Machine learning based air micro station sensor interference correction and drift calibration method

Abstract

The application provides a kind of air microstation sensor interference correction and drift calibration method based on machine learning, it is related to environmental monitoring and sensor signal processing technical field, the method includes: obtaining the measurement original value of multiple gas sensors and temperature data and humidity data, it is judged whether each sensor exists drift, if exists, then the drift is corrected, and the measured concentration value of multiple gases is calculated back through the trained calibration model.According to the application, it can be judged whether multiple sensors exist drift based on the measurement original value, if there is drift, subsequent calibration processing is carried out after the drift is corrected, the data accuracy and processing accuracy are improved, and the accuracy of calibration processing and the accuracy of concentration data can be improved by correcting complex nonlinear mixed interference through neural network model.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and sensor signal processing technology, and in particular to a machine learning-based method for interference correction and drift calibration of air micro-station sensors. Background Technology

[0002] Air micro-stations are widely used due to their low cost and flexible deployment. However, the electrochemical sensors used in air micro-stations have two major drawbacks: 1. Severe cross-interference: O3 sensors typically have a strong response to NO2; SO2 sensors are highly susceptible to the superimposed interference of NO2 and O3; CO sensors are easily affected by TVOC (Total Volatile Organic Compounds). Linear matrix correction methods in related technologies struggle to handle this complex nonlinear mixed interference.

[0003] 2. Temperature and humidity drift and aging: The sensor sensitivity and zero point are affected by temperature and humidity, exhibiting non-linear changes, and baseline drift occurs over time.

[0004] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a machine learning-based method for interference correction and drift calibration of air microstation sensors, which can solve the technical problems of baseline drift correction and complex nonlinear mixed interference that are difficult to handle in related technologies.

[0006] According to a first aspect of the present invention, a method for interference correction and drift calibration of air micro-station sensors based on machine learning is provided, comprising: Real-time acquisition of raw measurement values, temperature data, and humidity data from SO2, NO2, O3, CO, and TVOC sensors; Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor, determine whether each sensor is drifting. If drift exists, the amount of drift is calculated based on the original measurement value; A correction value is obtained based on the original measurement value of the drifting sensor and the amount of drift. The correction values, temperature data, and humidity data are input into the trained calibration model for processing to obtain the measured concentration values ​​of SO2, NO2, O3, CO, and TVOC.

[0007] According to the present invention, determining whether each sensor exhibits drift based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: Based on the raw measurement values ​​of the O3 sensor and the NO2 sensor, the concentration readings of the O3 sensor and the NO2 sensor are obtained. According to the formula ΔO3=C measured OX –C measured NO2 Obtain the virtual ozone residual ΔO3, where C measured OX For the concentration reading of the O3 sensor, C measured NO2 This is the concentration reading from the NO2 sensor; If the virtual ozone residual is less than -ε, it is determined that the O3 sensor has a negative drift or has failed, where ε is a preset judgment threshold. If the concentration reading of the NO2 sensor is lower than or equal to the first preset value, and the concentration reading of the O3 sensor is higher than or equal to the second preset value for multiple consecutive moments, then it is determined that the O3 sensor has a positive drift.

[0008] According to the present invention, Based on the raw measurement values ​​from the SO2, NO2, O3, CO, and TVOC sensors, determine whether each sensor exhibits drift, including: Based on the raw measurement values ​​of the SO2 sensor and the NO2 sensor, obtain the concentration readings of the SO2 sensor and the NO2 sensor. According to the formula Bias SO2 =C measured SO2 -(C measured NO2 ×K {SO2|NO2} ) Determine the theoretical baseline deviation of SO2. SO2 , where C measured SO2 For the concentration reading of the SO2 sensor, C measured NO2 K represents the concentration reading of the NO2 sensor. {SO2|NO2} The cross-interference coefficient generated by NO2 on the SO2 sensor; If the concentration reading of the NO2 sensor is higher than or equal to the third preset value, and the concentration reading of the SO2 sensor is higher than or equal to 0, then it is determined that the SO2 sensor has a positive drift. If the concentration reading of the NO2 sensor is less than or equal to the fourth preset value, and the concentration reading of the SO2 sensor is less than -ε for multiple consecutive moments, then it is determined that the SO2 sensor has a negative drift, where ε is a preset judgment threshold.

[0009] According to the present invention, determining whether each sensor exhibits drift based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, and O3 sensor, obtain the concentration readings of the SO2 sensor, NO2 sensor, and O3 sensor. If the concentration reading of the NO2 sensor is higher than or equal to the fifth preset value, the concentration reading of the O3 sensor is lower than or equal to the sixth preset value, and the concentration reading of the SO2 sensor is greater than or equal to 0, then it is determined that the NO2 sensor has a positive drift. If the concentration reading of the SO2 sensor is less than -ε and the concentration reading of the NO2 sensor is 0, then the NO2 sensor is determined to have negative drift or failure.

[0010] According to the present invention, determining whether each sensor exhibits drift based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: Based on the raw measurement values ​​of the CO sensor and the NO2 sensor, obtain the concentration readings of the CO sensor and the NO2 sensor. The Pearson correlation coefficients of the CO and NO2 sensors were determined based on the concentration readings of the CO sensor and the NO2 sensor at multiple time points. If the Pearson correlation coefficient is greater than or equal to a preset coefficient, then the CO sensor is determined to be drifting.

