A liquid nitrogen cold energy recovery process temperature monitoring method and system

Abstract

This invention belongs to the field of cold energy recovery technology, specifically relating to a temperature monitoring method and system for liquid nitrogen cold energy recovery processes. The method includes: acquiring the temperature time series of heat exchange nodes; determining the number of decomposition layers and performing wavelet packet decomposition based on the sequence sample entropy; reconstructing a denoised temperature series based on the energy entropy of sub-band nodes; correcting the chi-square test critical threshold using a distribution skewness factor and shortening the sliding window length; determining the true independent allocation weights based on the distance from the sample to the median and the rate of change of the local temperature gradient, and constructing a weighted chi-square statistic; if the threshold is exceeded, performing secondary cross-validation using adjacent sliding windows and removing true outliers; determining the spline node density based on the local temperature gradient and introducing sub-band node energy entropy correction interpolation; updating the monitoring interval and triggering an alarm using the cold energy utilization rate index, historical fluctuation standard deviation, and dimensionless adjustment coefficient. This invention solves the problems of high false alarm rate and distorted energy efficiency indicators in traditional monitoring modes.

Description

Technical Field

[0001] This invention relates to the field of cold energy recovery technology, and more specifically, to a method and system for temperature monitoring in a liquid nitrogen cold energy recovery process. Background Technology

[0002] Liquid nitrogen cold energy recovery is an important technological direction in the field of industrial energy conservation, with applications in various industries such as food freezing, chemical production, superconducting technology, and biological sample preservation. In liquid nitrogen cold energy recovery systems, heat exchange nodes, as the core components for cold energy transfer and energy exchange, directly determine the efficiency of cold energy recovery and the safety and stability of system operation based on their temperature status. To ensure efficient and reliable system operation, it is necessary to monitor the temperature of heat exchange nodes in real time, continuously, and with high precision. This involves collecting temperature data to calculate key operating indicators such as cold energy utilization rate, and promptly identifying system anomalies based on changes in these indicators.

[0003] However, the liquid nitrogen cold energy recovery process exhibits significant non-stationarity and strong interference characteristics. The vaporization of liquid nitrogen generates violent phase change shocks, while system operation is accompanied by pressure fluctuations, fluid turbulence disturbances, and complex electromagnetic interference in industrial environments. These factors inevitably introduce a large amount of high-frequency noise into the raw data collected by temperature sensors, resulting in outliers that significantly deviate from the true temperature change trend. Failure to effectively identify and process these abnormal data will directly lead to distortions in the calculation of operational indicators such as cold energy utilization, resulting in false alarms or missed detection of genuine anomalies, severely impacting the stable operation and energy efficiency of the cold energy recovery device.

[0004] Existing technologies typically employ wavelet denoising combined with data cleaning to preprocess temperature sequences and calculate efficiency indicators based on the cleaned data. For example, Chinese patent application CN119803995A discloses a method for detecting the cold storage efficiency of a coil heat exchanger. This method acquires the inlet temperature signal, outlet temperature signal, and real-time flow signal of the cooling medium in real time, performs wavelet multi-scale decomposition denoising and analog-to-digital conversion on each signal, then cleans the acquired data, and finally determines the cold storage efficiency based on the cleaned data.

[0005] While the aforementioned existing technologies can achieve temperature data preprocessing and efficiency calculation to a certain extent, they have significant limitations when applied to liquid nitrogen cold energy recovery processes. The method employs a fixed-level wavelet multi-scale decomposition, which cannot adaptively adjust to changes in the complexity of temperature sequences under different operating conditions. This easily leads to problems such as insufficient denoising resulting in residual noise, or excessive denoising weakening the true temperature fluctuation characteristics. Furthermore, the method uses a simple data cleaning process, failing to consider the skewed distribution characteristics of data during liquid nitrogen heat exchange, and lacking a differentiated weighting mechanism based on the degree of sample deviation and the rate of change of local temperature gradients. This makes it difficult to accurately distinguish between outliers and normal operating condition fluctuations, easily leading to false positives and false negatives. In addition, the method does not perform targeted interpolation to complete missing data, nor does it dynamically adjust monitoring thresholds based on the historical fluctuation characteristics of efficiency indicators, failing to meet the requirements of liquid nitrogen cold energy recovery systems for high-precision, low-false-report intelligent monitoring. Summary of the Invention

[0006] To address the technical problems in existing technologies where the liquid nitrogen cold energy recovery process exhibits significant non-stationarity and strong interference characteristics, resulting in severe noise interference in temperature monitoring data and difficulty in identifying true outliers, this invention provides a temperature monitoring method and system for the liquid nitrogen cold energy recovery process.

[0007] In a first aspect, the present invention provides a temperature monitoring method for a liquid nitrogen cold energy recovery process, comprising: S1: acquiring a time series of temperatures at heat exchange nodes, determining the number of decomposition layers based on the entropy of the sequence samples for wavelet packet decomposition, and obtaining a denoised temperature sequence based on weighted reconstruction of sub-band node energy entropy; calculating the distribution skewness factor of the denoised temperature sequence within a sliding window to correct the chi-square test critical threshold, and shortening the sliding window length according to the degree to which the distribution skewness factor deviates from a preset benchmark value; S2: combining the distance from the temperature sample to the median and the rate of change of the local temperature gradient and globally normalizing the two to determine the weighted independent allocation weights, thereby constructing a weighted chi-square statistic of the temperature samples within the sliding window; when the weighted chi-square statistic of a suspicious sample exceeds the corrected value... When performing the chi-square test for the critical threshold, a second cross-validation is performed using adjacent sliding windows. The confidence level of the second cross-validation is adjusted according to the number of consecutive occurrences of the suspicious sample. When both validations exceed the limit, the true outlier is removed. S3: The local temperature gradient is used to determine the spline node density for preliminary interpolation, and the sub-band node energy entropy is used as the residual correction factor to correct the preliminary interpolation temperature to complete the sequence. The cold energy utilization rate index is calculated based on the completed sequence. The current actual flow rate change rate and the preset benchmark flow rate change rate are obtained. The dimensionless adjustment coefficient is obtained by normalizing the flow rate change rate with the preset benchmark flow rate change rate. The monitoring interval is updated by combining the historical fluctuation standard deviation of the cold energy utilization rate index with the dimensionless adjustment coefficient. An alarm is triggered when the cold energy utilization rate index exceeds the monitoring interval.

[0008] By adopting the above technical solution, this invention achieves precise management of non-stationary data in the liquid nitrogen cold energy recovery process by constructing a full-link monitoring logic from wavelet adaptive denoising, dynamic threshold recognition to residual feature compensation interpolation. Through the synergistic effect of sequence sample entropy and sub-band node energy entropy, the denoising parameters are adaptively adjusted according to the complexity of the operating conditions. Combined with the chi-square test and two-way verification mechanism for dynamic skewness factor correction, it can accurately extract random outliers from strong interference backgrounds and identify real operating condition jumps. This ensures the authenticity of the cold energy utilization rate calculation, improves the monitoring accuracy of the system under harsh operating conditions such as phase change shocks, and solves the problems of high false alarm rate and distorted energy efficiency index calculation in traditional monitoring methods.

