Internet of Things test box heat dissipation efficiency evaluation method based on big data analysis

Abstract

The invention discloses an Internet of Things test box heat dissipation efficiency evaluation method based on big data analysis, and the method comprises the steps: collecting the data of each temperature measurement point, and carrying out the preprocessing of the data to obtain standardized time series data; a pseudo thermal excitation signal is generated based on the standardized time sequence data, and starting and stopping are controlled when abnormity occurs; performing time-frequency and cross-spectrum analysis on the standardized time sequence data and the pseudo-thermal excitation signal, and extracting a thermal admittance spectrum and a heat dissipation efficiency index; constructing a thermal network diagram and a physical constraint twinborn model in combination with the heat dissipation efficiency index, the thermal admittance spectrum and the structural information; performing dynamic prediction by using the optimized twinborn model to obtain a temperature field and heat dissipation efficiency prediction result; and performing small-amplitude intervention on the control variable, performing anti-fact reasoning and sensitivity analysis, outputting a regulation and control instruction, and issuing the regulation and control instruction to a control device when the regulation and control instruction is abnormal or degraded. Through big data acquisition and analysis, physical constraint twin modeling and intelligent regulation and control, dynamic and accurate evaluation and optimization of the heat dissipation efficiency of the Internet of Things test box in a complex operation environment are realized.

Description

Technical Field

[0001] This invention relates to the field of IoT testing and heat dissipation performance analysis technology, and in particular to a method for evaluating the heat dissipation efficiency of IoT test chambers based on big data analysis. Background Technology

[0002] Currently, with the rapid development of IoT devices and intelligent testing systems, test chambers are widely used in verifying the heat dissipation performance of electronic devices, studying thermal management characteristics, and assessing reliability under extreme environments. Existing methods for evaluating the heat dissipation efficiency of test chambers mainly rely on traditional sensor measurements and empirical model analysis. This process typically involves setting up temperature measurement points and collecting time-series temperature data, then combining this data with basic physical principles such as heat conduction, convection, and radiation to extrapolate heat dissipation performance. While these methods can reflect the overall thermal response of devices to some extent, they often suffer from insufficient data utilization, limited model generalization ability, and delayed prediction results, making it difficult to meet the needs of dynamic control and real-time optimization.

[0003] Existing research has attempted to improve the accuracy of temperature field and heat dissipation efficiency predictions by introducing thermal network models or numerical simulations. However, these methods typically require a large number of prior parameters and complex calculations, and are highly dependent on the internal topology, material parameters, and boundary conditions of the test chamber, making it difficult to achieve stable applicability in variable experimental environments. Traditional data-driven models, while capable of fitting multi-source time-series data, lack necessary physical constraints, often resulting in predictions that do not conform to thermodynamic laws, leading to distorted evaluations of heat dissipation efficiency. Especially when facing heat dissipation performance degradation or anomalies, existing methods are insufficient in characterizing causal relationships, making it difficult to verify the true mechanism of control variables through local intervention and counterfactual reasoning.

[0004] Existing technologies generally suffer from the following shortcomings: First, they lack a twin modeling method that effectively combines physical constraints with data-driven models, resulting in insufficient prediction accuracy and interpretability. Second, they lack a comprehensive evaluation and optimization mechanism for the dynamic heat dissipation efficiency index of the test chamber, making it impossible to form a stable prediction and control closed loop. Third, existing evaluation methods rely on thresholds or empirical rules for anomaly detection and degradation judgment, lacking quantitative judgment methods based on causal efficiency sensitivity. These problems restrict the further development of test chamber heat dissipation efficiency evaluation methods in terms of intelligence and high precision.

[0005] Therefore, how to provide a method for evaluating the heat dissipation efficiency of IoT test chambers based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for evaluating the heat dissipation efficiency of an IoT test chamber based on big data analysis. This invention fully integrates technologies such as multi-source sensor data acquisition from the Internet of Things, pseudo-thermal excitation injection mechanism, thermal admittance spectrum estimation, multi-dimensional heat dissipation index extraction, physical constraint twin model construction and structural optimization, and intelligent control tuning based on counterfactual reasoning. It describes in detail the entire process of interpretable, multi-dimensional, and high-precision modeling and evaluation of the dynamic heat dissipation performance of the test chamber under complex operating conditions. It has the advantages of high accuracy of evaluation results, physically interpretable model structure, and strong control strategy optimization capability. It breaks through the limitations of traditional heat dissipation evaluation that only relies on static temperature difference and energy efficiency ratio, and is applicable to thermal management efficiency monitoring and control optimization in multiple scenarios.

[0007] The method for evaluating the heat dissipation efficiency of an IoT test chamber based on big data analysis according to embodiments of the present invention includes:

[0008] Multi-source data from various temperature measurement points inside and outside the test chamber are collected by multiple types of sensors deployed through the Internet of Things. The multi-source data is preprocessed to obtain standardized time-series data.

[0009] Based on standardized time-series data, a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value is generated and superimposed on the heater or fan channel. At the same time, the parameters of the pseudo-thermal excitation signal are recorded, and the start and stop of the pseudo-thermal excitation signal are controlled according to the abnormal state detected in real time.

[0010] Time-frequency analysis and cross-spectral calculation are performed on standardized time-series data and pseudo-thermal excitation signals to obtain the thermal admittance spectrum of each temperature measurement point and extract heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, effective admittance bandwidth and equivalent heat dissipation emission coefficient.

[0011] Based on the heat dissipation efficiency index and thermal admittance spectrum, combined with the structural topology information of the test chamber, a thermal network graph containing nodes and edges is constructed, and a physical constraint twin model is established. The physical constraint twin model is then adaptively updated and its structure is optimized.

[0012] Based on the optimized physical constraint twin model, dynamic prediction is performed on standardized time series data to obtain prediction results of temperature field and heat dissipation efficiency;

[0013] During the operation of the test chamber, the control variables in the test chamber are slightly intervened. Based on the prediction results, counterfactual reasoning is performed. Combined with the heat dissipation efficiency index, the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency index is calculated, and the optimization control command is output. The abnormal or degraded state of heat dissipation performance is determined, and the optimization control command is sent to the control device of the test chamber.

[0014] Optionally, the multi-source data specifically includes temperature, wind speed, power, ambient temperature, ambient humidity, control component operating status data, and sensor data related to heat dissipation efficiency inside and outside the test chamber.

[0015] Optionally, the preprocessing of multi-source data specifically includes time synchronization, outlier removal, missing data imputation, and normalization of the multi-source data.

[0016] Optionally, the step of generating a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value based on standardized time-series data, superimposing it onto the heater or fan channel, simultaneously recording the parameters of the pseudo-thermal excitation signal, and controlling the start and stop of the pseudo-thermal excitation signal according to real-time detected abnormal states includes:

[0017] Excitation parameters are determined based on standardized time-series data. These excitation parameters include upper and lower limits of the excitation frequency band, sampling period, duration, amplitude ratio, random seed, symbol duration, and initial phase of each component.

