A refrigerant gas sensor zero drift self-adaptive correction method and system

Abstract

The application discloses a kind of refrigerant gas sensor zero drift self-adaptive correction method and system, the method is by double time scale decomposition framework, sensor zero drift is explicitly separated into fast variable hot drift component and slow variable aging drift component (modeling is adopted first-order Gauss Markov process), respectively for the different dynamic characteristics of two design modeling strategy;In online running phase, based on extended Kalman filter, the drift state of double component is recursively estimated, and the comprehensive drift estimation value and its uncertainty interval are output in real time, to realize the adaptive compensation of sensor reading.The application does not need frequent manual calibration, has low computing resource demand, is suitable for real-time operation on embedded sensor node, and is especially suitable for data center, cold chain warehouse, commercial refrigeration and other places with large temperature and humidity fluctuations, long-term unattended refrigerant leakage monitoring sites.

Description

Technical Field

[0001] This invention belongs to the field of sensor calibration technology, and particularly relates to an adaptive calibration method and system for zero-point drift of a refrigerant gas sensor. Background Technology

[0002] Leakage monitoring of refrigerant gases (such as HFCs like R410A, R32, and R134a, and HFOs like R1234yf) is of great significance for the safe operation of equipment and environmental protection. On the one hand, the accumulation of high concentrations of refrigerant may lead to personnel asphyxiation and electrical equipment failure; on the other hand, the greenhouse gas components in refrigerants have high global warming potential (GWP), and leaks will cause significant environmental hazards. Therefore, reliable online refrigerant gas leakage monitoring systems need to be deployed in all types of application sites.

[0003] Zero-point drift is a core technical challenge affecting the accuracy and reliability of refrigerant gas sensors during long-term operation. Zero-point drift refers to the phenomenon where the sensor's output signal deviates from the theoretical zero point when the target gas is absent. Zero-point drift in refrigerant gas sensors mainly originates from rapid thermal drift and slow aging drift. Frequent temperature and humidity fluctuations in application environments such as refrigeration rooms and data centers can induce reversible changes in the physical properties of the sensor's sensitive materials over timescales ranging from minutes to hours, leading to rapid zero-point signal drift. The temperature drift coefficient of catalytic combustion and semiconductor refrigerant sensors can reach 0.5-2.0 ppm / ℃, and the zero-point error can reach ±20 ppm under temperature fluctuations of ±10℃, far exceeding the sensor's nominal accuracy. Furthermore, the sensor's sensitive elements experience irreversible performance degradation during long-term operation due to material fatigue, catalyst poisoning, or degradation of the sensitive film properties, manifesting as monotonous drift or slow fluctuations in the zero-point baseline over timescales ranging from days to months. If left uncorrected, aging drift can accumulate to exceed the system alarm threshold within weeks, causing false alarms or missed alarms. Existing technologies mainly correct this error by periodically calibrating manually or fixing temperature compensation curves, failing to clearly distinguish between rapid thermal drift and slow aging drift, making it difficult to balance drift estimation response speed and long-term stability.

[0004] Therefore, there is an urgent need for an adaptive zero-point drift correction method that can simultaneously handle both fast and slow drifts without requiring frequent manual calibration. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for adaptive zero-point drift correction of refrigerant gas sensors. Through a framework combining dual-timescale drift decomposition and extended Kalman filtering, real-time online estimation and adaptive compensation of rapid thermal drift and slow aging drift are achieved.

[0006] The specific technical solution is as follows:

[0007] On one hand, the present invention provides an adaptive correction method for zero-point drift of a refrigerant gas sensor, the method comprising the following steps:

[0008] Step S1: Construct a refrigerant gas sensor monitoring system. The monitoring system includes at least one refrigerant gas sensor as the main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units.

[0009] Step S2: In the initial calibration stage of the system, historical synchronous data of the main detection unit and the environmental auxiliary sensing unit are collected. A zero-point drift feature model is established based on the dual time scale decomposition framework, and the sensor zero-point drift is decomposed into fast thermal drift component and slow aging drift component.

[0010] Step S3: During the online operation phase of the system, the zero-point drift state is recursively estimated using an extended Kalman filter, and the fast-changing thermal drift component and the slow-changing aging drift component are tracked in real time to obtain the comprehensive drift estimate at the current moment.

[0011] Step S4: Use the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and output the corrected refrigerant gas concentration value.