[0011] According to the present invention, If drift exists, the amount of drift is calculated based on the original measurement value, including: Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, and O3 sensor, obtain the concentration readings of the SO2 sensor, NO2 sensor, and O3 sensor. According to the formula V theory =S SO2 ×(K {SO2|NO2} ×C measured NO2 +K {SO2|O3} ×C measured O3 )+V base(T) Obtain the theoretical noise floor voltage V of the SO2 sensor theory , among which, S SO2 K is the conversion coefficient of the sensor. {SO2|NO2} K is the cross-interference coefficient generated by NO2 on the SO2 sensor.{SO2|O3} C represents the cross-interference coefficient generated by ozone on the SO2 sensor. measured NO2 For the concentration reading of the NO2 sensor, C measured O3 V represents the concentration reading from the O3 sensor. base(T) This is the reference voltage for the SO2 sensor; According to the formula ΔV=V measure -V theory Determine the voltage deviation value ΔV of the SO2 sensor, where V measure These are the raw measurement values ​​from the SO2 sensor; Determine whether the current environment is a clean background environment based on the current time or the concentration reading of the CO sensor; If the current environment is a clean background, then according to the formula V bias(t) =α×ΔV+(1-α)×V bias(t-1) Obtain the current drift V of the SO2 sensor. bias(t) , where V bias(t-1) Let α be the drift of the SO2 sensor at the previous moment, where α is the smoothing factor.

[0012] According to the present invention, the method further includes: Standard gas was measured using a standard station to obtain multiple sets of calibrated concentration values. These calibrated concentration values ​​included SO2, NO2, O3, CO, and TVOC calibrated concentration values ​​at multiple times under standard conditions. Construct orthogonal adversarial samples, wherein the concentration of the target gas in the orthogonal adversarial samples is lower than a preset inhibition value, and the concentration of the gas that has an interfering relationship with the target gas is higher than a preset interference value; Construct a full-environment sample, which includes calibration data for various gases under multiple temperatures and humidity levels; The loss function of the calibration model is determined using the sample calibration data, orthogonal adversarial samples and full environmental samples, as well as the original measurement values ​​of the calibration model and standard gas, temperature data and humidity data. The calibration model is trained based on its loss function to obtain the trained calibration model.

[0013] According to a second aspect of the present invention, a machine learning-based air micro-station sensor interference correction and drift calibration system is provided, comprising: The real-time acquisition module is used to acquire the raw measurement values, temperature data, and humidity data of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor in real time. The judgment module is used to determine whether each sensor has drifted based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor and TVOC sensor; A drift measurement module is used to calculate the drift amount based on the original measurement value if drift exists. The correction value module is used to obtain a correction value based on the original measurement value of the sensor with drift and the amount of drift; The calibration module is used to input the correction values, temperature data, and humidity data into the trained calibration model for processing, and to obtain the measured concentration values ​​of SO2, NO2, O3, CO, and TVOC.

[0014] According to a third aspect of the present invention, a machine learning-based air micro-station sensor interference correction and drift calibration device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the machine learning-based air micro-station sensor interference correction and drift calibration method.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the machine learning-based air microstation sensor interference correction and drift calibration method.

[0016] By adopting the above technical solution, the present invention can achieve the following technical effects: According to this invention, the presence of drift in multiple sensors can be determined based on the original measured values. If drift exists, subsequent calibration is performed after correcting the drift amount, improving data accuracy and processing precision. Furthermore, complex nonlinear mixed interference can be corrected using a neural network model, improving the accuracy of calibration and concentration data. When determining whether an O3 sensor is drifting, the characteristic of the O3 sensor responding simultaneously to ozone and NO2 can be used to calculate the virtual ozone residual to determine whether the O3 sensor is experiencing negative drift. Conversely, the characteristics of the concentration readings of the NO2 and O3 sensors can be used to determine whether the O3 sensor is experiencing positive drift, improving the rationality and accuracy of the judgment. When determining whether an SO2 sensor is drifting, the theoretical baseline deviation of SO2 and the negative influence of NO2 gas on the SO2 sensor are used to comprehensively determine whether the SO2 sensor is drifting, improving the objectivity and accuracy of the judgment. When determining whether an NO2 sensor is drifting, the responses of the O3 and SO2 sensors to NO2 gas can be utilized, and the concentration readings of the SO2, O3, and NO2 sensors can be compared to determine whether the NO2 sensor is drifting, improving the objectivity and accuracy of the judgment. Based on the detection principles of CO and TVOC sensors, the correlation and comparison between sensor concentration readings can determine whether CO and TVOC sensors exhibit drift, improving the objectivity and accuracy of the judgment. Furthermore, by using conversion coefficients and cross-interference coefficients to analyze the mutual influence of different types of gases on the sensor, the theoretical noise floor voltage can be determined, and the voltage deviation value can be calculated. The drift amount under clean background conditions can be calculated using exponentially weighted moving average processing to accurately correct the drift, providing more accurate data for subsequent calibration and improving data processing precision. During training, multiple training samples are constructed, enabling the calibration model to accurately identify low-concentration target gases and determine their precise concentrations under strong interference environments, as well as accurately calculate measured concentration values ​​in the entire environment. This improves the model's robustness and accuracy under various conditions and enhances the accuracy of calibration processing.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. Figure 1 An exemplary flowchart of a machine learning-based air micro-station sensor interference correction and drift calibration method according to an embodiment of the present invention is shown. Figure 2 A block diagram of a machine learning-based air microstation sensor interference correction and drift calibration system according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 An exemplary flowchart illustrates a machine learning-based method for interference correction and drift calibration of air micro-station sensors according to an embodiment of the present invention. The method includes: Step S1: Real-time acquisition of raw measurement values, temperature data, and humidity data from SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor; Step S2: Based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor, determine whether each sensor is drifting. Step S3: If drift exists, calculate the drift amount based on the original measurement value; Step S4: Obtain a correction value based on the original measurement value of the drifting sensor and the amount of drift; Step S5: Input the correction value, temperature data and humidity data into the trained calibration model for processing to obtain the measured concentration values ​​of SO2, NO2, O3, CO and TVOC.

