Consistency temperature interval estimation method based on temperature difference grading

Abstract

The invention belongs to the technical field of engineering data processing, and particularly relates to a consistency temperature interval estimation method based on temperature difference grading. According to the method, collaborative optimization of interval estimation precision and cross-domain consistency can be realized, dependence on artificial experience parameter adjustment and equipment specific calibration is reduced, and the method has good expandability and interpretability; meanwhile, the method can effectively reduce the operation and maintenance cost of power equipment and improve the operation safety of a power grid by improving the grading accuracy and consistency, and has remarkable technical and economic values.

Description

Technical Field

[0001] This invention belongs to the field of engineering data processing technology, and specifically relates to a method for estimating consistent temperature ranges based on temperature difference grading. Background Technology

[0002] Infrared thermography-based grading and assessment of power equipment is a key technology for ensuring the safe operation of the power grid. Its core requirement is to accurately grade equipment defects by using the temperature difference (ΔT) between hot spots and reference zones. Currently, there are two main technical approaches in this field: one is a point estimation grading method based on fixed thresholds, and the other is an interval estimation method based on statistical models.

[0003] While point estimation methods are simple to calculate and intuitive to make decisions, they are prone to misjudgment when measurement noise is high or the ΔT value is close to the classification threshold. While interval estimation methods can quantify measurement uncertainty, they rely heavily on strong assumptions such as Gaussian distribution and homoscedasticity, making them less adaptable to complex engineering environments.

[0004] To address measurement uncertainty, existing technologies often employ a hybrid approach combining multi-source calibration and parametric modeling. This involves reducing systematic biases through preprocessing such as distance compensation and angle correction, followed by constructing confidence intervals based on a fixed distribution assumption. However, these methods lack adaptability to variations in data distribution. Modeling requirements differ significantly across scenarios, including short-distance high-precision measurements, long-distance inspections, and environments with strong electromagnetic interference, making it difficult for fixed parametric assumptions to comprehensively cover all scenarios.

[0005] Regarding cross-device consistency, different infrared thermal imagers exhibit variations in performance parameters such as noise equivalent temperature difference and response nonlinearity. Changes in measurement distance and incident angle lead to different temperature mixing effects within the instantaneous field of view. Environmental factors such as atmospheric transmittance and background radiation further exacerbate the systematic bias in ΔT measurements. Existing correction methods mostly rely on simple linear transformations or lookup table compensation, lacking a systematic calibration within a unified probabilistic framework, which can easily result in overcompensation or undercompensation.

[0006] Traditional confidence interval construction methods often violate the strict assumptions about data distribution in practical applications: non-Gaussian distributions, heavy-tailed noise, outliers, and heteroscedasticity caused by changes in distance and angle render the normality and homoscedasticity assumptions invalid; the linear model assumption also fails to capture complex nonlinear dependencies. When equipment aging, seasonal changes, load fluctuations, and other factors cause shifts in the data distribution domain, the actual coverage of traditional methods will deviate significantly from the nominal confidence level.

[0007] In handling grade boundaries, existing methods suffer from decision instability. When the ΔT value approaches the grade threshold, small measurement fluctuations can lead to frequent grade jumps. When the confidence interval spans multiple grade boundaries, there is a lack of effective conflict resolution mechanisms, making it difficult to provide stable and interpretable judgment results. Furthermore, existing uncertainty analysis methods suffer from component fragmentation and incomplete propagation. Multiple sources of uncertainty, such as instrument noise and spatial solution mixing errors, are not co-modeled, and the fusion mechanism between measurement uncertainty and model uncertainty is inadequate.

[0008] The aforementioned problems make it difficult for existing infrared temperature measurement grading methods to provide reliable uncertainty quantification while ensuring accuracy. They also lack consistency and stability in complex engineering environments, failing to fully meet the high-standard requirements of power equipment operation and maintenance, quality inspection, and other scenarios. Summary of the Invention

[0009] The present invention aims to solve the technical problem of insufficient accuracy in defect classification in the existing infrared thermometry of power equipment due to measurement noise, data distribution differences, heteroscedasticity characteristics and unstable handling of classification threshold boundaries.

