processing on the structured thermal response data units associated with the operating conditions. Bind and package each independent thermal response data fragment with its corresponding load condition identifier, ambient temperature reference and equipment unique code, and write it into a non-volatile storage medium to form the original calibration dataset used to build physical constraint reference, so as to support the feature extraction calculation of the terminal thermal inertial fingerprint vector.

[0045] The original calibration dataset refers to the data set formed by writing the structured thermal response data units associated with the operating conditions into a non-volatile storage medium after encapsulating the metadata. This data set contains the dynamic trajectory of terminal temperature under each load condition and the corresponding environmental parameter records, which are used to build the benchmark for subsequent terminal thermal inertia fingerprint extraction and physical parameter initialization.

[0046] like Figure 2 As shown, step S2: Based on the terminal temperature dynamic trajectory in the original calibration dataset, extract the characteristic inflection point time, slope change rate, plateau maintenance duration, and steady-state temperature rise difference of the temperature rise and fall segments to generate a terminal thermal inertia fingerprint vector characterizing the inherent thermal response delay characteristics of the terminal. Specifically, this includes: S2.1: Perform time series segmentation processing on the terminal temperature dynamic trajectory in the original calibration dataset. Based on the inflection point of the current amplitude change in the stepped load sequence, divide the continuous temperature curve into a temperature rise segment and a temperature fall segment to obtain a standardized temperature segment sequence containing the complete thermal response process.

[0047] Read the multi-dimensional time-series data frames that have undergone time synchronization processing from the original calibration dataset, and extract the terminal temperature instantaneous value sequence and the corresponding loop current amplitude sequence that are strictly aligned with the stepped load sequence, as the input source for segmented processing.

[0048] Perform a first-order difference operation on the loop current amplitude sequence to calculate the rate of change of current between adjacent sampling points and generate a current gradient vector that reflects the moment of load jump.

[0049] Set a current gradient threshold, traverse the current gradient vector, identify extreme points where the absolute value exceeds the threshold, and mark them as critical time points for load step change, including the loading start point, the full-load stabilization point, and the unloading start point.

[0050] Based on the identified critical time points of load step change, the continuous instantaneous terminal temperature sequence is logically segmented on the time axis to divide the temperature rise range corresponding to zero load to half load, half load to full load, and full load to zero load.

[0051] Boundary smoothing is performed on each temperature range after division to remove edge distortion data caused by sensor response lag, and retain the intermediate segment data with complete thermodynamic evolution characteristics to form a standardized temperature segment sequence.

[0052] Through the above segmented processing, the continuous and mixed temperature data is transformed into independent thermal response segments with clear structures, achieving precise isolation of the terminal thermal inertial physical processes and providing a high signal-to-noise ratio data foundation for subsequent feature inflection point extraction.

[0053] For example, the current gradient threshold is set to 5A / s. During calibration, the current jumps from 0A to 60A. A current gradient of 20A / s is detected at t=10s, marked as the loading start point; the current gradient approaches 0 at t=120s, marked as the full-load stable point. Based on this, temperature data between t=10s and t=120s is extracted as the temperature rise segment. Edge data within 0.5s before and after are removed, resulting in a standard rise sequence of 109s in length, effectively eliminating the influence of electromagnetic interference during load switching on temperature feature extraction.

[0054] S2.2: Based on the standardized temperature segmented sequence, perform first-order and second-order derivative operations to calculate the slope change rate in each segment and locate the curvature extremum point, thereby extracting the characteristic inflection point time that characterizes the thermal response transition moment and the slope change rate that characterizes the change in heat transfer rate.

[0055] The system receives a standardized temperature segment sequence output from S2.1. This sequence contains transient terminal temperature values ​​after time alignment and noise filtering, along with their corresponding timestamps. Discrete difference operations are performed on the temperature segment sequence to calculate the ratio of temperature change to time interval between adjacent sampling points, generating a first-order derivative sequence characterizing the thermal response rate. The central difference method is used to optimize the derivative calculation accuracy at boundary points, eliminating phase lag errors introduced by numerical differentiation and ensuring the real-time performance and accuracy of slope changes. A second-order discrete difference operation is performed based on the first-order derivative sequence to obtain the acceleration index of the temperature change rate, generating a second-order derivative sequence characterizing the intensity of the thermal inertia transition. A local extremum search algorithm is performed on the second-order derivative sequence, setting a curvature threshold to filter out false extrema caused by high-frequency sensor noise, locating critical moments when the concavity / convexity of the temperature curve undergoes a fundamental change. Second-order derivative extrema points that meet the confidence requirements are mapped back to the original time axis, and the corresponding absolute time coordinates are extracted as feature inflection point times, marking physical nodes where the thermal response transitions from accelerated heating to decelerated heating or from accelerated cooling to decelerated cooling. A subset of the first derivative sequence is extracted within the time neighborhood of the characteristic inflection point. The arithmetic mean and peak deviation of this subset are calculated to generate a slope change rate index characterizing the stability of the heat transfer rate. Through the above differential operations and extreme value localization processing, continuous temperature time series data are transformed into discrete characteristic inflection point times and slope change rate parameters, realizing the quantitative extraction of the dynamic trajectory morphological characteristics of the terminal thermal response, and providing core geometric constraints for constructing an anti-interference thermal inertial fingerprint vector.

[0056] For example, a central difference operation is performed on a temperature rise sequence with a sampling frequency of 10Hz and a step size of 0.1s. The calculated first derivative sequence shows a maximum slope of 0.8℃ / s, occurring at 45 seconds. Further difference operations are performed on the first derivative sequence to obtain the second derivative sequence. A significant positive extremum of 0.15℃ / s² is detected at 30 seconds, identified as an acceleration point in the thermal response; a significant negative extremum of -0.12℃ / s² is detected at 60 seconds, identified as a slowing point. These two time points are extracted as feature inflection points. Within the interval from 30 to 45 seconds, the average rate of change of the first derivative is calculated to be 0.05℃ / s², serving as the slope change rate index. This processing method effectively eliminates noise interference with amplitudes less than 0.02℃ / s², significantly improving the signal-to-noise ratio and robustness of feature extraction.

[0057] S2.3: The standardized temperature segment sequence is truncated and steady-state is determined by using the characteristic inflection point time. The duration before the temperature reaches the equilibrium state is counted as the plateau maintenance time, and the difference between the initial temperature and the final stable temperature is calculated as the steady-state temperature rise difference, so as to form a multidimensional thermal response characteristic parameter set.

[0058] The system receives the characteristic inflection point time series and standardized temperature segmented series output from S2.2 as input data sources for interval truncation and steady-state determination. Based on the rising and falling edge times in the characteristic inflection point time series, boundary indexing is performed on the standardized temperature segmented series to extract independent temperature subsequences containing the complete thermal response process. Sliding window variance calculation is performed on the extracted independent temperature subsequences, and a small threshold is set to identify continuous time periods with temperature fluctuation rates below this threshold, which are determined as thermal equilibrium steady-state intervals. The time span from the moment the load excitation is applied to the start time of entering the thermal equilibrium steady-state interval is statistically analyzed and quantified as the platform maintenance duration, characterizing the inertial delay characteristics of the terminal reaching thermal steady state. The average temperature within the thermal equilibrium steady-state interval is extracted as the final stable temperature, and the instantaneous temperature value at the moment before the load excitation is applied is read as the starting temperature. The algebraic difference between the final stable temperature and the starting temperature is calculated to obtain the steady-state temperature rise difference, eliminating the influence of ambient temperature reference drift on the absolute temperature reading. The platform sustaining time and steady-state temperature rise difference are encapsulated into structured data units and incorporated into a multi-dimensional thermal response feature parameter set. Through the aforementioned interval truncation, steady-state determination, and difference calculation processing methods, the feature inflection point of the previous step is transformed into the platform sustaining time and steady-state temperature rise difference index of quantified thermal inertia. This achieves accurate digital characterization of the terminal thermal response physical characteristics and provides a highly robust feature basis for subsequent fingerprint vector generation.

[0059] For example, for the copper terminals of a certain type of power metering box, under a 60A full-load calibration condition, the rising edge inflection point time of the S2.2 output is t=15s, and the falling edge inflection point time is t=120s. The system extracts the temperature subsequence from t=0s to t=150s, and calculates the variance using a sliding window with a length of 10 sampling points. When the variance is below 0.01℃² for 50 consecutive points, the steady-state interval is determined to be from t=85s to t=115s. The time span from t=0s to t=85s is statistically analyzed, and the platform maintenance time is found to be 85s. The average temperature of the steady-state interval is read as 75.4℃ as the final stable temperature, and the temperature at t=0s is read as 25.0℃ as the initial temperature. The steady-state temperature rise difference is calculated as... =50.4℃. This process effectively eliminated the interference caused by the fluctuation of ambient temperature of 25℃±2℃, and significantly improved the reproducibility of characteristic parameters.

[0060] S2.4: Perform normalization and time-series alignment on the multidimensional thermal response feature parameter set, and encode the feature inflection point time, slope change rate, platform maintenance duration and steady-state temperature rise difference according to the preset weight order to generate a unique terminal thermal inertial fingerprint vector that is resistant to environmental drift.

[0061] Receive the multidimensional thermal response characteristic parameter set generated by S2.3, which includes the original data of characteristic inflection point time, slope change rate, plateau maintenance duration and steady-state temperature rise difference.

[0062] The original data is checked for dimensional consistency to identify the differences in physical units and numerical magnitudes of each characteristic parameter, and a normalized mapping model based on the maximum and minimum value method is established.

