A method for correcting time interval temperature drift based on regression prediction

By using regression prediction and bidirectional temperature drift extrapolation, a time interval temperature trajectory and thermal memory residence fingerprint are generated, and a trajectory regressor and a correction amount regressor are trained. This solves the instability problem of temperature drift correction in existing technologies and achieves higher accuracy and more stable temperature drift correction.

CN123116209APending Publication Date: 2026-11-17BEIJING YIPU TIME FREQUENCY TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611262196.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing temperature drift correction methods fail to effectively characterize temperature residence, duration of change, directional change, and gaps between adjacent time intervals. They also struggle to incorporate the migration relationships of historical correction residuals, resulting in insufficient identification of differences in temperature drift trajectory closure and inadequate stability of the correction results.

Method used

The method of regression prediction and bidirectional temperature drift extrapolation is adopted. By generating time interval temperature trajectories and thermal memory residence fingerprints, trajectory regressors and correction regressors are trained to perform forward temperature drift extrapolation and reverse temperature drift backtracking. Temperature drift correction is generated by combining residual migration characteristics.

Benefits of technology

It improves the accuracy and stability of temperature drift correction, reduces the cumulative propagation of historical correction errors during continuous measurement, and enhances the reliability and specificity of correction results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116209A_ABST
    Figure CN123116209A_ABST
Patent Text Reader

Abstract

This invention discloses a time interval temperature drift correction method based on regression prediction, comprising the following steps: collecting data and preprocessing to generate standard temperature measurement operation data; dividing historical time interval segments and current time interval segments and generating time interval temperature trajectories; generating historical thermal memory fingerprints and current thermal memory fingerprints; generating a trajectory regressor and a correction amount regressor, and generating interval endpoint measurement states and current measurement outputs; performing forward temperature drift extrapolation and reverse temperature drift backtracking to generate forward temperature drift extrapolation trajectories and reverse temperature drift backtracking trajectories; mapping the forward temperature drift extrapolation trajectories and reverse temperature drift backtracking trajectories to time interval coordinates, and combining them with correction record data to generate residual migration features; performing regression mapping to generate temperature drift correction amounts, and correcting and generating temperature drift correction results. This invention employs regression prediction and bidirectional temperature drift extrapolation to achieve time interval temperature drift correction, possessing the advantages of high accuracy and strong stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of time interval temperature drift correction technology, and more particularly to a time interval temperature drift correction method based on regression prediction. Background Technology

[0002] Existing temperature drift correction methods typically establish a correlation between temperature and measurement deviation by collecting equipment temperature data, measurement output data, and calibration reference data. Then, they use linear fitting, empirical compensation tables, or ordinary regression models to generate correction values ​​to compensate for the current measurement output, thereby reducing the impact of temperature changes on measurement accuracy.

[0003] However, the above methods mostly focus on a single measurement moment or the overall temperature change trend, lacking the characterization of temperature residence, change duration, direction change and interval gap within adjacent time intervals. They are also difficult to combine the migration relationship of historical correction residuals to predict the correction amount. At the same time, existing methods usually do not perform forward extrapolation and reverse backtracking verification of the measurement status at both ends of the time interval, resulting in the inability to fully identify the differences in temperature drift trajectory closure and insufficient stability of the correction results.

[0004] Therefore, how to provide a time interval temperature drift correction method based on regression prediction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a time interval temperature drift correction method based on regression prediction. This invention uses regression prediction and two-way temperature drift extrapolation to achieve time interval temperature drift correction, which has the advantages of high accuracy and strong stability.

[0006] A time interval temperature drift correction method based on regression prediction according to an embodiment of the present invention includes the following steps: Collect basic temperature measurement data of the device to be corrected during continuous measurement, and preprocess the basic temperature measurement data to generate standard temperature measurement operation data; Based on the measurement time data in the standard temperature measurement operation data, time interval segments are divided, historical time interval segments and current time interval segments are divided, and time interval temperature trajectories are generated; Thermal memory fingerprints are generated based on time interval temperature trajectories, and historical and current thermal memory fingerprints are generated according to the time interval segments corresponding to the thermal memory fingerprints. Based on the standard temperature measurement operation data, historical thermal memory resident fingerprints and correction record data corresponding to the historical time intervals, a trajectory regressor and a correction amount regressor are trained and generated, and the measurement status of the interval endpoints and the current measurement output are generated. Combining the current thermal memory-residual fingerprint, a trajectory regressor is used to perform forward temperature drift extrapolation and reverse temperature drift backtracking on the starting and ending measurement states, respectively, generating forward temperature drift extrapolation trajectory and reverse temperature drift backtracking trajectory; Based on the measurement time data corresponding to the current time interval, time interval coordinates are generated. The forward temperature drift extrapolation trajectory and the reverse temperature drift backtracking trajectory are mapped to the time interval coordinates, and residual migration features are generated by combining the corrected record data. A correction regression model is used to regress and map the current thermal memory residence fingerprint, temperature drift suture crack features and residual migration features to generate the temperature drift correction amount corresponding to the current time interval, and then corrects it to generate the temperature drift correction result.

[0007] Optionally, the generation of the standard temperature measurement operation data includes: The operational data acquisition data and correction record data are time-aligned according to the measurement time data, so that the temperature data, measurement output data, calibration reference data, historical temperature drift correction, historical calibration reference deviation and historical correction residual corresponding to the same measurement time are recorded at the same time. Anomalies are sampled and removed from temperature data and measurement output data recorded at the same time, and the removed temperature data, measurement output data, measurement time data, calibration reference data and correction record data are retained. The removed temperature data, measurement output data, calibration reference data, and correction record data are scaled and normalized to generate normalized temperature measurement data. The measurement cycle is divided according to the measurement time data in the normalized temperature measurement data to generate standard temperature measurement operation data.

[0008] Optionally, the generation of the time interval temperature trajectory includes: The temperature data and measurement output data are arranged in ascending order of time according to the measurement time data in the standard temperature measurement operation data to generate a measurement time sequence. Time interval segments are generated by using two adjacent measurement times in the measurement time sequence as the start and end boundaries of the interval; The time interval segment at the end of the measurement time sequence is determined as the current time interval segment, and the time interval segments before the current time interval segment are determined as historical time interval segments. Based on the start and end boundaries of each time interval, the corresponding temperature data, measurement output data, and measurement time data are collected from the standard temperature measurement operation data to generate the time interval temperature trajectory.

[0009] Optionally, the generation of the historical hot memory fingerprint and the current hot memory fingerprint includes: Arrange the temperature data according to the measurement time data in the time interval temperature trajectory, calculate the temperature difference, sampling time difference and temperature change direction between adjacent measurement times, and generate an interval temperature change sequence; A thermal memory fragment chain is generated based on the interval temperature change sequence, and a thermal memory feature set is generated based on the thermal memory fragment chain. The thermal memory feature set is combined according to the measurement time sequence within the time interval to generate thermal memory residency fingerprint. The thermal memory residency fingerprint corresponding to the historical time interval is determined as the historical thermal memory residency fingerprint, and the thermal memory residency fingerprint corresponding to the current time interval is determined as the current thermal memory residency fingerprint.