[0010] To address the aforementioned technical problems, this invention provides a method for estimating a consistent temperature range based on temperature difference grading, comprising the following steps: S1. Obtain the hot spot temperature and reference area temperature of the power equipment by infrared thermometry, calculate the original temperature difference based on the hot spot temperature and the reference area temperature, and correct the original temperature difference based on the infrared thermometry acquisition parameters to obtain a standardized temperature difference. S2. Using the collected parameters and the standardized temperature difference as raw data, training data is extracted from the raw data and used as input to the quantile regression model. The quantile estimation boundary of the standardized temperature difference is calculated through the quantile regression model. S3. Extract calibration data from the original data, calculate the sample inconsistency score based on the calibration data, determine the corresponding calibration amount, and construct the basic prediction interval based on the calibration amount and the quantile estimation boundary. S4. The calibration data is grouped according to different acquisition parameters to obtain multiple sets of calibration group data. The sample inconsistency score is calculated for each set of calibration group data, and the corresponding group calibration amount is determined. The basic prediction interval is updated according to the group calibration amount and the quantile estimation boundary to obtain the updated prediction interval corresponding to the calibration group data. S5. The model uncertainty and measurement uncertainty are used as variances to calculate the safety inflation amount, and the basic prediction interval is expanded according to the safety inflation amount to obtain the expansion prediction interval; S6. Determine the preset power equipment defect classification threshold, and determine the overlap between the preset power equipment defect classification threshold and the expansion prediction interval. Output the defect level determination result containing the interval with the same level based on the overlap. S7. Obtain the standard risk cost matrix, divide the verification data from the original data, and calculate the optimal confidence parameter based on the verification data and the standard risk cost matrix. Output the defect level determination result and the optimal confidence parameter as the consistency temperature range estimation result.

[0011] Furthermore, step S1 includes the following sub-steps: The hotspot temperature and the reference area temperature of the power equipment infrared thermometry are obtained, wherein the hotspot temperature is defined as... The reference region temperature is defined as

[0012] Calculate the original temperature difference, and define it as... And satisfy: ; Determine the acquisition parameters and calculate the standardized temperature difference value, defining the standardized temperature difference value as... It satisfies: ; in To measure distance, Angle of incidence For equipment identification, For scaling correction function, For equipment bias, , , , , Let the collected parameters be: ; ; (i=0,1,2,3) are the distance and angle correction coefficients. This represents the relative scale factor of the equipment. For device system bias, parameters It is learned through the optimization objective of minimizing the cross-domain residual variance, and: ; The standardized temperature difference of sample i, and Representing the same physical quantity, For the corresponding domain The mean within.

[0013] Furthermore, step S2 includes the following sub-steps: The acquisition parameters are defined as follows: The quantile regression model is trained using the Pinball loss function: ; ; in Let quantile regression be the total loss function. and These correspond to the quantile levels. and The asymmetric loss function, For confidence level parameters, Let be the covariate vector of the i-th sample; With the acquisition parameters and the standardized temperature difference value The standardized temperature difference is calculated as the input to the quantile regression model. The quantile estimation boundary ,in Represents a given covariate The standardized temperature difference under the conditions The Quantile estimates Represents the standardized temperature difference value The Quantile estimates.

[0014] Furthermore, step S3 includes the following sub-steps: The calibration data is extracted from the raw data. In the calibration data The CQR method is used to calculate the corresponding sample inconsistency score, and this process satisfies the following relationship:

[0015] in For calibration data The sample inconsistency score of the i-th calibration sample in the dataset represents the actual value of that sample. The extent to which it exceeds the quantile prediction interval; The calibration quantity is used to calculate the number of inconsistent samples. The process satisfies the following relationship:

[0016] in The empirical quantile function is used to calculate the set of inconsistent scores for all said samples from the calibration set. Extract the first Percentage value; Construct for any new sample Under the assumption of no distribution The basic prediction interval satisfies:

[0017] in This is the lower bound of the basic prediction interval. This is the upper bound of the basic prediction interval.