[0018] The excitation mode is selected to generate a pseudo-thermal excitation signal. The excitation mode includes a multi-frequency sinusoidal superposition mode and a pseudo-random binary sequence mode. In the multi-frequency sinusoidal superposition mode, several discrete frequency points are selected within the excitation frequency band, and the amplitude and initial phase of each frequency component are set respectively. The superposition forms a multi-frequency sinusoidal excitation signal. In the pseudo-random binary sequence mode, a pseudo-random sequence is generated based on a preset random seed, and the sequence is shaped according to the set symbol duration to obtain the corresponding perturbation waveform.

[0019] The generated pseudo thermal excitation signal is subjected to amplitude constraint and amplitude limit verification. The maximum power disturbance amplitude when applied to the heater power channel is no higher than 1% to 5% of the rated power, and the maximum duty cycle disturbance amplitude when applied to the fan pulse width modulation channel is no higher than 1% to 5% of the rated duty cycle. Slope limits are applied to the rising and falling edges.

[0020] The verified pseudo-thermal excitation signal is superimposed onto the heater power command or the fan pulse width modulation duty cycle command, and the superimposed object, start time, end time and excitation parameters are recorded.

[0021] Establish start-stop gating and read four monitoring quantities in real time: temperature, power, current and vibration. Stop the pseudo-thermal excitation signal when any of the monitoring quantities reaches the preset abnormal threshold. Allow the pseudo-thermal excitation signal to be restarted when the monitoring quantity returns to the normal range and meets the preset minimum interval time.

[0022] Optionally, the step of performing time-frequency analysis and cross-spectral calculation on standardized time-series data and pseudo-thermal excitation signals to obtain the thermal admittance spectrum at each temperature measurement point, and extracting heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, effective admittance bandwidth, and equivalent heat dissipation emission coefficient, includes:

[0023] Based on the upper and lower limits of the frequency band, symbol boundaries, and start and end times of the pseudo-thermal excitation signal, the power time series and temperature measurement point time series in the standardized time series data are segmented into fixed lengths, the overlap rate is set, and multi-window weighting is applied to generate segmented windowed power sequences and temperature sequences. Time-frequency transformation is performed on each segment of the power sequence and temperature sequence to obtain a set of corresponding frequency points.

[0024] The power auto-power spectrum and temperature-power cross-power spectrum are calculated for the power sequence and temperature sequence. A multi-window segmented averaging strategy is used to obtain the frequency domain spectral quantity, and a timestamp and segmentation identifier are added to each frequency domain spectral quantity.

[0025] Based on the occupancy distribution and symbol boundary of the pseudo thermal excitation signal in the frequency band, weights are assigned to frequency points that are in the occupancy band and whose coherence with the excitation meets the threshold requirement, and frequency points that do not meet the requirement are set to zero, thus forming the power auto-power spectrum and temperature-power cross-power spectrum after coherent phase-locked processing.

[0026] For each temperature measurement point, the thermal admittance spectrum of each temperature measurement point is obtained based on the power self-power spectrum and the temperature-power cross-power spectrum. Non-negativity constraints are applied to the real components of the thermal admittance spectrum and frequency points that do not meet the constraints are shielded.

[0027] Within the set of surface temperature measurement points, the thermal admittance spectra of each temperature measurement point are weighted and aggregated with coherence weights to obtain the surface synthetic thermal admittance spectrum, and the contribution of outlier temperature measurement points is suppressed by robust quantile rules.

[0028] The dynamic heat dissipation efficiency is obtained by weighted summation of the real components of the surface synthetic thermal admittance spectrum within the excitation frequency band. The effective admittance bandwidth is taken as the continuous frequency span exceeding a fixed threshold. Within the time window synchronized with the pseudo thermal excitation signal, the input power and the average surface temperature difference are bandpass processed and least squares fitting is performed to obtain the equivalent heat dissipation emission coefficient. At the same time, the corresponding confidence intervals are generated for the three heat dissipation efficiency indicators.

[0029] Optionally, based on heat dissipation efficiency indicators and thermal admittance spectra, combined with the structural topology information of the test chamber, a thermal network graph containing nodes and edges is constructed, and a physical constraint twin model is established. The physical constraint twin model is then adaptively updated and its structure optimized, including:

[0030] Determine the input data, which includes thermal admittance spectra at each temperature measuring point, surface-synthesized thermal admittance spectra, dynamic heat dissipation efficiency, effective admittance bandwidth, equivalent heat dissipation emission coefficient, test chamber structural topology information, and standardized time series data.

[0031] A thermal network diagram is constructed based on the structural topology information of the test chamber. Temperature measurement points and heat dissipation-related parts inside and outside the test chamber are defined as nodes, and the connection between conduction and convection is defined as edges. Parameters are set for each edge and each node.

[0032] A physical constraint twin model is constructed based on a thermal network graph. Standardized time-series data, three heat dissipation efficiency indicators, and thermal admittance spectrum are used as inputs. The nodes and edges of the thermal network graph are used as the structural basis of the physical constraint twin model. Physical constraints are applied to the parameters during the training and inference process of the physical constraint twin model. The physical constraints include node energy balance constraints, parameter non-negativity constraints, and thermal admittance spectrum real part non-negativity constraints.

[0033] The physical constraint twin model is updated, and the real-time data is incrementally calibrated using a sliding time window to reduce temperature prediction errors and energy budget imbalances. The parameters are range-projected according to the update cycle, and parameters that exceed the physical feasible range are adjusted to the allowable range. The updated parameter set is recorded.

[0034] Structural optimization is performed on the heat network graph. Based on the statistical results of the contribution and sensitivity of each edge within the long-term window, edges whose contribution is consistently below the threshold are removed. For node pairs with stable residual patterns and adjacent to the structural topology, new candidate edges are added and included in the heat network graph after physical feasibility verification. The parameters that have converged are no longer updated.

[0035] Optionally, the predicted results of obtaining the temperature field and heat dissipation efficiency include:

[0036] Load the optimized physical constraint twin model, heat network diagram, and parameter set;

[0037] From the standardized time series data, power time series, temperature measurement point time series, wind speed time series, ambient temperature and humidity time series, and command time series related to control variables are extracted by sliding time window and organized into an input data matrix corresponding to the nodes and edges of the heat network diagram.

[0038] Set the prediction step size, prediction time domain length and update period, and use a rolling method to predict the temperature field for the next time step, outputting the temperature prediction value and timestamp of each node;

[0039] After each temperature field prediction is completed, the physical constraint twin model generates the predicted results of the heat dissipation efficiency for the corresponding time step based on three heat dissipation efficiency indicators and the surface synthetic thermal admittance spectrum as auxiliary inputs.

[0040] Temperature field and heat dissipation efficiency are predicted in a loop for multiple consecutive time steps. The predicted results are then summarized in chronological order to form a complete temperature field prediction sequence and heat dissipation efficiency prediction sequence. The prediction sequences are then cached and labeled.