[0012] Furthermore, historical synchronous data from the main detection unit and the environment-assisted perception unit are collected, and a zero-point drift feature model is established based on a dual-timescale decomposition framework, specifically as follows:

[0013] The dual-timescale decomposition framework reduces sensor zero-point drift. Modeling: ;in, For discrete sampling times, for Total zero-point drift of the time sensor The rapid thermal drift component reflects the instantaneous effect of temperature and its rate of change on the sensor zero point. This is the slow-varying aging drift component, reflecting the irreversible degradation of the sensor's sensitive material over time. Zero-mean Gaussian white noise represents measurement random error.

[0014] Furthermore, the rapidly varying thermal drift component The mathematical model is as follows:

[0015] ;in, for The ambient temperature at any given time Calibrate the reference temperature for the sensor; for Relative humidity at any given time For calibrating reference humidity; It is the first-order temperature drift coefficient; This is the second-order temperature drift coefficient, characterizing the nonlinear thermal response; The temperature change rate drift coefficient characterizes the sensor's thermal inertia effect. The temperature and humidity coupled drift coefficients characterize the modulation effect of humidity on thermal drift. Each coefficient is determined by least squares estimation from historical calibration data.

[0016] Furthermore, the slow-varying aging drift component Modeled using a first-order Gaussian Markov process: ;in, The aging drift state transition coefficient has a value range of 0.9990-0.9999, reflecting the high persistence of aging drift. for The slow aging drift component over time; The process noise has a mean of zero and a variance of . Gaussian distribution, The range of values ​​is to ppm².

[0017] Furthermore, the state vector of the extended Kalman filter is defined as: The prediction steps of the filter are as follows:

[0018] ;

[0019] ;

[0020] in, For based on Time information The prior estimate vector of the state at time step; for The posterior state estimate vector at time t. for The prior error covariance matrix at time t; for The posterior error covariance matrix at time t. Here is the state transition matrix. The process noise covariance matrix is:

[0021] ;

[0022] diag is a diagonal matrix operator, representing a diagonal matrix formed by the elements within the parentheses as diagonal elements; The linearized state transition coefficient for thermal drift is determined by the temperature change. and humidity calculate; This represents the noise variance during the thermal drift process.

[0023] Furthermore, the filter update steps are as follows:

[0024] ;

[0025] ;

[0026] ;

[0027] in, This refers to the zero-point deviation observation of the sensor's raw readings under the assumption of no refrigerant leakage. For the observation matrix, To observe the noise variance, The Kalman gain matrix; For integration The posterior state estimation vector following the observation at time step; for The posterior error covariance matrix at time t; To and Same-dimensional identity matrix; combined drift estimate: .

[0028] Furthermore, the effectiveness of the drift model is continuously monitored by accumulating and testing the CUSUM statistic, and parameter updates are triggered when the model degenerates significantly.

[0029] The CUSUM statistic is defined as follows: ;

[0030] ;

[0031] in, for The positive cumulative sum statistic at time step is used to detect the positive persistent shift of the residual mean; for The negative cumulative sum statistic at time step is used to detect the negative persistent shift of the residual mean; and They are respectively The positive and negative cumulative sums and statistics corresponding to each moment; for Time residuals , for Raw readings from the time sensor; The true value of the refrigerant concentration estimated by the model; The historical residual mean; The CUSUM sensitivity parameter is typically the standard deviation of historical residuals. 0.5 times; when or Exceeding the decision threshold When the drift model is deemed to have significantly degraded, a model parameter update is triggered: Where ξ is the decision threshold coefficient, ranging from 4.0 to 6.0; once an update is triggered, the system re-collects data from the recent calibration window and re-estimates the thermal drift coefficient. and the noise variance of the aging drift process And reset the CUSUM statistic. .

[0032] Furthermore, the adaptive compensation correction output: ; ;in, This is the corrected refrigerant gas concentration value. The raw sensor readings. The overall drift estimate of the extended Kalman filter output; when When the value is negative, it is truncated to zero. ;

[0033] Meanwhile, the correction results include the drift estimation uncertainty range. ,in The standard deviation of the overall drift estimate is used to assess the confidence level of the correction results.

[0034] Furthermore, based on the corrected refrigerant gas concentration value And the uncertainty of drift estimation, and implement alarm strategies.