[0009] Preferably, the step of calculating the distribution skewness factor of the denoised temperature sequence within the sliding window to correct the chi-square test critical threshold includes: sorting all temperature samples within the sliding window by value and extracting the median; calculating the absolute difference between all temperature samples within the sliding window and the median, and sorting all absolute differences to extract the absolute deviation of the median; calculating the standard deviation of all temperature samples within the sliding window; dividing the standard deviation by the absolute deviation of the median to obtain the distribution skewness factor; extracting a preset basic chi-square test critical threshold, using the ratio of the distribution skewness factor to a preset benchmark value as a threshold correction coefficient, and multiplying the basic chi-square test critical threshold by the threshold correction coefficient to obtain the corrected chi-square test critical threshold.

[0010] By employing the above technical solution, this invention uses the ratio of the standard deviation to the absolute deviation of the median to calculate the distribution skewness factor, enabling the critical threshold of the chi-square test to be dynamically scaled according to the real-time distribution shape of the monitoring data. This real-time tracking of the distribution pattern shifts the statistical discrimination benchmark from the ideal normal distribution assumption to a skewed distribution logic that fits the actual situation on site, improving the targeting of outlier identification during non-stationary operation phases.

[0011] Preferably, the step of determining the true independent allocation weight based on the distance from the temperature sample to the median and the local temperature gradient change rate, combined with global normalization, includes: calculating the absolute difference between each temperature sample value within the sliding window and the median; adding a minimum normal number to the absolute difference and taking the reciprocal as the first weight base; calculating the temperature difference between the temperature sample and the previous adjacent temperature sample in the time series; dividing the temperature difference by the system sampling time interval to obtain the local temperature gradient change rate; adding a minimum normal number to the absolute value of the local temperature gradient change rate and taking the reciprocal as the second weight base; multiplying the first weight base and the second weight base point by point to obtain the joint initial weight; and dividing the joint initial weight of each temperature sample within the sliding window by the sum of the joint initial weights of all temperature samples within the sliding window to perform global normalization processing to obtain the true independent allocation weight of each temperature sample.

[0012] Preferably, constructing the weighted chi-square statistic for temperature samples within the sliding window includes: using the true independent allocation weights to perform a weighted summation of the temperature samples within the sliding window to obtain a weighted mean; using the true independent allocation weights to calculate the sum of squares of the differences between the temperature samples and the weighted mean to obtain a weighted variance; and based on the weighted mean and the weighted variance, calculating the ratio of the squared deviation of each temperature sample relative to the weighted mean to the weighted variance to construct the weighted chi-square statistic.

[0013] Preferably, the secondary cross-validation using adjacent preceding and following sliding windows includes: when the weighted chi-square statistic of a suspicious sample is greater than the corrected chi-square test threshold, extracting the adjacent previous historical sliding window based on the timestamp corresponding to the suspicious sample, incorporating the suspicious sample data into the previous historical sliding window to recalculate the forward weighted chi-square statistic; simultaneously extracting the adjacent next future sliding window immediately following the suspicious sample, incorporating the suspicious sample data into the next future sliding window to recalculate the backward weighted chi-square statistic.

[0014] Preferably, the confidence level of the secondary cross-validation is adjusted according to the number of consecutive occurrences of the suspicious sample. When both validations exceed the limit, the true outlier is removed. This includes: recording the number of consecutive occurrences of the suspicious sample exceeding the limit; multiplying the number of occurrences by a preset step size coefficient and adding it to the initial confidence level; and truncating the upper limit of the calculation result within the effective interval to adjust the confidence level; calculating the increased validation threshold using the inverse chi-square distribution function based on the adjusted confidence level; determining whether both the forward-weighted chi-square statistic and the backward-weighted chi-square statistic are greater than the increased validation threshold; if both validations exceed the limit, the timestamp and value corresponding to the true outlier are removed; if at least one does not exceed the limit, it is retained.

[0015] By adopting the above technical solution, this invention utilizes the inverse chi-square distribution function to increase the verification threshold of repeated anomalies in real time, maintaining high-intensity detection while leaving a reasonable tolerance space for normal operating condition steps, balancing recognition accuracy and system stability, and avoiding mass false deletions.

[0016] Preferably, the step of determining the spline node density using the local temperature gradient for preliminary interpolation, and correcting the preliminary interpolation temperature using the sub-band node energy entropy as a residual correction factor to complete the sequence includes: extracting the absolute value of the local temperature gradient of adjacent retained points before and after the location of the removed outlier; determining the spline node density of the completed interpolation interval based on the direct proportional mapping relationship between the absolute value of the local temperature gradient and the spline node density; establishing a cubic spline mapping function using at least four adjacent retained points before and after the location of the removed outlier; calculating the preliminary interpolation temperature at the determined spline nodes; extracting the node energy entropy of the sub-band corresponding to the time period of the denoising stage; multiplying the node energy entropy by the sign of the local temperature gradient at the location of the removed outlier and a preset constant as a residual correction factor; and adding the preliminary interpolation temperature and the residual correction factor for fine-tuning and correction to obtain the corrected interpolation result as the completed sequence.

[0017] By adopting the above technical solution, this invention compensates for the local fluctuations of the spline curve by utilizing the useful energy features filtered out in the noise reduction stage, restores the temperature change details that were masked by traditional smoothing interpolation, and makes the completed sequence conform to the kinetic characteristics of liquid nitrogen heat transfer, thus ensuring the data fidelity of subsequent energy efficiency analysis.

[0018] Preferably, the step of calculating the cold energy utilization rate index based on the completed sequence, obtaining the current actual flow rate change rate and the preset benchmark flow rate change rate, and normalizing the flow rate change rate to the preset benchmark flow rate change rate to obtain a dimensionless adjustment coefficient includes: synchronously acquiring the corresponding pipeline pressure parameters, extracting the corresponding absolute temperature based on the completed sequence, and calculating the ratio of the actual recovered enthalpy difference to the theoretical maximum recoverable enthalpy difference of the cold energy recovery heat exchange node as the cold energy utilization rate index in combination with the pipeline pressure parameters; calculating the average value and historical fluctuation standard deviation of the cold energy utilization rate index within the historical time window; acquiring the current actual flow rate change rate, and normalizing it relative to the preset benchmark flow rate change rate to obtain a dimensionless adjustment coefficient.

[0019] Preferably, the step of updating the monitoring interval by combining the historical fluctuation standard deviation of the cold energy utilization rate index with the dimensionless adjustment coefficient, and triggering an alarm when the cold energy utilization rate index exceeds the monitoring interval, includes: adding the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient to the average value of the cold energy utilization rate index within the historical time window to obtain the upper monitoring limit; subtracting the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient from the average value to obtain the lower monitoring limit; constructing a dynamic monitoring interval from the upper and lower monitoring limits; determining whether the cold energy utilization rate index at the current moment exceeds the monitoring interval; and triggering an alarm if the cold energy utilization rate index at the current moment is greater than the upper monitoring limit or less than the lower monitoring limit.