[0041] Optionally, the step of implementing minor intervention on the control variables in the test chamber, performing counterfactual reasoning based on the prediction results, calculating the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency index in conjunction with the heat dissipation efficiency index, and outputting optimization control commands to determine the abnormal or degraded state of heat dissipation performance includes:

[0042] Select control variables in the test chamber, including fan speed and guide valve opening. Set upper limits for intervention amplitude and rate of change. The intervention amplitude of the fan and the guide valve shall not exceed 3% of the rated value, and the rate of change shall not exceed 2% per minute.

[0043] Within the running window, apply a small-amplitude step or gradual disturbance to a single control variable, record the start and end times, amplitude, target channel, and sequence number of the disturbance, and keep the other control variables unchanged;

[0044] Based on the predicted results of temperature field and heat dissipation efficiency, counterfactual reasoning is performed on two states, with and without disturbance, to obtain the efficiency change and temperature change within the same time window.

[0045] By combining three heat dissipation efficiency indicators—dynamic heat dissipation efficiency, effective admittance bandwidth, and equivalent heat dissipation emission coefficient—the local causal efficiency sensitivity of each control variable to dynamic heat dissipation efficiency is calculated, and the effective improvement interval and ineffective interval are determined.

[0046] Based on the local causal efficiency sensitivity and the set constraints of energy consumption, noise and temperature fluctuation, the system generates optimized control commands. It determines the abnormal or degraded state of heat dissipation performance according to the threshold and trend of the three heat dissipation efficiency indicators, outputs alarm information and command priority, and sends the optimized control commands to the control device of the test chamber.

[0047] The beneficial effects of this invention are:

[0048] This invention, by introducing a pseudo-thermal excitation signal and a thermal admittance spectral analysis mechanism, breaks away from the existing technology's reliance on single indicators such as static temperature difference and energy efficiency ratio. It can proactively apply low-amplitude perturbations during test chamber operation and dynamically mine multi-source response data, achieving proactive excitation and response analysis of the heat dissipation process. Compared to traditional methods, this invention can identify heat dissipation performance fluctuations in real time under different operating conditions, improving the proactiveness and dynamism of thermal management assessment.

[0049] This invention constructs a physically constrained twin model that integrates the thermal network topology and heat dissipation indicators, and jointly optimizes its structure and parameters, effectively solving the problem of the "black box and uninterpretable" nature of traditional models. Based on the thermal admittance spectrum, this model introduces various physical consistency constraints, such as node energy balance, parameter non-negativity, and non-negativity of the real part of the spectrum, ensuring that the predicted structure has real physical meaning. Furthermore, it allows for dynamic updates to uncertain parameters in the model, thereby improving the accuracy and robustness of heat dissipation performance prediction.

[0050] This invention proposes a method for counterfactual reasoning based on prediction results. This method enables small-scale adjustments to the control variables of the test chamber, calculates the local causal efficiency sensitivity of each control variable by combining three key heat dissipation indicators, and outputs precise control and optimization commands. This allows for intelligent identification and response to heat dissipation efficiency degradation. This mechanism makes heat dissipation control interpretable, adjustable, and forward-looking, providing a scientific basis and reliable guarantee for the energy efficiency optimization of thermal management systems. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is a flowchart of the IoT test chamber heat dissipation efficiency evaluation method based on big data analysis proposed in this invention;

[0053] Figure 2 This is a schematic diagram of the physical constraint twin model structure based on the thermal network diagram for the IoT test chamber heat dissipation efficiency evaluation method based on big data analysis proposed in this invention. Detailed Implementation

[0054] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0055] refer to Figure 1 and Figure 2 A method for evaluating the heat dissipation efficiency of IoT test chambers based on big data analysis includes:

[0056] Multi-source data from various temperature measurement points inside and outside the test chamber are collected by multiple types of sensors deployed through the Internet of Things. The multi-source data is preprocessed to obtain standardized time-series data.

[0057] Based on standardized time-series data, a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value is generated and superimposed on the heater or fan channel. At the same time, the parameters of the pseudo-thermal excitation signal are recorded, and the start and stop of the pseudo-thermal excitation signal are controlled according to the abnormal state detected in real time.

[0058] Time-frequency analysis and cross-spectral calculation are performed on standardized time-series data and pseudo-thermal excitation signals to obtain the thermal admittance spectrum of each temperature measurement point and extract heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, effective admittance bandwidth and equivalent heat dissipation emission coefficient.

[0059] Based on the heat dissipation efficiency index and thermal admittance spectrum, combined with the structural topology information of the test chamber, a thermal network graph containing nodes and edges is constructed, and a physical constraint twin model is established. The physical constraint twin model is then adaptively updated and its structure is optimized.

[0060] Based on the optimized physical constraint twin model, dynamic prediction is performed on standardized time series data to obtain prediction results of temperature field and heat dissipation efficiency;

[0061] During the operation of the test chamber, the control variables in the test chamber are slightly intervened. Based on the prediction results, counterfactual reasoning is performed. Combined with the heat dissipation efficiency index, the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency index is calculated, and the optimization control command is output. The abnormal or degraded state of heat dissipation performance is determined, and the optimization control command is sent to the control device of the test chamber.

[0062] In this embodiment, the multi-source data specifically includes temperature, wind speed, power, ambient temperature, ambient humidity, control component operating status data, and sensor data related to heat dissipation efficiency inside and outside the test chamber.

[0063] In this embodiment, the preprocessing of multi-source data specifically includes time synchronization, outlier removal, missing data imputation, and normalization of the multi-source data.

[0064] In this embodiment, the step of generating a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value based on standardized time-series data, superimposing it onto the heater or fan channel, simultaneously recording the parameters of the pseudo-thermal excitation signal, and controlling the start and stop of the pseudo-thermal excitation signal according to real-time detected abnormal states includes:

[0065] Excitation parameters are determined based on standardized time-series data. These excitation parameters include upper and lower limits of the excitation frequency band, sampling period, duration, amplitude ratio, random seed, symbol duration, and initial phase of each component.

[0066] A pseudo-thermal excitation signal is generated by selecting an excitation mode, which includes a multi-frequency sinusoidal superposition mode and a pseudo-random binary sequence mode. In the multi-frequency sinusoidal superposition mode, several discrete frequency points are selected within the excitation frequency band, and the amplitude and initial phase of each frequency component are set respectively, and superimposed to form a multi-frequency sinusoidal excitation signal. In the pseudo-random binary sequence mode, a pseudo-random sequence is generated based on a preset random seed, and the sequence is shaped according to a set symbol duration to obtain the corresponding perturbation waveform, wherein:

[0067] The process of generating a pseudo-thermal excitation signal by selecting a multi-frequency sinusoidal superposition mode includes selecting several excitation frequency points within a preset excitation frequency band, setting the corresponding component amplitude and initial phase for each excitation frequency point, superimposing the sinusoidal components of all excitation frequency points to form a multi-frequency sinusoidal excitation signal, and inputting the multi-frequency sinusoidal excitation signal into the heater or fan channel.