[0063] The range standardization algorithm is used to perform dimensionless processing on the features of each dimension. The original feature values ​​are mapped to the interval [0, 1] by taking the minimum and maximum values ​​of each feature in the statistical window of the calibration dataset as the boundary, thus eliminating the numerical scale imbalance caused by the time unit being seconds and the temperature unit being degrees Celsius.

[0064] The normalized feature inflection point time, slope change rate, platform maintenance duration, and steady-state temperature rise difference are weighted and combined according to preset weight coefficients to construct a high-dimensional feature space vector.

[0065] The preset weighting coefficients are allocated based on the sensitivity of each feature to changes in terminal contact state, with the steady-state temperature rise difference having the highest weight and the feature inflection point time having the second highest weight.

[0066] Perform timing alignment operation, using the current excitation transition moment as the time zero point, to correct the feature phase offset caused by sensor response delay, and ensure the consistency of the fingerprint vector's time base.

[0067] The weighted feature vector is serialized into a binary data stream, and a unique device identifier and calibration version number are added to generate a unique terminal thermal inertial fingerprint vector that is resistant to environmental drift.

[0068] The terminal thermal inertial fingerprint vector is a multi-dimensional numerical vector that can characterize the inherent thermal response delay characteristics of the terminal after normalizing and weighting the characteristic inflection point time, slope change rate, plateau maintenance time and steady-state temperature rise difference. This vector serves as a benchmark for subsequent anomaly identification and deviation comparison.

[0069] By using normalization and time-series alignment, the multidimensional heterogeneous features from the previous step are transformed into standardized terminal thermal inertial fingerprint vectors, thereby achieving physical uniqueness and environmental robustness of the anomaly identification benchmark.

[0070] For example, for a certain type of electricity metering box terminal, the feature inflection point time is extracted to be 15.2s, the slope change rate is 0.8℃ / s, the platform maintenance time is 45s, and the steady-state temperature rise difference is 12.5℃. The minimum and maximum values ​​of the corresponding features in the calibration dataset are set to [10s, 0.5℃ / s, 30s, 8℃] and [20s, 1.2℃ / s, 60s, 15℃], respectively. Using the range standardization formula, the normalized feature values ​​are calculated to be 0.52, 0.43, 0.50, and 0.64. These are then weighted and combined according to weights [0.2, 0.2, 0.2, 0.4] to generate a fingerprint vector [0.104, 0.086, 0.100, 0.256]. This vector eliminates the influence of dimensions, accurately characterizes the thermal inertia characteristics of the terminal, and significantly improves the accuracy of subsequent anomaly comparisons.

[0071] Step S3: Using the prior constraints of physical properties in the terminal thermal inertia fingerprint vector, initialize the three adjustable physical parameters of the thermal response simulator: equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient, to construct a heat conduction differential model capable of reproducing the theoretical temperature response curve in real time. Specifically, this includes: S3.1: Based on the steady-state temperature rise difference characteristics and platform maintenance duration characteristics extracted from the terminal thermal inertia fingerprint vector, perform heat capacity parameter inversion calculation to derive and generate the equivalent heat capacity initial value characterizing the heat storage capacity of the terminal component.

[0072] The steady-state temperature rise difference and platform maintenance duration feature data from the terminal thermal inertia fingerprint vector are received as input boundary conditions for the equivalent heat capacity inversion calculation. The constant load current amplitude and nominal terminal contact resistance under calibration conditions are extracted, and the constant heating power per unit time is calculated using Joule's law to establish the heat source intensity benchmark.

[0073] A thermal response simulator is a numerical calculation unit that uses the physical properties in the terminal thermal inertia fingerprint vector as constraints, and constructs and runs a thermal conduction differential model by initializing three adjustable physical parameters—equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient—to reproduce the theoretical temperature response curve of the terminal in real time.

[0074] Based on the principle of the lumped-parameter thermal circuit model, a first-order linear ordinary differential equation is constructed to describe the dynamic change of terminal temperature over time, and heat capacity is defined as the core physical parameter to be solved. During the temperature rise phase, when the system reaches thermal equilibrium, the rate of temperature change is zero, and the heating power equals the heat dissipation power. A static correlation equation between thermal resistance and heat capacity is established using the steady-state temperature rise difference. For the non-steady-state process, time points where the temperature reaches a specific proportion of the steady-state value within the platform's maintenance duration are selected, and these are substituted into the exponential response function for inverse solution. The initial value of the equivalent heat capacity is calculated using the following formula:

[0075] in, For equivalent heat capacity, For constant heating power, To maintain the platform's duration, This represents the steady-state temperature rise difference. The calculated equivalent heat capacity value undergoes a physical rationality check, eliminating outliers that exceed the theoretical range of the specific heat capacity of metallic materials. Through the above-mentioned heat capacity parameter inversion calculation, the thermal response time-series characteristics of the previous step are transformed into initial values ​​of equivalent heat capacity characterizing the heat storage capacity of the terminal assembly. This achieves a quantitative mapping from macroscopic temperature trajectory to microscopic physical properties, providing a definite mass-energy conservation constraint basis for subsequent decoupled estimation of contact thermal resistance and convective heat transfer coefficient.

[0076] For example, under full-load calibration conditions, the load current is 60A, the terminal contact resistance is 50μΩ, and the calculated heat generation power P is 0.18W. The measured steady-state temperature rise difference ΔT is 15K, and the plateau duration t is 120s. Substituting the parameters into the formula, the exponent e raised to the power of -1 is approximately 0.368, the denominator is 15 multiplied by (1-0.368), which is 9.48, and the numerator is 0.18 multiplied by 120, which is 21.6. The final calculated equivalent heat capacity C is approximately 2.28J / K. This value is within the physical characteristic range of copper-aluminum composite terminals, significantly improving the accuracy of the simulation model in representing thermal inertia and effectively avoiding the phase lag problem of the theoretical curve caused by deviations in heat capacity parameters.

[0077] S3.2: Utilizing the temperature rise slope change rate and inflection point time characteristics contained in the terminal thermal inertia fingerprint vector, combined with the generated initial value of equivalent heat capacity, perform contact impedance mapping estimation processing to calculate and output the initial value of contact thermal resistance characterizing the thermal conductivity of the terminal connection interface.

[0078] The characteristic values ​​of the slope change rate of the temperature rise segment and the characteristic inflection point time data in the terminal thermal inertia fingerprint vector are obtained as boundary input conditions for the contact thermal resistance calculation.

[0079] Based on the lumped parameter thermal circuit model theory, a first-order linear differential equation containing equivalent heat capacity, contact thermal resistance and heat source terms is constructed to describe the transient temperature rise dynamics of the terminals under stepped load excitation.

[0080] Using the initial value of the generated equivalent heat capacity and the real-time heat power density calculated by Joule's law, the differential equation is solved analytically, and the expression for the time constant of the theoretical temperature rise curve is derived. This time constant is equal to the product of the contact thermal resistance and the equivalent heat capacity.

[0081] The time interval from the initial moment to the point where the steady-state temperature rise of 63.2% is reached in the measured temperature rise segment is extracted and calibrated as the measured thermal time constant.

[0082] Substitute the measured thermal time constant into the above time constant expression, perform algebraic inversion, isolate and calculate the numerical solution of the contact thermal resistance.

[0083] The physical rationality of the calculated initial value of contact thermal resistance is verified, and abnormal solutions that exceed the reasonable range of thermal conductivity of metallic materials are eliminated to ensure that the parameters meet the physical constraints.

[0084] Through the above contact impedance mapping estimation process, the thermal capacity parameters and measured dynamic response characteristics of the previous step are transformed into initial values ​​of contact thermal resistance that characterize the thermal conductivity of the terminal connection interface, thereby achieving accurate initialization of key parameters of the thermal circuit model.

[0085] For example, for a copper terminal block with a rated current of 60A, the equivalent heat capacity C has been calculated to be 15.2 J / K in step S3.1. During the standardization calibration process, a 30A step current is applied, and the time required for the measured temperature to rise from the ambient temperature of 25°C to 63.2% of the steady-state temperature rise (i.e., the temperature rise reaches 12.4°C) is 45 seconds, meaning the measured thermal time constant τ is 45 s. The calculated contact thermal resistance R = 45 / 15.2 ≈ 2.96 K / W is obtained. This value is within the typical thermal resistance range of copper-copper contact interfaces (1-5 K / W) and is determined to be a valid initial value. If the measured time is 120 s, the calculated R is 7.89 K / W, indicating that the contact surface may have oxidation or loosening. The system records this high resistance value as the initial state for subsequent anomaly comparison benchmarks, significantly improving the model's sensitivity to poor contact conditions.

[0086] S3.3: Based on the cooling rate characteristics of the temperature drop segment recorded in the terminal thermal inertial fingerprint vector and the ambient airflow velocity data, and in conjunction with the generated initial values ​​of equivalent heat capacity and contact thermal resistance, perform convective heat transfer coefficient fitting optimization processing to determine and generate the initial value of convective heat transfer coefficient characterizing the heat dissipation efficiency of the terminal surface.

[0087] The cooling rate characteristics of the temperature drop segment in the terminal thermal inertia fingerprint vector and the ambient airflow velocity data during the calibration period are obtained. Combined with the initial values ​​of the equivalent heat capacity and the initial values ​​of the contact thermal resistance generated in the previous steps, they are used as the input boundary conditions for the fitting and optimization of the convective heat transfer coefficient.

[0088] Based on Newton's law of cooling, a thermal balance differential equation for the natural cooling stage of the terminal is constructed. The heat dissipation process on the terminal surface is modeled as a first-order linear system dominated only by convection heat transfer. The secondary influence of radiation heat transfer in the low temperature difference range is ignored, thus establishing the physical model basis.