[0035] On the other hand, the present invention provides a refrigerant gas sensor zero-point drift adaptive correction system for performing the method of the first aspect of the present invention. The system includes: a refrigerant gas sensor monitoring system, which includes at least one refrigerant gas sensor as a main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units.

[0036] Furthermore, the system also includes: a data acquisition module, a drift state estimation module, and an adaptive compensation module.

[0037] During the initial calibration phase of the system, the data acquisition module collects historical synchronous data from the main detection unit and the environmental auxiliary sensing unit. Based on the dual time-scale decomposition framework, a zero-point drift characteristic model is established, decomposing the sensor zero-point drift into a fast-changing thermal drift component and a slow-changing aging drift component.

[0038] During the online operation phase of the system, the drift state estimation module uses an extended Kalman filter to recursively estimate the zero-point drift state, tracks the fast-changing thermal drift component and the slow-changing aging drift component in real time, and obtains the comprehensive drift estimate at the current moment.

[0039] The adaptive compensation module uses the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and outputs the corrected refrigerant gas concentration value.

[0040] Compared with the prior art, the beneficial effects of this invention are:

[0041] This invention clearly separates the zero-point drift of refrigerant sensors into fast-changing thermal drift and slow-changing aging drift components using a dual-timescale drift decomposition framework. Combined with an extended Kalman filter, it achieves online adaptive collaborative estimation and compensation for both types of drift, improving thermal drift compensation accuracy by over 40% and reducing aging drift tracking latency to the second level. Simultaneously, it introduces CUSUM statistics to monitor the validity of the drift model and trigger parameter updates as needed. A multi-sensor node mutual verification mechanism further enhances the system's redundancy and fault tolerance, significantly reducing the number of false alarms throughout the year. It eliminates the need for frequent manual calibration, has low computational resource requirements, and is suitable for real-time operation on embedded sensor nodes, significantly improving the long-term accuracy and reliability of refrigerant leak monitoring systems. Attached Figure Description

[0042] Figure 1 This is a flowchart of an adaptive zero-point drift correction method for a refrigerant gas sensor according to the present invention;

[0043] Figure 2 This is a schematic diagram of the composition of a refrigerant gas sensor zero-point drift adaptive correction system according to the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of this invention, not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0045] Example 1

[0046] like Figure 1 The diagram shown is a flowchart of an adaptive zero-point drift correction method for a refrigerant gas sensor according to the present invention. The method includes the following steps:

[0047] Step S1: Construct a refrigerant gas sensor monitoring system. The monitoring system includes at least one refrigerant gas sensor as the main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units.

[0048] The monitoring system's hardware components include: a refrigerant gas sensor (the main detection unit, which can be either infrared absorption or catalytic combustion type, detecting common refrigerants such as R410A, R32, R134a, and R1234yf), a platinum resistance temperature sensor (resolution ≥ 0.1℃), a capacitive humidity sensor (resolution ≥ 1%RH), and an embedded processing unit with real-time computing capabilities. The sampling frequency of each sensor is uniformly set to 1 time / 10 seconds, and data is transmitted to the processing unit via RS485 or I²C bus. Sensor placement principles: The refrigerant gas sensor should be installed 0.3-0.5 m below (for refrigerants denser than air) or 0.2-0.4 m above (for refrigerants less dense than air) at pipe joints, valves, and compressor units where leaks may occur; the installation distance between the temperature and humidity sensors and the refrigerant sensor should not exceed 0.5 m, and direct heat radiation sources should be avoided to ensure the representativeness of environmental parameters.

[0049] Taking a data center server room as an example, the room covers an area of ​​approximately 600 m² and is equipped with 22 precision air conditioners using R410A refrigerant. Following the aforementioned layout principles, each precision air conditioner is designated as a monitoring node. A refrigerant gas sensor is installed 0.4 m below each air conditioner compressor unit, along with a platinum resistance temperature sensor and a capacitive humidity sensor. The sampling frequency is set to once every 10 seconds. Data is aggregated to the central processing unit via an RS485 bus, resulting in a total of 22 monitoring nodes. Status data is then aggregated to the central management platform via a local area network, enabling cross-node mutual verification.

[0050] Step S2: In the initial calibration stage of the system, historical synchronous data of the main detection unit and the environmental auxiliary sensing unit are collected. A zero-point drift feature model is established based on the dual time scale decomposition framework, and the sensor zero-point drift is decomposed into a fast thermal drift component and a slow aging drift component.