[0068] The pseudo-random binary sequence mode for generating pseudo-thermal excitation signals includes generating a pseudo-random binary sequence based on a preset random seed, shaping the pseudo-random sequence according to a set symbol duration to obtain a continuous perturbation waveform, which serves as the pseudo-random binary excitation signal. The preset random seed is 202501, which corresponds to the starting point for initializing the pseudo-random binary sequence. The symbol duration is set to 100ms, which corresponds to each pseudo-random symbol being maintained for 100ms before being flipped or held, so that the generated pseudo-random binary excitation signal has sufficient bandwidth coverage and uniform energy distribution.

[0069] The generated pseudo thermal excitation signal is subjected to amplitude constraint and amplitude limit verification. The maximum power disturbance amplitude when applied to the heater power channel is no higher than 1% to 5% of the rated power, and the maximum duty cycle disturbance amplitude when applied to the fan pulse width modulation channel is no higher than 1% to 5% of the rated duty cycle. Slope limits are applied to the rising and falling edges.

[0070] The verified pseudo-thermal excitation signal is superimposed onto the heater power command or the fan pulse width modulation duty cycle command, and the superimposed object, start time, end time and excitation parameters are recorded.

[0071] Establish start-stop gating and read four monitoring quantities in real time: temperature, power, current and vibration. Stop the pseudo-thermal excitation signal when any of the monitoring quantities reaches the preset abnormal threshold. Allow the pseudo-thermal excitation signal to be restarted when the monitoring quantity returns to the normal range and meets the preset minimum interval time.

[0072] This invention achieves dual coverage of the frequency and time domain characteristics of thermal disturbances by introducing a dual-mode design of multi-frequency sinusoidal superposition mode and pseudo-random binary sequence mode in the pseudo-thermal excitation signal generation stage. The multi-frequency sinusoidal superposition mode can form a controllable spectrum distribution within the preset excitation frequency band, facilitating system identification and frequency domain characteristic extraction; while the pseudo-random binary sequence mode enhances the randomness and richness of the excitation signal through a reproducible disturbance generation mechanism based on a preset random seed, taking into account both the requirements of repeatable experiments and the diversity of excitation signals. This invention sets amplitude constraints and amplitude limiting verification during signal injection to ensure that the maximum power or duty cycle disturbance amplitude does not exceed 1% to 5% of the rated value, and limits the slope of the rising and falling edges, thereby effectively avoiding damage to the heater and fan actuators, and improving experimental safety and equipment reliability. By using a start-stop gating mechanism to monitor temperature, power, current, and vibration in real time, and promptly cutting off the excitation signal when an abnormality occurs, risks can be significantly reduced and operational stability improved. This not only enhances the controllability and diversity of pseudo-thermal excitation signals, but also ensures the safety of the experimental process and the accuracy of the system identification results, demonstrating significant engineering application value and promotional significance.

[0073] In this embodiment, the step of performing time-frequency analysis and cross-spectral calculation on standardized time-series data and pseudo-thermal excitation signals to obtain the thermal admittance spectrum at each temperature measurement point and extracting heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, effective admittance bandwidth, and equivalent heat dissipation emission coefficient, includes:

[0074] Based on the upper and lower limits of the frequency band, symbol boundaries, and start and end times of the pseudo-thermal excitation signal, the power time series and temperature measurement point time series in the standardized time series data are segmented into fixed lengths, the overlap rate is set, and multi-window weighting is applied to generate segmented windowed power sequences and temperature sequences. Time-frequency transformation is performed on each segment of the power sequence and temperature sequence to obtain a set of corresponding frequency points.

[0075] The power auto-power spectrum and temperature-power cross-power spectrum are calculated for the power and temperature sequences. A multi-window segmented averaging strategy is used to obtain the frequency domain spectral quantities, and a timestamp and segment identifier are added to each frequency domain spectral quantity. Specifically, the calculation of the power auto-power spectrum and temperature-power cross-power spectrum for the power and temperature sequences involves:

[0076] The standardized power sequence and the temperature measurement point sequence are segmented according to a fixed length and a preset overlap rate. A window function is applied to each segment and the corresponding timestamp and frequency resolution are recorded.

[0077] Frequency domain transformation is performed on each power and temperature sequence to obtain the complex amplitude value at each frequency point. The power self-power spectrum within the segment is obtained by using the power complex amplitude value and its conjugate. The temperature-power cross-power spectrum within the segment is obtained by using the temperature complex amplitude value and its conjugate with the corresponding power complex amplitude value. The consistency correction of amplitude and energy is then completed.

[0078] The spectral quantity of all segments is averaged and smoothed at each frequency point, and the frequency points are weighted according to the coherence with pseudo-thermal excitation and in-band occupancy; low coherence or noise-dominant frequency points are deweighted or shielded, and the corrected power auto-power spectrum and temperature-power cross-power spectrum are output.

[0079] The multi-window segmented averaging strategy refers to dividing the time-series data of power and temperature into several segments, applying multiple different window functions to each segment, averaging the spectral results obtained under each window function, and then fusing the average results of all segments to obtain the final spectral estimation result. The window function is a weighting function constructed within the segmented data range, formed by assigning different weights to each sampling point.

[0080] Based on the occupancy distribution and symbol boundary of the pseudo thermal excitation signal in the frequency band, weights are assigned to frequency points that are in the occupancy band and whose coherence with the excitation meets the threshold requirement, and frequency points that do not meet the requirement are set to zero, thus forming the power auto-power spectrum and temperature-power cross-power spectrum after coherent phase-locked processing.

[0081] For each temperature measurement point, the thermal admittance spectrum of each temperature measurement point is obtained based on the power auto-power spectrum and the temperature-power cross-power spectrum. Non-negativity constraints are applied to the real components of the thermal admittance spectrum, and frequency points that do not meet the constraints are masked. Specifically, obtaining the thermal admittance spectrum of each temperature measurement point based on the power auto-power spectrum and the temperature-power cross-power spectrum involves:

[0082] Power spectrum analysis is performed on the power time-series signal to obtain the power auto-power spectrum, which characterizes the energy distribution of the power signal in the frequency domain;

[0083] Joint spectral analysis of temperature and power time series signals yields the temperature-power cross-power spectrum, reflecting the correlation between the two at various frequency points.

[0084] Using the temperature-power cross-power spectrum as the numerator and the power self-power spectrum as the denominator, the ratio is calculated at each frequency point to obtain the corresponding thermal admittance spectrum, thus forming a description of the heat transfer characteristics of each temperature measurement point in the frequency domain.

[0085] Within the set of surface temperature measurement points, the thermal admittance spectra of each temperature measurement point are weighted and aggregated using coherence weights to obtain the surface composite thermal admittance spectrum. A robust quantile rule is then applied to suppress the contribution of outlier temperature measurement points. Specifically, the application of the robust quantile rule to suppress the contribution of outlier temperature measurement points involves:

[0086] The numerical distribution of thermal admittance spectra at the same frequency point in the statistical set of surface temperature measurement points;

[0087] In the numerical distribution of each frequency point, the quantile range is calculated to identify temperature measurement points with large deviations.