[0020] By adopting the above technical solution, the present invention tightens the boundary to improve sensitivity when the system is stable, and loosens the tolerance to eliminate interference when the operating conditions change. This dynamic boundary mechanism eliminates false alarms while maintaining efficient monitoring.

[0021] Secondly, the present invention provides a temperature monitoring system for a liquid nitrogen cold energy recovery process, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned method for monitoring the temperature of a liquid nitrogen cold energy recovery process is implemented.

[0022] By adopting the above technical solution, a computer program is generated for the temperature monitoring method of liquid nitrogen cold energy recovery process, and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0023] The technical solution of the present invention has the following beneficial technical effects:

[0024] This invention employs wavelet packet decomposition to determine the number of decomposition levels based on sequence sample entropy, and performs weighted reconstruction based on sub-band node energy entropy. This effectively filters out high-frequency environmental noise while fully preserving the local fluctuation characteristics of the true temperature. By extracting the distribution skewness factor of the denoised temperature sequence within a sliding window, the critical threshold of the basic chi-square test is corrected. The sliding window length is dynamically adjusted according to the degree to which the distribution skewness factor deviates from the preset benchmark value. Combined with a weighted chi-square statistic constructed based on the distance from the sample to the median and the rate of change of the local temperature gradient, and a secondary cross-validation mechanism that flexibly adjusts the confidence level based on the number of consecutive occurrences, the accuracy of identifying sudden outliers under complex operating conditions is improved, effectively avoiding the risk of misjudgment caused by normal fluctuations in operating conditions. Interpolation is performed using spline node density controlled by the local temperature gradient, and a residual correction factor is used to reduce the bias during the missing data completion process, ensuring the fidelity and completeness of the temperature time series. Based on this, the cold energy utilization rate index is calculated using the completed sequence, and a real-time updated dynamic monitoring interval is constructed by combining historical fluctuation characteristics, ultimately providing technical assurance for the safe and stable operation of the liquid nitrogen cold energy recovery process. Attached Figure Description

[0025] Figure 1 This is a flowchart of a temperature monitoring method for a liquid nitrogen cold energy recovery process according to an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the denoised temperature sequence with the original temperature in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the distribution of weighted chi-square statistics and multi-level critical thresholds in an embodiment of the present invention; Figure 4 This is a schematic diagram comparing the ablation experiment results of different detection schemes in the embodiments of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0027] This invention discloses a temperature monitoring method for a liquid nitrogen cold energy recovery process, referring to... Figure 1 This includes steps S1-S3: S1: Temperature sequence adaptive noise reduction and sliding window dynamic adjustment.

[0028] It should be noted that, in order to avoid the high-frequency abrupt changes caused by the liquid nitrogen vaporization phase change impact and the turbulent disturbance of the pipeline fluid from masking the true temperature gradual change characteristics, thereby causing processor operation blockage or front-end temperature measurement unit crash, this step uses wavelet packet energy weighted reconstruction combined with an adaptive window adjustment mechanism based on the distribution skewness factor to remove physical noise and achieve accurate characterization of the underlying temporal state.

[0029] Preferably, as an example, the heat exchange node temperature time series is obtained; the number of decomposition levels is determined based on the sequence sample entropy for wavelet packet decomposition; and a denoised temperature series is obtained by weighted reconstruction based on the energy entropy of the sub-band nodes; the distribution skewness factor of the denoised temperature series within the sliding window is calculated to correct the chi-square test critical threshold; and the sliding window length is shortened according to the degree to which the distribution skewness factor deviates from the preset benchmark value, including: First, the data acquisition device collects temperature data from the liquid nitrogen cold energy recovery heat exchange node in real time using sensors, and converts the analog signals into digital signals, which are then input into the processor to form a continuous temperature time series. For example, the data acquisition device uses a PT100 platinum resistance temperature sensor to collect the outlet temperature data of the heat exchanger in real time at a sampling frequency of 1Hz, which is then processed by the lower-level computer via an analog-to-digital conversion module. Next, the processor sets the embedding dimension and similarity tolerance to calculate the sequence sample entropy of the temperature time series. The sample entropy is divided by a preset logarithmic basis coefficient and rounded down to limit the result to a preset upper and lower limit range to obtain the target wavelet packet decomposition layer number. Assuming the embedding dimension is set to 2 and the similarity tolerance is set to 0.2 times the sequence standard deviation, the preset logarithmic basis coefficient is in the range of 0.3 to 0.5. In this embodiment, the preset logarithmic basis coefficient is preferably 0.4, and the corresponding preset upper and lower limit range is between 3 and 6 layers. If the initial result after division is 4.2 and the result after rounding down is 4, then the final target wavelet packet decomposition layer number limited and output is 4 layers. By limiting the upper and lower limits mentioned above, over-decomposition or under-decomposition of the temperature time series is prevented.

[0030] Subsequently, an orthogonal wavelet basis is used to perform wavelet packet decomposition on the temperature time series at the corresponding target wavelet packet decomposition level. Wavelet coefficient sequences of each sub-band in the bottom layer are extracted, the squared energy of each wavelet coefficient within the sub-band is calculated, and the node energy entropy is calculated based on the proportion of energy within the sub-band. For example, the orthogonal wavelet basis chosen is the db4 wavelet basis or the sym8 wavelet basis, which have good compact support characteristics. Specifically, the corresponding wavelet coefficient sequences satisfy the following relationship:

[0031] In the formula, Characterizing the energy percentage, Characterize wavelet coefficients, where the set of wavelet coefficients is constrained to satisfy... .

[0032] Therefore, the corresponding frequency domain energy distribution state satisfies the following relationship:

[0033] In the formula, Characterizes the energy entropy of a node, where a given energy percentage satisfies .

[0034] Next, the proportion of the reciprocal of the energy entropy of each sub-band node in the sum of all energy entropy reciprocals is calculated, and this proportion is normalized as the denoising weight for the corresponding sub-band. This denoising weight is then multiplied point-by-point with the wavelet coefficients of the corresponding sub-band to perform coefficient scaling. The scaled coefficients are then used to perform inverse wavelet packet transform reconstruction, combined with... Figure 2 This allows for the generation of denoised temperature sequences that filter out high-frequency background noise while retaining the characteristics of local cold energy temperature abrupt changes. For example, the normalization operation strictly limits the range of denoising weights to 0 to 1. This normalized denoising weight mechanism effectively enhances the characteristics of low-entropy subbands containing useful signals and suppresses high-entropy subbands with high noise.