[0022] According to an embodiment of the present invention, the machine learning-based air microstation sensor interference correction and drift calibration method can determine whether multiple sensors are drifting based on the original measurement values. If drift exists, subsequent calibration processing is performed after correcting the drift amount, thereby improving data accuracy and processing accuracy. Furthermore, complex nonlinear mixed interference can be corrected through a neural network model, thereby improving the accuracy of calibration processing and the accuracy of concentration data.

[0023] Example 1: According to one embodiment of the present invention, in step S1, data can be collected in real time by an SO2 sensor, a NO2 sensor, an O3 sensor, a CO sensor, and a TVOC sensor to obtain the raw measurement values. Besides detecting gas concentration data corresponding to their own functions, these sensors may also respond to other gases due to the influence of their detection principles. For example, an O3 sensor readily responds to NO2, an SO2 sensor readily responds to both NO2 and O3, and a CO sensor is easily affected by TVOC. Furthermore, humidity and temperature may also have a certain impact on the sensors. Therefore, in addition to using the above-mentioned raw measurement values ​​for mutual calibration, temperature data and humidity data can also be acquired and incorporated into the calibration process to further improve calibration accuracy.

[0024] Example 2: According to an embodiment of the present invention, in step S2, it can be first determined whether there is drift in the SO2 sensor, NO2 sensor, O3 sensor, CO sensor and TVOC sensor. If there is drift, the amount of drift is corrected, and then the corrected data can be further calibrated to improve the data accuracy.

[0025] According to one embodiment of the present invention, it is possible to determine whether the O3 sensor is drifting based on the original measurement values ​​of the O3 sensor and the NO2 sensor. Determining whether each sensor is drifting based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: obtaining the concentration readings of the O3 sensor and the NO2 sensor based on the original measurement values ​​of the O3 sensor and the NO2 sensor; obtaining the virtual ozone residual ΔO3 according to formula (1). ΔO3=C measured OX –C measured NO2 (1) Among them, C measured OX For the concentration reading of the O3 sensor, C measured NO2The concentration reading of the NO2 sensor is used; if the virtual ozone residual is less than -ε, it is determined that the O3 sensor has negative drift or failure, where ε (10ppb) is a preset judgment threshold; if the concentration reading of the NO2 sensor is lower than or equal to the first preset value, and the concentration reading of the O3 sensor is higher than or equal to the second preset value for multiple consecutive times, it is determined that the O3 sensor has positive drift.

[0026] According to one embodiment of the present invention, the O3 sensor can be used to detect ozone concentration data, but it also responds to NO2 gas. Therefore, the NO2 gas concentration will affect the ozone concentration reading of the O3 sensor. Thus, the concentration reading of the O3 sensor should not be significantly lower than the concentration reading of the NO2 sensor. Therefore, when making a judgment, the virtual ozone residual ΔO3 can be obtained by subtracting the concentration reading of the O3 sensor from the concentration reading of the NO2 sensor based on formula (1). If the virtual ozone residual is less than -ε, it indicates that the O3 sensor has a negative drift or has failed, that is, the concentration reading of the O3 sensor does not match the theoretical condition. On the other hand, if it is nighttime and the NO2 sensor concentration reading is very low, for example, below or equal to the first preset value (10 ppb), then if the O3 sensor concentration reading remains at a high level for several consecutive moments (e.g., 10 consecutive moments), for example, above or equal to the second preset value (100 ppb), it can be determined that the O3 sensor has a positive zero-point drift. In other words, the pollutant concentration level is low at night, and the very low NO2 sensor concentration reading further confirms the low pollutant concentration level. In this case, if the O3 sensor concentration reading is above or equal to the second preset value for several consecutive moments, then the O3 sensor concentration reading does not match the theory and there may be a positive drift.

[0027] In this way, when determining whether the O3 sensor is drifting, the characteristic that the O3 sensor responds to both ozone and NO2 can be used to calculate the virtual ozone residual to determine whether the O3 sensor is drifting negatively, and the characteristics of the concentration readings of the NO2 sensor and the O3 sensor can be used to determine whether the O3 sensor is drifting positively, thus improving the rationality and accuracy of the judgment.