[0089] The least squares method was used to fit the measured temperature sequence of the temperature drop segment to an exponential decay curve, and the time constant τ, which characterizes the heat dissipation rate, was extracted. This time constant directly reflects the overall efficiency of the terminal assembly in dissipating heat to the surrounding environment.

[0090] Using the extracted time constant τ, combined with the known equivalent heat capacity C eq and contact thermal resistance R c The initial value of the convective heat transfer coefficient h is calculated by inverting the algebraic relationship between the heat dissipation time constant and thermal resistance and heat capacity.

[0091] By introducing the ambient airflow velocity v as a correction factor, a nonlinear mapping relationship h=k·v between the convective heat transfer coefficient and wind speed is established. n By using multiple sets of wind speeds and corresponding h values ​​from the calibration data for regression analysis, the proportionality coefficient k and the exponent n are determined to enhance the model's adaptability to different ventilation conditions.

[0092] The calculated initial value of the convective heat transfer coefficient is stored in the parameter register of the thermal response simulator to complete the initial configuration of the heat dissipation boundary conditions.

[0093] By using the above-mentioned inversion and fitting process based on physical laws, the thermal response characteristics of the previous step are transformed into the initial value of the convective heat transfer coefficient characterizing the heat dissipation efficiency of the terminal surface. This achieves accurate calibration of key heat dissipation parameters in the terminal thermodynamic model and significantly improves the prediction accuracy of the theoretical temperature response curve under dynamic load.

[0094] For example, during the calibration process, data was collected showing the terminal temperature dropping from 60℃ to 40℃, while the ambient airflow velocity remained stable at 0.5 m / s. An exponential fitting yielded a time constant τ of 120 s, and the equivalent heat capacity C was known. eq The thermal resistance is 50 J / K, and the contact thermal resistance is R. c The initial heat transfer coefficient (H) is 0.2 K / W, and the terminal heat dissipation area (A) is 0.01 m². The calculated basic convective heat transfer coefficient (h) is approximately 0.83 W / (m²·K). Combining calibration data under multiple wind speeds, regression analysis yields k=1.5 and n=0.6. This initial value is written into the simulator, reducing the root mean square error between the theoretical and measured temperature drop curves at a wind speed of 0.5 m / s to within 0.5℃, effectively eliminating the risk of false alarms caused by deviations in the heat dissipation model.

[0095] S3.4: Integrate the generated initial values ​​of equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient as physical constraint boundary conditions, substitute them into the preset first-order lumped parameter thermal network topology, and perform instantiation of the differential equation system to form a heat conduction differential model with real-time dynamic deduction capability.

[0096] The initial values ​​of equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient generated in steps S3.1 to S3.3 are obtained as the physical parameter basis for constructing the differential heat conduction model.

[0097] Based on the lumped parameter method, the terminals of the power metering box are abstracted as a thermodynamic system with a single temperature node. The microscopic temperature gradient inside the terminal is ignored, and a first-order linear non-homogeneous differential equation topology is established.

[0098] Equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient are collectively referred to as adjustable physical parameters. These three parameters together determine the thermal response behavior of the terminal under given load and environmental conditions. They are variables in the heat conduction differential model that need to be initialized based on calibration results and corrected based on aging conditions.

[0099] Based on the law of conservation of energy, the rate of change of heat at the terminal node is defined as equal to the difference between the input heat power and the dissipated heat power, and a heat balance equation including the time derivative term is constructed.

[0100] The differential equation is expressed using the MathML scheme:

[0101] in For equivalent heat capacity, For terminal temperature, For time, For current, For contact resistance, For ambient temperature, For contact thermal resistance, The convective heat transfer coefficient is... This represents the heat dissipation area.

[0102] Assign the initial value of the equivalent heat capacity obtained in S3.1 to C in the equation. eq The parameters are used to establish the inertial response benchmark of the terminals to thermal excitation.

[0103] Assign the initial value of the contact thermal resistance obtained in S3.2 to R in the equation. c Parameters define the thermal resistance characteristics of the terminal connection interface.

[0104] The initial value of the convective heat transfer coefficient obtained from S3.3 is combined with the geometric surface area of ​​the terminal to calculate the comprehensive heat dissipation coefficient, which is then assigned to the hA term in the equation to quantify the heat exchange efficiency between the terminal surface and the environment.

[0105] The constructed system of differential equations is discretized, and the forward Euler method or Runge-Kutta method is used to determine the numerical integration step size to ensure the real-time solution stability of the simulator on the embedded terminal.

[0106] Instantiate and generate a heat conduction differential model with dynamic extrapolation capabilities. This model can output a theoretical temperature trajectory based on real-time input current and environmental parameters.

[0107] By substituting the above parameters and instantiating the equations, the static physical fingerprint features are transformed into a dynamic thermal response simulation kernel, achieving high-fidelity reproduction of the terminal thermal behavior and providing a physical consistency comparison benchmark for subsequent abnormal pattern recognition.

[0108] For example, the equivalent heat capacity C is set. eq It is 15.2 J / K, and the contact thermal resistance R is 15.2 J / K. c The heat transfer coefficient is 0.8 K / W, the convective heat transfer coefficient h is 12.5 W / (m²·K), and the heat dissipation area A is 0.005 m². When the input current I is 50 A, the contact resistance R is 0.0002 Ω, and the ambient temperature T... env At 25℃, the model calculates a heat dissipation power of 500W based on the equations, with the heat dissipation temperature difference driving term dynamically changing with temperature T. At t=0, if the terminal temperature is 25℃, the model calculates an initial value of dT / dt of 28.4K / s. After 10 seconds of iterative solving, the theoretical temperature rises to 68.5℃. This trajectory closely matches the measured thermal inertial fingerprint from the calibration phase, verifying the accuracy and extrapolation capability of the model parameters and significantly improving the physical interpretability of anomaly detection.

[0109] S3.5: Input the real-time collected current amplitude and ambient temperature and humidity parameters into the constructed heat conduction differential model, perform numerical integration and iterative solution processing, and output the theoretical temperature response curve reflecting the current load and environment coupling state in real time.

[0110] Obtain the instantiated heat conduction differential model, which contains the initial values ​​of three physical parameters determined by steps S3.1 to S3.4: equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient.

[0111] It receives real-time current amplitude data under current operating conditions, calculates the current square by multiplying it by the terminal equivalent resistance according to Joule's law, and derives the instantaneous heat power density sequence, which is used as the input of the heat source term in the differential equation system.

[0112] The system synchronously reads data from sensors on temperature, humidity, and ambient airflow velocity inside the chamber. It then dynamically corrects the boundary heat dissipation conditions in the model using empirical formulas for convective heat transfer, and constructs an instance of a set of non-homogeneous linear differential equations that vary over time.

[0113] The fourth-order Runge-Kutta numerical integration algorithm is used to iteratively solve the differential equation system. A fixed time step is set, and the calculated temperature value at the previous moment is used as the initial state to gradually deduce the theoretical node temperature value at the current moment.

[0114] During each iteration, the physical rationality of the temperature change rate is verified in real time. If a non-physical temperature change is detected, a damping factor is introduced to smooth the transition and ensure the stability and convergence of the numerical solution.

[0115] The theoretical temperature point sequence calculated at discrete time steps is linearly interpolated and reconstructed into a continuous time domain signal, generating a theoretical temperature response curve that reflects the current load and environmental coupling state.

[0116] By using numerical integration and iterative solution, the physical parameters and real-time operating data from the previous step are transformed into theoretical temperature response curve data with physical consistency, realizing real-time high-fidelity reproduction of the terminal thermal behavior and providing an accurate benchmark for subsequent abnormal pattern recognition.

[0117] Step S4: Input the real-time acquired multi-dimensional parameter flow of the current operating condition into the heat conduction differential model, calculate and output the theoretical temperature response curve reflecting the current load and environmental coupling state, as the benchmark reference signal for subsequent morphological consistency comparison. Specifically, this includes: S4.1: The multi-dimensional operating condition parameter streams, such as the real-time acquired current amplitude, temperature and humidity inside the chamber, and ambient airflow velocity, are time-stamped and noise-filtered to generate standardized real-time operating condition driving vectors, ensuring that the timing consistency and signal-to-noise ratio of the input data meet the simulation calculation requirements.

[0118] It receives raw sampling data streams from a high-precision Hall current sensor, a temperature and humidity monitoring module, and a wind speed sensor. The data streams include timestamp tags and corresponding physical quantity values.

[0119] The multi-source heterogeneous data stream is subjected to timestamp alignment processing based on hardware clock synchronization signal. The asynchronously sampled current amplitude, chamber temperature, relative humidity and ambient airflow speed are resampled to a unified standard time base sequence using a linear interpolation algorithm to eliminate timing misalignment caused by sensor response delay.

[0120] A sliding window mid-range filtering algorithm is used to process the resampled current amplitude sequence. The window length is set to 5 sampling points to remove instantaneous spike noise caused by electromagnetic interference and retain the smooth current curve that reflects the true trend of load change.

[0121] Low-pass Butterworth filtering was applied to the temperature and relative humidity data inside the chamber, with a cutoff frequency set to 0.1Hz, to filter out high-frequency measurement noise and retain the slow drift characteristics of environmental parameters, thus ensuring the stability of environmental boundary conditions.

[0122] The filtered current amplitude, chamber temperature, relative humidity, and ambient airflow velocity are concatenated into vectors according to time index to construct a multi-dimensional real-time operating condition driving vector that includes electrical excitation and environmental constraints.

[0123] The current operating condition multidimensional parameter stream refers to the standardized real-time operating condition data vector generated after performing timestamp alignment and noise filtering on the raw sampling data such as real-time collected current amplitude, temperature and humidity inside the chamber, and ambient airflow velocity, which serves as the input driving the heat conduction differential model.