[0088] Temperature measurement points falling outside the quantile range are marked as outliers, and their weight contribution is reduced during the weighted aggregation process.

[0089] The dynamic heat dissipation efficiency is obtained by weighted summation of the real components of the surface synthetic thermal admittance spectrum within the excitation frequency band. The effective admittance bandwidth is defined as the continuous frequency span exceeding a fixed threshold. Within a time window synchronized with the pseudo-thermal excitation signal, the input power and the average surface temperature difference are bandpass processed, and least squares fitting is performed to obtain the equivalent heat dissipation emission coefficient. At the same time, corresponding confidence intervals are generated for the three heat dissipation efficiency indicators. The bandpass processing of the input power and the average surface temperature difference refers to the pre-setting of the upper and lower limits of the bandpass filter frequency band. The input power signal and the average surface temperature difference signal are respectively input to the bandpass filter. During the filtering process, only the frequency components within the preset frequency band are retained, DC and slowly varying components below the lower limit frequency are suppressed, and high-frequency noise components above the upper limit frequency are weakened to obtain the input power bandpass signal and the average surface temperature difference bandpass signal containing only the target frequency band components.

[0090] This invention employs a multi-window segmented averaging strategy and a coherence weighting method, combined with time-frequency analysis and cross-spectrum calculation driven by pseudo-thermal excitation signals. This enables the acquisition of high-resolution, low-bias power auto-power spectra and temperature-power cross-power spectra under noisy and dynamic disturbance environments, improving the accuracy and stability of frequency domain feature extraction. Through frequency domain point coherence screening and weighting, it accurately distinguishes physical responses strongly correlated with the excitation signal, shielding uncorrelated or noise-dominated frequencies and effectively suppressing interference from environmental noise and non-ideal equipment responses. Robust quantile rules are used to suppress the influence of outlier temperature measurement points during thermal admittance spectrum aggregation, improving the representativeness and anomaly resistance of the surface-synthesized thermal admittance spectrum. It innovatively integrates power spectrum analysis, cross-spectrum estimation, frequency domain weighting, and quantile robust processing, achieving accurate extraction and confidence interval generation of multi-dimensional indicators such as dynamic heat dissipation efficiency, effective admittance bandwidth, and equivalent heat dissipation emission coefficient. The overall scheme not only improves the sensitivity and anti-interference capability of heat dissipation efficiency assessment but also enhances the physical interpretability and practical engineering applicability of the assessment results.

[0091] In this embodiment, the process of constructing a thermal network graph containing nodes and edges based on heat dissipation efficiency indicators and thermal admittance spectra, combined with the structural topology information of the test chamber, and establishing a physical constraint twin model, followed by adaptive updates and structural optimization of the physical constraint twin model, includes:

[0092] Determine the input data, which includes thermal admittance spectra at each temperature measuring point, surface-synthesized thermal admittance spectra, dynamic heat dissipation efficiency, effective admittance bandwidth, equivalent heat dissipation emission coefficient, test chamber structural topology information, and standardized time series data.

[0093] A thermal network diagram is constructed based on the structural topology information of the test chamber. Temperature measurement points and heat dissipation-related parts inside and outside the test chamber are defined as nodes, and conduction and convection connections are defined as edges. Parameters are set for each edge and each node. The initial parameters are set based on the estimated values ​​of nominal material parameters, geometric dimensions and equivalent heat dissipation emission coefficient. The structural topology information of the test chamber refers to the formal description of the geometric configuration, component connection relationship and heat transfer coupling channel of the test chamber.

[0094] A physically constrained twin model is constructed based on a thermal network graph. Standardized time-series data, three heat dissipation efficiency indicators, and thermal admittance spectrum are used as inputs. The nodes and edges of the thermal network graph form the structural basis of the physically constrained twin model. Physical constraints are applied to the parameters during the training and inference process of the physically constrained twin model. These physical constraints include node energy balance constraints, parameter non-negativity constraints, and thermal admittance spectrum real part non-negativity constraints. Applying physical constraints to the parameters during the training and inference process of the physically constrained twin model refers to:

[0095] Node energy balance constraint: At every moment, the input energy and output energy of each node in the thermal network diagram must be equal to ensure that the node temperature rise calculation conforms to the energy conservation relationship;

[0096] Non-negativity constraint for parameters: During training and inference, the values ​​of thermal resistance, heat capacity and heat transfer coefficient of the heat dissipation path are restricted to be no less than zero to ensure the rationality and interpretability of the physical parameters;

[0097] Non-negativity constraint of the real part of thermal admittance spectrum: During frequency domain calculation, the real part of the thermal admittance spectrum is constrained to remain non-negative to avoid spurious responses that do not conform to the heat flow-temperature rise relationship;

[0098] The physical constraint twin model is updated, and the real-time data is incrementally calibrated using a sliding time window to reduce temperature prediction errors and energy budget imbalances. The parameters are range-projected according to the update cycle, and parameters that exceed the physical feasible range are adjusted to the allowable range. The updated parameter set is recorded.

[0099] Structural optimization is performed on the heat network graph. Based on the statistical results of the contribution and sensitivity of each edge within the long-term window, edges whose contribution is consistently below the threshold are removed. For node pairs with stable residual patterns and adjacent to the structural topology, new candidate edges are added and included in the heat network graph after physical feasibility verification. The parameters that have converged are no longer updated.

[0100] This invention achieves a deep integration of structural topology and time-series data-driven approaches by introducing a physically constrained twin model onto a thermal network graph, demonstrating significant innovation and practical effectiveness. Firstly, it uses multi-dimensional indicators such as thermal admittance spectrum, dynamic heat dissipation efficiency, and effective admittance bandwidth, along with standardized time-series data, as input. Training and inference are conducted through node energy balance constraints, parameter non-negativity constraints, and real part non-negativity constraints of the thermal admittance spectrum. This not only ensures the physical rationality of the model's predictions but also avoids the non-physical interpretations that may occur with existing purely data-driven methods. Secondly, it employs incremental calibration using a sliding time window, combined with a parameter range projection mechanism, enabling the model to adapt to real-time data, reducing temperature prediction errors and energy budget deviations, and improving dynamic adaptability. Thirdly, it introduces a structural optimization strategy based on contribution and sensitivity, eliminating inefficient edges and dynamically adding candidate edges according to residual patterns, thereby achieving the evolution and simplification of the thermal network graph, effectively avoiding overfitting and enhancing model interpretability. This invention can significantly improve the accuracy of heat dissipation efficiency assessment and prediction while maintaining physical consistency, forming a stable, reliable, and scalable thermal network twin framework.

[0101] In this embodiment, obtaining the predicted results of the temperature field and heat dissipation efficiency includes:

[0102] Load the optimized physical constraint twin model, heat network diagram, and parameter set;

[0103] From the standardized time series data, power time series, temperature measurement point time series, wind speed time series, ambient temperature and humidity time series, and command time series related to control variables are extracted by sliding time window and organized into an input data matrix corresponding to the nodes and edges of the heat network diagram.