[0010] Optionally, the generation of the interval endpoint measurement status and the current measurement output includes: Historical endpoint measurement status is generated based on the standard temperature measurement operation data and correction record data corresponding to the historical time interval. The difference is calculated based on the measurement output data and calibration reference data corresponding to each measurement time within the historical time interval, and the difference calculation results are arranged according to the measurement time data to generate a historical temperature drift difference sequence. Trajectory training samples are generated based on historical endpoint measurement status, historical thermal memory residency fingerprints, and historical temperature drift difference sequences, and the trajectory training samples are used to train and generate a trajectory regressor. Correction training samples are generated based on historical thermal memory fingerprints, historical temperature drift difference sequences, and correction record data, and a correction regressor is trained using the correction training samples. The interval endpoint measurement status and current measurement output are generated based on the standard temperature measurement operation data and correction record data corresponding to the current time interval.

[0011] Optionally, the generation of the reverse temperature drift backtracking trajectory includes: Read the measurement time data from the time interval temperature trajectory corresponding to the current time interval, generate a forward extrapolation sampling point sequence according to the increasing direction of measurement time, and generate a reverse backtracking sampling point sequence according to the decreasing direction of measurement time; Based on the forward inference sampling point sequence, the starting measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the forward inference input sequence. Based on the reverse backtracking sampling point sequence, the ending measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the reverse backtracking input sequence. A trajectory regressor is used to perform a sampling-point regression on the forward extrapolation input sequence to generate a forward temperature drift state sequence corresponding to the forward extrapolation sampling point sequence. A trajectory regressor is also used to perform a sampling-point regression on the reverse backtracking input sequence to generate a reverse temperature drift state sequence corresponding to the reverse backtracking sampling point sequence. The forward temperature drift state sequence is arranged according to the forward extrapolation sampling point sequence to generate the forward temperature drift extrapolation trajectory, and the reverse temperature drift state sequence is arranged according to the reverse backtracking sampling point sequence to generate the reverse temperature drift backtracking trajectory.

[0012] Optionally, the generation of the residual migration features includes: Using the starting measurement time of the current time interval as the coordinate starting point, the time interval coordinates are generated according to the measurement time data corresponding to the current time interval. The forward temperature drift trajectory is mapped to the time interval coordinate according to the increasing direction of measurement time, and the reverse temperature drift back trajectory is rearranged and mapped to the time interval coordinate after being arranged according to the increasing direction of measurement time, thus generating a pair of temperature drift trajectories with the same coordinates. Based on the same coordinate temperature drift trajectory pair, calculate the trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement between the forward temperature drift extrapolation trajectory and the reverse temperature drift back trajectory at the same time interval coordinate position, and generate the temperature drift stitching area according to the time interval coordinate order. Based on the continuous positional correlation of trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement in the temperature drift suture region, the temperature drift suture crack characteristics are generated. Based on the correction record data corresponding to the historical time intervals, the historical temperature drift correction, historical calibration reference deviation and historical correction residual are associated in the order of measurement time of the historical time intervals to generate the interval residual migration chain. Based on the interval residual migration chain, the residual cumulative change, residual decay change, residual direction change and residual abrupt change magnitude between adjacent historical time intervals are calculated to generate residual migration features.

[0013] Optionally, the generation of the temperature drift correction result includes: The current thermal memory fingerprint, temperature drift suture crack features, and residual migration features are combined according to the current time interval to generate the correction amount regression input features; A correction regression analyzer is used to perform regression mapping on the correction regression input features to generate the temperature drift correction corresponding to the current time interval. The current measurement output is processed to cancel out drift based on the temperature drift correction amount, and a corrected measurement output is generated. Combine the current time interval, temperature drift correction amount, and corrected measurement output to generate the temperature drift correction result.

[0014] The beneficial effects of this invention are: This invention generates standard temperature measurement operation data based on basic temperature measurement data during continuous measurement processes. It divides the measurement time data into time intervals, transforming temperature drift correction from instantaneous deviation compensation at a single measurement moment into a dynamic correction process oriented towards time intervals. By generating time interval temperature trajectories within each time interval and further forming a thermal memory fingerprint encompassing temperature dwell characteristics, temperature change persistence characteristics, temperature direction change characteristics, temperature plateau characteristics, interval gap characteristics, and temperature change rate characteristics, it can more comprehensively characterize the hysteretic, cumulative, and directional effects of temperature changes on the measurement output, thereby improving the targeting and precision of temperature drift correction prediction.

[0015] This invention trains and generates a trajectory regressor and a correction regressor using standard temperature measurement data corresponding to historical time intervals, historical thermal memory fingerprints, and correction record data. This allows historical temperature drift corrections, historical calibration reference deviations, and historical correction residuals to participate in the generation of the current correction. Compared to compensation methods based solely on the current temperature value or simple temperature changes, this invention combines historical residual migration relationships to characterize the continuation, reverse change, and abrupt change amplitude of residual errors during the temperature drift correction process. This reduces the accumulation and propagation of historical correction errors in subsequent measurements and improves the stability of correction results in continuous measurement scenarios.

[0016] This invention also uses a trajectory regressor to perform forward temperature drift extrapolation and reverse temperature drift backtracking on the initial and final measurement states, respectively. The forward temperature drift extrapolation trajectory and the reverse temperature drift backtracking trajectory are mapped to the same time interval coordinates, generating temperature drift stitching regions and temperature drift stitching crack features. This processing can reflect the closure difference between the extrapolation results at both ends of the time interval, enabling the correction regressor to simultaneously consider temperature thermal memory, trajectory stitching differences, and residual migration characteristics when generating temperature drift correction values. This improves the reliability of the temperature drift correction results and reduces measurement output deviations caused by the combined effects of temperature fluctuations, sampling intervals, and historical residuals. Attached Figure Description

[0017] 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: Figure 1 This is an overall flowchart of a time interval temperature drift correction method based on regression prediction proposed in this invention; Figure 2 This is a schematic diagram of the generation of thermal memory residency fingerprints in a time interval temperature drift correction method based on regression prediction proposed in this invention. Detailed Implementation

[0018] 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.