[0124] Through the aforementioned time alignment and multi-level filtering methods, the multi-source raw sampling data from the previous step is transformed into a standardized real-time operating condition driving vector with consistent timing and a signal-to-noise ratio that meets the simulation accuracy requirements, thereby achieving the expected technical effects of input data quality purification and spatiotemporal alignment.

[0125] For example, the system sampling frequency is set to 100Hz, and the time alignment window is 1 second. For the collected current data, an abnormal spike with an amplitude of 200A (normal range 60A) is detected at the 50th sampling point. After median filtering, this is corrected to 60.2A. For the ambient temperature data, a 0.1Hz low-pass filter is applied to remove high-frequency fluctuations with an amplitude of 0.5℃, resulting in a smooth temperature curve. The final generated real-time operating condition driving vector includes: current 60.2A, temperature 25.1℃, humidity 45%, and wind speed 0.2m / s. This vector is directly used as the input for the thermal power calculation in step S4.2, significantly improving the convergence speed and prediction accuracy of the subsequent thermal simulation model.

[0126] S4.2: Based on the current amplitude component in the real-time operating condition driving vector and combined with the Joule law calculation logic, the real-time heat power density sequence characterizing the intensity of the terminal heat source is derived, and the electrical excitation quantity is converted into the thermodynamic boundary condition input quantity.

[0127] The current amplitude component at the current moment is extracted from the standardized real-time operating condition drive vector and used as the electrical excitation input source for calculating the Joule heating effect of the terminals. The equivalent resistance parameter of the power metering box terminals is obtained; this parameter is determined based on the contact thermal resistance calculated in step S3 and the inherent resistivity of the terminal material, characterizing the impedance characteristics when current flows through the terminals. Based on Joule's law, a mapping relationship between electrical power and thermal power is established. The instantaneous heating power on the terminal surface is calculated by multiplying the square of the instantaneous current by the equivalent resistance.

[0128] Based on Joule's law, the instantaneous heat generation power on the terminal surface is calculated by multiplying the square of the real-time current amplitude by the equivalent resistance of the terminal. The calculated instantaneous heat generation power is then spatially normalized, and combined with the effective heat dissipation surface area of ​​the terminal, a real-time heat power density sequence per unit area is derived.

[0129] A time-smoothing filter mechanism is introduced to eliminate glitches in thermal power calculation caused by high-frequency noise in current sampling, ensuring the stability of the heat source term input. The processed real-time thermal power density sequence is then converted into the boundary conditions of the internal heat source terms in the thermal conduction differential equation system, realizing the energy form conversion from the electrical domain to the thermodynamic domain.

[0130] Through the above processing method, the current amplitude data of the previous step is transformed into a real-time heat power density sequence that characterizes the intensity of the terminal heat source, realizing the accurate mapping of electrical excitation quantity to thermodynamic boundary condition input quantity, and providing accurate heat source driving data for subsequent dynamic heat balance equation solution.

[0131] For example, under a rated current of 60A, the real-time current amplitude is extracted to be 60.5A, and the equivalent resistance of the terminals is set to 0.0002Ω. Based on Joule's law, the instantaneous heating power is calculated to be 0.732W. Combined with the terminal surface area of ​​0.001 square meters, the heat power density is calculated to be 732W / m². After sliding window filtering, a smooth heat power density sequence is output as the input for the heat source term. This process significantly improves the accuracy of heat source calculation, eliminates thermal simulation errors caused by current fluctuations, and ensures a high degree of consistency between the theoretical temperature response curve and the actual physical process.

[0132] S4.3: By utilizing the real-time heat power density sequence and the ambient airflow velocity and chamber temperature and humidity components in the real-time operating condition driving vector, the convective heat transfer coefficient boundary conditions and heat source term parameters in the heat conduction differential model are dynamically updated to construct an instance of a dynamic heat balance equation set containing time-varying physical parameters.

[0133] Read the air temperature and relative humidity data inside the chamber from the real-time operating condition driving vector, calculate the specific heat capacity and thermal conductivity of the air at the current moment based on the empirical formula of the thermal properties of humid air, and generate a set of air property parameters that reflect the heat transfer capacity of the medium.

[0134] The ambient airflow velocity component in the real-time operating condition driving vector is extracted. Combined with the terminal geometric surface area characteristics, the local heat transfer enhancement factor under forced convection conditions is calculated using the Nusselt number correlation formula, and the basic natural convection heat transfer coefficient is corrected.

[0135] The modified local heat transfer enhancement factor is multiplied by the air thermal conductivity and divided by the terminal characteristic length to calculate the instantaneous convective heat transfer coefficient at the current time step, which is used as the input quantity of the dynamic boundary condition.

[0136] Obtain the real-time thermal power density sequence generated in the previous step, map it to the internal heat source term in the thermal conduction differential equation system, and establish the energy injection rate of electrical energy loss converted into thermal energy.

[0137] Based on the aforementioned lumped-parameter heat balance equation, the instantaneous convective heat transfer coefficient, real-time thermal power, and ambient temperature are substituted to complete the parameter instantiation of the dynamic heat balance equation set instance, forming a thermodynamic state equation with time-varying characteristics.

[0138] Substituting the instantaneous convective heat transfer coefficient, real-time thermal power, and ambient temperature into the above equations, the parameters of the dynamic thermal balance equation set are instantiated, forming a thermodynamic state equation with time-varying characteristics.

[0139] By dynamically updating the convective heat transfer coefficient and heat source parameters, the static physical model is transformed into an instance of a dynamic thermal balance equation set that adapts to real-time environmental fluctuations, thereby achieving a high-fidelity reproduction of the theoretical temperature response curve for complex operating conditions.

[0140] For example, the current temperature inside the chamber is set to 35℃ and the relative humidity to 60%. The air thermal conductivity, obtained from a table, is 0.026 W / (m·K). The monitored ambient airflow velocity is 0.5 m / s, and the local heat transfer enhancement factor, calculated using the Nusselt number correlation, is 1.2. The basic natural convection heat transfer coefficient is 5 W / (m²·K), and the corrected instantaneous convection heat transfer coefficient h(t) is 6 W / (m²·K). The real-time thermal power P(t) is 15 W, and the contact thermal resistance R... contact The heat dissipation area A is 0.01m², and the ambient temperature T is 0.5K / W. env The temperature is 35℃. Substituting these parameters into the dynamic heat balance equation, the heat exchange boundary conditions at the current moment are accurately defined, significantly improving the robustness of the simulation model to environmental disturbances.

[0141] S4.4: Perform explicit finite difference numerical iterative solution on the dynamic heat balance equation set instance, using three adjustable physical parameters, equivalent heat capacity, contact thermal resistance and convective heat transfer coefficient, as the constraint basis, and gradually deduce the theoretical nodal temperature distribution sequence under discrete time step.

[0142] The time-varying physical parameters in the dynamic thermal equilibrium equations, including equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient, are obtained as the constraint basis for numerical solution. The discrete time step Δt is set, and the maximum allowable step size is determined according to the Courant-Friedrich-Levy stability condition to ensure the numerical stability of the explicit difference scheme.

[0143] A difference iterative scheme for a first-order lumped-parameter thermal network is constructed to transform the continuous differential equation into a discrete algebraic equation. The forward Euler method is used to discretize the temperature derivative with respect to time, and a recursive relationship is established between the node temperature at the current time step and the node temperature and heat source term at the previous time step.

[0144] The theoretical node temperature at the (k+1)th time step is calculated using the following formula:

[0145] in, For the first The theoretical terminal temperature over a time step For the first The theoretical terminal temperature over a time step For ambient temperature, For real-time Joule thermal power, For equivalent heat capacity, For contact thermal resistance, The convective heat transfer coefficient is... For heat dissipation surface area, denoted as the discrete time step.

[0146] The Joule thermal power P calculated by substituting the current amplitude into the real-time operating condition drive vector Joule The calculation results T from the previous time step are used to perform a single-step iterative calculation, along with the dynamically updated convective heat transfer coefficient h. k As initial conditions, and combining the heat source term and heat dissipation boundary conditions of the current step, the theoretical temperature value T of the current step is calculated. k+1 .

[0147] The calculated T k+1 Store in a temporary buffer and update T for the next iteration. k The above differential iteration process is repeated until the entire monitoring time window is covered. Through the step-by-step derivation of the explicit finite difference method, the continuous heat conduction differential equation is transformed into a theoretical node temperature distribution sequence under discrete time series, realizing lightweight real-time simulation of the terminal thermal response process, and providing a high-precision theoretical reference signal for subsequent morphological consistency comparison.

[0148] For example, the equivalent heat capacity C is set. eq It is 15.2 J / K, and the contact thermal resistance R is 15.2 J / K. c The heat transfer coefficient is 0.8 K / W, the convective heat transfer coefficient h is 12.5 W / (m²·K), the heat dissipation area A is 0.005 m², and the ambient temperature T is 0.8 K / W. env The initial temperature of the terminal is 25℃, and the time step Δt is 0.1s. At t=0, the initial terminal temperature T0 is 25℃, and the applied current generates Joule heat power P. Joule The value is 50W. Substituting into the formula, the theoretical temperature for the first step is calculated as: T1 = 25 + (0.1 / 15.2) * (50 - (25 - 25) / (0.8 + 1 / (12.5 * 0.005))) ≈ 25.329℃. As the iteration progresses, at the 100th step (t = 10s), the theoretical temperature converges to near the steady-state value. This process does not require complex matrix inversion, and the calculation time is less than 1ms, significantly improving the response speed of real-time simulation on the edge side.