[0051] The dual-timescale decomposition framework reduces sensor zero-point drift. The model is as follows: ;in, The rapid thermal drift component reflects the instantaneous effect of temperature and its rate of change on the sensor zero point. This is the slow-varying aging drift component, reflecting the irreversible degradation of the sensor's sensitive material over time. Zero-mean Gaussian white noise represents measurement random error.

[0052] The rapid thermal drift component The mathematical model is as follows:

[0053] ;in, for The ambient temperature at any given time Calibrate the reference temperature for the sensor; for Relative humidity at any given time For calibrating reference humidity; It is the first-order temperature drift coefficient; This is the second-order temperature drift coefficient, characterizing the nonlinear thermal response; The temperature change rate drift coefficient characterizes the sensor's thermal inertia effect. The temperature and humidity coupled drift coefficients characterize the modulation effect of humidity on thermal drift. Each coefficient is determined by least squares estimation from historical calibration data.

[0054] The slow aging drift component Modeled using a first-order Gaussian Markov process: ;in, The aging drift state transition coefficient has a value range of 0.9990-0.9999, reflecting the high persistence of aging drift. The process noise has a mean of zero and a variance of . Gaussian distribution, The range of values ​​is to ppm²;

[0055] Relative Time of a First-Order Gaussian Markov Process satisfy: ;in, The sampling time interval, The value ranges from 10 to 30 days, reflecting the time-scale characteristics of sensor aging drift.

[0056] During the calibration phase following initial system deployment (recommended to last 14-21 days), ensure no refrigerant leaks occur in the monitored area and simultaneously collect sensor readings. Ambient temperature and relative humidity This constitutes the calibration dataset. Taking the aforementioned data center server room as an example, the calibration phase lasted for 14 days. During this period, the server room temperature fluctuated normally within the range of 20-26℃, and the humidity varied within the range of 40%-65%. The calibration dataset contains approximately 120,960 sets of synchronous data points (14 days × 24 hours × 360 times / hour), covering the typical temperature and humidity fluctuation range of the target location, providing a sufficient sample basis for high-precision estimation of the thermal drift coefficient.

[0057] Using the least squares method to analyze the parameters of the rapidly changing thermal drift model To make an estimate, the following objective function is established:

[0058] ;

[0059] in, , ℃, , RH, The rate of temperature change (approximately calculated using a difference approximation) is given. This is the initial aging drift estimate (extracted from the low-frequency component of the readings during calibration).

[0060] Solving this least squares problem yields an estimate of the thermal drift coefficient for this computer room: ppm / ℃ ppm / ℃², ppm·s / ℃ ppm / (℃·%RH), model fit This indicates that the thermal drift model can explain 87% of the rapid drift variance.

[0061] The Kalman filter initialization parameters are set as follows: observation noise variance ppm² (determined based on nominal noise from the sensor datasheet); noise variance during thermal drift. ppm²; Noise variance during aging drift process ppm² (determined based on the aging rate specifications provided by the sensor manufacturer); aging drift state transition coefficient (corresponding to relevant time) The initial value of the state vector is determined by fitting the low-frequency drift autocorrelation function during the calibration period. Initial covariance matrix .

[0062] Step S3: During the online operation phase of the system, the zero-point drift state is recursively estimated using an extended Kalman filter, and the fast-changing thermal drift component and the slow-changing aging drift component are tracked in real time to obtain the comprehensive drift estimate at the current moment.

[0063] The state vector of the extended Kalman filter is defined as follows:

[0064] ;

[0065] The prediction steps of the filter are as follows:

[0066] ;

[0067] ;

[0068] in, Here is the state transition matrix. The process noise covariance matrix is:

[0069] ;

[0070] The linearized state transition coefficient for thermal drift is determined by the temperature change. and humidity calculate; This represents the noise variance during the thermal drift process.

[0071] The filter update steps are as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] in, This refers to the zero-point deviation observation of the sensor's raw readings under the assumption of no refrigerant leakage. For the observation matrix, To observe the noise variance, The Kalman gain matrix;

[0076] The overall drift estimate is: .

[0077] Taking a typical moment Taking an example, we will illustrate the single-step calculation process of the extended Kalman filter. Let... The posterior state estimate at time step is ppm, posterior covariance is ; Temperature is monitored at all times ℃ (temperature rises by approximately 0.6℃ / 10s compared to the previous moment), humidity RH, raw sensor reading ppm.