[0018] Furthermore, step S4 includes the following sub-steps: The calibration data are grouped according to different acquisition parameters x using a grouping function. The data is divided into M calibration groups, and the corresponding calibration amount for each group is calculated. The process satisfies the following relationship: ; The basic prediction interval is updated based on the grouped calibration pairs and the quantile estimation boundaries to obtain the updated prediction interval corresponding to the calibration group data.

[0019] Furthermore, step S5 includes the following sub-steps: Calculate the total variance based on the model uncertainty and the measurement uncertainty. The process satisfies the following relationship: ; in The model predicts variance. For instrument noise variance, Solve for the mixed variance of IFOV. Choose the variance for the reference region; Total variance Standard deviation Convert to safe expansion amount ,in As a conservative coefficient, the basic prediction interval is expanded to obtain the expanded prediction interval. This process satisfies the following relationship: ; in This is the lower bound of the expansion prediction interval. This is the upper bound of the expansion prediction interval.

[0020] Furthermore, step S6 includes the following sub-steps: The preset power equipment defect classification threshold is determined, and the classification threshold boundary is defined as follows: Then the intervals of consistent levels satisfy the following relationship: ; in For cost-sensitive contraction functions; Calculate the overlap ratio between the expansion prediction interval and the level-consistent interval, and combine it with the quantile distribution output by the base learner to determine the defect level determination result using Bayesian posterior probability estimation: ; in This represents the k-th level label. Let be a conditional probability function. and The interval boundaries are defined by the consistent projection of the levels. This refers to the defect level determination result.

[0021] Furthermore, step S7 includes the following sub-steps: Determining the cost of false alarms based on power equipment standards Cost of underreporting Confusion Cost Matrix with Rank These are combined to form the standard risk cost matrix; The validation data is extracted from the original data, and the optimal confidence parameter is calculated based on the validation data and the standard risk cost matrix. This process satisfies the following relationship: ; in This is the optimal confidence parameter. The true level label in the verification data. This represents the expectation based on the verification data; The defect level determination result and the optimal confidence parameter The output is the estimated result of the consistent temperature range.

[0022] The beneficial effects achieved by this invention are that it proposes a consistent temperature range estimation method based on temperature difference grading. This method can achieve synergistic optimization of range estimation accuracy and cross-domain consistency, reduce dependence on manual experience parameter tuning and equipment-specific calibration, and has good scalability and interpretability. At the same time, by improving the accuracy and consistency of grading, this method can effectively reduce the operation and maintenance costs of power equipment and improve the safety of power grid operation, which has significant technical and economic value. Attached Figure Description

[0023] The present invention will now be described in detail with reference to the accompanying drawings. The above and other aspects of the present invention will become clearer and more readily understood through the detailed description following the accompanying drawings. In the drawings: Figure 1 This is a flowchart of the steps of the consistent temperature range estimation method based on temperature difference grading provided in the embodiments of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of the consistent temperature range estimation method based on temperature difference grading provided in this embodiment of the invention. The consistent temperature range estimation method based on temperature difference grading includes the following steps: S1. Obtain the hot spot temperature and reference area temperature of the power equipment by infrared thermometry, calculate the original temperature difference based on the hot spot temperature and the reference area temperature, and correct the original temperature difference based on the infrared thermometry acquisition parameters to obtain a standardized temperature difference.

[0026] Step S1 is used to construct a standardized target quantity for temperature difference to address systematic deviations across devices, distances, and environmental conditions. Step S1 includes the following sub-steps: The hotspot temperature and the reference area temperature of the power equipment infrared thermometry are obtained, wherein the hotspot temperature is defined as... The reference region temperature is defined as

[0027] Calculate the original temperature difference, and define it as... And satisfy: ; Determine the acquisition parameters and calculate the standardized temperature difference value, defining the standardized temperature difference value as... It satisfies: ; in To measure distance, Angle of incidence For equipment identification, For scaling correction function, For equipment bias, , , , , Let the collected parameters be: ; ; (i=0,1,2,3) are the distance and angle correction coefficients. This represents the relative scale factor of the equipment. For device system bias, parameters It is learned through the optimization objective of minimizing the cross-domain residual variance, and: ; The standardized temperature difference of sample i, and Representing the same physical quantity, For the corresponding domain The mean within.