[0019] refer to Figures 1-2 A time interval temperature drift correction method based on regression prediction includes the following steps: Collect basic temperature measurement data of the device to be corrected during continuous measurement. The basic temperature measurement data includes operational acquisition data and correction record data. The operational acquisition data includes temperature data, measurement output data, measurement time data and calibration reference data. The correction record data includes historical temperature drift correction amount, historical calibration reference deviation and historical correction residual. The basic temperature measurement data is preprocessed to generate standard temperature measurement operational data. The time interval is divided based on the measurement time data in the standard temperature measurement operation data. The time interval is divided into historical time intervals and current time intervals according to the measurement time sequence. The time interval temperature trajectory is generated based on the temperature data, measurement output data and measurement time data corresponding to each time interval. Thermal memory residency fingerprints are generated based on time interval temperature trajectories, and historical thermal memory residency fingerprints and current thermal memory residency fingerprints are generated according to the time interval segments corresponding to the thermal memory residency fingerprints. The thermal memory residency fingerprints include temperature residency features, temperature change duration features, temperature direction change features, temperature plateau features, interval blank features, and temperature change rate features. Based on the standard temperature measurement operation data, historical thermal memory fingerprint and correction record data corresponding to the historical time interval, a trajectory regressor and a correction regressor are trained and generated. Based on the standard temperature measurement operation data and correction record data corresponding to the current time interval, the interval endpoint measurement status and the current measurement output are generated. The interval endpoint measurement status includes the start measurement status and the end measurement status. A trajectory regressor is used to perform forward temperature drift extrapolation on the initial measurement state and the current thermal memory fingerprint, generating a forward temperature drift extrapolation trajectory. A trajectory regressor is also used to perform reverse temperature drift backtracking on the final measurement state and the current thermal memory fingerprint, generating a reverse temperature drift backtracking trajectory. Based on the measurement time data corresponding to the current time interval, time interval coordinates are generated. The forward temperature drift extrapolation trajectory and the reverse temperature drift backtracking trajectory are mapped to the time interval coordinates to generate the temperature drift suture area. Based on the temperature drift suture area, temperature drift suture crack features are generated. Based on the correction record data corresponding to the historical time interval, the interval residual migration chain is generated. Based on the interval residual migration chain, residual migration features are generated. A correction regression analyzer is used to regress and map the current thermal memory residence fingerprint, temperature drift suture crack features and residual migration features to generate the temperature drift correction amount corresponding to the current time interval. The current measurement output is then corrected according to the temperature drift correction amount to generate the temperature drift correction result.

[0020] In this embodiment, the generation of standard temperature measurement operation data includes: The operational data acquisition data and correction record data are time-aligned according to the measurement time data, so that the temperature data, measurement output data, calibration reference data, historical temperature drift correction, historical calibration reference deviation and historical correction residual corresponding to the same measurement time are recorded at the same time. Using measurement time data as the time axis, the operational acquisition data and correction record data are merged according to the measurement time number. Temperature data, measurement output data, calibration reference data, historical temperature drift correction, historical calibration reference deviation, and historical correction residual at the same measurement time are written into the same time record. In the implementation data, the continuous measurement process originally forms 27 time records. Each time record contains 1 temperature value, 1 measurement output value, 1 calibration reference value, 1 historical temperature drift correction, 1 historical calibration reference deviation, and 1 historical correction residual. Anomalies are sampled and removed from temperature data and measurement output data recorded at the same time, and the removed temperature data, measurement output data, measurement time data, calibration reference data and correction record data are retained. The integrity of temperature data, measurement output data, and measurement time data in the same time record after time alignment is checked. Anomaly sampling marks are generated based on repeated measurement time, null temperature data, and null measurement output data. Time records with anomaly sampling marks are removed from the time-aligned time records. Temperature data, measurement output data, measurement time data, calibration reference data, and correction record data in the remaining time records are retained. In the implementation data, 1 record of 27 time records has repeated measurement time and 1 record has null measurement output data. After removing these, 25 valid time records are formed. The removed temperature data, measurement output data, calibration reference data, and correction record data are scaled and normalized to generate normalized temperature measurement data. The mean and standard deviation of the removed temperature data, measurement output data, calibration reference data, historical temperature drift correction, historical calibration reference deviation, and historical correction residual were calculated respectively. Then, the mean of each data value was subtracted and the standard deviation was divided to generate normalized temperature measurement data. In the actual data, the mean of the temperature data was 32.6℃ and the standard deviation was 1.8℃. The temperature value of 35.3℃ was normalized to 1.50. The measurement cycle is divided according to the measurement time data in the normalized temperature measurement data to generate standard temperature measurement operation data; Based on the measurement time data in the normalized temperature measurement data, the time records in the continuous measurement process are divided into measurement cycle records in ascending order of measurement time. A measurement cycle number is generated for each measurement cycle record. Each measurement cycle record retains the cycle start time, cycle end time, temperature data within the cycle, measurement output data within the cycle, calibration reference data within the cycle, and correction record data within the cycle. In the implementation data, the measurement time corresponding to the 25 valid time records is from 08:00 to 10:00. These are divided into 4 measurement cycle records in 30-minute intervals. All measurement cycle records together constitute the standard temperature measurement operation data.

[0021] In this embodiment, the generation of the time interval temperature trajectory includes: The temperature data and measurement output data are arranged in ascending order of time according to the measurement time data in the standard temperature measurement operation data to generate a measurement time sequence. The measurement time data in the standard temperature measurement operation data are arranged in ascending order of time, and the temperature data and measurement output data corresponding to the same measurement time are moved synchronously to form a measurement time sequence with temperature data and measurement output data. In the implementation data, the measurement time data corresponding to the 25 valid time records are 08:00, 08:05, 08:10 to 10:00. After being arranged, 25 measurement times are formed, and each measurement time corresponds to 1 temperature value and 1 measurement output value. Time interval segments are generated by using two adjacent measurement times in the measurement time sequence as the start and end boundaries of the interval; The two adjacent measurement times in the measurement time sequence are used as the start and end boundaries of the interval. The temperature data, measurement output data and measurement time data corresponding to the two measurement times are written into the same time interval. In the implementation data, 25 measurement times form 24 time intervals. The first time interval is from 08:00 to 08:05, and the 24th time interval is from 09:55 to 10:00. The time interval segment at the end of the measurement time sequence is determined as the current time interval segment, and the time interval segments before the current time interval segment are determined as historical time interval segments. All time intervals are arranged in the order of measurement time. The time interval at the end of the arrangement is written into the current time interval, and the time intervals before the current time interval are written into the historical time interval. In the implementation data, after the 24 time intervals are arranged in the order of measurement time, the 24th time interval is written into the current time interval, and the 1st to 23rd time intervals are written into the historical time interval. Based on the start and end boundaries of each time interval, the corresponding temperature data, measurement output data, and measurement time data are collected from the standard temperature measurement operation data to generate the time interval temperature trajectory. The temperature data, measurement output data, and measurement time data within each time interval are combined in order of the start and end boundaries of the interval to form a time interval temperature trajectory with time coordinates, temperature values, and measurement output values. In the implemented data, the starting temperature of the time interval from 09:55 to 10:00 is 31.20℃, the ending temperature is 31.34℃, the starting measurement output is 100.018, and the ending measurement output is 100.026. This time interval temperature trajectory records a temperature change of 0.14℃ and a measurement output change of 0.008.