[0035] Further, a sliding window is initialized and shifted step-by-step along the time axis of the denoised temperature time series. Within the current sliding window, all temperature samples are sorted by numerical value to extract the median. The absolute difference between all temperature samples within the window and the median is calculated, and all absolute differences are sorted to extract the median absolute deviation. For example, the initial length of the sliding window contains 50 to 100 sampling points. Next, the standard deviation of all temperature samples within the same sliding window is calculated. To avoid division by zero, the standard deviation is divided by the median absolute deviation, provided the median absolute deviation is greater than zero, to calculate the distribution skewness factor. Then, a preset basic chi-square test critical threshold is extracted. The ratio of the distribution skewness factor to the preset benchmark value is calculated as the threshold correction coefficient. The basic chi-square test critical threshold is multiplied by the threshold correction coefficient for scaling adjustment, thereby outputting the corrected chi-square test critical threshold. Specifically, under ideal high-frequency white noise normal distribution conditions, the preset benchmark value is fixed at a constant of 1.4826. Based on an initial significance level of 0.05 and a physical setting of 1 degree of freedom, the system calculates the static basic chi-square test critical threshold by referring to a table according to the upper-side test quantile rule of the chi-square distribution. Using this distribution skewness factor for correction reduces the misclassification rate of true outliers caused by the non-stationarity of the liquid nitrogen heat exchange process data.

[0036] Finally, the absolute value of the difference between the distribution skewness factor and the preset benchmark value is calculated and used as a variable indicating the degree of high skewness in the data distribution. Next, the system's preset maximum initial window length is obtained as the benchmark boundary value. The high skewness variable is multiplied by a preset adjustment coefficient to obtain a penalty length, which is then subtracted from the maximum initial window length. The calculation result is rounded down, and the rounded result is compared with the preset minimum window length to extract the maximum value. A slicing operation is then used to truncate the original sequence to generate a shortened sliding window length. For example, in actual production line applications, window adjustment is triggered when the absolute value is greater than 0.5. The maximum initial window length is set to 120 sampling points, the preset minimum window length is 30 sampling points, and the preset adjustment coefficient is between 10 and 25. In this embodiment, the preset adjustment coefficient is preferably 15.5. Assuming the absolute difference triggers a threshold, the window adaptive calculation process is completed by subtracting the rounded result of this difference multiplied by 15.5 from 120.

[0037] It should be added that the above-mentioned procedure for obtaining the pre-set logarithmic basis coefficients includes: introducing a constant reference heat source for excitation during the factory no-load calibration stage of the liquid nitrogen system and applying a sinusoidal sweep frequency electromagnetic disturbance of rated power; continuously recording no less than 500 sets of temperature sequence sample entropy extreme values ​​parsed by the hardware processor; and then performing a quotient operation between the median of the extreme value set and the maximum number of available decomposition layers supported by the system main control board to reverse-solidify and extract the logarithmic basis coefficients.

[0038] It should be added that the above-mentioned procedures for obtaining the embedding dimension and similarity tolerance include: capturing the complete temperature decay physical waveform of a single liquid nitrogen valve operation cycle from fully open to fully closed using a high-frequency oscilloscope, extracting the time taken for the waveform's autocorrelation function to first cross zero as the background time delay parameter, converting it into the current discrete number of sensor sampling points to calibrate the embedding dimension, and multiplying the standard deviation of the temperature amplitude of the stable segment of the waveform by the sensor's maximum permissible tolerance limit to output the similarity tolerance.

[0039] It should be added that the above-mentioned procedure for obtaining the preset adjustment coefficient includes: under the standard operating conditions of maintaining an absolutely constant pipeline flow and no obvious wind load vibration interference in the industrial site, the adaptive optimization logic of the microcontroller is forcibly shielded to maintain the operation of the minimum observation window and the frequency of high-frequency false alarm pulses at this time is counted. With the convergence of this frequency to one ten-thousandth as the iteration target, the value that minimizes the global error is locked in the constant range of 10 to 25 by the hardware-in-the-loop simulation system as the final adjustment coefficient.

[0040] In this way, by extracting node energy entropy, the underlying time-series signal can be effectively preserved and noise suppressed. The sliding window length and static judgment boundary can be dynamically adjusted based on the skewed distribution evolution law. This reduces the masking effect of drastic phase transitions and pipeline jitter on the real weak temperature gradual change law, and provides a high signal-to-noise ratio basic data support base for high-precision statistical real outlier identification under subsequent complex working conditions.

[0041] S2: Construct weighted statistics and cross-validate to remove true outliers.

[0042] It should be noted that, in order to avoid the failure of the system to falsely report shutdown due to the inability of a single static statistical quantity to accurately distinguish between true outliers and normal thermal shock caused by complex electromagnetic interference and sudden changes in liquid nitrogen turbulence in industrial sites, this step implements the true outlier stripping operation for stability by using a dual-base weighting mechanism based on temperature distance and gradient characteristics in conjunction with time-series causal cross-validation.

[0043] Preferably, as an example, the weighted chi-square statistic of temperature samples within a sliding window is constructed by combining the distance from the temperature sample to the median and the rate of change of the local temperature gradient, and then globally normalizing both to determine the weighted independent values. When the weighted chi-square statistic of a suspected sample exceeds the modified chi-square test threshold, a second cross-validation is performed using adjacent sliding windows. The confidence level of the second cross-validation is adjusted according to the number of consecutive occurrences of the suspected sample. When both validations exceed the limit, the true outlier is removed, including: First, the processor calculates the absolute difference between each temperature sample value within the sliding window and the median. It then adds a minimum positive constant to this absolute difference and takes its reciprocal, using this reciprocal as the first weighting base. For example, in the underlying mathematical logic, to avoid the denominator becoming invalid due to a zero absolute difference, the minimum positive constant added to the denominator is represented as... Next, the temperature difference between the temperature sample and its preceding adjacent temperature sample in the time series is calculated. This temperature difference is divided by a fixed system sampling time interval to obtain the local temperature gradient rate of change. The absolute value of the local temperature gradient rate of change is added to the same minimum normal number, and the reciprocal is taken as the second weighting base. Under this setting, the first weighting base assigns high weights to stable samples close to the overall median, and the second weighting base assigns high weights to samples with smooth temperature changes and no abrupt changes. Subsequently, the first weighting base and the second weighting base are multiplied point by point to obtain the joint initial weight. The joint initial weight of each temperature sample within the window is then divided by the sum of the joint initial weights of all temperature samples in the entire window to perform global normalization, resulting in the true independent weight assigned to each temperature sample.

[0044] Specifically, the truly independent weight allocation satisfies the following relation:

[0045] In the formula, Characterizing the first within the sliding window Each temperature sample has a truly independent weighting; The total number of samples representing the sliding window is the window length; This represents the first weighting cardinality corresponding to the sample; This represents the second weight base corresponding to the sample. Through this global normalization process, it is ensured that the sum of the weights of all samples within the window is always equal to 1.

[0046] Furthermore, by using real independent weights instead of equal weights to perform weighted summation on the temperature samples within the sliding window, a weighted mean that can characterize the background benchmark is obtained. Specifically, the weighted mean satisfies the following relationship:

[0047] In the formula, Represents the weighted mean; The total number of samples representing the sliding window is the window length; Characterizing the first within the sliding window Each temperature sample has a truly independent weighting; Characterizing the first within the sliding window The value of each temperature sample.

[0048] The weighted variance, which characterizes the degree of local dispersion of the data, is calculated by summing the squares of the weighted mean values ​​of each temperature sample value and multiplying them by the corresponding true independent allocation weights.