[0104] The prediction step size, prediction time domain length, and update period are set. A rolling method is used to predict the temperature field for the next time step, outputting the predicted temperature value and timestamp for each node. Specifically, the rolling method for predicting the temperature field for the next time step involves:

[0105] Under the constraint of a predetermined prediction step size, the prediction result at the current moment is used as the starting condition for the prediction input at the next moment, so as to realize the step-by-step advancement of time series data;

[0106] Within the prediction time domain, the prediction input and output for each time step are generated sequentially and iteratively, so that the prediction process covers the complete time evolution interval.

[0107] In each update cycle, the latest predicted value and timestamp are used to replace the old input to maintain the continuity and consistency of the prediction sequence, forming a rolling prediction chain;

[0108] After each temperature field prediction is completed, based on three heat dissipation efficiency indices and the surface synthesized thermal admittance spectrum as auxiliary inputs, the physical constraint twin model generates the predicted heat dissipation efficiency for the corresponding time step. Specifically, the generation of the predicted heat dissipation efficiency for the corresponding time step by the physical constraint twin model is as follows:

[0109] The temperature field prediction results at the current time step are combined with three heat dissipation efficiency indicators and the surface synthesized thermal admittance spectrum to construct a complete input vector.

[0110] Under the constraints of energy balance, parameter non-negativity, and real part non-negativity of thermal admittance spectrum in the physical constraint twin model, the calculation and reasoning of heat dissipation path and node state are performed.

[0111] Generate prediction results including dynamic heat dissipation efficiency, effective admittance bandwidth and equivalent heat dissipation emission coefficient, and output them in conjunction with the predicted temperature value and timestamp of the corresponding time step;

[0112] Temperature field and heat dissipation efficiency are predicted in a loop for multiple consecutive time steps. The predicted results are then summarized in chronological order to form a complete temperature field prediction sequence and heat dissipation efficiency prediction sequence. The prediction sequences are then cached and labeled.

[0113] In this embodiment, the step of implementing minor intervention on the control variables in the test chamber, performing counterfactual reasoning based on the prediction results, calculating the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency index in conjunction with the heat dissipation efficiency index, and outputting optimization control instructions to determine the abnormal or degraded state of heat dissipation performance includes:

[0114] Select control variables in the test chamber, including fan speed and guide valve opening. Set upper limits for intervention amplitude and rate of change. The intervention amplitude of the fan and the guide valve shall not exceed 3% of the rated value, and the rate of change shall not exceed 2% per minute.

[0115] Within the running window, apply a small-amplitude step or gradual disturbance to a single control variable, record the start and end times, amplitude, target channel, and sequence number of the disturbance, and keep the other control variables unchanged;

[0116] Based on the predicted results of the temperature field and heat dissipation efficiency, counterfactual reasoning is performed for both disturbed and undisturbed states to obtain the efficiency change and temperature change within the same time window. Specifically, the counterfactual reasoning for both disturbed and undisturbed states involves:

[0117] Under the same input conditions, the temperature field and heat dissipation efficiency prediction results for the disturbed state and the undisturbed state are constructed respectively. The disturbed state is simulated by setting the change of specific edge or node parameters to simulate the external disturbance effect.

[0118] By comparing the predicted node temperature and heat dissipation efficiency values ​​within the same time window under both disturbed and undisturbed conditions, the difference between the temperature change and the efficiency change is calculated.

[0119] The changes in efficiency and temperature are output as counterfactual inference results and associated with timestamps.

[0120] Combining three heat dissipation efficiency indicators—dynamic heat dissipation efficiency, effective admittance bandwidth, and equivalent heat dissipation emission coefficient—the local causal efficiency sensitivity of each control variable to dynamic heat dissipation efficiency is calculated, and the effective improvement interval and ineffective interval are determined. Specifically, the calculation of the local causal efficiency sensitivity of each control variable to dynamic heat dissipation efficiency is as follows:

[0121] While keeping other variables constant, a small perturbation is applied to a single control variable to generate a prediction result of dynamic heat dissipation efficiency under the perturbation state;

[0122] The dynamic heat dissipation efficiency results under disturbed and undisturbed states are compared, and the difference between the two is calculated and normalized to the local causal efficiency sensitivity of the control variable.

[0123] The local causal efficiency sensitivity of each control variable is statistically analyzed, the changing trend in different value ranges is identified, and the effective improvement range and ineffective range are divided.

[0124] Based on the local causal efficiency sensitivity and the set constraints of energy consumption, noise and temperature fluctuation, an optimized control command is generated. The optimized control command includes the target increment, execution time, rise and fall slope limits and effective time. Based on the threshold and trend of the three heat dissipation efficiency indicators, the abnormal or degraded state of heat dissipation performance is determined, alarm information and command priority are output, and the optimized control command is sent to the control device of the test chamber.

[0125] This invention, by introducing refined perturbation settings and counterfactual reasoning mechanisms for control variables into the test chamber, can accurately quantify the causal effects of control variables such as fan speed and deflector valve opening on dynamic heat dissipation efficiency without relying on large-scale actual intervention. By setting upper limits for perturbation amplitude and rate of change, the safety and controllability of the experimental process are ensured, avoiding the risk of equipment overload caused by large-scale adjustments in traditional methods. Furthermore, this invention combines the counterfactual reasoning results with three heat dissipation efficiency indicators, innovatively proposing a method for calculating local causal efficiency sensitivity, which can accurately identify effective and ineffective improvement intervals, achieving high-resolution causal attribution analysis of control variables. The optimized control instructions generated based on the sensitivity results not only include target increment, duration, and slope limits, but also comprehensively consider energy consumption, noise, and temperature fluctuation constraints, thereby maintaining system stability and energy efficiency while improving heat dissipation performance. Under the premise of ensuring safe operation, this invention improves the dynamic control capability of the test chamber's heat dissipation efficiency and can detect abnormal or degraded states of heat dissipation performance in real time, achieving both early warning and adaptive optimization effects.

[0126] Example 1:

[0127] To verify the feasibility of this invention in practice, it was applied to the heat dissipation performance evaluation scenario of an IoT environmental monitoring laboratory built by a research institute. The test chamber was a high power density power supply chamber with built-in multi-point temperature sensors, air flow rate sensors and humidity sensors. It was also equipped with an adjustable-speed cooling fan and a programmable electric heating module to simulate the heat dissipation load and environmental disturbances under actual working conditions.

[0128] In the experimental scenario, multiple types of sensors were first deployed to collect real-time temperature, humidity, and current load curves inside and outside the cabin. After standardized preprocessing, these data were converted into multi-source heterogeneous data streams and used as input to the constructed physical constraint twin model. The physical constraint twin model established a thermal network graph based on a graph structure, and combined with physical boundary conditions, modeled and optimized the heat dissipation path. A deep learning module was used to dynamically predict heat dissipation efficiency indicators, and further, a counterfactual reasoning mechanism was incorporated to evaluate the local causal efficiency sensitivity of different control variables (such as fan speed, electric heating power, and inlet valve opening).