[0022] In this embodiment, the generation of historical hot memory fingerprints and current hot memory fingerprints includes: Arrange the temperature data according to the measurement time data in the time interval temperature trajectory, calculate the temperature difference, sampling time difference and temperature change direction between adjacent measurement times, and generate an interval temperature change sequence; Temperature data are arranged according to the measurement time data in the time interval temperature trajectory. Temperature difference is generated by subtracting the temperature data corresponding to adjacent measurement times, and sampling time difference is generated by subtracting the measurement time data corresponding to adjacent measurement times. The direction of temperature change is generated according to the positive and negative directions of the temperature difference. The temperature difference, sampling time difference and temperature change direction are combined in the order of measurement time to generate an interval temperature change sequence. In the actual data, the temperature is 31.20℃ at 09:55 and 31.34℃ at 10:00, corresponding to a temperature difference of 0.14℃, a sampling time difference of 5min, and an upward temperature change direction. A thermal memory segment chain is generated based on the interval temperature change sequence. The thermal memory segment chain includes temperature dwell segments, temperature change duration segments, temperature direction change segments, temperature plateau segments, and interval blank segments. Temperature dwell segments are formed by sampling segments where the temperature difference remains constant. Temperature change duration segments are formed by sampling segments where the temperature change direction remains consistent. Temperature direction change segments are formed by positions where the temperature change direction reverses. Temperature plateau segments are formed by connecting multiple consecutive temperature dwell segments in the order of measurement time. Interval blank segments are formed by the sampling time difference between adjacent measurement times within the time interval. Using adjacent measurement moments in the interval temperature change sequence as basic sampling pairs, each basic sampling pair is written with temperature difference, sampling time difference, and temperature change direction. Continuous basic sampling pairs with a temperature difference of 0 are written into the temperature dwell segment. Basic sampling pairs with a continuous temperature difference of the same non-zero value are written into the temperature change duration segment. Basic sampling pairs with a continuous temperature change direction of either rising or falling are written into the temperature change duration segment. The position where the temperature change direction between adjacent basic sampling pairs changes from rising to falling or from falling to rising is written into the temperature direction change segment. Continuously connected temperature dwell segments are merged into a temperature plateau segment. The sampling time difference between adjacent measurement moments within the time interval is written into the interval blank segment. A thermal memory feature set is generated based on the thermal memory fragment chain. The thermal memory feature set includes temperature dwell characteristics, temperature change duration characteristics, temperature direction change characteristics, temperature plateau characteristics, interval gap characteristics, and temperature change rate characteristics. The temperature dwell time segment in the thermal memory fragment chain is accumulated to generate a temperature dwell time feature. The temperature change duration and the temperature change duration are accumulated separately for the temperature change duration segment to generate a temperature change duration feature. The number of direction reversals and the position of the reversal measurement time are counted for the temperature direction change segment to generate a temperature direction change feature. The plateau duration and the average plateau temperature are accumulated for the temperature plateau segment to generate a temperature plateau feature. The sampling time difference is accumulated for the interval blank segment to generate an interval blank feature. The ratio of the temperature difference to the sampling time difference in the interval temperature change sequence is calculated to generate a temperature change rate feature. The thermal memory feature set is combined according to the measurement time sequence within the time interval to generate thermal memory fingerprint. The thermal memory fingerprint corresponding to the historical time interval is determined as the historical thermal memory fingerprint, and the thermal memory fingerprint corresponding to the current time interval is determined as the current thermal memory fingerprint. Temperature dwell characteristics, temperature change duration characteristics, temperature direction change characteristics, temperature plateau characteristics, interval blank characteristics, and temperature change rate characteristics are written into the same fingerprint record in the order of measurement time to form a thermal memory dwell fingerprint. In the implementation data, a thermal memory dwell fingerprint includes dwell duration of 0 min, rise duration of 5 min, fall duration of 0 min, number of direction reversals of 0 times, plateau duration of 0 min, cumulative interval blank time of 5 min, and average temperature change rate of 0.028℃ / min. The thermal memory dwell fingerprints corresponding to historical time intervals are written into historical thermal memory dwell fingerprints, and the thermal memory dwell fingerprints corresponding to the current time interval are written into the current thermal memory dwell fingerprint.

[0023] In this embodiment, the generation of the interval endpoint measurement status and the current measurement output includes: Historical endpoint measurement status is generated based on the standard temperature measurement operation data and correction record data corresponding to the historical time interval. The historical endpoint measurement status includes historical beginning measurement status and historical end measurement status. The historical beginning measurement status is formed by the beginning temperature data, beginning measurement output data and beginning historical correction residual of the historical time interval. The historical end measurement status is formed by the end temperature data, end measurement output data and the difference between the end measurement output data and the end calibration reference data of the historical time interval. The starting temperature data, starting measurement output data, and corresponding historical correction residual are read from the starting measurement time of each historical time interval and combined to form the historical starting measurement state. The ending temperature data, ending measurement output data, and ending calibration reference data are read from the ending measurement time of each historical time interval and the difference between the ending measurement output data and the ending calibration reference data is calculated and combined to form the historical ending measurement state. In the implementation data, the starting temperature of a certain historical time interval is 30.80℃, the starting measurement output is 100.012, and the starting historical correction residual is 0.004, forming one historical starting measurement state record. The difference is calculated based on the measurement output data and calibration reference data corresponding to each measurement time within the historical time interval, and the difference calculation results are arranged according to the measurement time data to generate a historical temperature drift difference sequence. The difference between the measurement output data and calibration reference data corresponding to the start and end measurement times within the historical time interval is calculated, and the difference is arranged in ascending order of measurement time data to form a historical temperature drift difference sequence. In the implementation data, a certain historical time interval includes the start and end measurement times, with corresponding differences of 0.004 and 0.006, which are arranged to form a historical temperature drift difference sequence of length 2. Trajectory training samples are generated based on historical endpoint measurement status, historical thermal memory residency fingerprints, and historical temperature drift difference sequences, and the trajectory training samples are used to train and generate a trajectory regressor. The input samples in the trajectory training samples are formed by splicing historical endpoint measurement states and historical thermal memory fingerprints. The output samples are formed by the historical temperature drift difference sequence. In one implementation, the trajectory regressor uses a randomly configured network regressor with 80 hidden layer nodes. The input weights are taken from a uniform sequence of -1 to 1, and the biases are taken from a uniform sequence of -0.5 to 0.5. The output weights are solved by ridge regression. In the implementation data, the historical sample library contains 300 historical time intervals, of which 23 historical time intervals are provided by the current continuous measurement process. The remaining historical time intervals are from the standard temperature measurement operation data of the device to be corrected in the previous continuous measurement process. The historical sample library forms 300 sets of trajectory training samples. The output error is calculated using the mean square error between the historical temperature drift difference sequence and the regression output sequence. Correction training samples are generated based on historical thermal memory fingerprints, historical temperature drift difference sequences, and correction record data, and a correction regressor is trained using the correction training samples. The sample input in the correction training sample is formed by splicing historical thermal memory fingerprints, historical temperature drift difference sequences, and correction record data. The sample output is the historical temperature drift correction amount. In one implementation, the correction regressor adopts a correlation vector regressor, the kernel function adopts a radial basis kernel, the kernel width is taken as the median value of the historical sample input distance, and the initial value of the noise variance is 0.01. In the implementation data, one correction training sample is generated for each historical time interval in the historical sample library, and 300 historical time intervals form 300 correction training samples. The correction regressor after training is used to output the temperature drift correction amount corresponding to the current time interval. Based on the standard temperature measurement operation data and correction record data corresponding to the current time interval, the interval endpoint measurement status and current measurement output are generated. The interval endpoint measurement status includes the start measurement status and the end measurement status. The start measurement status is formed by the start temperature data, start measurement output data and historical correction residuals corresponding to the start measurement time of the current time interval in the correction record data of the current time interval. The end measurement status is formed by the end temperature data, end measurement output data and the difference between the end measurement output data and the end calibration reference data of the current time interval. The current measurement output is formed by the end measurement output data of the current time interval. The starting temperature data, starting measurement output data, and historical correction residuals from the correction record data are read from the starting measurement time of the current time interval and combined to form the starting measurement state. The ending temperature data, ending measurement output data, and ending calibration reference data are read from the ending measurement time of the current time interval and the difference between the ending measurement output data and the ending calibration reference data are calculated and combined to form the ending measurement state. In the implementation data, the starting measurement output of the current time interval is 100.018 and the ending measurement output is 100.026. The difference between the ending measurement output data and the ending calibration reference data is written into the ending measurement state, and the ending measurement output is written into the current measurement output.