[0049] Specifically, the weighted variance satisfies the following relationship:

[0050] In the formula, Represents the weighted mean; Characterizes the weighted variance; The total number of samples representing the sliding window is the window length; Characterizing the first within the sliding window The value of each temperature sample; Characterizing the first within the sliding window Each temperature sample is assigned a truly independent weight.

[0051] Next, based on the weighted mean and weighted variance, to ensure the validity of the statistical calculation, when the weighted variance is greater than zero, the value of the current detected sample is extracted, the weighted mean is subtracted, the square is calculated, and then divided by the weighted variance. This calculates the ratio of the squared deviation of each temperature sample relative to the weighted mean to the weighted variance, thus constructing the weighted chi-square statistic. The distribution diagram of this statistic and the multi-level critical threshold is shown in the reference diagram. Figure 3 .

[0052] Specifically, the weighted chi-square statistic satisfies the following relationship:

[0053] In the formula, Characterized by the weighted chi-square statistic; The temperature value representing the current suspected sample to be tested; Characterizes the weighted variance; It represents the weighted mean.

[0054] Next, the weighted chi-square statistic of the current sample is compared with the aforementioned modified chi-square test critical threshold. When the weighted chi-square statistic of the suspected sample calculated from a certain temperature sampling point is greater than the modified chi-square test critical threshold, the secondary cross-validation process is triggered. Specifically, the previous historical sliding window adjacent to the suspected sample's timestamp is extracted, and the suspected sample data is incorporated into the dataset of the previous historical sliding window as an incremental test point for recalculation to obtain the forward weighted chi-square statistic. Simultaneously, the next adjacent future sliding window sequence immediately following the suspected sample is extracted, and the suspected sample data is also incorporated into the next future sliding window for recalculation to obtain the backward weighted chi-square statistic.

[0055] Subsequently, an internal counter records the number of times a suspicious sample exceeds the revised chi-square test threshold consecutively within the current detection period. This number is multiplied by a preset step size coefficient and added to the initial confidence level. The calculated result is then truncated to an upper limit between 0 and 1 to update the confidence level. Based on this updated confidence level, the increased verification threshold is calculated using the inverse chi-square distribution function. Specifically, the initial confidence level for secondary verification is set to 95%, and the preset step size coefficient is set to 0.01. Each time consecutive suspicious samples appear, the basic initial confidence level is added to the adjustment value obtained by multiplying the number of consecutive occurrences by 0.01 to obtain the adjusted new confidence level.

[0056] Finally, it is determined whether both the forward-weighted chi-square statistic and the backward-weighted chi-square statistic are greater than the adjusted verification threshold. If both verifications exceed the threshold, the cause is confirmed to be non-operating condition-related, the suspicious sample is identified as a true outlier, and the timestamp and value corresponding to the true outlier are removed from the temperature time series. If at least one of the forward-weighted chi-square statistic or the backward-weighted chi-square statistic does not exceed the adjusted verification threshold, the point is considered to be a compliant operating condition disturbance such as thermal shock in the cooling system, the true outlier label is removed, and it is determined to be an operating condition disturbance and retained for subsequent sequences.

[0057] It should be added that the above-mentioned procedure for obtaining the smallest normal number includes: reading the resolution configuration word of the underlying analog-to-digital converter of the hardware system and the physical temperature conversion constant corresponding to the least significant bit of the front-end temperature transmitter, multiplying the two and introducing a safety margin correction factor of 0.5 to strictly define the hardware physical bottom line threshold to ensure that the division operation does not cause denominator failure.

[0058] It should be added that the above-mentioned procedure for obtaining the initial confidence level includes: collecting a historical data set of the liquid nitrogen heat exchange system under fault-free and stable operation for a period of no less than six months; conducting basic chi-square test backtesting on the extracted local stationary segment sequence; recording the maximum statistical boundary percentile constant that prevents the real normal jump sample from being mistakenly intercepted; and directly sending it to the microcontroller's static memory for solidification.

[0059] It should be added that the above-mentioned procedure for obtaining the preset step size coefficient includes: on a manually set fixed frequency abnormal pulse injection test platform, the step size variable is dynamically adjusted by gradually increasing the microcontroller simulation interface, recording the critical occurrence value of each time that the normal disturbance of the real working condition is mistakenly deleted as a real outlier due to premature saturation of the confidence level, and calculating the statistical median, thereby establishing the benchmark parameters amplified by the underlying verification anti-shake mechanism through order reduction mapping.

[0060] Thus, by integrating local temperature deviation distance and gradient evolution characteristics to construct an adaptive weighted statistic, and combining it with a secondary verification and correction mechanism that combines the causal relationship between the preceding and following time series, the system can sensitively filter out transient electromagnetic spikes and outlier disordered data nodes while successfully avoiding the detection cutoff disaster caused by thermal shock of the compliant system, effectively consolidating the continuity guarantee of the data flow link for thermodynamic time-domain modeling of liquid nitrogen pipeline network.

[0061] S3: Sequence residual interpolation completion and dynamic energy efficiency monitoring alarm.

[0062] It should be noted that, in order to avoid the distortion of downstream thermodynamic index calculations caused by the time-series discrepancy faults resulting from the removal of true outliers, and the objective physical failure consequences of false alarms induced by static monitoring thresholds under variable load conditions, this step uses gradient-driven cubic spline adaptive interpolation combined with a dynamic interval reconstruction mechanism based on fluctuation characteristics to ensure the continuity of efficiency assessment and the reliability of alarm triggering.

[0063] Preferably, as an example, the spline node density is determined using a local temperature gradient for preliminary interpolation, and the sub-band node energy entropy is used as a residual correction factor to correct the preliminary interpolation temperature to complete the sequence; the cold energy utilization rate index is calculated based on the completed sequence, the current actual flow rate change rate and the preset benchmark flow rate change rate are obtained, and the dimensionless adjustment coefficient is obtained by normalization with the preset benchmark flow rate change rate. The monitoring interval is updated by combining the historical fluctuation standard deviation of the cold energy utilization rate index with the dimensionless adjustment coefficient, and an alarm is triggered when the cold energy utilization rate index exceeds the monitoring interval, including: First, the processor extracts the absolute values ​​of the local temperature gradients of adjacent retained points before and after the location of the removed outlier. Based on the direct proportionality between the absolute values ​​of the local temperature gradients and the spline node density, the spline node density for completing the interpolation interval is determined. For example, in actual time-axis mesh generation, when the temperature gradient is below 0.5℃ / s, the default number of mapped nodes is 2, and adjacent retained sampling points with a time interval of no more than 5s are selected as spline nodes. When the gradient rises to 2℃ / s or higher in the drastic temperature change range, the spline node density is mapped and increased to 4 to 6, with adjacent retained sampling points with a time interval of no more than 1s preferentially selected as spline nodes. This establishes a non-uniform spline node set, ensuring sufficient information anchor points to constrain the direction of the interpolation curve in the high-frequency temperature change range. Next, a cubic spline mapping function is established using at least four adjacent retained points before and after the location of the removed outlier. The system of equations is solved at the determined spline nodes to calculate the preliminary interpolation temperature. Subsequently, the node energy entropy of the sub-band corresponding to the denoising stage is extracted from memory. The node energy entropy is multiplied by the sign of the local temperature gradient at the location of the removed outlier and a preset constant with temperature dimension conversion function as the residual correction factor. This residual correction factor is specifically used to compensate for the weakening of local temperature fluctuation characteristics caused by conventional cubic spline interpolation. Then, the initial interpolated temperature is added to the residual correction factor for single-step local fluctuation consistency fine-tuning correction. The corrected result is written to the memory block at the missing position of the original time series to complete the completion, thus obtaining the corrected interpolation result. This completes the closed-loop correction interpolation completion that takes into account both macroscopic smoothness and local fluctuation characteristics.