[0078] Prediction steps: Based on the current temperature change ℃ Using ℃ / s and humidity, calculate the prior predictions for rapid thermal drift:

[0079] ;

[0080] make (by the pair) (obtained by first-order linearization), then:

[0081] ;

[0082] ;

[0083] Update steps: Construct observation points under conditions of no refrigerant leakage. ppm, calculate the covariance of new information:

[0084] ;

[0085] Kalman gain:

[0086] ;

[0087] New information: ppm;

[0088] Posterior state estimation:

[0089] ;

[0090] Overall drift estimate: ppm, corrected refrigerant concentration: ppm, cut off to zero, indicates that there is no refrigerant leakage at the current moment.

[0091] In multi-sensor network deployment scenarios, a cross-verification mechanism between nodes is introduced to improve the robustness of drift estimation. This mechanism is applied to spatially adjacent sensor nodes. and nodes Calculate the difference in readings between nodes:

[0092] ;in, and These are the raw sensor readings for nodes i and j, respectively; if there is no spatial concentration gradient of refrigerant gas within the target area, then It should fluctuate within a range where the mean is zero; when Exceeding the mutual verification threshold When the drift estimate of at least one node is determined to be biased, the drift re-estimation process of that node is triggered.

[0093] Mutual verification threshold Determined through historical data statistics:

[0094] ;in, and They are respectively The historical mean and standard deviation, where κ is the cross-validation sensitivity coefficient, ranging from 2.0 to 3.0.

[0095] The effectiveness of the drift model is continuously monitored by accumulating and testing the CUSUM statistic, and parameter updates are triggered when the model degenerates significantly.

[0096] The CUSUM statistic is defined as follows: ;

[0097] ;

[0098] in, for Time residuals , The true value of the refrigerant concentration estimated by the model; The historical residual mean; The CUSUM sensitivity parameter is typically the standard deviation of historical residuals. 0.5 times; when or Exceeding the decision threshold When the drift model is deemed to have significantly degraded, a model parameter update is triggered:

[0099] ;

[0100] Wherein, ξ is the decision threshold coefficient, with a value ranging from 4.0 to 6.0; once an update is triggered, the system re-collects data from the recent calibration window and re-estimates the thermal drift coefficient. and the noise variance of the aging drift process And reset the CUSUM statistic. .

[0101] Taking the aforementioned data center node as an example, test data from 90 days of continuous system operation shows that the extended Kalman filter's tracking delay for thermal drift caused by short-term temperature changes (±5℃ / 10 min) does not exceed 30 seconds; and its tracking error for aging drift accumulating to 12 ppm does not exceed ±0.8 ppm. On the 63rd day of continuous operation, Continuously exceeding the decision threshold (assuming) , ppm, (ppm), the system determined that the sensor's negative aging drift accelerated, successfully triggering a parameter update, and changing the aging drift state transition coefficient from (The relevant time is approximately 24.97 days) adjusted to (The relevant time was approximately 16.64 days), after which the drift tracking accuracy returned to normal levels.

[0102] Step S4: Use the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and output the corrected refrigerant gas concentration value.

[0103] The correction output of adaptive compensation: ; ;in, This is the corrected refrigerant gas concentration value. The raw sensor readings. The overall drift estimate of the extended Kalman filter output; when When the value is negative, it is truncated to zero. Meanwhile, the correction results include the drift estimation uncertainty range. ,in The standard deviation of the overall drift estimate is used to assess the confidence level of the correction results.

[0104] Taking a significant temperature adjustment event in a temperature control system as an example, when the ambient temperature suddenly rose from 22°C to 28°C within 15 minutes, the sensor's original reading... The instantaneous increase is approximately 5 ppm. This is because the extended Kalman filter has been tracking the rapidly changing thermal drift component in real time. Increase accordingly and correct the output. Maintaining a level close to zero, the drift estimate uncertainty The temperature briefly rose to approximately 1.5 ppm due to an increased rate of temperature change. The system automatically raised the first-level warning threshold temporarily to avoid false alarms; the temperature stabilized after 15 minutes. Once the concentration returns to normal (approximately 0.8 ppm), the warning threshold also returns to normal, ensuring reliable detection of low-concentration real leak signals.