[0024] In this embodiment, the generation of the reverse temperature drift tracing trajectory includes: Read the measurement time data from the time interval temperature trajectory corresponding to the current time interval, generate a forward extrapolation sampling point sequence according to the increasing direction of measurement time, and generate a reverse backtracking sampling point sequence according to the decreasing direction of measurement time; The starting and ending measurement times are read from the time interval temperature trajectory corresponding to the current time interval. The data are arranged in the direction of increasing measurement time to form a forward projection sampling point sequence. The same set of measurement time data is arranged in the direction of decreasing measurement time to form a reverse backtracking sampling point sequence. In the actual data, the current time interval is from 09:55 to 10:00. The forward projection sampling point sequence is 09:55 and 10:00, and the reverse backtracking sampling point sequence is 10:00 and 09:55. Based on the forward inference sampling point sequence, the starting measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the forward inference input sequence. Based on the reverse backtracking sampling point sequence, the ending measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the reverse backtracking input sequence. The current hot memory fingerprint is copied into a measurement-time hot memory record according to the number of measurement moments in the current time interval. Each measurement-time hot memory record is combined with the corresponding measurement time data. The forward extrapolation input sequence is formed by combining the starting measurement state, the measurement-time hot memory record, and the forward extrapolation sampling point sequence. The reverse backtracking input sequence is formed by combining the ending measurement state, the measurement-time hot memory record, and the reverse backtracking sampling point sequence. In the implementation data, the current time interval contains 2 measurement moments. Both the forward extrapolation input sequence and the reverse backtracking input sequence contain 2 measurement-time input records. A trajectory regressor is used to perform a sampling-point regression on the forward extrapolation input sequence to generate a forward temperature drift state sequence corresponding to the forward extrapolation sampling point sequence. A trajectory regressor is also used to perform a sampling-point regression on the reverse backtracking input sequence to generate a reverse temperature drift state sequence corresponding to the reverse backtracking sampling point sequence. The trajectory regressor outputs a positive temperature drift difference value for each measurement-time input record in the forward extrapolation input sequence, and generates a positive temperature drift rate based on the positive temperature drift difference value corresponding to adjacent measurement times and the sampling time difference. The positive temperature drift difference value and the positive temperature drift rate are combined to form a positive temperature drift state. The trajectory regressor outputs a negative temperature drift difference value for each measurement-time input record in the reverse backtracking input sequence, and generates a negative temperature drift rate based on the negative temperature drift difference value corresponding to adjacent measurement times and the sampling time difference. The negative temperature drift difference value and the negative temperature drift rate are combined to form a negative temperature drift state. In the implementation data, two measurement times correspond to two positive temperature drift states and two negative temperature drift states. Arrange the forward temperature drift state sequence according to the forward extrapolation sampling point sequence to generate the forward temperature drift extrapolation trajectory; arrange the reverse temperature drift state sequence according to the reverse backtracking sampling point sequence to generate the reverse temperature drift backtracking trajectory. The positive temperature drift difference and positive temperature drift rate in the positive temperature drift state sequence are arranged according to the positive inference sampling point sequence to form a positive temperature drift inference trajectory. The negative temperature drift difference and negative temperature drift rate in the negative temperature drift state sequence are arranged according to the negative backtracking sampling point sequence to form a negative temperature drift backtracking trajectory. In the implemented data, two positive temperature drift states form two positive temperature drift inference trajectories, and two negative temperature drift states form two negative temperature drift backtracking trajectories.

[0025] In this embodiment, the generation of residual migration features includes: Using the starting measurement time of the current time interval as the coordinate starting point, the time interval coordinates are generated according to the measurement time data corresponding to the current time interval. Using the starting measurement time of the current time interval as the coordinate starting point, the starting measurement time and the ending measurement time within the current time interval are converted into coordinate values ​​to form time interval coordinates. In the implemented data, the current time interval is from 09:55 to 10:00. The starting measurement time 09:55 corresponds to the time interval coordinate of 0min, and the ending measurement time 10:00 corresponds to the time interval coordinate of 5min. The forward temperature drift trajectory is mapped to the time interval coordinate according to the increasing direction of measurement time, and the reverse temperature drift back trajectory is rearranged and mapped to the time interval coordinate after being arranged according to the increasing direction of measurement time, thus generating a pair of temperature drift trajectories with the same coordinates. Each positive temperature drift state in the positive temperature drift extrapolation trajectory is mapped to the corresponding time interval coordinate according to its measurement time. The reverse temperature drift backtracking trajectory is rearranged according to the increasing direction of measurement time, and each rearranged reverse temperature drift state is mapped to the corresponding time interval coordinate. In the implementation data, positive temperature drift state 0.032 and reverse temperature drift state 0.026 are written at the time interval coordinate of 5min, thus forming a temperature drift trajectory pair with the same coordinate. Based on the same coordinate temperature drift trajectory pair, calculate the trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement between the forward temperature drift extrapolation trajectory and the reverse temperature drift back trajectory at the same time interval coordinate position, and generate the temperature drift stitching area according to the time interval coordinate order. The positive and negative temperature drift states in the same coordinate temperature drift trajectory pair are paired point by point according to the same time interval coordinates to form a coordinate stitching record. In each coordinate stitching record, the trajectory difference, endpoint closure error, adjacent coordinate slope difference, and trajectory lag displacement are calculated. The trajectory difference is formed by subtracting the positive and negative temperature drift differences under the same time interval coordinates. The endpoint closure error is formed by adding the absolute values ​​of the trajectory differences corresponding to the starting and ending coordinate positions. The adjacent coordinate slope difference is formed by subtracting the rate of change of the temperature drift difference between adjacent time interval coordinates of the positive temperature drift projection trajectory and the rate of change of the temperature drift difference between adjacent time interval coordinates of the negative temperature drift retrospective trajectory. The trajectory lag displacement is formed by the time interval coordinate difference of the same temperature drift state value in the positive temperature drift projection trajectory and the negative temperature drift retrospective trajectory. In the implementation data, when the positive temperature drift difference is 0.032 and the negative temperature drift difference is 0.026, the trajectory difference at that coordinate position is 0.006. Based on the continuous positional correlation of trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement in the temperature drift suture region, the temperature drift suture crack characteristics are generated. The coordinate-by-coordinate suturing records in the temperature drift suturing area are connected in time interval coordinate order to form a suturing difference sequence. The maximum value, cumulative value, mean value and position distribution of the trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement in the suturing difference sequence are calculated respectively to generate temperature drift suturing crack features. In the implementation data, two coordinate-by-coordinate suturing records are formed in the current time interval. The temperature drift suturing crack features include the maximum trajectory difference of 0.009, the cumulative endpoint closure difference of 0.015, the average adjacent coordinate slope difference of 0.004 and the maximum trajectory lag displacement of 1 sampling interval. Based on the correction record data corresponding to the historical time intervals, the historical temperature drift correction, historical calibration reference deviation and historical correction residual are associated in the order of measurement time of the historical time intervals to generate the interval residual migration chain. The historical temperature drift correction, historical calibration reference deviation, and historical correction residual corresponding to each historical time interval are combined into interval residual nodes. The interval residual nodes are connected according to the measurement time sequence of the historical time interval to form an interval residual migration chain. The previous historical correction residual, the next historical correction residual, and the difference between the two are recorded between adjacent interval residual nodes in the chain. In the implementation data, 20 interval residual nodes are generated for 20 consecutive historical time intervals, and 19 adjacent node connection relationships are formed. Based on the interval residual migration chain, the residual cumulative change, residual decay change, residual direction change and residual abrupt change magnitude between adjacent historical time intervals are calculated to generate residual migration features; Based on the historical corrected residual difference between adjacent residual nodes in the interval residual migration chain, the residual cumulative change, residual decay change, residual direction change, and residual abrupt change magnitude are calculated. The residual cumulative change is formed by summing the residual differences in the same direction. The residual decay change is formed by the decrease in the absolute value of the residuals of adjacent nodes. The residual direction change is formed by the number of sign changes of the residuals of adjacent nodes. The residual abrupt change magnitude is formed by the maximum absolute value of the residual difference between adjacent nodes. In the implemented data, when the residual sequence is 0.010, 0.014, 0.011, and -0.006, the residual cumulative change is 0.004, the residual decay change is 0.003, the residual direction change occurs once, and the residual abrupt change magnitude is 0.017.