[0028] Example 3: According to one embodiment of the present invention, the SO2 sensor responds to NO2 gas, and NO2 gas has a strong negative interference to the SO2 sensor. Therefore, it is possible to determine whether the SO2 sensor is drifting based on the concentration readings of the SO2 sensor and the NO2 sensor.

[0029] According to one embodiment of the present invention, determining whether each sensor has drift based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: obtaining the concentration readings of the SO2 sensor and the NO2 sensor based on the original measurement values ​​of the SO2 sensor and the NO2 sensor; and determining the theoretical baseline deviation Bias of SO2 according to formula (2). SO2 , Bias SO2 =C measured SO2 -(C measured NO2 ×K {SO2|NO2} (2) Among them, C measured SO2 For the concentration reading of the SO2 sensor, C measured NO2 K represents the concentration reading of the NO2 sensor. {SO2|NO2} The cross-interference coefficient generated by NO2 on the SO2 sensor; if the concentration reading of the NO2 sensor is higher than or equal to the third preset value, and the concentration reading of the SO2 sensor is higher than or equal to 0, then the SO2 sensor is determined to have positive drift; if the concentration reading of the NO2 sensor is less than or equal to the fourth preset value, and the concentration reading of the SO2 sensor is less than -ε for multiple consecutive moments, then the SO2 sensor is determined to have negative drift, where ε is a preset judgment threshold.

[0030] According to one embodiment of the present invention, K {SO2|NO2} Typically, the value is negative; for example, it can be -1.2. Based on this SO2 theoretical baseline deviation, if the NO2 sensor concentration reading is higher than or equal to the third preset value (e.g., 50 ppb), the SO2 theoretical baseline deviation is usually negative. If, in this case, the SO2 sensor concentration reading is 0 or positive, then the SO2 sensor is determined to have a positive drift; that is, a positive zero-point drift masks the negative interference. On the other hand, if the NO2 sensor concentration reading is very low (less than or equal to the fourth preset value (10 ppb), i.e., close to 0), but the SO2 sensor concentration reading is less than -ε for multiple consecutive moments, i.e., the magnitude of the negative degree of the SO2 sensor concentration reading is large, then the SO2 sensor is determined to have a negative drift.

[0031] In this way, when determining whether SO2 sensor is drifting, the determination is based on the theoretical baseline deviation of SO2 and the negative influence of NO2 gas on SO2 sensor, thereby improving the objectivity and accuracy of the judgment.

[0032] Example 4: According to an embodiment of the present invention, based on the above analysis, both the O3 sensor and the SO2 sensor can respond to NO2 gas. Therefore, the concentration readings of the O3 sensor and the SO2 sensor can be used to verify whether the NO2 sensor is drifting.

[0033] According to one embodiment of the present invention, determining whether each sensor is drifting based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: obtaining the concentration readings of the SO2 sensor, NO2 sensor, and O3 sensor based on the original measurement values ​​of the SO2 sensor, NO2 sensor, and O3 sensor; if the concentration reading of the NO2 sensor is higher than or equal to a fifth preset value, the concentration reading of the O3 sensor is lower than or equal to a sixth preset value, and the concentration reading of the SO2 sensor is greater than or equal to 0, then it is determined that the NO2 sensor has positive drift; if the concentration reading of the SO2 sensor is less than -ε and the concentration reading of the NO2 sensor is 0, then it is determined that the NO2 sensor has negative drift or is malfunctioning.

[0034] According to one embodiment of the present invention, as analyzed above, when the concentration reading of the O3 sensor is low (e.g., below or equal to the sixth preset value (10 ppb)) and the concentration reading of the SO2 sensor is greater than or equal to 0, it indicates that the NO2 gas concentration in the air is low. If, in this case, the concentration reading of the NO2 sensor is high (e.g., above or equal to the fifth preset value (50 ppb)), it indicates that the NO2 sensor is experiencing positive drift. On the other hand, if the concentration reading of the SO2 sensor is less than -ε, that is, it can be determined that NO2 gas is present in the air and has a certain concentration, in this case, if the concentration reading of the NO2 sensor is 0, it is determined that the NO2 sensor is experiencing negative drift or is malfunctioning.

[0035] In this way, when determining whether the NO2 sensor is drifting, the responses of the O3 and SO2 sensors to NO2 gas can be used, and the concentration readings of the SO2, O3 and NO2 sensors can be compared to determine whether the NO2 sensor is drifting, thus improving the objectivity and accuracy of the judgment.

[0036] Example 5: According to one embodiment of the present invention, if the filter membrane of the CO sensor fails, it will respond to NO2 gas and exhibit a high correlation with the concentration reading of the NO2 sensor. Determining whether each sensor is drifting based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor includes: obtaining the concentration readings of the CO sensor and the NO2 sensor based on the raw measurement values ​​of the CO sensor and the NO2 sensor; determining the Pearson correlation coefficient between the CO sensor and the NO2 sensor based on the concentration readings of the CO sensor and the NO2 sensor at multiple time points; and determining that the CO sensor is drifting if the Pearson correlation coefficient is greater than or equal to a preset coefficient.