[0105] Based on the corrected refrigerant gas concentration value And the uncertainty of drift estimation, and implement alarm strategies.

[0106] when Exceeding the leakage warning threshold However, the drift uncertainty When the value is large, a Level 1 warning will be issued, indicating that manual verification is required.

[0107] when Exceeding the leakage alarm threshold And drift uncertainty When the alarm level is low, a level 2 alarm is issued, triggering a coordinated response.

[0108] when If N consecutive sampling points exceed the leakage alarm threshold At any time, regardless of the magnitude of the drift uncertainty, the highest level emergency alarm will be issued; the value of N ranges from 3 to 5. and Determined according to current refrigerant leakage safety standards.

[0109] The criterion for determining the magnitude of drift uncertainty is defined as the uncertainty threshold. The value is ;when When the drift uncertainty is large, it indicates that the current drift estimate has insufficient confidence, and the correction result may misjudge the drift residual as a leakage signal; when When the drift uncertainty is small, it indicates that the extended Kalman filter's estimate of the current zero-point drift has converged, and the correction output has high reliability.

[0110] Leakage warning threshold With leakage alarm threshold Determined based on current national standards and industry specifications. For HFCs / HFOs refrigerants such as R410A and R32, the setting shall be determined with reference to relevant standards. ppm (approximately 8.4% of the acceptable concentration limit (ACL) of R410A). ppm (approximately 33.6% of ACL); for refrigerants such as R404A used in low-temperature cold chains, the setting is... ppm, ppm.

[0111] The above thresholds can be adjusted within the range of ±20% / pm based on the ventilation conditions, personnel density, and rated charge capacity of the equipment in the specific application environment. The ratio should not be lower than 2.5 to ensure sufficient window for manual response between Level 1 warning and Level 2 alarm. The uncertainty threshold should be adjusted synchronously to always meet the requirements. .

[0112] Taking a large pharmaceutical cold chain warehouse (approximately 2000 m², storage temperature −18℃-4℃, using R404A refrigerant, with a total of 15 monitoring nodes) as an example, the technical effect of the present invention is illustrated: Before applying the method of the present invention, the warehouse monitoring system generated 3-5 false alarms per day during the frequent start-up and shutdown of the refrigeration system (temperature fluctuation of ±8℃). The measurement deviation caused by the accumulation of zero-point drift reached an average of 28 ppm after 6 months of operation, which is equivalent to 2.8% of the R404A alarm threshold (1000 ppm), seriously affecting the early detection capability of low-concentration leaks.

[0113] After applying the method of this invention, under the same temperature fluctuation conditions: the average residual value of rapid thermal drift compensation decreased from ±6.3 ppm to ±1.2 ppm, a reduction of 81%; after 6 months of continuous operation, the aging drift tracking error remained within ±1.8 ppm (without triggering CUSUM updates); the number of false alarms throughout the year decreased from over 1500 to 47, a reduction of 97%; during the same period, 3 real refrigerant micro-leakage events were successfully captured (leakage concentrations were all below 50 ppm), avoiding refrigerant loss and potential safety risks. The system algorithm's single calculation latency on the ARM Cortex-M4 processor does not exceed 15 milliseconds, fully meeting the requirements for real-time correction.

[0114] Example 2

[0115] like Figure 2 The diagram shown is a schematic of the composition of a refrigerant gas sensor zero-point drift adaptive correction system according to the present invention. The system includes: a refrigerant gas sensor monitoring system, which includes at least one refrigerant gas sensor as a main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units.

[0116] The system also includes: a data acquisition module, a drift state estimation module, and an adaptive compensation module.

[0117] During the initial calibration phase of the system, the data acquisition module collects historical synchronous data from the main detection unit and the environmental auxiliary sensing unit. Based on the dual time scale decomposition framework, it establishes a zero-point drift characteristic model, decomposing the sensor zero-point drift into a fast-changing thermal drift component and a slow-changing aging drift component.