[0026] In this embodiment, the generation of the temperature drift correction result includes: The current thermal memory fingerprint, temperature drift suture crack features, and residual migration features are combined according to the current time interval to generate the correction amount regression input features; The current thermal memory fingerprint, temperature drift suture crack feature, and residual migration feature are combined according to the current time interval number and arranged in the order of thermal memory field, suture crack field, and residual migration field to form a correction regression input feature. In the implementation data, the correction regression input feature contains 7 thermal memory numerical fields, 4 temperature drift suture crack fields, and 4 residual migration fields, forming a total of 15-dimensional input features. This correction regression input feature serves as the current sample input for the correction regression device. A correction regression analyzer is used to perform regression mapping on the correction regression input features to generate the temperature drift correction corresponding to the current time interval. The correction regression unit reads the correction regression input features and outputs the single temperature drift correction corresponding to the current time interval. In the actual data, the average temperature change rate corresponding to the current thermal memory fingerprint is 0.028℃ / min, the maximum trajectory difference is 0.009, the residual mutation amplitude is 0.017, and the correction regression unit outputs a temperature drift correction of 0.031. This temperature drift correction is used to correct the numerical value of the current measurement output. The current measurement output is processed to cancel out drift based on the temperature drift correction amount, and a corrected measurement output is generated. The current measurement output is subtracted from the temperature drift correction amount to generate the corrected measurement output. In the actual data, the current measurement output is 100.026, the temperature drift correction amount is 0.031, and the corrected measurement output is 99.995. The corrected measurement output, the current time interval, and the temperature drift correction amount are all written into the temperature drift correction result. Combine the current time interval, temperature drift correction amount, and corrected measurement output to generate the temperature drift correction result; The current time interval, temperature drift correction amount, and corrected measurement output are combined according to the current time interval number to generate the temperature drift correction result. In the actual data, the current time interval is from 09:55 to 10:00, the temperature drift correction amount is 0.031, and the corrected measurement output is 99.995. The three together form the temperature drift correction result corresponding to the current time interval.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a continuous measurement scenario of a precision weighing sensor in a constant-temperature metrology room of a certain enterprise. This device is used to continuously output data for minute mass changes, and its internal sensing elements are highly sensitive to ambient temperature and the device's own heat generation. During continuous operation, although the ambient temperature is within a controllable range, it still experiences slow rises, brief pauses, and directional fluctuations, causing the measurement output to gradually deviate from the calibration reference. Traditional correction methods typically involve lookup table compensation based on the current temperature value or correction based on the difference between the current measurement output and the calibration reference. These methods struggle to reflect the lag effect of temperature changes within adjacent time intervals and also fail to address the issue of historical correction residuals being passed on to subsequent measurements. Therefore, during long-term continuous measurements, the device is prone to insufficient correction in previous time intervals, leading to continued accumulation of deviations in subsequent time intervals, affecting the consistency of measurement results.

[0028] In this scenario, basic temperature measurement data of the device to be corrected during continuous measurement is first collected. The collected data includes temperature data, measurement output data, measurement time data, and calibration reference data. Correction record data includes historical temperature drift corrections, historical calibration reference deviations, and historical correction residuals. After collection, the measurement time data is used as the time axis to align the collected data and correction record data, ensuring that temperature data, measurement output data, calibration reference data, historical temperature drift corrections, historical calibration reference deviations, and historical correction residuals at the same measurement moment are written to the same time record. This initial run generated 27 time records, one of which contained duplicate measurement times, and one contained null values ​​in the measurement output data. After removing these, 25 valid time records remain. Subsequently, the temperature data, measurement output data, calibration reference data, and correction record data in the valid time records are scaled and normalized to generate standard temperature measurement operation data. This standard temperature measurement operation data covers 08:00 to 10:00, a total of 25 measurement moments, providing a unified data foundation for subsequent time interval division and temperature drift correction.

[0029] After generating standard temperature measurement data, time intervals are generated by using two adjacent measurement times in the measurement time sequence as the start and end boundaries. The 25 measurement times form 24 time intervals, with the first interval from 08:00 to 08:05 and the 24th interval from 09:55 to 10:00. Following the measurement time sequence, the 24th time interval is written into the current time interval, and the first 23 intervals are written into the historical time interval. For each time interval, the corresponding temperature data, measurement output data, and measurement time data are combined according to the start and end boundaries to form a time interval temperature trajectory. In the current time interval, the temperature is 31.20℃ at 09:55 and 31.34℃ at 10:00. The starting measurement output is 100.018, and the ending measurement output is 100.026, forming a time interval temperature trajectory with a temperature change of 0.14℃ and a measurement output change of 0.008. Therefore, the correction process can revolve around the temperature and measurement output changes within the time interval, rather than compensating only for a single measurement time.

[0030] After generating the time interval temperature trajectory, the temperature data are arranged according to the measurement time sequence. The temperature difference, sampling time difference, and temperature change direction between adjacent measurement moments are calculated to generate an interval temperature change sequence. The current time interval corresponds to a temperature difference of 0.14℃, a sampling time difference of 5 minutes, and an upward temperature change direction. Further, a thermal memory fragment chain is generated based on the interval temperature change sequence. From this chain, temperature dwell characteristics, temperature change duration characteristics, temperature direction change characteristics, temperature plateau characteristics, interval blank characteristics, and temperature change rate characteristics are generated, and finally combined to form a thermal memory dwell fingerprint. In this operation, the current thermal memory dwell fingerprint includes a dwell duration of 0 minutes, an upward duration of 5 minutes, a downward duration of 0 minutes, 0 direction reversals, a plateau duration of 0 minutes, a cumulative interval blank time of 5 minutes, and an average temperature change rate of 0.028℃ / min, which can characterize the hysteresis effect of temperature rise on the measurement output within the current time interval.