[0037] According to one embodiment of the present invention, the correlation between the concentration readings of the CO sensor and the NO2 sensor can be reflected by the Pearson correlation coefficient. If the correlation is high, that is, the Pearson correlation coefficient is greater than or equal to a preset coefficient (e.g., 0.9), and there is no combustion source, it indicates that the concentration readings obtained by the CO sensor fluctuate with the concentration of NO2, and it is determined that the CO sensor filter membrane is faulty and drift exists.

[0038] According to one embodiment of the present invention, it can be further determined whether the TVOC sensor is drifting. For example, the concentration readings of the CO sensor and the TVOC sensor are compared. If the concentration reading of the CO sensor is high (e.g., above a first threshold) while the concentration reading of the TVOC sensor is low (e.g., below a second threshold) or even close to 0, then the TVOC sensor is experiencing negative drift. If the concentration reading of the CO sensor is low (e.g., below the second threshold) or even close to 0 while the concentration reading of the TVOC sensor is high (e.g., above the first threshold), then the TVOC sensor is experiencing positive drift.

[0039] In this way, based on the detection principles of CO and TVOC sensors, the correlation and comparison between the concentration readings of the sensors can be used to determine whether there is drift in CO and TVOC sensors, thus improving the objectivity and accuracy of the judgment.

[0040] Example 6: According to an embodiment of the present invention, in step S3, after the above judgment process, it can be determined whether each sensor has drift. If drift exists, the drift amount is calculated based on the concentration reading, and the drift is calibrated by the drift amount to improve the accuracy of each concentration value.

[0041] According to one embodiment of the present invention, if drift exists, the drift amount is calculated based on the original measurement values, including: obtaining the concentration readings of the SO2 sensor, the NO2 sensor, and the O3 sensor based on the original measurement values ​​of the SO2 sensor, the NO2 sensor, and the O3 sensor; and obtaining the theoretical noise floor voltage V of the SO2 sensor according to formula (3). theory , V theory =S SO2 ×(K {SO2|NO2} ×C measured NO2 +K {SO2|O3} ×C measured O3 )+V base(T) (3) Among them, S SO2 K is the conversion coefficient of the sensor. {SO2|NO2} K is the cross-interference coefficient generated by NO2 on the SO2 sensor. {SO2|O3} C represents the cross-interference coefficient generated by ozone on the SO2 sensor. measured NO2 For the concentration reading of the NO2 sensor, C measured O3 V represents the concentration reading from the O3 sensor. base(T) The reference voltage of the SO2 sensor is used; the voltage deviation value ΔV of the SO2 sensor is determined according to formula (4). ΔV=V measure -V theory (4) Among them, V measure The original measurement value of the SO2 sensor is used; based on the original measurement value of the CO sensor at the current time, it is determined whether the current time is in a clean background environment; if the current time is in a clean background environment, the drift amount V of the SO2 sensor at the current time is obtained according to formula (5). bias(t) , V bias(t) =α×ΔV+(1-α)×V bias(t-1) (5) Among them, V bias(t-1) Let α be the drift of the SO2 sensor at the previous moment, where α is the smoothing factor.

[0042] According to one embodiment of the present invention, taking the calculation of the zero-point drift of an SO2 sensor as an example, both ozone and NO2 affect the SO2 sensor. The theoretical noise floor voltage V of the SO2 sensor can be determined based on the influence of both ozone and NO2 on the SO2 sensor. theory In formula (3), K {SO2|NO2} ×C measured NO2 +K {SO2|O3} ×C measured O3S represents the sum of the interference values ​​generated by NO2 and ozone on the SO2 sensor. SO2 ×(K {SO2|NO2} ×C measured NO2 +K {SO2|O3} ×C measured O3 This indicates that the overall interference value is converted into the voltage value of the SO2 sensor, and after being added to the reference voltage, the theoretical noise floor voltage V of the SO2 sensor is obtained. theory , can represent the theoretical voltage under the influence of NO2 and ozone. Formula (4) can be used to calculate the difference between the original measured value of the SO2 sensor and the noise interference value, that is, the voltage deviation value ΔV of the SO2 sensor. This voltage deviation value can be used to represent the difference between the measured value and the theoretical value of the SO2 sensor. This difference can be used as the zero-point drift. However, if the current time is in a clean background environment, for example, at night, or the current pollutant concentration is determined to be low based on the concentration reading of the CO sensor, then an exponentially weighted moving average can be processed by formula (5), combined with the drift amount V from the previous time. bias(t-1) Using a smoothing factor (between 0.01 and 0.05), calculate the current drift V of the SO2 sensor. bias(t) The zero-point drift is the drift of the detection voltage value of the SO2 sensor. In step S4, the SO2 concentration correction value can be obtained by subtracting this drift from the original measurement value of the SO2 sensor. The process of obtaining correction values ​​for other sensors is similar to the above. That is, firstly, the theoretical noise floor voltage is calculated based on the concentration data of the gas associated with the sensor, the cross-interference coefficient, and the conversion coefficient. Then, the voltage deviation value is calculated based on the original measurement value of the sensor and the theoretical noise floor voltage, which is used as the voltage drift. Furthermore, if the current time is in a clean background environment, the drift value at the current time is calculated by using exponentially weighted moving average processing, combined with the zero-point drift value and smoothing factor from the previous time. Furthermore, in step S4, the correction value can be obtained by subtracting the original measurement value from the drift value.