[0028] Scale correction function The types of acquired parameters are determined based on their type, including distance, angle, device scale correction functions, and device bias functions. These parameters are learned through a criterion of minimizing cross-domain residual variance. Step S1, by constructing standardized temperature difference values ​​and device correction functions, enables consistent calibration of measurement results across devices, distances, and environmental conditions. This effectively reduces the variability in temperature difference measurements across different infrared cameras, measurement distances, and environmental conditions, making the grading thresholds comparable between different stations and providing a technical foundation for standardized temperature measurement assessment.

[0029] S2. Using the collected parameters and the standardized temperature difference as raw data, training data is extracted from the raw data and used as input to the quantile regression model. The quantile estimation boundary of the standardized temperature difference is calculated through the quantile regression model.

[0030] In this embodiment of the invention, the acquisition parameters include distance. Angle of incidence Environmental quantity Device ID, etc. Step S2 includes the following sub-steps: The acquisition parameters are defined as follows: The quantile regression model is trained using the Pinball loss function: ; ; in Let quantile regression be the total loss function. and These correspond to the quantile levels. and The asymmetric loss function, For confidence level parameters, Let be the covariate vector of the i-th sample; With the acquisition parameters and the standardized temperature difference value The standardized temperature difference is calculated as the input to the quantile regression model. The quantile estimation boundary ,in Represents a given covariate The standardized temperature difference under the conditions The Quantile estimates Represents the standardized temperature difference value The Quantile estimates.

[0031] In an alternative implementation, the quantile regression model can be implemented using linear regression, decision trees, or neural networks to overcome the distribution assumptions imposed by traditional parameterization methods.

[0032] S3. Extract calibration data from the original data, calculate the sample inconsistency score based on the calibration data, determine the corresponding calibration amount, and construct the basic prediction interval based on the calibration amount and the quantile estimation boundary.

[0033] Step S3 includes the following sub-steps: The calibration data is extracted from the raw data. In the calibration data The CQR (Conformalized Quantile Regression) method is used to calculate the corresponding sample inconsistency score, and this process satisfies the following relationship:

[0034] in For calibration data The sample inconsistency score of the i-th calibration sample in the dataset represents the actual value of that sample. The extent to which it exceeds the quantile prediction interval; The calibration quantity is used to calculate the number of inconsistent samples. The process satisfies the following relationship:

[0035] in The empirical quantile function is used to calculate the set of inconsistent scores for all said samples from the calibration set. Extract the first Percentage value; Construct for any new sample Under the assumption of no distribution The basic prediction interval satisfies:

[0036] in This is the lower bound of the basic prediction interval. This is the upper bound of the basic prediction interval.

[0037] In step S3, CQR is an uncertainty quantification method that integrates quantile regression (QR) and conformal prediction (CP). It aims to address the shortcomings of traditional prediction intervals, such as insufficient adaptability or unreliable coverage, ultimately generating prediction intervals that combine locally adaptive interval width with strict statistical coverage guarantees. Step S3 ensures that the method proposed in this embodiment can adaptively construct distribution-independent prediction intervals, eliminating the dependence of traditional parameterization methods on strong assumptions such as Gaussian distribution and homoscedasticity, thus improving robustness in complex engineering environments. In other optional implementations, the conformal interval construction required in step S3 can also be achieved using the Jackknife+ or CV+ methods.

[0038] S4. The calibration data is grouped according to different acquisition parameters to obtain multiple sets of calibration group data. The sample inconsistency score is calculated for each set of calibration group data, and the corresponding group calibration amount is determined. The basic prediction interval is updated according to the group calibration amount and the quantile estimation boundary to obtain the updated prediction interval corresponding to the calibration group data.