[0031] To ensure the correction process inherits historical measurement patterns, standard temperature measurement data from previous continuous measurements of the equipment to be corrected are selected and combined with 23 historical time intervals provided by the current continuous measurement process to form a historical sample library. This library contains 300 historical time intervals, each generating historical endpoint measurement states, historical thermal memory fingerprints, and historical temperature drift difference sequences. The trajectory training samples are input from historical endpoint measurement states and historical thermal memory fingerprints, and output from historical temperature drift difference sequences, training a trajectory regressor. The correction quantity training samples are input from historical thermal memory fingerprints, historical temperature drift difference sequences, and correction record data, and output as historical temperature drift correction quantities, training a correction quantity regressor. Through this process, historical temperature drift correction quantities, historical calibration reference deviations, and historical correction residuals can participate in the generation of the current correction quantity, ensuring that the correction result is not only affected by the current temperature value but also reflects the continuity of historical residuals.

[0032] When applying the current time interval, the starting temperature data, starting measurement output data, and corresponding historical correction residuals are read from 09:55 to form the starting measurement status. From 10:00, the ending temperature data, ending measurement output data, and ending calibration reference data are read, and the difference between the ending measurement output data and the ending calibration reference data is written into the ending measurement status. The ending measurement output 100.026 is written into the current measurement output. Subsequently, a trajectory regressor is used to perform forward temperature drift extrapolation on the starting measurement status and the current thermal memory fingerprint, generating a forward temperature drift extrapolation trajectory. A trajectory regressor is also used to perform reverse temperature drift backtracking on the ending measurement status and the current thermal memory fingerprint, generating a reverse temperature drift backtracking trajectory. After mapping the two trajectories to the same time interval coordinates, a temperature drift suture region is generated, and further, temperature drift suture crack features are generated. In this operation, a positive temperature drift value of 0.032 and a negative temperature drift value of 0.026 were recorded at the 5-minute time interval coordinate, corresponding to a trajectory difference of 0.006. The temperature drift suture crack characteristics include a maximum trajectory difference of 0.009, a cumulative endpoint closure error of 0.015, an average adjacent coordinate slope difference of 0.004, and a maximum trajectory lag displacement of 1 sampling interval. This feature can reflect the closure difference between the initial extrapolation and the final retracement, providing more stable error information for correction prediction.

[0033] Simultaneously, based on the correction record data corresponding to the historical time intervals, the historical temperature drift correction, historical calibration reference deviation, and historical correction residuals are combined into interval residual nodes according to the measurement time sequence, and connected to form an interval residual migration chain. Based on the interval residual migration chain, the cumulative change, decay change, direction change, and abrupt change amplitude of the residuals are calculated to generate residual migration characteristics. In this operation, the residual sequences involved in the current correction are 0.010, 0.014, 0.011, and -0.006, corresponding to a cumulative residual change of 0.004, a decay change of 0.003, a direction change of 1, and an abrupt change amplitude of 0.017. Therefore, the current correction process can consider the trend of historical correction residuals propagating from the previous time interval to the current time interval, reducing the impact of continuous residual accumulation on the measurement output.

[0034] Finally, the current thermal memory fingerprint, temperature drift suture crack features, and residual migration features are combined according to the current time interval segment number to form the correction amount regression input feature. After reading this input feature, the correction amount regressor outputs the temperature drift correction amount of 0.031 corresponding to the current time interval segment, and performs drift cancellation processing on the current measurement output according to this temperature drift correction amount, correcting the current measurement output from 100.026 to 99.995. Continuous operation records show that before using this invention, the deviation between the device's end measurement output and the calibration reference reached 0.031, and there was cumulative transmission of residuals between consecutive time interval segments. After using this invention, the measurement output after correction of the current time interval segment is consistent with the calibration reference, the residual abrupt change amplitude between consecutive time interval segments is reduced from 0.017 to 0.006, and the measurement output fluctuation range of five consecutive subsequent measurement moments is reduced from 0.024 to 0.008. It can be seen that this invention can stably correct temperature drift and improve the consistency of measurement output in continuous measurement scenarios with slow temperature changes, continuous sampling intervals, and continuous migration of historical correction residuals.

[0035] Table 1 Comparison of performance improvements in time interval temperature drift correction

[0036] As shown in Table 1, under the uncorrected baseline, the absolute deviation at the end of the current time interval is 0.031, the output fluctuation range for five consecutive subsequent measurement times is 0.024, and the residual abrupt change amplitude is 0.017, indicating that the equipment exhibits significant temperature drift and residual propagation phenomena in the later stages of continuous measurement. Although the temperature lookup table compensation method can reduce the absolute deviation at the end to 0.020, achieving a reduction rate of 35.5%, it mainly compensates based on the current temperature value and cannot fully reflect the direction of temperature change, the rate of temperature change, and historical residual migration within the time interval. Therefore, the residual abrupt change reduction rate is only 17.6%.

[0037] The multivariate ordinary regression method outperforms the temperature lookup table compensation method overall, reducing the terminal absolute bias to 0.012 (a reduction of 61.3%) and the mean absolute corrected residual to 0.010 (a reduction of 61.5%). This indicates that incorporating temperature data, measurement output data, and correction record data into the regression process can improve the shortcomings of a single temperature compensation method. However, this method still belongs to direct regression correction and lacks forward temperature drift extrapolation, reverse temperature drift backtracking, and temperature drift suture crack characteristics. Therefore, its ability to identify differences in the closure of drift trajectories at both ends of the time interval is limited, and the residual abrupt change amplitude remains at 0.010.

[0038] The method of this invention reduces the terminal absolute deviation from 0.031 to 0.002, the output fluctuation range over five consecutive subsequent measurement times from 0.024 to 0.008, a reduction rate of 66.7%; the residual abrupt change amplitude from 0.017 to 0.006, a reduction rate of 64.7%; and the average absolute corrected residual from 0.026 to 0.004, a reduction rate of 84.6%. These data correspond to the current measurement output of 100.026, the temperature drift correction of 0.031, and the corrected measurement output of 99.995 in the embodiment, and there are no numerical conflicts.

[0039] The main reason for the performance improvement of this invention is that it does not directly generate the correction amount based solely on the current temperature value or multivariate input. Instead, it first generates a thermal memory residence fingerprint based on the temperature trajectory over a time interval, so that temperature residence, temperature change persistence, temperature direction change, interval gap, and temperature change rate all participate in temperature drift correction. At the same time, it generates temperature drift stitching crack features through the forward temperature drift extrapolation trajectory and the reverse temperature drift backtracking trajectory, enabling the correction amount regressor to identify the trajectory closure differences at both ends of the time interval. In addition, the interval residual migration chain associates the historical temperature drift correction amount, historical calibration reference deviation, and historical correction residual in chronological order, so that the cumulative change, decay change, and abrupt change amplitude of the historical residual participate in the current correction. Therefore, it can simultaneously reduce the end bias, subsequent fluctuations, and residual abrupt changes.