[0118] The data acquisition module consists of a refrigerant gas sensor, a platinum resistance temperature sensor (resolution ≥ 0.1℃), and a capacitive humidity sensor (resolution ≥ 1%RH) at the hardware level. Each sensor has a sampling frequency of 1 time / 10 seconds. Data is transmitted in frame format via RS485 or I²C bus to an embedded processing unit with real-time computing capabilities. The module synchronously aligns the timestamps of each sensor and outputs aligned data triplets. This data is intended for use by subsequent modules. During the initial calibration phase, the data acquisition module is also responsible for accumulating the calibration dataset (recommended for 14-21 days) and establishing the thermal drift coefficient through least squares estimation. The polynomial mapping relationship between the parameters and environmental parameters was established, and the aging drift parameters were determined by autocorrelation analysis of the low-frequency drift components during the calibration period. and The above model parameters are stored in the non-volatile storage area of ​​the processing unit for subsequent modules to call.

[0119] During the online operation phase of the system, the drift state estimation module uses an extended Kalman filter to recursively estimate the zero-point drift state, tracks the rapidly changing thermal drift component and the slowly changing aging drift component in real time, and obtains the comprehensive drift estimate at the current moment.

[0120] The drift state estimation module maintains the state estimation vector. Covariance Matrix A prediction-update cycle is completed every 10 seconds. When a potential leakage signal is detected ( The current observation is marked as suspicious and the observation noise variance is increased accordingly. This prevents leakage signals from being incorrectly absorbed into the drift estimate, thus ensuring the consistency of the drift model's estimation when a real leakage occurs. The module's single-processor computation latency on the ARM Cortex-M4 processor does not exceed 15 milliseconds, fully meeting the real-time requirements of embedded nodes.

[0121] The adaptive compensation module uses the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and outputs the corrected refrigerant gas concentration value.

[0122] The adaptive compensation module outputs to the upper-level system. The binary tuple is used by the hierarchical alarm logic and continuously performs CUSUM model validity monitoring. If and only if At that time (ensuring no leakage signal interference), the current correction residual will be... Included in CUSUM statistics, updated and Once the statistic exceeds the decision threshold... The data acquisition module is automatically triggered to re-acquire data from the recent calibration window to check the thermal drift coefficient. and aging drift parameters Perform a reassessment and reset the CUSUM statistic to enable on-demand, accurate updates to the drift model.

[0123] In the multi-node mutual verification mode, the central management platform performs a difference check on all node pairs every 5 minutes. The calculation and comparison. When a node is determined to be abnormal by mutual verification of multiple neighboring nodes (i.e., ... When multiple node pairs are established simultaneously, the drift estimation weight of the node is reduced to 0.3, and an equipment inspection reminder is issued to the maintenance personnel, which significantly enhances the system's redundancy and fault tolerance capabilities in single-point failure scenarios.

[0124] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive zero-point drift correction of a refrigerant gas sensor, characterized in that, The method includes the following steps: Step S1: Construct a refrigerant gas sensor monitoring system. The monitoring system includes at least one refrigerant gas sensor as the main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units. Step S2: In the initial calibration stage of the system, historical synchronous data of the main detection unit and the environmental auxiliary sensing unit are collected. A zero-point drift feature model is established based on the dual time scale decomposition framework, and the sensor zero-point drift is decomposed into fast thermal drift component and slow aging drift component. Step S3: During the online operation phase of the system, the zero-point drift state is recursively estimated using an extended Kalman filter, and the fast-changing thermal drift component and the slow-changing aging drift component are tracked in real time to obtain the comprehensive drift estimate at the current moment. Step S4: Use the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and output the corrected refrigerant gas concentration value.

2. The method according to claim 1, characterized in that, In step S2, the historical synchronization data of the main detection unit and the environment auxiliary perception unit are collected, and a zero-point drift feature model is established based on a dual-time-scale decomposition framework, specifically as follows: The dual-timescale decomposition framework reduces sensor zero-point drift. Modeling: ;in, For discrete sampling times, for Total zero-point drift of the time sensor The rapid thermal drift component reflects the instantaneous effect of temperature and its rate of change on the sensor zero point. This is the slow-varying aging drift component, reflecting the irreversible degradation of the sensor's sensitive material over time. Zero-mean Gaussian white noise represents random measurement error.

3. The method according to claim 2, characterized in that, The rapid thermal drift component The mathematical model is as follows: ； in, for The ambient temperature at any given time Calibrate the reference temperature for the sensor; for Relative humidity at any given time For calibrating reference humidity; It is the first-order temperature drift coefficient; This is the second-order temperature drift coefficient, characterizing the nonlinear thermal response; The temperature change rate drift coefficient characterizes the sensor's thermal inertia effect. The temperature and humidity coupled drift coefficients characterize the modulation effect of humidity on thermal drift. Each coefficient is determined by least squares estimation from historical calibration data.