[0149] S4.5: Based on the theoretical node temperature distribution sequence, perform time-series reconstruction and interpolation smoothing to output a continuous theoretical temperature response curve that reflects the current load and environmental coupling state, which serves as a benchmark reference signal for subsequent morphological consistency comparison.

[0150] The system receives the theoretical node temperature distribution sequence at discrete time steps from the numerical iterative solution output by the explicit finite difference method. This sequence contains temperature sampling points at several non-uniform time intervals.

[0151] A time-axis linear interpolation process is performed on the theoretical node temperature distribution sequence. Based on a preset fixed sampling frequency, an equally spaced time grid is constructed. The slope of temperature change between adjacent discrete nodes is calculated to fill the data gaps caused by variable step-size integration and generate a theoretical temperature intermediate sequence with a unified time base.

[0152] A cubic spline interpolation algorithm is used to smoothly fit the intermediate theoretical temperature sequence, constructing a piecewise cubic polynomial function to ensure the continuity of the first and second derivatives of the curve at the connection points, eliminating the high-frequency jagged noise introduced by numerical calculation, and forming a smooth and continuous theoretical temperature response trajectory.

[0153] Boundary condition constraint verification is performed on the smooth and continuous theoretical temperature response trajectory to eliminate abnormal fluctuation points that exceed the physical limit temperature. The edge data is corrected using a local weighted regression method, and the final output is a standard theoretical temperature response curve that reflects the current load and environmental coupling state.

[0154] The theoretical temperature response curve is a continuous curve that reflects the change of the terminal theoretical temperature over time under the current load and environmental coupling state. It is obtained by numerically iterating the discrete temperature node sequence from the heat conduction differential model and then processing it through time-series reconstruction and interpolation smoothing.

[0155] By using time-series reconstruction and interpolation smoothing, the discrete temperature distribution sequence from the previous step is transformed into continuous theoretical temperature response curve data with a high signal-to-noise ratio, thus achieving the expected technical effect of providing a high-precision reference signal for subsequent dynamic time-warping comparison.

[0156] For example, the simulation step size is set to 0.1 seconds and the acquisition period to 1 second. Linear interpolation is performed on the discrete temperature sequence [25.0, 25.8, 26.5...] to generate an intermediate sequence with 10 times the density. Cubic spline interpolation is applied to construct a smooth curve that satisfies the continuity of S''(x), eliminating numerical oscillations. After boundary verification, non-physical spikes greater than 150 degrees Celsius are removed. The final output is a continuous theoretical temperature curve with a sampling rate of 10Hz, and its root mean square error is less than 0.05 degrees Celsius, significantly improving the morphological consistency accuracy with the measured curve.

[0157] like Figure 3 As shown, step S5: Perform dynamic time warping comparison operation on the real-time measured temperature curve and the theoretical temperature response curve, calculate the dynamic time warping similarity score representing the morphological similarity between the two, and simultaneously extract the thermal inertia fingerprint vector corresponding to the real-time measured temperature curve. Specifically, this includes: S5.1: Perform time-axis resampling on the real-time measured temperature curve and the theoretical temperature response curve to unify their sampling frequency and time length, generating standardized measured temperature sequences and standardized theoretical temperature sequences with the same time dimension, providing an aligned input basis for subsequent dynamic time warping comparison operations.

[0158] The theoretical temperature response curve output by S4.5 and the real-time measured temperature curve generated by S1.3 are received and used as the original input data sequence for time axis resampling processing.

[0159] Timestamp analysis was performed on the theoretical temperature response curve and the measured temperature curve respectively to extract their respective time vector sequences and determine the start time, end time and original sampling interval of both.

[0160] Based on a preset unified target sampling frequency, a standard equidistant time axis vector is constructed. The time step of this vector is determined by the system clock reference to ensure the timing alignment accuracy of subsequent comparison operations.

[0161] For the theoretical temperature response curve, a cubic spline interpolation algorithm is used to map the discrete theoretical node temperature distribution onto a standard equidistant time axis, generating a continuously differentiable theoretical temperature interpolation function.

[0162] At each discrete point on the standard time axis, the function value of the theoretical temperature interpolation function is calculated to form a standard theoretical temperature sequence with a fixed length N.

[0163] For real-time measured temperature curves, cubic spline interpolation or linear interpolation algorithms are also applied, and resampling is performed based on a standard equidistant time axis to eliminate time jitter caused by asynchronous sensor acquisition.

[0164] At the corresponding discrete points on the standard time axis, the measured temperature values ​​after resampling are obtained to form a standard measured temperature sequence with the same length as the theoretical sequence.

[0165] Zero-mean normalization was performed on the generated standard theoretical temperature sequence and the standard measured temperature sequence to eliminate the interference of absolute temperature benchmark differences on morphological comparison.

[0166] Through the above resampling and normalization processes, heterogeneous time-series data are transformed into standard temperature sequences with consistent dimensions and time base alignment, providing a high-quality input basis for dynamic time warping algorithms and significantly improving the robustness of morphological similarity calculation.

[0167] For example, a unified target sampling frequency of 10Hz is set, i.e., a time step of 0.1s. The original sampling rate of the theoretical curve is 1Hz, while the original sampling rate of the measured curve is 5Hz with ±20ms jitter. A standard time axis of 1200 points (covering 120s) is constructed. Cubic spline interpolation is used on the theoretical curve to calculate the interpolated temperature at t=0.1s, 0.2s...120.0s, resulting in a 1200-dimensional standard theoretical sequence. The measured curve is denoised and then linearly resampled to the same time point to obtain a 1200-dimensional standard measured sequence. The mean of each sequence is subtracted from the two sequences and divided by the standard deviation to complete normalization. This process eliminates the DTW path distortion caused by the sampling rate mismatch, reducing the subsequent similarity score calculation error to within 0.01, significantly improving the anomaly identification accuracy.

[0168] S5.2: Construct a local distance matrix based on the standardized measured temperature sequence and the standardized theoretical temperature sequence, and use the dynamic time warping algorithm to calculate the minimum cumulative distance path to eliminate the nonlinear drift of the time axis caused by the difference in heat conduction delay, and generate a dynamic time warping alignment path that characterizes the degree of morphological matching between the two.

[0169] The system receives standardized measured temperature sequences and standardized theoretical temperature sequences after S5.1 resampling processing, serving as the input data source for constructing the local distance matrix. A two-dimensional distance matrix D with dimension M×N is initialized, where M is the length of the measured sequence and N is the length of the theoretical sequence. The matrix elements are initially set to infinity to store the Euclidean distance cost between each pair of time points. The system iterates through each time point i in the measured sequence and each time point j in the theoretical sequence, extracting the temperature value at the corresponding time.

[0170] Calculate the instantaneous Euclidean distance between two points, and fill the calculation result into the i-th row and j-th column of the distance matrix D to complete the local cost mapping.

[0171] Based on the recursive logic of the cumulative distance algorithm of dynamic time warping, a cumulative distance matrix C is constructed, and the starting point C(1,1) is set to be equal to d(1,1). The matrix is ​​traversed in the order from left to right and from top to bottom. For any non-boundary point (i,j), according to the step size constraint, the minimum cumulative distance value among the three adjacent predecessor nodes C(i,j-1) to the left, C(i-1,j) above, and C(i-1,j-1) to the upper left is selected.

[0172] The selected minimum cumulative distance of the predecessor is added to the current local distance d(i,j) to obtain the cumulative distance C(i,j) of the current point, thereby eliminating the nonlinear drift of the time axis caused by the difference in thermal conduction delay.

[0173] Starting from the bottom right corner endpoint C(M,N) of the matrix, backtrack backward to the top left corner starting point C(1,1), recording the path index pair (i,j) that minimizes the cumulative distance at each backtracking step. Arrange all recorded path index pairs in chronological order to generate a dynamic time-warped alignment path that represents the optimal shape matching relationship between the measured curve and the theoretical curve.

[0174] By using the dynamic programming and backtracking methods described above, the time-series data from the previous step is transformed into aligned path data that eliminates phase deviation, thus achieving a high-precision quantification foundation for consistent thermal response morphology.

[0175] For example, suppose both the measured and theoretical sequence lengths are 100 points, with a sampling interval of 1 second. At t=10s, the measured temperature lags due to increased contact resistance; the theoretical value is 45℃, while the measured value is 42℃. The local distance d(10,10)=9 is calculated. In the cumulative calculation, the algorithm automatically maps the measured point t=10 to the theoretical point t=12, minimizing the cumulative distance increment. The final generated alignment path exhibits a slope change in the lag interval, effectively compensating for thermal inertia differences. The output alignment path contains 100 coordinate pairs, providing an accurate benchmark for subsequent similarity score calculations.

[0176] S5.3: Extract the normalized cumulative distance value based on the dynamic time warping alignment path, and perform a reciprocal weighted operation in combination with the sequence length factor to convert the distance metric into a similarity index, generating a dynamic time warping similarity score that quantitatively reflects the consistency between the measured curve and the theoretical curve shape.

[0177] The dynamic time-warped alignment path and local distance matrix are obtained, and the minimum cumulative distance value is extracted as the original metric for morphological difference. This cumulative distance value reflects the sum of the total Euclidean distances between the measured temperature sequence and the theoretical temperature sequence under the optimal alignment path; the larger the value, the greater the degree of morphological deviation between the two.

[0178] The minimum cumulative distance value is normalized and divided by the alignment path length factor to eliminate the influence of sequence length on the dimension of the distance metric. The path length factor is set to the total number of matching point pairs in the optimal warpath generated by the dynamic time warping algorithm, ensuring that similarity scores are comparable under different sampling durations.

[0179] A reciprocal weighting function is used to convert the normalized distance metric into a similarity index, constructing a nonlinear mapping relationship from the distance space to the similarity space. A smoothing coefficient is introduced to avoid computational overflow caused by zero denominator, while enhancing the sensitivity of small distance changes to the similarity score.