[0043] In this way, the theoretical noise floor voltage can be determined by the interaction of different types of gases on the sensor through the conversion coefficient and cross-interference coefficient. Then, the voltage deviation value can be calculated, and the drift amount in a clean background environment can be calculated by exponential weighted moving average processing to accurately correct the drift, provide more accurate data for subsequent calibration, and improve the data processing accuracy.

[0044] Example 7: According to an embodiment of the present invention, in step S5, in order to solve the nonlinear mixing interference problem between multiple gases and the nonlinear interference problem between temperature and humidity, a BP neural network model can be used as a calibration model to comprehensively calibrate multiple factors. This leverages the BP neural network model's ability to comprehensively process multiple data nonlinearly to calibrate mutually influencing data. The calibration model is a 3-layer BP neural network model, which can use hidden layer neurons to fit the nonlinear cross-interference relationship between sensors. Its input layer contains 7 input nodes; the hidden layer contains 15-64 nodes, using the ReLU activation function to remove negative voltage noise and fit the temperature and humidity index characteristics; the output layer contains 5 nodes, corresponding to the corrected concentrations of 5 gases. During input, the data after the above zero-point drift correction (i.e., the corrected value), temperature data, and humidity data can be input into the trained calibration model. The output values ​​are the measured concentrations of SO2, NO2, O3, CO, and TVOC. Furthermore, the detection voltage value of the TVOC sensor can be taken as the natural logarithm to linearize its power response characteristics. In other words, [V SO2 V NO2 V O3 V CO ,ln(V TVOC Using [),T,RH] as the input vector, that is, each data in this input vector is input into the above 7 input nodes for calculation, and finally the measured concentration values ​​of SO2, NO2, O3, CO and TVOC are obtained, V SO2 V is the correction value for the SO2 sensor. NO2 V is the correction value for the NO2 sensor. O3 V is the correction value for the O3 sensor. CO V is the correction value for the CO sensor. TVOC Here are the correction values ​​for the TVOC sensor, where T represents temperature data and RH represents humidity data.

[0045] According to an embodiment of the present invention, the training steps of the above-mentioned calibration model include: measuring standard gases using a standard station to obtain multiple sets of calibration concentration values, wherein the calibration concentration values ​​include SO2 calibration concentration values, NO2 calibration concentration values, O3 calibration concentration values, CO calibration concentration values, and TVOC calibration concentration values ​​at multiple times under standard conditions; constructing orthogonal adversarial samples, wherein the concentration of the target gas in the orthogonal adversarial samples is lower than a preset suppression value, and the concentration of gases that have an interfering relationship with the target gas is higher than a preset interference value; constructing a full-environment sample, wherein the full-environment sample includes calibration data of multiple gases under various temperatures and humidity conditions; determining the loss function of the calibration model using the sample calibration data, orthogonal adversarial samples, and full-environment samples, as well as the calibration model, the original measurement values ​​of the standard gases, temperature data, and humidity data; and training the calibration model according to the loss function of the calibration model to obtain a trained calibration model.

[0046] According to one embodiment of the present invention, the number of samples consisting of calibrated concentration values ​​can account for 20% of the total number of samples, which can represent the accurate concentration value of a single gas under standard conditions. These accurate concentration values ​​can be used to train a calibration model, thereby reducing the deviation between the output value of the calibration model and the accurate concentration value.

[0047] According to one embodiment of the present invention, orthogonal adversarial samples account for 40% of the total number of samples. In the orthogonal adversarial samples, the concentration of the target gas is very low, close to 0, and the concentration of the gas that interferes with the target gas is higher than a preset interference value. For example, the SO2 sensor is interfered with by NO2 gas. Therefore, SO2 can be set as the target gas (concentration close to 0) and NO2 as the interfering gas (concentration exceeding 100 ppb) as the sample. This enables the calibration model to acquire the ability to accurately identify low-concentration target gases and determine the accurate concentration of target gases in a strong interference environment during the training process. That is, the ability to identify target gas concentrations close to 0 in a strong interference environment.

[0048] According to one embodiment of the present invention, the total environmental samples account for 40% of the total number of samples. The total environmental samples may include samples ranging from -20°C to 50°C and humidity from 10% to 95%RH, including high temperature and high humidity (drift zone), in order to train the calibration model to accurately calculate the measured concentration values ​​in the total environment.

[0049] According to one embodiment of the present invention, a microstation can be used to measure the same gas as the aforementioned standard station. The raw measurement values ​​from each sensor in the microstation are then corrected for drift, and the corrected values ​​are combined into an input vector in the aforementioned format. This vector is then input into a calibration model, causing the calibration model to output measured concentration values ​​for multiple gases. The measured concentration values ​​are compared with the accurate calibrated concentration values ​​in each sample, and the loss function of the calibration model is calculated. For example, the mean square error between the measured and calibrated concentration values ​​is used as the loss function. The parameters of the calibration model are adjusted using gradient descent. During the adjustment process, the Adam optimizer can be used, with an initial learning rate set to 0.001. During training, the maximum number of training epochs is set to 2000-5000 epochs, and 20% of the total samples are allocated as a validation set. If, during validation in the validation set, the loss function no longer decreases significantly within 50 consecutive epochs, training can be stopped, and the trained calibration model is obtained.