[0064] Specifically, the residual correction factor satisfies the following relationship:

[0065] In the formula, Characterizes the residual correction factor; The node energy entropy characterizing the corresponding sub-band; The sign function represents the local temperature gradient at the location of the removed true outlier, taking a positive sign when the temperature is increasing and a negative sign when the temperature is decreasing. A preset constant characterizing the temperature dimension conversion function is preferably 0.2℃ in this embodiment.

[0066] Furthermore, based on the completed and corrected continuous temperature time series, the corresponding absolute temperature and pipeline pressure parameters are extracted. The inlet state specific enthalpy, outlet state specific enthalpy, and the specific enthalpy corresponding to the theoretical maximum recovery final state are queried or calculated. Under the premise that the difference between the specific enthalpy corresponding to the theoretical maximum recovery final state and the inlet state specific enthalpy is greater than zero, the ratio of the actual recovery enthalpy difference of the cold energy recovery heat exchange node to the theoretical maximum recoverable enthalpy difference is calculated as the cold energy utilization rate index.

[0067] Specifically, the corresponding cold energy utilization rate index satisfies the following relationship:

[0068] In the formula, Indicators characterizing cold energy utilization rate Specific enthalpy characterizing the export state Enthalpy characterizing the state of import. Characterize the enthalpy corresponding to the final state of maximum recovery in the theory.

[0069] Next, the average value and historical standard deviation of the cold energy utilization rate index within the historical time window are calculated, and the actual flow rate change rate of the current flow meter output is obtained. This rate is then normalized relative to a preset benchmark flow rate change rate to obtain a dimensionless adjustment coefficient. This dimensionless adjustment coefficient is introduced to effectively track the efficiency safety boundary under different operating load conditions. For example, the system establishes a shifted historical time window with a length of 200 sampling points, and then iteratively calculates the historical standard deviation and historical average value of the past 200 cold energy utilization rate indices.

[0070] Specifically, the dimensionless adjustment coefficient satisfies the following relationship:

[0071] in, This represents a dimensionless adjustment factor; This represents the current actual rate of change in flow rate; This is the preset baseline flow rate change rate.

[0072] Finally, the upper limit of monitoring is obtained by adding the average value of the cold energy utilization rate index within the historical time window to the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient. At the same time, the lower limit of monitoring is obtained by subtracting the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient from the average value. This constructs a closed dynamic monitoring interval consisting of the upper and lower limits, which gives the monitoring interval a strong self-adaptive capability to operating conditions and effectively overcomes the problem that traditional static settings cannot adapt to fluctuations.

[0073] Specifically, the boundary calculation of this monitoring interval satisfies the following relationship:

[0074]

[0075] In the formula, Characterizes the upper limit of monitoring; Characterizes the lower limit of monitoring; The average value of the cold energy utilization rate index within a historical time window; Characterizing the standard deviation of historical fluctuations; This represents a dimensionless adjustment factor.

[0076] Next, it checks whether the current energy efficiency index exceeds the monitoring range defined by the upper and lower limits. If the current energy efficiency index exceeds the upper limit or falls below the lower limit (i.e., exceeds the monitoring range), an alarm is triggered, and a risk of abnormal temperature is indicated. Specifically, when the index exceeds the limit, the processor outputs a high-level signal to the alarm pin of the Raspberry Pi control board, driving an external relay to close and execute the alarm action. If the energy efficiency index is within the monitoring range, the system is determined to be operating within the normal energy efficiency tolerance range, and the monitoring process continues for the next cycle.

[0077] In the experimental setup, the processor selected a specific sequence of temperature data points as the base dataset, and artificially injected random high-frequency true outliers and continuously fluctuating pseudo-true outliers as test samples. For example, the processor selected 100,000 continuously collected temperature data points from a liquid nitrogen cooling heat exchange system in a cold storage facility as the base dataset, and artificially injected 500 random high-frequency true outliers and 200 continuously fluctuating pseudo-true outliers as test samples. The hardware environment for performing data processing operations used an industrial control computer, specifically, an industrial control computer containing a specific model processor and 16GB of memory.

[0078] It should be added that the calibration procedure for the gradient threshold of the spline mapping mesh division mentioned above includes: cyclically impacting the temperature probe on a bidirectional rapid cooling and heating destructive test bench from room temperature water bath to deep-freezing liquid nitrogen, using a high-frequency data logger to capture the critical physical heat flux slope of the sensor's temperature sensing element as it moves away from the linear conduction region and approaches the response saturation dead zone, and then reducing this physical measurement slope by 30% as the high-frequency dense mesh activation judgment node.

[0079] It should be added that the calibration procedure for the above-mentioned preset constant includes: measuring the energy entropy substrate distribution of the underlying amplifier circuit itself excited by semiconductor thermal noise in an ultra-low temperature constant temperature bath device where the pump is stopped and stationary to ensure no passive fluctuations; performing first-order least squares fitting on the dimensionless distribution curve and the difference between the substrate thermal drift physical temperature measured by the probe platinum resistance to obtain the corresponding slope as the constant proportional factor for the physical dimension conversion.

[0080] It should be added that the calibration procedure for the above-mentioned preset benchmark flow rate change rate includes: the on-site process personnel adjust the main control nitrogen supply regulating valve to carry out repeated climb tests within the maximum variable load safety slope allowed by the safety production manual, and extract the maximum permissible value of the highest per unit time flow rate increase that will not cause abnormal vibration and whistling in the downstream heat exchange finned pipe network from the continuously accumulated feedback messages of the mass flow meter as the benchmark anchoring parameter.

[0081] It should be added that the above-mentioned procedure for obtaining the length of the historical time window includes: injecting an artificially constructed micro-perturbation heat source within the standard liquid nitrogen supply full-load stable cycle, starting the high-frequency timer counter inside the underlying microcontroller to monitor the exact physical hysteresis clock cycles required for the heat to diffuse completely from the heat exchanger inlet to the outlet and cause the temperature to fall back and stabilize, and converting it into the number of discrete sequences under the current sampling interval of the sensor as the window size.