[0180] The dynamic time-warped similarity score is calculated using the following formula:

[0181] in, For dynamic time-warped similarity scores, The minimum cumulative distance value. To align path length, This is the sensitivity adjustment coefficient.

[0182] The calculated similarity score is compared with a preset morphological threshold. This score quantitatively characterizes the degree of morphological consistency between the measured thermal response curve and the theoretical curve derived from the physical model. The closer the score is to 1, the more the measured behavior conforms to the physical laws of thermal inertia of a normal terminal.

[0183] By using a reciprocal weighted operation, the distance metric from the previous step is transformed into a similarity index in the range of 0 to 1, thereby achieving a quantitative assessment of the consistency of the terminal thermal response morphology and effectively improving the robustness of abnormal pattern recognition to nonlinear time drift.

[0184] For example, the sensitivity adjustment coefficient α is set to 0.5. In a certain real-time monitoring, the minimum cumulative distance value D is obtained through dynamic time warping calculation. DTW The value is 12.8, and the alignment path length L is 64 sampling points. Substituting into the formula, the normalized distance is calculated as 12.8 / 64 = 0.2. Further calculation of the similarity score S = 1 / (1 + 0.5 + 0.2) = 1 / 1.1 ≈ 0.909. If the preset morphological threshold is 0.85, then 0.909 is greater than 0.85, and the morphological consistency is deemed acceptable. If the thermal resistance increases due to loose terminals, the measured curve lag will worsen, and D... DTW If the value is increased to 25.6, then S decreases to 1 / (1+0.5-0.4)=0.833, which is below the threshold of 0.85, triggering the out-of-limit flag. This calculation process is completed within milliseconds, significantly improving the ability to detect early contact failures.

[0185] S5.4: Perform segmented slope extreme value detection processing on the real-time measured temperature curve to identify the inflection point position and steady-state plateau interval of the temperature rise and fall segments, so as to extract a time-series structured parameter set containing characteristic inflection point time, slope change rate, plateau maintenance duration and steady-state temperature rise difference.

[0186] Receive the standardized measured temperature sequence output from step S5.1. This sequence has a unified time dimension and sampling frequency, serving as the raw data basis for feature extraction.

[0187] A sliding window first-order difference operation is performed on the standardized measured temperature sequence to calculate the rate of temperature change at adjacent time steps, generating a first-order derivative sequence that reflects the thermal response rate, so as to quantify the dynamic trend of temperature change.

[0188] A local extremum search algorithm is executed based on the first derivative sequence. A slope change threshold is set to identify the zero-crossing point where the derivative changes from positive to negative or from negative to positive. The turning point between the temperature rise and fall segments is located, and the time coordinate of the characteristic inflection point is determined.

[0189] Using the determined characteristic inflection point time coordinates, the original temperature sequence is segmented and extracted to separate the temperature rise interval, the high temperature plateau maintenance interval, and the temperature fall interval, thus constructing a set of segmented temperature subsequences.

[0190] Linear regression fitting was performed on the temperature rise and fall intervals respectively, and the average slope value in each interval was calculated as the slope change rate index characterizing the heat transfer efficiency, thus eliminating the influence of instantaneous noise on a single sampling point.

[0191] The duration for which the temperature fluctuation within the high-temperature platform maintenance range is less than the preset stability threshold is recorded as the platform maintenance duration, reflecting the stability characteristics of the terminal reaching thermal equilibrium.

[0192] Extract the initial temperature value of the temperature rise range and the average steady-state temperature value of the platform maintenance range, calculate the difference between the two, and obtain the steady-state temperature rise difference, which represents the final heat accumulation under the current load condition.

[0193] The characteristic inflection point time, slope change rate, plateau duration, and steady-state temperature rise difference are arranged in a fixed order and combined to form a time-series structured parameter set, thus completing the transformation from continuous curves to discrete feature vectors.

[0194] By employing segmented slope extremum detection and interval statistical processing, the continuous temperature sequence from the previous step is transformed into a time-series structured parameter set containing key thermodynamic features. This achieves a digital abstraction of the terminal thermal response morphology, providing highly discriminative feature input for subsequent fingerprint vector encapsulation. For example, a standardized measured temperature sequence with a sampling frequency of 1Hz and a length of 120 points is used, and a sliding window with a window size of 5 points is employed to calculate the first-order difference. A slope abrupt change threshold of 0.5℃ / s is set, identifying the 30th and 90th seconds as feature inflection points. The 0-30 second segment is extracted as the rising segment, and the fitted slope change rate is 0.8℃ / s; the 30-90 second segment is the plateau segment, with a statistical duration of 60 seconds; the initial temperature of 25℃ and the steady-state temperature of 55℃ are extracted, and the steady-state temperature rise difference is calculated to be 30℃. The final output parameter set {30s, 0.8℃ / s, 60s, 30℃} significantly improves the noise resistance and physical interpretability of feature extraction.

[0195] S5.5: Based on the time-series structured parameter set, perform vector encapsulation operation, combine feature parameters of each dimension according to the preset fingerprint encoding rules, so as to form a thermal inertia fingerprint vector that can characterize the inherent thermal response characteristics of the terminal under the current load and environment coupling state.

[0196] Receive the time-series structured parameter set output from step S5.4. This parameter set contains raw physical quantity data in four dimensions: characteristic inflection point time, slope change rate, platform maintenance duration, and steady-state temperature rise difference.

[0197] The original physical quantity data is subjected to dimensionless normalization to eliminate the order-of-magnitude differences between different physical dimensions and ensure the weight balance of each dimension feature in the vector space. The min-max normalization algorithm is used to map each feature value to the interval [0,1] to construct a standardized feature component sequence.

[0198] Based on the preset fingerprint encoding rules, the arrangement order and bit width allocation of each feature component are determined. The feature inflection point time occupies the high bit segment, the slope change rate occupies the middle bit segment, and the platform maintenance time and steady-state temperature rise difference occupies the low bit segment, forming a fixed vector topology structure.

[0199] The normalized feature components are serially spliced ​​according to a predetermined topology to generate an initial thermal inertial fingerprint data stream, ensuring the consistency of byte alignment and storage format of the data stream.

[0200] A unique device identifier and an environmental compensation coefficient are introduced as verification suffixes and appended to the end of the initial thermal inertial fingerprint data stream to enhance the uniqueness and environmental adaptability of the fingerprint vector.

[0201] A hash digest operation is performed on the concatenated data stream to generate a checksum and embed it into the vector header, forming the final data structure with integrity protection mechanism.

[0202] Through the above vector encapsulation operation, the multi-dimensional temporal structured parameters extracted in the previous step are transformed into thermal inertial fingerprint vectors with uniqueness and anti-drift characteristics, realizing the morphological transformation from discrete physical features to standardized digital fingerprints, and providing a unified data benchmark for subsequent deviation comparison.

[0203] For example, the feature inflection point time is set to 12.5 seconds, the slope change rate is 0.8℃ / s, the platform maintenance time is 45 seconds, and the steady-state temperature rise difference is 15℃. Using minimum-maximum normalization, assuming the maximum values ​​of each feature are 20 seconds, 1.0℃ / s, 60 seconds, and 20℃, the normalized values ​​are calculated to be 0.625, 0.8, 0.75, and 0.75, respectively. According to the preset encoding rules, 0.625 is encoded as a 16-bit binary number 1010000000000000, 0.8 is encoded as 1011001100110011, 0.75 is encoded as 11000000000000000, and 0.75 is encoded as 11000000000000000. After serial concatenation, a 64-bit initial fingerprint data stream is obtained. The device ID hash value and an environmental compensation coefficient of 0.98 are added to generate the final fingerprint vector. This vector accurately characterizes the thermal response characteristics under the current operating conditions, significantly improving the accuracy and robustness of anomaly identification.

[0204] Step S6: Determine whether the dynamic time warping similarity score is lower than a preset morphological threshold and whether the deviation between the thermal inertial fingerprint vector and the terminal thermal inertial fingerprint vector exceeds a preset tolerance band. If both conditions are met, generate a terminal abnormal mode trigger signal; otherwise, generate a normal operation status indicator. Specifically, this includes: S6.1: Obtain the real-time measured temperature curve and the theoretical temperature response curve as input objects, perform nonlinear time axis alignment processing based on the dynamic time warping algorithm to eliminate the time phase difference caused by load transients, calculate and output the dynamic time warping similarity score that represents the consistency of the two shapes.

[0205] The standardized measured temperature sequence and the standardized theoretical temperature sequence, after time axis resampling, are obtained as input data objects for dynamic time warping comparison.

[0206] A local distance matrix is ​​constructed, and the Euclidean distance between each time point in the measured sequence and all time points in the theoretical sequence is calculated to form a cost matrix that reflects the local morphological differences.

[0207] A dynamic programming algorithm is used to find the path with the minimum cumulative distance. The algorithm traverses from the top left corner to the bottom right corner of the matrix to find the alignment path that minimizes the total cumulative distance, thereby eliminating the nonlinear time drift caused by thermal inertia differences.

[0208] The normalized cumulative distance value is extracted based on the optimal alignment path. This value represents the degree of deviation of the overall shape of the two curves under the optimal time mapping.

[0209] By combining the sequence length factor with the cumulative distance and performing a reciprocal weighted operation, the distance metric is converted into a similarity index in the range of 0 to 1, generating a dynamic time-warped similarity score.

[0210] By using the nonlinear alignment mechanism of the dynamic time warping algorithm, the resampled data from the previous step is transformed into a similarity score for quantized morphological consistency, thereby achieving accurate comparison of thermal response trajectories under load transients.