[0039] To maintain coverage under conditions of device grouping, distance tiering, and environmental drift, Mondrian-conditional conformal prediction is introduced in step S4. Step S4 includes the following sub-steps: The calibration data are grouped according to different acquisition parameters x using a grouping function. The data is divided into M calibration groups, and the corresponding calibration amount for each group is calculated. The process satisfies the following relationship: ; The basic prediction interval is updated based on the grouped calibration pairs and the quantile estimation boundaries to obtain the updated prediction interval corresponding to the calibration group data.

[0040] When predicting new samples, the group to which the sample belongs is used. Select the corresponding When there is covariate distribution drift (such as changes in the covariate distribution between test data and calibration data, such as data distribution domain shift caused by equipment aging, seasonal changes, etc.), a weighted conformal prediction method is used: Estimated density ratio (in To test the set covariate distribution density, To calibrate the set of covariate distribution density, density ratios (which can be estimated via kernel mean matching or adversarial estimation) are used, weighted quantiles are employed. Perform calibration.

[0041] In step S4, Mondrian-conditional conformal prediction avoids global recalibration, requiring only local updates when grouping conditions change, significantly reducing computational complexity. The distribution-independent coverage guarantee provided by conformal prediction theory enables the method proposed in this embodiment to maintain stable performance even when facing practical engineering problems such as data distribution changes, equipment aging, and environmental drift.

[0042] S5. The model uncertainty and measurement uncertainty are used as variances to calculate the safety expansion amount, and the basic prediction interval is expanded according to the safety expansion amount to obtain the expansion prediction interval.

[0043] Step S5 includes the following sub-steps: Calculate the total variance based on the model uncertainty and the measurement uncertainty. The process satisfies the following relationship: ; in The model predicts variance. For instrument noise variance, Solve for the mixed variance of IFOV. Choose the variance for the reference region; Total variance Standard deviation Convert to safe expansion amount ,in As a conservative coefficient, the basic prediction interval is expanded to obtain the expanded prediction interval. This process satisfies the following relationship: ; in This is the lower bound of the expansion prediction interval. This is the upper bound of the expansion prediction interval.

[0044] S6. Determine the preset power equipment defect classification threshold, and determine the overlap between the preset power equipment defect classification threshold and the expansion prediction interval. Based on the overlap, output the defect level determination result that includes the interval with the same level.

[0045] Step S6 is used to approximate the probability by the overlap ratio between the interval and each level interval or the quantile distribution of the base learner. Step S6 includes the following sub-steps: The preset power equipment defect classification threshold is determined. The classification threshold is set to avoid decision instability caused by intervals crossing multiple classification boundaries. The classification threshold boundary is defined as follows: Then the intervals of consistent levels satisfy the following relationship: ; in For cost-sensitive contraction functions; Calculate the overlap ratio between the expansion prediction interval and the level-consistent interval, and combine it with the quantile distribution output by the base learner to determine the defect level determination result using Bayesian posterior probability estimation: ; in This represents the k-th level label. Let be a conditional probability function. and The interval boundaries are defined by the consistent projection of the levels. This refers to the defect level determination result.

[0046] The level-consistent interval projection mechanism in step S6 effectively solves the decision jitter problem of traditional point estimation methods near the level boundary. When the temperature difference to be evaluated is close to the level threshold, the level jump frequency is greatly reduced through the cost-sensitive contraction strategy, thereby providing continuous and reliable level determination results.

[0047] S7. Obtain the standard risk cost matrix, divide the verification data from the original data, and calculate the optimal confidence parameter based on the verification data and the standard risk cost matrix. Output the defect level determination result and the optimal confidence parameter as the consistency temperature range estimation result.

[0048] Step S7 includes the following sub-steps: Determining the cost of false alarms based on power equipment standards Cost of underreporting Confusion Cost Matrix with Rank These are combined to form the standard risk cost matrix; The validation data is extracted from the original data, and the optimal confidence parameter is calculated based on the validation data and the standard risk cost matrix. This process satisfies the following relationship: ; in This is the optimal confidence parameter. The true level label in the verification data. This represents the expectation based on the verification data; The defect level determination result and the optimal confidence parameter The output is the estimated result of the consistent temperature range.