[0050] In this way, multiple training samples are constructed during the training process, enabling the calibration model to accurately identify low-concentration target gases and determine their accurate concentrations in environments with strong interference, as well as to accurately calculate measured concentration values ​​in the entire environment. This improves the robustness and accuracy of the model under various conditions and enhances the accuracy of the calibration process.

[0051] The machine learning-based air microstation sensor interference correction and drift calibration method according to embodiments of the present invention can determine whether multiple sensors are drifting based on the original measurement values. If drift exists, subsequent calibration processing is performed after correcting the drift amount, improving data accuracy and processing accuracy. Furthermore, complex nonlinear mixed interference can be corrected through a neural network model, improving the accuracy of calibration processing and concentration data accuracy. When determining whether the O3 sensor is drifting, the characteristic of the O3 sensor responding to both ozone and NO2 can be used to calculate the virtual ozone residual to determine whether the O3 sensor is drifting negatively, and the characteristics of the concentration readings of the NO2 sensor and the O3 sensor can be used to determine whether the O3 sensor is drifting positively, improving the rationality and accuracy of the judgment. When determining whether the SO2 sensor is drifting, the theoretical baseline deviation of SO2 and the negative influence of NO2 gas on the SO2 sensor are used to comprehensively determine whether the SO2 sensor is drifting, improving the objectivity and accuracy of the judgment. When determining whether an NO2 sensor is drifting, the responses of O3 and SO2 sensors to NO2 gas can be utilized. Comparison of the concentration readings from these sensors improves the objectivity and accuracy of the assessment. Similarly, based on the detection principles of CO and TVOC sensors, the correlation and comparison between their concentration readings can determine whether they are drifting, further enhancing objectivity and accuracy. Furthermore, the theoretical noise floor voltage can be determined by analyzing the interaction of different gases on the sensor through conversion coefficients and cross-interference coefficients. This allows for the calculation of voltage deviation, and the drift amount under clean background conditions can be calculated using exponentially weighted moving averages to accurately correct for drift, providing more accurate data for subsequent calibration and improving data processing precision. During training, multiple training samples are constructed to enable the calibration model to accurately identify low-concentration target gases and determine their precise concentrations under strong interference environments, as well as accurately calculate measured concentrations in all environments. This improves the model's robustness and accuracy under various conditions and enhances the accuracy of calibration processing.

[0052] Figure 2 An exemplary block diagram of a machine learning-based air micro-station sensor interference correction and drift calibration system according to an embodiment of the present invention is shown, the system comprising: The real-time acquisition module is used to acquire the raw measurement values, temperature data, and humidity data of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor in real time. The judgment module is used to determine whether each sensor has drifted based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor and TVOC sensor; A drift measurement module is used to calculate the drift amount based on the original measurement value if drift exists. The correction value module is used to obtain a correction value based on the original measurement value of the sensor with drift and the amount of drift; The calibration module is used to input the correction values, temperature data, and humidity data into the trained calibration model for processing, and to obtain the measured concentration values ​​of SO2, NO2, O3, CO, and TVOC.

[0053] According to one embodiment of the present invention, a machine learning-based air micro-station sensor interference correction and drift calibration device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the machine learning-based air micro-station sensor interference correction and drift calibration method.

[0054] According to one embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions, when executed by a processor, implement the machine learning-based air micro-station sensor interference correction and drift calibration method.

[0055] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0056] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for interference correction and drift calibration of air micro-station sensors based on machine learning, characterized in that, include: Real-time acquisition of raw measurement values, temperature data, and humidity data from SO2, NO2, O3, CO, and TVOC sensors; Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor, determine whether each sensor is drifting. If drift exists, the amount of drift is calculated based on the original measurement value; A correction value is obtained based on the original measurement value of the drifting sensor and the amount of drift. The correction values, temperature data, and humidity data are input into the trained calibration model for processing to obtain the measured concentration values ​​of SO2, NO2, O3, CO, and TVOC.

2. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, Based on the raw measurement values ​​from the SO2, NO2, O3, CO, and TVOC sensors, determine whether each sensor exhibits drift, including: Based on the raw measurement values ​​of the O3 sensor and the NO2 sensor, the concentration readings of the O3 sensor and the NO2 sensor are obtained. According to the formula ΔO3=C measured OX –C measured NO2 Obtain the virtual ozone residual ΔO3, where C measured OX For the concentration reading of the O3 sensor, C measured NO2 This is the concentration reading from the NO2 sensor; If the virtual ozone residual is less than -ε, it is determined that the O3 sensor has a negative drift or has failed, where ε is a preset judgment threshold. If the concentration reading of the NO2 sensor is lower than or equal to the first preset value, and the concentration reading of the O3 sensor is higher than or equal to the second preset value for multiple consecutive moments, then it is determined that the O3 sensor has a positive drift.

3. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, Based on the raw measurement values ​​from the SO2, NO2, O3, CO, and TVOC sensors, determine whether each sensor exhibits drift, including: Based on the raw measurement values ​​of the SO2 sensor and the NO2 sensor, obtain the concentration readings of the SO2 sensor and the NO2 sensor. According to the formula Bias SO2 =C measured SO2 -(W measured NO2 ×K {SO2|NO2} ) Determine the theoretical baseline deviation of SO2. SO2 , where C measured SO2 For the concentration reading of the SO2 sensor, C measured NO2 K represents the concentration reading of the NO2 sensor. {SO2|NO2} The cross-interference coefficient generated by NO2 on the SO2 sensor; If the concentration reading of the NO2 sensor is higher than or equal to the third preset value, and the concentration reading of the SO2 sensor is higher than or equal to 0, then it is determined that the SO2 sensor has a positive drift. If the concentration reading of the NO2 sensor is less than or equal to the fourth preset value, and the concentration reading of the SO2 sensor is less than -ε for multiple consecutive moments, then it is determined that the SO2 sensor has a negative drift, where ε is a preset judgment threshold.

4. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, Based on the raw measurement values ​​from the SO2, NO2, O3, CO, and TVOC sensors, determine whether each sensor exhibits drift, including: Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, and O3 sensor, obtain the concentration readings of the SO2 sensor, NO2 sensor, and O3 sensor. If the concentration reading of the NO2 sensor is higher than or equal to the fifth preset value, the concentration reading of the O3 sensor is lower than or equal to the sixth preset value, and the concentration reading of the SO2 sensor is greater than or equal to 0, then it is determined that the NO2 sensor has a positive drift. If the concentration reading of the SO2 sensor is less than -ε and the concentration reading of the NO2 sensor is 0, then the NO2 sensor is determined to have negative drift or failure.

5. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, Based on the raw measurement values ​​from the SO2, NO2, O3, CO, and TVOC sensors, determine whether each sensor exhibits drift, including: Based on the raw measurement values ​​of the CO sensor and the NO2 sensor, obtain the concentration readings of the CO sensor and the NO2 sensor. The Pearson correlation coefficients of the CO and NO2 sensors were determined based on the concentration readings of the CO sensor and the NO2 sensor at multiple time points. If the Pearson correlation coefficient is greater than or equal to a preset coefficient, then the CO sensor is determined to be drifting.

6. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, If drift exists, the amount of drift is calculated based on the original measurement value, including: Based on the raw measurement values ​​of the SO2 sensor, NO2 sensor, and O3 sensor, obtain the concentration readings of the SO2 sensor, NO2 sensor, and O3 sensor. According to the formula V theory =S SO2 ×(K {SO2|NO2} ×C measured NO2 +K {SO2|O3} ×C measured O3 )+V base(T) Obtain the theoretical noise floor voltage V of the SO2 sensor theory , of which S SO2 K is the conversion coefficient of the sensor. {SO2|NO2} K is the cross-interference coefficient generated by NO2 on the SO2 sensor. {SO2|O3} C represents the cross-interference coefficient generated by ozone on the SO2 sensor. measured NO2 For the concentration reading of the NO2 sensor, C measured O3 V represents the concentration reading from the O3 sensor. base(T) This is the reference voltage for the SO2 sensor; According to the formula ΔV=V measure -V theory Determine the voltage deviation value ΔV of the SO2 sensor, where V measure These are the raw measurement values ​​from the SO2 sensor; Determine whether the current environment is a clean background environment based on the current time or the concentration reading of the CO sensor; If the current environment is a clean background, then according to the formula V bias(t) =α×ΔV+(1-α)×V bias(t-1) Obtain the current drift V of the SO2 sensor. bias(t) , where V bias(t-1) Let α be the drift of the SO2 sensor at the previous moment, where α is the smoothing factor.

7. The method for interference correction and drift calibration of air micro-station sensors based on machine learning according to claim 1, characterized in that, The method further includes: Standard gas was measured using a standard station to obtain multiple sets of calibrated concentration values. These calibrated concentration values ​​included SO2, NO2, O3, CO, and TVOC calibrated concentration values ​​at multiple times under standard conditions. Construct orthogonal adversarial samples, wherein the concentration of the target gas in the orthogonal adversarial samples is lower than a preset inhibition value, and the concentration of the gas that has an interfering relationship with the target gas is higher than a preset interference value; Construct a full-environment sample, which includes calibration data for various gases under multiple temperatures and humidity levels; The loss function of the calibration model is determined using the sample calibration data, orthogonal adversarial samples and full environmental samples, as well as the original measurement values ​​of the calibration model and standard gas, temperature data and humidity data. The calibration model is trained based on its loss function to obtain the trained calibration model.

8. A machine learning-based air micro-station sensor interference correction and drift calibration system, characterized in that, include: The real-time acquisition module is used to acquire the raw measurement values, temperature data, and humidity data of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor, and TVOC sensor in real time. The judgment module is used to determine whether each sensor has drifted based on the original measurement values ​​of the SO2 sensor, NO2 sensor, O3 sensor, CO sensor and TVOC sensor; A drift measurement module is used to calculate the drift amount based on the original measurement value if drift exists. The correction value module is used to obtain a correction value based on the original measurement value of the sensor with drift and the amount of drift; The calibration module is used to input the correction values, temperature data, and humidity data into the trained calibration model for processing, and to obtain the measured concentration values ​​of SO2, NO2, O3, CO, and TVOC.

9. A machine learning-based air micro-station sensor interference correction and drift calibration device, characterized in that, include: processor; A memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions that, when executed by a processor, implement the method as described in any one of claims 1-7.

Images