[0211] For example, both the measured and theoretical sequence lengths are set to 120 points, with a sampling interval of 1 second. When calculating the local distance matrix, the absolute value of the temperature difference between the i-th measured point and the j-th theoretical point is used as the cost. Dynamic programming is used to solve for the minimum path, assuming the final normalized cumulative distance is 15.6. The sequence length weight coefficient is set to 0.8, and the calculated similarity score is approximately 0.075. This score directly reflects the degree of agreement between the current terminal thermal behavior and the standard fingerprint, significantly improving the anti-interference capability of anomaly identification.

[0212] S6.2: Obtain the thermal inertial fingerprint vector and the original terminal thermal inertial fingerprint vector as input objects, perform multi-dimensional feature space difference calculation based on the Euclidean distance metric algorithm to quantify the physical property drift caused by device aging or poor contact, calculate and output the fingerprint deviation quantification value that characterizes the degree of fingerprint deviation.

[0213] The thermal inertial fingerprint vector generated by S5.5 and the original terminal thermal inertial fingerprint vector stored in S2 are obtained as inputs for calculating the difference in the multidimensional feature space. A dimensional consistency check is performed on the two fingerprint vectors to ensure data alignment across four dimensions: feature inflection point time, slope change rate, platform sustaining duration, and steady-state temperature rise difference. A geometric distance model of the multidimensional feature space is constructed using an Euclidean distance metric algorithm to quantify the deviation between the measured thermal response characteristics and the baseline physical model. The fingerprint deviation quantification value is calculated using the following formula:

[0214] in, This is the quantification value of fingerprint deviation. The current running fingerprint vector is the first One portion, The first of the original calibrated fingerprint vectors Each component has a variance. The difference between each component is squared to eliminate the canceling effect of positive and negative values, and the sum of the squared differences across all dimensions is obtained as the total variance. The total variance is then squared to obtain a scalar distance index characterizing the degree of physical property drift. Dimensional weighting coefficients are introduced to weight specific sensitive features, enhancing sensitivity to key anomalies such as changes in contact thermal resistance. The calculated fingerprint deviation quantization value is output to S6.3 for subsequent boundary determination.

[0215] By using Euclidean distance metric processing, the multidimensional feature vector from the previous step is transformed into a single fingerprint deviation quantification value, which enables accurate quantification of physical property drift caused by terminal aging or poor contact, and improves the ability of anomaly identification to capture progressive faults.

[0216] For example, the current running fingerprint vector is [12.5s, 0.8℃ / s, 300s, 15.2℃], and the original fingerprint vector is [12.0s, 0.9℃ / s, 290s, 14.5℃]. The differences in each dimension are calculated to be 0.5, -0.1, 10, and 0.7, respectively. The squared values ​​are 0.25, 0.01, 100, and 0.49, respectively. The sum is 100.75. The square root yields a fingerprint deviation quantization value of approximately 10.04. If the preset tolerance threshold is 5.0, this value exceeds the threshold, triggering the fingerprint over-limit flag, indicating a significant deterioration in the terminal contact state.

[0217] S6.3: Obtain the preset morphology threshold and preset tolerance band as reference parameters, and perform boundary judgment processing on the dynamic time warping similarity score and fingerprint deviation quantization value based on the comparison logic circuit to identify single-dimensional abnormal signs and generate morphology limit-out flag and fingerprint limit-out flag.

[0218] Read the preset morphological similarity threshold Th from the storage unit sim Fingerprint deviation tolerance upper limit D tol It is loaded into the comparator's reference register to establish a static baseline for exception detection.

[0219] The preset morphology threshold refers to a similarity judgment boundary value pre-set and stored in the system, used to compare with the dynamic time warping similarity score. When the similarity score is lower than the threshold, it is determined that the morphology of the measured temperature curve deviates from the theoretical expectation. The preset tolerance band refers to the fingerprint deviation allowable range pre-set and stored in the system, used to compare with the fingerprint deviation quantification value. When the deviation quantification value exceeds the tolerance band, it is determined that the physical properties of the terminal have drifted significantly.

[0220] Receive the dynamic time-warped similarity score S from the output of step S6.1. dtw This score characterizes the degree of agreement between the measured temperature curve and the theoretical simulation curve after alignment in the time domain.

[0221] Execute the first-way comparison logic, and set S dtw With Th sim Perform a numerical comparison; if S dtw Less than Th sim If the thermal response pattern is distorted, the flag_shape of the out-of-limit setting will be set to a high level.

[0222] Receive the fingerprint deviation quantization value Dev from the output of step S6.2. finger This value reflects the Euclidean distance drift of the current operating thermal inertial fingerprint vector relative to the original calibrated fingerprint vector.

[0223] Execute the second-way comparison logic, and compare Dev finger With D tol Perform a numerical comparison; if Dev finger Greater than D tol If the physical thermal properties of the terminal deviate significantly, the fingerprint over-limit flag will be set. drift It is in a high-level active state.

[0224] Through the parallel comparison processing described above, the continuous similarity scores and deviation quantization values ​​are transformed into discrete binary logic flags, realizing dual independent monitoring of thermal behavior anomalies and physical property drift, and providing clear logical input signals for subsequent joint verification.

[0225] For example, setting a morphological similarity threshold Th sim The upper limit of fingerprint deviation tolerance is 0.85. tol The value is 0.15. When a terminal block is detected in real-time under a sudden load change condition, the calculated dynamic time warping similarity score S is... dtw The value is 0.72. Since 0.72 is less than 0.85, the comparator outputs a Flag. shape =1 indicates an unexpected lag or oscillation in the thermal response trajectory. Simultaneously, the Euclidean distance Dev between the extracted current running-state fingerprint vector and the original fingerprint vector is... finger The value is 0.08. Since 0.08 is less than 0.15, the comparator outputs a Flag. drift =0 indicates that the thermal inertia of the terminal body has not undergone structural changes. This combination of flags indicates that the anomaly at this moment is mainly caused by transient load disturbances or momentary arcing at the contact point, rather than terminal aging and loosening, thus avoiding false alarms for normal transient processes and significantly improving the anti-interference capability of the early warning system.

[0226] S6.4: Obtain the morphology over-limit flag and fingerprint over-limit flag as logic input, perform dual-condition joint verification processing based on Boolean AND gate logic to eliminate single-factor false alarms caused by environmental changes or sensor noise, and generate terminal abnormal mode trigger signal or normal operation status indicator.

[0227] The morphology over-limit flag and fingerprint over-limit flag output from step S6.3 are received as Boolean input signals for the logic verification unit. A level state detection is performed on the morphology over-limit flag. If the flag is high, it indicates that the dynamic time warping similarity score is lower than a preset morphology threshold, and it is determined that there is a significant nonlinear mismatch between the measured temperature curve and the theoretical response curve in the time domain morphology. A level state detection is also performed on the fingerprint over-limit flag. If the flag is high, it indicates that the Euclidean distance between the thermal inertial fingerprint vector and the original reference vector exceeds a preset tolerance band, and it is determined that the physical thermal properties of the terminal have abnormally drifted. The two flag signals are connected to a digital logic AND gate circuit, and a Boolean AND operation is performed. Only when both input signals are simultaneously high, a high-level trigger pulse is generated at the AND gate output. This trigger pulse is mapped to a terminal abnormal mode trigger signal, used to drive subsequent alarm modules or protection actions. If either input signal is low, the AND gate output remains low, generating a normal operation status indicator. By using a dual-condition joint verification mechanism, the single-dimensional criteria of the previous step are transformed into a comprehensive anomaly decision signal, thereby eliminating single-factor false alarms caused by environmental changes or sensor noise and significantly improving the robustness and accuracy of the early warning system.

[0228] For example, the morphology threshold is set to 0.85, and the fingerprint tolerance band radius is 0.12. Under a sudden load increase scenario, the measured DTW similarity score is 0.92 (above the threshold), and the morphology exceeding the limit flag is 0; the fingerprint deviation value is 0.15 (exceeding the tolerance), and the fingerprint exceeding the limit flag is 1. An AND gate with inputs of 0 and 1 outputs 0, generating a normal operating status indicator to avoid false alarms. Under a terminal loosening fault scenario, the DTW similarity score is 0.70 (below the threshold), and the morphology exceeding the limit flag is 1; the fingerprint deviation value is 0.20 (exceeding the tolerance), and the fingerprint exceeding the limit flag is 1. An AND gate with inputs of 1 and 1 outputs 1, generating a terminal abnormality mode trigger signal to accurately identify the fault.

[0229] Preferably, in response to a terminal abnormality mode trigger signal, or in response to a preset periodic stable low-load maintenance arrival time, the terminal thermal inertial fingerprint vector is updated or adjustable physical parameters are adjusted by comparing the current fingerprint with the aging model prediction range, thereby completing the adaptive closed-loop optimization of the abnormality identification model. Specifically: When the system determines that a terminal exhibits an abnormal mode, or reaches the preset periodic stable low-load maintenance time, it automatically initiates the model self-calibration process. Because the terminal contact interface undergoes irreversible physical degradation due to oxidation, mechanical vibration, thermal cycling fatigue, and other factors during long-term operation, a deviation arises between the baseline fingerprint vector and the actual physical state. If not corrected in time, the baseline of the anomaly identification model will gradually become ineffective, and the false alarm and false negative rates will increase over time. This step, through replay calibration, deviation quantification, and parameter inversion, enables the model to track the aging trajectory of the equipment, upgrading the baseline from a static calibration value to a dynamic tracking value, thereby maintaining the accuracy and reliability of anomaly detection throughout the entire lifecycle.