[0049] During implementation, the optimal confidence parameter can be calculated using methods such as grid search or Bayesian optimization, thereby achieving an optimal trade-off between coverage and interval width for the application scenario.

[0050] The defect level determination result output in step S7 and optimal confidence parameters During implementation, the post-processing module can statistically generate auditable reports of the corresponding data, including but not limited to global and group coverage, average interval width, ECE reliability map, cross-device distance consistency statistics, etc. The report is based on test set or actual application data and outputs various indicators in the form of tables and charts, which can ensure the engineering traceability and cost-sensitive decision-making of the method proposed in the embodiments of the present invention.

[0051] The beneficial effects achieved by this invention are that it proposes a consistent temperature range estimation method based on temperature difference grading. This method can achieve synergistic optimization of range estimation accuracy and cross-domain consistency, reduce dependence on manual experience parameter tuning and equipment-specific calibration, and has good scalability and interpretability. At the same time, by improving the accuracy and consistency of grading, this method can effectively reduce the operation and maintenance costs of power equipment and improve the safety of power grid operation, which has significant technical and economic value.

[0052] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by hardware related to computer programs or instructions. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0053] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0055] The embodiments of the present invention have been described above with reference to the accompanying drawings. The disclosed embodiments are merely preferred embodiments of the present invention. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many equivalent changes in form under the guidance of the present invention without departing from the spirit and scope of the claims. All such changes are within the protection scope of the present invention.

Claims

1. A method for estimating a consistent temperature range based on temperature difference grading, characterized in that, Includes the following steps: S1. Obtain the hot spot temperature and reference area temperature of the power equipment by infrared thermometry, calculate the original temperature difference based on the hot spot temperature and the reference area temperature, and correct the original temperature difference based on the infrared thermometry acquisition parameters to obtain a standardized temperature difference. S2. Using the collected parameters and the standardized temperature difference as raw data, training data is extracted from the raw data and used as input to the quantile regression model. The quantile estimation boundary of the standardized temperature difference is calculated through the quantile regression model. S3. Extract calibration data from the original data, calculate the sample inconsistency score based on the calibration data, determine the corresponding calibration amount, and construct the basic prediction interval based on the calibration amount and the quantile estimation boundary. S4. The calibration data is grouped according to different acquisition parameters to obtain multiple sets of calibration group data. The sample inconsistency score is calculated for each set of calibration group data, and the corresponding group calibration amount is determined. The basic prediction interval is updated according to the group calibration amount and the quantile estimation boundary to obtain the updated prediction interval corresponding to the calibration group data. S5. The model uncertainty and measurement uncertainty are used as variances to calculate the safety inflation amount, and the basic prediction interval is expanded according to the safety inflation amount to obtain the expansion prediction interval; S6. Determine the preset power equipment defect classification threshold, and determine the overlap between the preset power equipment defect classification threshold and the expansion prediction interval. Output the defect level determination result containing the interval with the same level based on the overlap. S7. Obtain the standard risk cost matrix, divide the verification data from the original data, and calculate the optimal confidence parameter based on the verification data and the standard risk cost matrix. Output the defect level determination result and the optimal confidence parameter as the consistency temperature range estimation result.

2. The method for estimating a consistent temperature range based on temperature difference grading according to claim 1, characterized in that, Step S1 includes the following sub-steps: The hotspot temperature and the reference area temperature of the power equipment infrared thermometry are obtained, wherein the hotspot temperature is defined as... The reference region temperature is defined as Calculate the original temperature difference, and define it as... And satisfy: ； Determine the acquisition parameters and calculate the standardized temperature difference value, defining the standardized temperature difference value as... It satisfies: ； in To measure distance, Angle of incidence For equipment identification, For scaling correction function, For equipment bias, , , , , Let the collected parameters be: ； ； (i=0,1,2,3) are the distance and angle correction coefficients. This represents the relative scale factor of the equipment. For device system bias, parameters It is learned through the optimization objective of minimizing the cross-domain residual variance, and: ； The standardized temperature difference of sample i, and Representing the same physical quantity, For the corresponding domain The mean within.