4. The method according to claim 3, characterized in that, The slow aging drift component Modeled using a first-order Gaussian Markov process: ;in, The aging drift state transition coefficient has a value range of 0.9990-0.9999, reflecting the high persistence of aging drift. for The slow aging drift component over time; The process noise has a mean of zero and a variance of . Gaussian distribution, The range of values ​​is to ppm².

5. The method according to claim 4, characterized in that, In step S3, the state vector of the extended Kalman filter is defined as follows: The prediction steps of the filter are as follows: ； ； in, Based on Time information The prior estimate vector of the state at time step; for The posterior state estimate vector at time t. for The prior error covariance matrix at time t; for The posterior error covariance matrix at time t. Here is the state transition matrix. The process noise covariance matrix is: ； diag is a diagonal matrix operator, representing a diagonal matrix formed by the elements within the parentheses as diagonal elements; The linearized state transition coefficient for thermal drift is determined by the temperature change. and humidity calculate; This represents the noise variance during the thermal drift process.

6. The method according to claim 5, characterized in that, The filter update steps are as follows: ； ； ； in, This refers to the zero-point deviation observation of the sensor's raw readings under the assumption of no refrigerant leakage. For the observation matrix, To observe the noise variance, The Kalman gain matrix; For integration The posterior state estimation vector following the observation at time step; for The posterior error covariance matrix at time t; To and Same-dimensional identity matrix; combined drift estimate: .

7. The method according to claim 6, characterized in that, The effectiveness of the drift model is continuously monitored by accumulating and testing the CUSUM statistic, and parameter updates are triggered when the model degenerates significantly. The CUSUM statistic is defined as follows: ; ； in, for The positive cumulative sum statistic at time step is used to detect the positive persistent shift of the residual mean; for The negative cumulative sum statistic at time step is used to detect the negative persistent shift of the residual mean; and They are respectively The positive and negative cumulative sums and statistics corresponding to each moment; for Time residuals , for Raw readings from the time sensor; The true value of the refrigerant concentration estimated by the model; The historical residual mean; The CUSUM sensitivity parameter is typically the standard deviation of historical residuals. 0.5 times; when or Exceeding the decision threshold When the drift model is deemed to have significantly degraded, a model parameter update is triggered: Where ξ is the decision threshold coefficient, ranging from 4.0 to 6.0; once an update is triggered, the system re-collects data from the recent calibration window and re-estimates the thermal drift coefficient. and the noise variance of the aging drift process And reset the CUSUM statistic. .

8. The method according to claim 7, characterized in that, In step S4, the adaptive compensation correction output is: ; ;in, This is the corrected refrigerant gas concentration value. The raw sensor readings. The overall drift estimate of the extended Kalman filter output; when When the value is negative, it is truncated to zero. ; Meanwhile, the correction results include the drift estimation uncertainty range. ,in The standard deviation of the overall drift estimate is used to assess the confidence level of the correction results.

9. The method according to claim 8, characterized in that, Based on the corrected refrigerant gas concentration value And the uncertainty of drift estimation, and implement alarm strategies.

10. A refrigerant gas sensor zero-point drift adaptive correction system, used to perform the method according to any one of claims 1-9, characterized in that, The system includes: a refrigerant gas sensor monitoring system, which includes at least one refrigerant gas sensor as a main detection unit, and a temperature sensor and a humidity sensor arranged spatially adjacent to the main detection unit as environmental auxiliary sensing units. The system also includes: a data acquisition module, a drift state estimation module, and an adaptive compensation module; During the initial calibration phase of the system, the data acquisition module collects historical synchronous data from the main detection unit and the environmental auxiliary sensing unit, and establishes a zero-point drift feature model based on a dual time-scale decomposition framework, decomposing the sensor zero-point drift into a fast-changing thermal drift component and a slow-changing aging drift component. During the online operation phase of the system, the drift state estimation module uses an extended Kalman filter to recursively estimate the zero-point drift state, tracks the fast-changing thermal drift component and the slow-changing aging drift component in real time, and obtains the comprehensive drift estimate at the current moment. The adaptive compensation module uses the comprehensive drift estimate to adaptively compensate the real-time refrigerant gas concentration reading of the main detection unit and outputs the corrected refrigerant gas concentration value.

Images