[0230] When an abnormal trigger signal arrives or the scheduled maintenance time arrives, the system filters out the current operating data that meets the basic excitation conditions (i.e., the stepped load sequence segment consistent with the factory calibration stage) from the historical operating data, and uses this as the input for drift analysis.

[0231] The thermal response simulator is then invoked to reproduce the theoretical temperature response curve under the current physical parameters, and the current thermal inertia fingerprint vector is reconstructed simultaneously from the measured temperature dynamic trajectory. The reconstructed real-time fingerprint is compared with the original baseline fingerprint using a weighted Euclidean distance calculation to obtain the comprehensive drift index D. The system then queries the aging model prediction interval to obtain the maximum allowable theoretical drift threshold T at the current moment. max If D≤T max This indicates that the current drift is within the normal aging range, and the system only records drift data but does not update parameters; if D>T max This indicates an abnormal drift that exceeds normal aging expectations. The system then initiates a parameter inversion optimization algorithm based on gradient descent. Using the sum of squared deviations between the measured and simulated fingerprints as the loss function, it iteratively corrects the equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient to make the simulation output approximate the measured response. The updated physical parameter set is then output after convergence.

[0232] The simulator's initial state is reset using the updated parameters, and the current fingerprint is written to the storage unit to replace the original baseline, completing a full closed loop from perceived deviation to corrected baseline. This process leverages the causal determinism of thermodynamic systems, where a unique mapping exists between physical parameters and temperature response. Therefore, the changes in physical parameters can be inferred from the deviations in the measured response, thereby achieving physical consistency correction of the model baseline.

[0233] By enabling the anomaly identification benchmark to be adaptively updated as the equipment ages, the sensitivity to detecting changes in contact thermal resistance is maintained throughout the entire lifecycle. The aging model predicts intervals to distinguish between normal aging and abnormal drift, avoiding benchmark oscillations caused by frequent updates, while ensuring early detection of progressive faults.

[0234] A power metering box terminal has been initially calibrated, with the original fingerprint vector being 40.0℃, 10.0s, 1.0℃ / s. The initial equivalent heat capacity is 50J / K, the initial contact thermal resistance is 0.5K / W, and the initial convective heat transfer coefficient is 10W / (m²·K). After 18 months of operation, the system reaches the preset monthly low-load maintenance time (2:00 AM). It automatically selects a 200-second data segment from the historical data cache that meets the stepped load sequence conditions as the current operating condition benchmark dataset. The thermal response simulator is used to reproduce the theoretical temperature response curve with the current physical parameters. Simultaneously, the current fingerprint vector is reconstructed from the measured temperature dynamic trajectory, resulting in a steady-state temperature rise of 45.2℃, an inflection point time of 12.5s, a slope change rate of 0.8℃ / s, and a comprehensive drift index D=4.15. The aging model prediction interval is queried to obtain the current maximum allowable theoretical drift threshold T. max =3.8. Because D>T max The system initiated parameter inversion optimization, which converged after 22 gradient descent iterations. The updated physical parameters are: equivalent heat capacity 51.8 J / K, contact thermal resistance 0.43 K / W, and convective heat transfer coefficient 10.7 W / (m²·K). Using the updated parameters to reset the simulator, the theoretical curves were re-compared with the same set of measured temperature curves before the update using DTW. The similarity score improved from 0.76 to 0.97, indicating that the corrected model can more accurately reproduce the actual thermal response behavior, verifying the effectiveness of the model correction. The entire update process was completed on the local terminal, taking approximately 180 ms.

[0235] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0236] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the element or object preceding “comprising” or “including” encompasses the element or object listed following “comprising” or “including” and its equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0237] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for abnormal early warning and terminal temperature correction in an electricity metering box, characterized in that, include: Acquire multidimensional time-series data of the power metering box, which includes the dynamic trajectory of terminal temperature, current amplitude, temperature and humidity inside the box and ambient airflow speed corresponding to the stepped load sequence, to form the original calibration dataset. Based on the terminal temperature dynamic trajectory in the original calibration dataset, the characteristic inflection point time, slope change rate, platform maintenance duration and steady-state temperature rise difference of the temperature rise and fall segments are extracted to generate terminal thermal inertia fingerprint vectors. Using the prior constraints of physical properties in the terminal thermal inertial fingerprint vector, the three adjustable physical parameters of the thermal response simulator, namely the equivalent heat capacity, contact thermal resistance and convective heat transfer coefficient, are initialized to construct a differential model of heat conduction. The real-time collected multi-dimensional parameter stream of the current operating condition is input into the heat conduction differential model to calculate and output the theoretical temperature response curve reflecting the current load and environmental coupling state. A dynamic time warping comparison operation is performed on the real-time measured temperature curve and the theoretical temperature response curve to calculate the dynamic time warping similarity score, and the thermal inertia fingerprint vector corresponding to the real-time measured temperature curve is extracted simultaneously. If the dynamic time warping similarity score is lower than a preset morphological threshold and the deviation between the thermal inertial fingerprint vector and the terminal thermal inertial fingerprint vector exceeds a preset tolerance band, a terminal abnormal mode trigger signal is generated if both conditions are met; otherwise, a normal operation status identifier is generated.

2. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, After generating the terminal abnormal mode trigger signal or normal operation status indicator, it also includes: In response to the terminal abnormal mode trigger signal, or in response to the preset time of periodic stable low-load maintenance, the terminal thermal inertial fingerprint vector is updated or the adjustable physical parameters are adjusted by comparing the current fingerprint with the aging model prediction range, thereby completing the adaptive closed-loop optimization of the abnormal identification model.

3. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, Based on the terminal temperature dynamic trajectory in the original calibration dataset, the characteristic inflection point time, slope change rate, plateau maintenance duration, and steady-state temperature rise difference of the temperature rise and fall segments are extracted to generate a terminal thermal inertia fingerprint vector, including: The dynamic trajectory of the terminal temperature is segmented into time series, and the continuous temperature curve is divided into a temperature rising segment and a temperature falling segment based on the inflection point of the current amplitude change of the stepped load sequence. Perform first-order and second-order derivative operations on the temperature rise segment and the temperature fall segment, calculate the slope change rate in each segment and locate the extreme point, and extract the characteristic inflection point time and slope change rate; The characteristic inflection point time is used to perform interval interception and steady state determination for each segment. The duration before the temperature reaches steady state is counted as the platform maintenance time, and the difference between the initial temperature and the steady state temperature is calculated as the steady state temperature rise difference. The terminal thermal inertial fingerprint vector is generated by combining and encoding the characteristic inflection point time, the slope change rate, the platform maintenance duration, and the steady-state temperature rise difference.

4. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The process involves performing dynamic time warping comparison between the real-time measured temperature curve and the theoretical temperature response curve to calculate a dynamic time warping similarity score, and simultaneously extracting the thermal inertia fingerprint vector corresponding to the real-time measured temperature curve, including: The real-time measured temperature curve and the theoretical temperature response curve are resampled to unify their sampling frequency and time length, generating a standardized measured temperature sequence and a standardized theoretical temperature sequence. A local distance matrix is ​​constructed based on the standardized measured temperature sequence and the standardized theoretical temperature sequence. The minimum cumulative distance path is calculated using the dynamic time warping algorithm to generate a dynamic time warping alignment path. The normalized cumulative distance value is extracted based on the dynamic time warping alignment path, and a reciprocal weighted operation is performed in combination with the sequence length factor to generate the dynamic time warping similarity score. The real-time measured temperature curve is subjected to segmented slope extreme value detection to identify the inflection point position and steady-state plateau interval of the temperature rising and falling segments, and to extract the characteristic inflection point time, slope change rate, plateau maintenance duration and steady-state temperature rise difference. The thermal inertial fingerprint vector is generated by combining the characteristic inflection point time, the slope change rate, the platform maintenance duration, and the steady-state temperature rise difference according to a preset fingerprint encoding rule.

5. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The process of forming the original calibration dataset also includes: Based on the transition edges of the stepped load sequence, the multidimensional time series data is divided into data segments corresponding to each load condition to form the original calibration dataset.

6. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The method of initializing three adjustable physical parameters of the thermal response simulator—equivalent heat capacity, contact thermal resistance, and convective heat transfer coefficient—using the prior constraints of physical properties in the terminal thermal inertial fingerprint vector includes: The equivalent heat capacity is initialized using the steady-state temperature rise difference and platform maintenance time in the terminal thermal inertia fingerprint vector; the contact thermal resistance is initialized using the rate of change of the rising slope and the inflection point time in the terminal thermal inertia fingerprint vector; and the convective heat transfer coefficient is initialized using the cooling rate of the falling segment and the ambient airflow velocity in the terminal thermal inertia fingerprint vector.

7. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The equivalent heat capacity, the contact thermal resistance, and the convective heat transfer coefficient are substituted into the preset thermal network topology to instantiate the heat conduction differential model.

8. The method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The real-time acquired current amplitude, internal temperature and humidity, and ambient airflow velocity are time-stamp aligned and filtered to generate the real-time acquired multi-dimensional parameter stream of the current operating condition.

9. A method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 1, characterized in that, The calculation and output of the theoretical temperature response curve reflecting the current load and environmental coupling state includes: The real-time heat power density is derived using the current amplitude collected in real time, and the convective heat transfer coefficient and heat source term in the heat conduction differential model are dynamically updated using the ambient airflow velocity and the temperature and humidity inside the chamber. The dynamically updated heat conduction differential model is numerically iterated to generate a discrete temperature sequence, and the discrete temperature sequence is then time-series reconstructed and interpolated to output the theoretical temperature response curve.

10. A method for abnormal early warning and terminal temperature correction of an electricity metering box according to claim 2, characterized in that, The aging model prediction range is determined by the cumulative operating time of the equipment and the average ambient temperature.

Images