3. The method for estimating a consistent temperature range based on temperature difference grading according to claim 2, characterized in that, Step S2 includes the following sub-steps: The acquisition parameters are defined as follows: The quantile regression model is trained using the Pinball loss function: ； ； in Let quantile regression be the total loss function. and These correspond to the quantile levels. and The asymmetric loss function, For confidence level parameters, Let i be the covariate vector of the i-th sample; With the acquisition parameters and the standardized temperature difference value The standardized temperature difference is calculated as the input to the quantile regression model. The quantile estimation boundary ,in Represents a given covariate The standardized temperature difference under the conditions The Quantile estimates This represents the standardized temperature difference value. The Quantile estimates.

4. The method for estimating a consistent temperature range based on temperature difference grading according to claim 3, characterized in that, Step S3 includes the following sub-steps: Calibration data is extracted from the raw data. In the calibration data The CQR method is used to calculate the corresponding sample inconsistency score, and this process satisfies the following relationship: in For calibration data The sample inconsistency score of the i-th calibration sample in the dataset represents the actual value of that sample. The extent to which it exceeds the quantile prediction interval; The calibration quantity is used to calculate the number of inconsistent samples. The process satisfies the following relationship: in The empirical quantile function is used to calculate the set of inconsistent scores for all said samples from the calibration set. Extract the first Percentage value; Construct for any new sample Under the assumption of no distribution The basic prediction interval satisfies: in This is the lower bound of the basic prediction interval. This is the upper bound of the basic prediction interval.

5. The method for estimating a consistent temperature range based on temperature difference grading according to claim 4, characterized in that, Step S4 includes the following sub-steps: The calibration data are grouped according to different acquisition parameters x using a grouping function. The data is divided into M calibration groups, and the corresponding calibration amount for each group is calculated. The process satisfies the following relationship: ； The basic prediction interval is updated based on the grouped calibration pairs and the quantile estimation boundaries to obtain the updated prediction interval corresponding to the calibration group data.

6. The method for estimating a consistent temperature range based on temperature difference grading according to claim 5, characterized in that, Step S5 includes the following sub-steps: Calculate the total variance based on the model uncertainty and the measurement uncertainty. The process satisfies the following relationship: ； in The model predicts variance. For instrument noise variance, Solve for the mixed variance of IFOV. Choose the variance for the reference region; Total variance Standard deviation Convert to safe expansion amount ,in As a conservative coefficient, the basic prediction interval is expanded to obtain the expanded prediction interval. This process satisfies the following relationship: ； in This is the lower bound of the expansion prediction interval. This is the upper bound of the expansion prediction interval.

7. The method for estimating a consistent temperature range based on temperature difference grading according to claim 6, characterized in that, Step S6 includes the following sub-steps: The preset power equipment defect classification threshold is determined, and the classification threshold boundary is defined as follows: Then the intervals of consistent levels satisfy the following relationship: ； in For cost-sensitive contraction functions; Calculate the overlap ratio between the expansion prediction interval and the level-consistent interval, and combine it with the quantile distribution output by the base learner to determine the defect level determination result using Bayesian posterior probability estimation: ； in This represents the k-th level label. Let be a conditional probability function. and The interval boundaries are defined by the consistent projection of the levels. This refers to the defect level determination result.

8. The method for estimating a consistent temperature range based on temperature difference grading according to claim 7, characterized in that, Step S7 includes the following sub-steps: Determining the cost of false alarms based on power equipment standards Cost of underreporting Confusion Cost Matrix with Rank These are combined to form the standard risk cost matrix; The validation data is extracted from the original data, and the optimal confidence parameter is calculated based on the validation data and the standard risk cost matrix. This process satisfies the following relationship: ； in This is the optimal confidence parameter. The true level label in the verification data. This represents the expectation based on the verification data; The defect level determination result and the optimal confidence parameter The output is the estimated result of the consistent temperature range.

Images