[0082] Thus, by combining interpolation reconstruction under environmental gradient constraints with the physical residual recovery mechanism of the underlying high-frequency perturbation energy, the original physical characteristics of local hydrodynamic fluctuations in the pipeline network are resolutely defended while making up for the data vacuum derived from the real outlier removal operation. Furthermore, the terminal thermodynamic efficiency monitoring range is endowed with strong operating condition adaptive vitality, eliminating the chain of false alarms caused by the switching of operating loads, and realizing high-safety-level intelligent management and maintenance of the cold energy recovery cluster in a closed loop.

[0083] The processor divided the experiment into three groups for rigorous ablation comparison verification.

[0084] The first group serves as the baseline control group. Its execution logic uses a fixed sliding window length and a static basic chi-square test threshold to detect true outliers, and combines this with standard cubic spline interpolation for completion. For example, the sliding window length is fixed at 100.

[0085] The second group, as a partial ablation group, introduces a chi-square test critical threshold corrected by the distribution skewness factor and a sliding window length adjustment mechanism based on the execution logic of the first group, but lacks secondary cross-validation and residual correction operations based on node energy entropy.

[0086] The third group, as a complete scheme group, includes all the implementation steps involved in this invention, such as the modified chi-square test critical threshold double cross-validation mechanism and the node energy entropy residual correction interpolation.

[0087] After the system was fully operational, the experimental results for each group were as follows: Statistics show that in the first group, 126 outliers were missed and 184 were misidentified, with a root mean square error (RMSE) of 2.75 degrees Celsius and a 4.6 percentage point deviation in the cold energy utilization rate calculation. After the second group's test, the number of missed outliers decreased to 42, the number of misidentified outliers decreased to 57, the RMSE of the sequence interpolation narrowed to 1.82 degrees Celsius, and the 2.1 percentage point deviation in the cold energy utilization rate calculation. The third group, representing the complete solution, showed only 5 missed outliers and 2 misidentified outliers. The RMSE of the sequence interpolation further decreased to 0.45 degrees Celsius, and the 0.3 percentage point deviation in the cold energy utilization rate calculation was controlled within 0.3 percentage points. Furthermore, no mass outlier deletion failures caused by normal step conditions occurred. The specific trends are as follows: Figure 4 As shown.

[0088] Comparative data shows that the second group improved the accuracy of detecting true outliers compared to the first group, indicating that the distribution skewness factor and window shortening mechanism can cope with the statistical threshold failure caused by non-stationary data skewness. The third group exhibits the best performance across all error metrics, demonstrating that the quadratic cross-validation mechanism based on the causal direction of the preceding and following time series successfully filtered out pseudo-true outlier interference caused by the rapid fluctuations in liquid nitrogen heat exchange, improving the system's decision stability under harsh conditions. Furthermore, the residual correction factor based on the node energy entropy in the denoising stage helps compensate for the insufficient characterization of local temperature fluctuations by traditional smoothed cubic spline interpolation, improving the consistency between the completed denoised temperature sequence and the actual temperature change trend, and ensuring the reliability of the subsequent monitoring and alarm closed loop.

[0089] Figure 2 This is a schematic diagram comparing the denoised temperature sequence with the original temperature in an embodiment of the present invention. The light-colored thin line with high-frequency, violent fluctuations represents the original temperature directly collected by the sensor; the thick black solid line represents the denoised temperature sequence obtained after weighted reconstruction by node energy entropy.

[0090] As can be seen from the image, the black solid line runs smoothly through the background of the violently fluctuating light-colored thin line, which proves that the algorithm successfully identified and filtered out the high-frequency random noise generated by the liquid nitrogen phase transition impact. Observing the fluctuation trend of the denoised temperature sequence, the peaks and troughs highly coincide with the dense areas of the original signal. This corresponds to the technical effect of using node energy entropy weighted reconstruction to retain the local cold energy temperature change characteristics described in the specific implementation, ensuring that the signal does not lose the true temperature change information while filtering out background interference.

[0091] Figure 3This is a schematic diagram illustrating the distribution of weighted chi-square statistics and multi-level critical thresholds in an embodiment of the present invention. In the diagram, discrete solid dots represent the weighted chi-square statistics corresponding to each sampling point; horizontally distributed dashed lines represent the basic chi-square test critical thresholds; solid lines that dynamically fluctuate with the sample distribution characteristics represent the corrected chi-square test critical thresholds; and X-shaped symbols located in the high-value region of the statistics represent the increased validation thresholds in the secondary cross-validation stage.

[0092] As can be seen from the image, the corrected critical threshold is not fixed, but adaptively adjusted according to the fluctuation of the background statistic. This proves that the distribution skewness factor successfully corrects the statistical bias under non-stationary conditions. Observing the marked area, when the weighted chi-square statistic exceeds the correction threshold, the threshold corresponding to the X-shaped symbol rises synchronously. This corresponds to the defense mechanism described in the specific implementation that dynamically adjusts the secondary verification confidence level according to the number of consecutive occurrences of suspicious samples, and achieves accurate discrimination between true outliers and thermal shock disturbances of the operating conditions.

[0093] Figure 4 This is a schematic diagram comparing the ablation experimental results of different detection schemes in this embodiment of the invention. The horizontal axis represents the experimental groups, showing the basic control group, the partial ablation group, and the complete scheme group from left to right; the bar chart associated with the left vertical axis represents the number of fault judgments, where bars filled with diagonal lines represent the number of missed true outliers, and bars filled with intersecting grids represent the number of misjudgments; the line chart associated with the right vertical axis represents the error index, where solid lines marked with squares represent the root mean square error of sequence interpolation, and dashed lines marked with diamonds represent the calculation deviation of the cold energy utilization rate index.

[0094] Observing the data from each group reveals that, from the basic control group to the complete scheme group, the height of the bars representing incorrect decisions decreases in a stepwise manner, and the slope of the line representing the error index also decreases synchronously. In the complete scheme group, both the bar distribution and the deviation line are at the bottom of the coordinate system. This directly demonstrates the synergistic effect of the core components proposed in this invention, such as dynamic threshold correction and residual correction interpolation, which improves the monitoring accuracy and data fidelity of the system under complex operating conditions.

[0095] This invention also discloses a temperature monitoring system for a liquid nitrogen cold energy recovery process, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a method for monitoring the temperature of a liquid nitrogen cold energy recovery process according to the present invention.

[0096] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0097] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for temperature monitoring in a liquid nitrogen cold energy recovery process, characterized in that, include: S1: Obtain the time series of heat exchange node temperatures, determine the number of decomposition layers based on the sequence sample entropy, perform wavelet packet decomposition, and obtain the denoised temperature sequence by weighted reconstruction based on the energy entropy of sub-band nodes; calculate the distribution skewness factor of the denoised temperature sequence within the sliding window to correct the chi-square test critical threshold, and shorten the sliding window length according to the degree to which the distribution skewness factor deviates from the preset benchmark value. S2: Based on the distance from the temperature sample to the median and the local temperature gradient change rate, the two are combined and globally normalized to determine the weighted weights of the true independent samples, thus constructing a weighted chi-square statistic for the temperature samples within the sliding window. When the weighted chi-square statistic of a suspected sample exceeds the corrected chi-square test threshold, a second cross-validation is performed using adjacent sliding windows. The confidence level of the second cross-validation is adjusted according to the number of consecutive occurrences of the suspected sample. When both validations exceed the limit, the true outlier is removed. S3: The spline node density is determined using the local temperature gradient for preliminary interpolation, and the energy entropy of the sub-band nodes is used as a residual correction factor to correct the preliminary interpolated temperature to complete the sequence. The cold energy utilization rate index is calculated based on the completed sequence, and the current actual flow rate change rate and the preset baseline flow rate change rate are obtained. The current actual flow rate change rate and the preset baseline flow rate change rate are normalized to obtain a dimensionless adjustment coefficient. The monitoring interval is updated by combining the historical fluctuation standard deviation of the cold energy utilization rate index with the dimensionless adjustment coefficient. An alarm is triggered when the cold energy utilization rate index exceeds the monitoring interval.

2. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 1, characterized in that, The calculation of the distribution skewness factor of the denoised temperature sequence within the sliding window to correct the chi-square test critical threshold includes: sorting all temperature samples within the sliding window by value and extracting the median; calculating the absolute difference between all temperature samples within the sliding window and the median, sorting all absolute differences and extracting the absolute deviation of the median; calculating the standard deviation of all temperature samples within the sliding window; dividing the standard deviation by the absolute deviation of the median to obtain the distribution skewness factor; extracting a preset basic chi-square test critical threshold, using the ratio of the distribution skewness factor to a preset benchmark value as a threshold correction coefficient, and multiplying the basic chi-square test critical threshold by the threshold correction coefficient to obtain the corrected chi-square test critical threshold.

3. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 1, characterized in that, The method for determining the true independent weighting based on the distance from the temperature sample to the median and the local temperature gradient change rate, combined with global normalization, includes: calculating the absolute difference between each temperature sample value within the sliding window and the median; adding a minimum normal number to the absolute difference and taking the reciprocal as the first weighting base; calculating the temperature difference between the temperature sample and the previous adjacent temperature sample in the time series; dividing the temperature difference by the system sampling time interval to obtain the local temperature gradient change rate; adding a minimum normal number to the absolute value of the local temperature gradient change rate and taking the reciprocal as the second weighting base; multiplying the first weighting base and the second weighting base point by point to obtain the joint initial weight; and dividing the joint initial weight of each temperature sample within the sliding window by the sum of the joint initial weights of all temperature samples within the sliding window to perform global normalization processing to obtain the true independent weighting of each temperature sample.

4. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 3, characterized in that, The construction of the weighted chi-square statistic for temperature samples within the sliding window includes: using the true independent allocation weights to perform a weighted summation of the temperature samples within the sliding window to obtain a weighted mean; using the true independent allocation weights to calculate the sum of squares of the differences between the temperature samples and the weighted mean to obtain a weighted variance; and based on the weighted mean and the weighted variance, calculating the ratio of the squared deviation of each temperature sample relative to the weighted mean to the weighted variance to construct the weighted chi-square statistic.

5. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 1, characterized in that, The method of using adjacent sliding windows for secondary cross-validation includes: when the weighted chi-square statistic of a suspicious sample is greater than the corrected chi-square test threshold, extracting the adjacent previous historical sliding window based on the timestamp corresponding to the suspicious sample, incorporating the suspicious sample data into the previous historical sliding window to recalculate the forward weighted chi-square statistic; simultaneously extracting the adjacent next future sliding window immediately following the suspicious sample, incorporating the suspicious sample data into the next future sliding window to recalculate the backward weighted chi-square statistic.

6. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 5, characterized in that, The confidence level of the second cross-validation is adjusted according to the number of consecutive occurrences of the suspicious sample. When both validations exceed the limit, the true outlier is removed. This includes: recording the number of consecutive occurrences of the suspicious sample exceeding the limit; multiplying the number of occurrences by a preset step size coefficient and adding it to the initial confidence level; and truncating the upper limit of the calculation result within the effective interval to adjust the confidence level; calculating the increased validation threshold using the inverse chi-square distribution function based on the adjusted confidence level; determining whether both the forward-weighted chi-square statistic and the backward-weighted chi-square statistic are greater than the increased validation threshold; if both validations exceed the limit, the timestamp and value corresponding to the true outlier are removed; if at least one does not exceed the limit, it is retained.

7. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 1, characterized in that, The process of determining spline node density using local temperature gradients for preliminary interpolation and correcting the preliminary interpolation temperature using sub-band node energy entropy as a residual correction factor to complete the sequence includes: extracting the absolute values ​​of local temperature gradients of adjacent retained points before and after the location of the removed outlier; determining the spline node density of the completed interpolation interval based on the direct proportional mapping relationship between the absolute value of the local temperature gradient and the spline node density; establishing a cubic spline mapping function using at least four adjacent retained points before and after the location of the removed outlier; calculating the preliminary interpolation temperature at the determined spline nodes; extracting the node energy entropy of the sub-band corresponding to the time period of the denoising stage; multiplying the node energy entropy by the sign of the local temperature gradient at the location of the removed outlier and a preset constant as a residual correction factor; and adding the preliminary interpolation temperature and the residual correction factor for fine-tuning and correction to obtain the corrected interpolation result as the completed sequence.

8. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 1, characterized in that, The step of calculating the cold energy utilization rate index based on the completed sequence, obtaining the current actual flow rate change rate and the preset benchmark flow rate change rate, and normalizing the current actual flow rate change rate with the preset benchmark flow rate change rate to obtain a dimensionless adjustment coefficient includes: synchronously acquiring the corresponding pipeline pressure parameters, extracting the corresponding absolute temperature based on the completed sequence, and calculating the ratio of the actual recovered enthalpy difference to the theoretical maximum recoverable enthalpy difference of the cold energy recovery heat exchange node as the cold energy utilization rate index in combination with the pipeline pressure parameters; calculating the average value and historical fluctuation standard deviation of the cold energy utilization rate index within the historical time window; acquiring the current actual flow rate change rate, and normalizing it relative to the preset benchmark flow rate change rate to obtain a dimensionless adjustment coefficient.

9. The method for temperature monitoring in a liquid nitrogen cold energy recovery process according to claim 8, characterized in that, The method of updating the monitoring interval by combining the historical fluctuation standard deviation of the cold energy utilization rate index with the dimensionless adjustment coefficient, and triggering an alarm when the cold energy utilization rate index exceeds the monitoring interval, includes: adding the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient to the average value of the cold energy utilization rate index within the historical time window to obtain the upper monitoring limit; subtracting the product of the historical fluctuation standard deviation and the dimensionless adjustment coefficient from the average value to obtain the lower monitoring limit; constructing a dynamic monitoring interval from the upper and lower monitoring limits; determining whether the cold energy utilization rate index at the current moment exceeds the monitoring interval; and triggering an alarm if the cold energy utilization rate index at the current moment is greater than the upper monitoring limit or less than the lower monitoring limit.

10. A temperature monitoring system for a liquid nitrogen cold energy recovery process, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a temperature monitoring method for a liquid nitrogen cold energy recovery process according to any one of claims 1-9.

Images