[0040] The above are merely preferred embodiments 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 time interval temperature drift correction method based on regression prediction, characterized in that, Includes the following steps: Collect basic temperature measurement data of the device to be corrected during continuous measurement, and preprocess the basic temperature measurement data to generate standard temperature measurement operation data; Based on the measurement time data in the standard temperature measurement operation data, time interval segments are divided, historical time interval segments and current time interval segments are divided, and time interval temperature trajectories are generated; Thermal memory fingerprints are generated based on time interval temperature trajectories, and historical and current thermal memory fingerprints are generated according to the time interval segments corresponding to the thermal memory fingerprints. Based on the standard temperature measurement operation data, historical thermal memory resident fingerprints and correction record data corresponding to the historical time intervals, a trajectory regressor and a correction amount regressor are trained and generated, and the measurement status of the interval endpoints and the current measurement output are generated. Combining the current thermal memory-residual fingerprint, a trajectory regressor is used to perform forward temperature drift extrapolation and reverse temperature drift backtracking on the starting and ending measurement states, respectively, generating forward temperature drift extrapolation trajectory and reverse temperature drift backtracking trajectory; Based on the measurement time data corresponding to the current time interval, time interval coordinates are generated. The forward temperature drift extrapolation trajectory and the reverse temperature drift backtracking trajectory are mapped to the time interval coordinates, and residual migration features are generated by combining the corrected record data. A correction regression model is used to regress and map the current thermal memory residence fingerprint, temperature drift suture crack features and residual migration features to generate the temperature drift correction amount corresponding to the current time interval, and then corrects it to generate the temperature drift correction result.

2. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the standard temperature measurement operation data includes: The operational data acquisition data and correction record data are time-aligned according to the measurement time data, so that the temperature data, measurement output data, calibration reference data, historical temperature drift correction, historical calibration reference deviation and historical correction residual corresponding to the same measurement time are recorded at the same time. Anomalies are sampled and removed from temperature data and measurement output data recorded at the same time, and the removed temperature data, measurement output data, measurement time data, calibration reference data and correction record data are retained. The removed temperature data, measurement output data, calibration reference data, and correction record data are scaled and normalized to generate normalized temperature measurement data. The measurement cycle is divided according to the measurement time data in the normalized temperature measurement data to generate standard temperature measurement operation data.

3. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the time interval temperature trajectory includes: The temperature data and measurement output data are arranged in ascending order of time according to the measurement time data in the standard temperature measurement operation data to generate a measurement time sequence. Time interval segments are generated by using two adjacent measurement times in the measurement time sequence as the start and end boundaries of the interval; The time interval segment at the end of the measurement time sequence is determined as the current time interval segment, and the time interval segments before the current time interval segment are determined as historical time interval segments. Based on the start and end boundaries of each time interval, the corresponding temperature data, measurement output data, and measurement time data are collected from the standard temperature measurement operation data to generate the time interval temperature trajectory.

4. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the historical hot memory fingerprint and the current hot memory fingerprint includes: Arrange the temperature data according to the measurement time data in the time interval temperature trajectory, calculate the temperature difference, sampling time difference and temperature change direction between adjacent measurement times, and generate an interval temperature change sequence; A thermal memory fragment chain is generated based on the interval temperature change sequence, and a thermal memory feature set is generated based on the thermal memory fragment chain. The thermal memory feature set is combined according to the measurement time sequence within the time interval to generate thermal memory residency fingerprint. The thermal memory residency fingerprint corresponding to the historical time interval is determined as the historical thermal memory residency fingerprint, and the thermal memory residency fingerprint corresponding to the current time interval is determined as the current thermal memory residency fingerprint.

5. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the interval endpoint measurement status and the current measurement output includes: Historical endpoint measurement status is generated based on the standard temperature measurement operation data and correction record data corresponding to the historical time interval. The difference is calculated based on the measurement output data and calibration reference data corresponding to each measurement time within the historical time interval, and the difference calculation results are arranged according to the measurement time data to generate a historical temperature drift difference sequence. Trajectory training samples are generated based on historical endpoint measurement status, historical thermal memory residency fingerprints, and historical temperature drift difference sequences, and the trajectory training samples are used to train and generate a trajectory regressor. Correction training samples are generated based on historical thermal memory fingerprints, historical temperature drift difference sequences, and correction record data, and a correction regressor is trained using the correction training samples. The interval endpoint measurement status and current measurement output are generated based on the standard temperature measurement operation data and correction record data corresponding to the current time interval.

6. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the reverse temperature drift backtracking trajectory includes: Read the measurement time data from the time interval temperature trajectory corresponding to the current time interval, generate a forward extrapolation sampling point sequence according to the increasing direction of measurement time, and generate a reverse backtracking sampling point sequence according to the decreasing direction of measurement time; Based on the forward inference sampling point sequence, the starting measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the forward inference input sequence. Based on the reverse backtracking sampling point sequence, the ending measurement state is combined with the current hot memory fingerprint sampling point by sampling point to generate the reverse backtracking input sequence. A trajectory regressor is used to perform a sampling-point regression on the forward extrapolation input sequence to generate a forward temperature drift state sequence corresponding to the forward extrapolation sampling point sequence. A trajectory regressor is also used to perform a sampling-point regression on the reverse backtracking input sequence to generate a reverse temperature drift state sequence corresponding to the reverse backtracking sampling point sequence. The forward temperature drift state sequence is arranged according to the forward extrapolation sampling point sequence to generate the forward temperature drift extrapolation trajectory, and the reverse temperature drift state sequence is arranged according to the reverse backtracking sampling point sequence to generate the reverse temperature drift backtracking trajectory.

7. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the residual migration features includes: Using the measurement start time of the current time interval as the coordinate starting point, time interval coordinates are generated according to the measurement time data corresponding to the current time interval. The forward temperature drift trajectory is mapped to the time interval coordinate according to the increasing direction of measurement time, and the reverse temperature drift back trajectory is rearranged and mapped to the time interval coordinate after being arranged according to the increasing direction of measurement time, thus generating a pair of temperature drift trajectories with the same coordinates. Based on the same coordinate temperature drift trajectory pair, calculate the trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement between the forward temperature drift extrapolation trajectory and the reverse temperature drift back trajectory at the same time interval coordinate position, and generate the temperature drift stitching area according to the time interval coordinate order. Based on the continuous positional correlation of trajectory difference, endpoint closure difference, adjacent coordinate slope difference and trajectory lag displacement in the temperature drift suture region, the temperature drift suture crack characteristics are generated. Based on the correction record data corresponding to the historical time intervals, the historical temperature drift correction, historical calibration reference deviation and historical correction residual are associated in the order of measurement time of the historical time intervals to generate the interval residual migration chain. Based on the interval residual migration chain, the residual cumulative change, residual decay change, residual direction change and residual abrupt change magnitude between adjacent historical time intervals are calculated to generate residual migration features.

8. The time interval temperature drift correction method based on regression prediction according to claim 1, characterized in that, The generation of the temperature drift correction result includes: The current thermal memory fingerprint, temperature drift suture crack features, and residual migration features are combined according to the current time interval to generate the correction amount regression input features; A correction regression analyzer is used to perform regression mapping on the correction regression input features to generate the temperature drift correction corresponding to the current time interval. The current measurement output is processed to cancel out drift based on the temperature drift correction amount, and a corrected measurement output is generated. Combine the current time interval, temperature drift correction amount, and corrected measurement output to generate the temperature drift correction result.