[0129] In practical applications, by implementing small-scale interventions (within ±5%) in the fan speed, the system can automatically determine the direction and magnitude of the intervention's impact on heat dissipation efficiency. When the heat dissipation efficiency output by the predictive model is lower than a preset threshold, the system generates optimization control commands, automatically increasing the fan speed and adjusting the air inlet opening, thereby quickly suppressing the abnormal temperature rise trend inside the test chamber. Experiments show that, under the same load power conditions, compared with the traditional fixed threshold adjustment method, the method of this invention can detect signs of heat dissipation efficiency degradation more quickly, effectively reduce the overshoot of the chamber temperature, and improve system stability.

[0130] Table 1. Comparison Test Results of Heat Dissipation Efficiency of IoT Test Chambers

[0131] Test time (min) Ambient temperature (°C) Load power (kW) Fan speed (rpm) Traditional method chamber temperature (°C) The chamber temperature (°C) of the method of this invention Improved heat dissipation efficiency (%) 10 30.1 2.5 1800 38.2 36.9 3.4 20 30.3 2.5 1800 44.7 42.1 5.8 30 30.5 2.5 1850 49.2 46.0 6.5 40 30.6 2.5 1850 53.5 49.8 6.9 50 30.7 2.5 1900 56.8 51.7 9.0 60 30.8 2.5 1900 57.6 52.3 9.2

[0132] Based on the data in Table 1, it is clear that the method of the present invention exhibits significantly better heat dissipation performance than the traditional method at different test time points. During the experiment, the ambient temperature was maintained at around 30℃, the load power was kept constant at 2.5kW, and the fan speed was gradually increased from 1800rpm to 1900rpm during the experiment according to the temperature rise to adapt to changes in heat load.

[0133] In the first 10 minutes, the temperature difference between the two methods was small, but the temperature of the chamber using the present invention was already 1.3℃ lower than that of the traditional method, with a 3.4% improvement in heat dissipation efficiency. As the test time progressed, the temperature difference gradually widened. By the 20-minute mark, the chamber temperature under the traditional method rose to 44.7℃, while the temperature under the present invention was controlled at 42.1℃, resulting in a 5.8% improvement in efficiency. After 30 minutes, with a slight increase in fan speed, the chamber temperature under the present invention was controlled at 46.0℃, 3.2℃ lower than that of the traditional method, with a 6.5% improvement in heat dissipation efficiency. Between 40 and 50 minutes, as the heat load continued to accumulate, the effect of the present invention in suppressing the temperature rise of the chamber became more prominent, with the temperature difference maintained between 3.7℃ and 5.1℃, and the efficiency improvement reaching 6.9% to 9.0%.

[0134] By the 60th minute, the high-temperature phase of the experiment, the chamber temperature reached a maximum of 57.6℃ under the traditional method, while the temperature using the method of this invention was only 52.3℃, a temperature reduction of 5.3℃ and a 9.2% improvement in heat dissipation efficiency. Throughout the entire test cycle, the method of this invention was able to track the temperature rise trend well and make timely dynamic interventions, achieving effective control of the chamber temperature under high-load conditions.

[0135] The heat dissipation efficiency evaluation method based on big data analysis proposed in this invention can not only identify the changing trend of heat dissipation performance in advance, but also actively intervene through intelligent control strategies, significantly improving heat dissipation efficiency and system operation safety, and has good engineering practical value.

[0136] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating the heat dissipation efficiency of an Internet of Things test box based on big data analysis, characterized in that, The method comprises the following steps: Collecting multi-source data of each temperature measuring point in and outside the test box by multi-type sensors deployed through the Internet of Things, and pre-processing the multi-source data to obtain standardized time series data; Based on the standardized time series data, a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value is generated and superimposed on the heater or fan channel, the parameters of the pseudo-thermal excitation signal are recorded, and the start and stop of the pseudo-thermal excitation signal are controlled according to the real-time detected abnormal state; Performing time-frequency analysis and cross-spectrum calculation on the standardized time series data and the pseudo-thermal excitation signal to obtain the thermal admittance spectrum of each temperature measuring point, and extracting the heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, admittance effective bandwidth and equivalent heat dissipation emission coefficient; Based on the heat dissipation efficiency indicators and the thermal admittance spectrum, the structure topology information of the test box is combined to construct a thermal network graph containing nodes and edges, and a physically constrained twin model is established, and the physically constrained twin model is adaptively updated and optimized; Based on the optimized physically constrained twin model, the standardized time series data is dynamically predicted to obtain the prediction results of the temperature field and the heat dissipation efficiency; During the operation of the test box, small-amplitude intervention is performed on the control variables in the test box, counterfactual reasoning is performed based on the prediction results, the heat dissipation efficiency indicators are combined to calculate the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency indicators, and optimization control instructions are output, the abnormal or degraded state of the heat dissipation performance is judged, and the optimization control instructions are sent to the control device of the test box. 2.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, characterized in that, The multi-source data specifically includes temperature, wind speed, power, environment temperature, environment humidity, control component running state data and sensor data related to heat dissipation efficiency inside and outside the test box. 3.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, characterized in that, The pre-processing of the multi-source data specifically includes time synchronization, outlier rejection, missing data interpolation and normalization processing of the multi-source data. 4.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, wherein, Based on the standardized time series data, a pseudo-thermal excitation signal with an amplitude of 1% to 5% of the rated value is generated and superimposed on the heater or fan channel, the parameters of the pseudo-thermal excitation signal are recorded, and the start and stop of the pseudo-thermal excitation signal are controlled according to the real-time detected abnormal state, including: Determining excitation parameters based on the standardized time series data, the excitation parameters including upper and lower limits of the excitation frequency band, sampling period, duration, amplitude ratio, random seed, symbol duration and initial phase of each component; Selecting an excitation mode to generate a pseudo-thermal excitation signal, the excitation mode including a multi-frequency sine superposition mode and a pseudo-random binary sequence mode, in the multi-frequency sine superposition mode, a plurality of discrete frequency points are selected within the excitation frequency band, and the amplitude and initial phase of each frequency component are set respectively to form a multi-frequency sine excitation signal, in the pseudo-random binary sequence mode, a pseudo-random sequence is generated based on a preset random seed, and the sequence is shaped according to a set symbol duration to obtain a corresponding disturbance waveform; Performing amplitude constraint and amplitude limiting verification on the generated pseudo-thermal excitation signal, the maximum power disturbance amplitude acting on the heater power channel is not higher than 1% to 5% of the rated power, the maximum duty cycle disturbance amplitude acting on the fan pulse width modulation channel is not higher than 1% to 5% of the rated duty cycle, and the rising and falling edges are subjected to slope limitation; Superimpose the pseudo-thermal excitation signal that passes the verification to the heater power command or the fan pulse width modulation duty cycle command, record the superimposition object, start time, end time and excitation parameters; Establish start-stop door control, real-time read temperature, power, current and vibration four monitoring quantities, stop the pseudo-thermal excitation signal when any monitoring quantity reaches the preset abnormal threshold, and allow to restart the pseudo-thermal excitation signal when the monitoring quantity recovers to the normal interval and meets the preset minimum interval time. 5.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, wherein, The time-frequency analysis and cross-spectrum calculation of the standardized time series data and the pseudo-thermal excitation signal obtain the thermal admittance spectrum of each temperature measurement point, and extract the heat dissipation efficiency indicators, including dynamic heat dissipation efficiency, admittance effective bandwidth and equivalent heat dissipation discharge coefficient, including: According to the upper and lower limits of the frequency band of the pseudo-thermal excitation signal, the symbol boundary and the start and end time, the power time sequence and the temperature measurement point time sequence in the standardized time series data are segmented by fixed length, the overlap rate is set and the multi-window weighting is applied, the segmented and windowed power sequence and temperature sequence are generated, and the time-frequency transformation of each segment of the power sequence and the temperature sequence is performed to obtain a group of corresponding frequency points; The power self-power spectrum and temperature-power cross-power spectrum of the power sequence and the temperature sequence are calculated, the multi-window segmented average strategy is adopted to obtain the frequency domain spectrum, and a timestamp and a segment identifier are attached to each frequency domain spectrum; According to the occupation distribution and symbol boundary of the pseudo-thermal excitation signal in the frequency band, the frequency points that are in the occupied band and meet the threshold requirement of coherence with the excitation are assigned weights, and the frequency points that do not meet the requirement are set to zero, forming the power self-power spectrum and temperature-power cross-power spectrum after coherent phase-locked processing; For each temperature measurement point, the thermal admittance spectrum of each temperature measurement point is obtained based on the power self-power spectrum and the temperature-power cross-power spectrum, the real component of the thermal admittance spectrum is subjected to non-negativity constraint, and the frequency points that do not meet the constraint are shielded; In the set of surface temperature measurement points, the thermal admittance spectrum of each temperature measurement point is weighted and aggregated with the coherence weight to obtain the surface synthetic thermal admittance spectrum, and the contribution of the outlier temperature measurement point is suppressed by using the robust quantile rule; In the excitation frequency band, the real component of the surface synthetic thermal admittance spectrum is weighted and accumulated to obtain the dynamic heat dissipation efficiency, the continuous frequency span that exceeds the fixed threshold is taken as the admittance effective bandwidth, and in the time window synchronized with the pseudo-thermal excitation signal, the input power and the surface average temperature difference are band-pass processed, and the least square fitting is performed to obtain the equivalent heat dissipation discharge coefficient, and the corresponding confidence interval is generated for the three heat dissipation efficiency indicators. 6.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, wherein, The heat network graph containing nodes and edges is constructed based on the heat dissipation efficiency indicators and the thermal admittance spectrum, combined with the structure topology information of the test box, and the physical constraint twin model is established, and the physical constraint twin model is adaptively updated and structurally optimized, including: Determine the input data, including the thermal admittance spectrum of each temperature measurement point, the surface synthetic thermal admittance spectrum, the dynamic heat dissipation efficiency, the admittance effective bandwidth, the equivalent heat dissipation discharge coefficient, the test box structure topology information and the standardized time series data; constructing a thermal network graph based on the structural topology information of the test chamber, defining temperature measuring points inside and outside the test chamber and heat dissipation related parts as nodes, defining conduction and convection connections as edges, and setting parameters for each edge and each node; constructing a physical constraint twin model based on the thermal network graph, taking standardized time series data, three heat dissipation efficiency indicators and thermal admittance spectrum as inputs, taking nodes and edges of the thermal network graph as a structural basis of the physical constraint twin model, and applying physical constraints including node energy balance constraints, non-negative parameter constraints and non-negative real part of thermal admittance spectrum constraints during training and reasoning of the physical constraint twin model; updating the physical constraint twin model, incrementally calibrating real-time data using a sliding time window to reduce temperature prediction error and energy imbalance, projecting parameters within a range according to an update period, adjusting parameters outside the physically feasible range to the allowed interval, and recording the updated parameter set; performing structural optimization on the thermal network graph, removing edges with a contribution degree continuously below a threshold value based on statistical results of contribution degree and sensitivity of each edge within a long-term window, adding new candidate edges to nodes adjacent to the structural topology based on stable residual patterns, and incorporating the new candidate edges into the thermal network graph after physical feasibility verification, and stopping updating parameters that have converged. 7.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, wherein, The obtained prediction results of the temperature field and the heat dissipation efficiency include: loading the optimized physical constraint twin model, the thermal network graph and the parameter set; extracting power time series, temperature measuring point time series, wind speed time series, environmental temperature and humidity time series and instruction time series related to control variables from standardized time series data according to a sliding time window, and arranging them into input data matrices corresponding to nodes and edges of the thermal network graph; setting a prediction step, a prediction time domain length and an update period, and predicting the temperature field at the next time step in a rolling manner, and outputting temperature prediction values and time stamps of each node; after completing each temperature field prediction, generating prediction results of the heat dissipation efficiency at the corresponding time step based on the three heat dissipation efficiency indicators and the surface synthetic thermal admittance spectrum as auxiliary inputs from the physical constraint twin model; performing temperature field prediction and heat dissipation efficiency prediction for a plurality of consecutive time steps, arranging the obtained prediction results of the temperature field and the heat dissipation efficiency in chronological order to form complete temperature field prediction sequences and heat dissipation efficiency prediction sequences, and caching and marking the prediction sequences. 8.The big data analysis based evaluation method of heat dissipation efficiency of an Internet of Things test box according to claim 1, wherein, The small-amplitude intervention on the control variables in the test chamber is based on the prediction results, and the counterfactual reasoning is combined with the heat dissipation efficiency indicators to calculate the local causal efficiency sensitivity of the control variables to the dynamic heat dissipation efficiency indicators, and output optimization control instructions to determine the abnormal or degraded state of the heat dissipation performance, including: selecting control variables in the test chamber, including fan speed and flow valve opening, setting an upper limit of the intervention amplitude and an upper limit of the change rate, and the intervention amplitude of the fan and the flow valve does not exceed 3% of the rated value, and the change rate does not exceed 2% per minute; performing small-amplitude step or slow-changing disturbance on a single control variable within a running window, recording the start and end time, amplitude, target channel and sequence number of the disturbance, and keeping the remaining control variables unchanged; Based on the prediction results of temperature field and heat dissipation efficiency, counterfactual reasoning is performed on the two states of existence disturbance and non-disturbance to obtain the efficiency change and temperature change in the same time window; Combined with the dynamic heat dissipation efficiency, the admittance effective bandwidth and the equivalent heat dissipation emission coefficient, the local causal efficiency sensitivity of each control variable to the dynamic heat dissipation efficiency is calculated, and the effective promotion interval and the invalid interval are determined; According to the local causal efficiency sensitivity and the set energy consumption, noise and temperature fluctuation constraints, the optimization control instruction is generated, the heat dissipation performance abnormality or degradation state is judged according to the threshold and trend of the three heat dissipation efficiency indexes, the alarm information and instruction priority are output, and the optimization control instruction is sent to the control device of the test box.

Images