Sensor signal attenuation calibration method, apparatus, device, and storage medium

Abstract

The application provides a sensor signal attenuation calibration method, device, equipment and storage medium, and relates to the technical field of computers.The method comprises the following steps: acquiring a real-time detection signal of a target sensor; determining a target feature value corresponding to the real-time detection signal under each first feature dimension in a feature dimension set; the first feature dimension is strongly related to the concentration of a to-be-measured object; inputting the target feature value under each first feature dimension into a concentration estimation model for concentration calibration processing to determine a calibrated concentration corresponding to the to-be-measured object; the concentration estimation model is obtained by pre-training according to multiple sample feature values under the first feature dimension and the concentration of a sample object corresponding to each sample feature value; each sample feature value and the concentration of each sample object are determined according to sample detection signals of sample sensors with different aging degrees under different working conditions. The technical scheme can improve the accuracy of the sensor detection signal.

Description

Sensor signal attenuation calibration methods, devices, equipment and storage media Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a sensor signal attenuation calibration method, apparatus, device, and storage medium. Background Technology

[0002] Sensors, as a convenient measurement method, have a wide range of applications. Taking electrochemical sensors as an example, their advantages, such as high sensitivity, good selectivity, fast response speed, and low cost, have led to their widespread use in environmental monitoring, industrial safety, automotive battery management, and health monitoring. However, during long-term use, electrochemical sensors experience aging phenomena such as reduced electrode activity, electrolyte consumption, and membrane fouling. This leads to a gradual attenuation of the sensor's output signal. Signal attenuation directly alters the correlation between the sensor's response signal and the target gas concentration, distorting the concentration estimation results based on the original calibration model. Consequently, the sensor becomes unusable, affecting its lifespan and detection reliability.

[0003] In related technologies, in order to detect and calibrate sensors after they have aged, methods such as manually set thresholds, lookup tables, or empirical formulas are usually used to obtain the signal values ​​that need to be compensated, and then the measured values ​​of the sensors are compensated using the signal values ​​that need to be compensated.

[0004] However, under conditions of long-term sensor use, the above-mentioned technologies still suffer from the problem of low accuracy of the detection signal after calibration or compensation. Summary of the Invention

[0005] This invention provides a sensor signal attenuation calibration method, apparatus, device, and storage medium to address the shortcomings of existing technologies where the accuracy of sensor detection signals remains low even after long-term use. It achieves accurate calibration or compensation of sensor signal attenuation by calculating feature values ​​strongly correlated with aging characteristics and concentration from the measured sensor signal, and then inputting these feature values ​​into a pre-trained concentration estimation model for concentration calibration. This improves the accuracy of the obtained sensor detection signal, ensuring that the aged sensor can overcome the effects of signal attenuation and achieve accurate detection of gas concentration.

[0006] This invention provides a sensor signal attenuation calibration method, comprising:

[0007] Acquire the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object;

[0008] Based on the real-time detection signal, the target feature value corresponding to each first feature dimension in the feature dimension set is determined; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be measured and is determined according to the sample detection signal of the sample sensor with different aging degree under different operating conditions.

[0009] The target feature value under each first feature dimension is input into the concentration estimation model for concentration calibration to determine the calibration concentration corresponding to the object to be measured. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity and the concentration of the sample object.

[0010] According to a sensor signal attenuation calibration method provided by the present invention, before determining the target feature value corresponding to the real-time detection signal in each first feature dimension of the feature dimension set based on the real-time detection signal, the method further includes:

[0011] Acquire sample detection signals from sensors with different aging levels under different operating conditions;

[0012] Based on the detection signals of each group of samples, determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions; the multiple different feature dimensions include the first feature dimension;

[0013] For each aging level of the sample sensor, based on the correlation between each sample feature value of the detection signal of each group of samples corresponding to the aging level of the sample sensor and the concentration of different sample objects, the first feature dimension under the aging level of the sample sensor is determined among multiple different feature dimensions; the first feature dimension under each aging level of the sample sensor is strongly correlated with the concentration of the sample object under the corresponding aging level.

[0014] The set of feature dimensions is determined based on the first feature dimension of the sample sensors with various aging levels.

[0015] According to a sensor signal attenuation calibration method provided by the present invention, the method determines a first feature dimension under the aging degree of the sample sensor based on the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects among multiple different feature dimensions, including:

[0016] Calculate the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects, and obtain multiple first correlation coefficients corresponding to the aging degree sample sensor;

[0017] Based on multiple first correlation coefficients corresponding to the aging degree of the sample sensor, determine at least one set of sample feature values ​​and the corresponding concentration of the sample object whose first correlation coefficients satisfy the strong concentration correlation condition.

[0018] The feature dimension corresponding to the feature value of the sample that meets the strong correlation condition is determined as the first feature dimension under the aging degree of the sample sensor.

[0019] According to a sensor signal attenuation calibration method provided by the present invention, the method further includes:

[0020] Acquire sample detection signals from sensors with different aging levels under different operating conditions;

[0021] Based on the detection signals of each group of samples, determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions;

[0022] Based on the correlation between the sample feature values ​​at the concentration of each sample object and different aging degrees, a set of sample feature values ​​that meet the strong correlation condition at the concentration of each sample object are determined.

[0023] A set of sample feature values ​​that meet the strong correlation condition at the concentration of each sample object are used as the label sample feature values ​​at the concentration of the corresponding sample object; the above label sample feature values ​​are used to characterize the aging degree of the sample sensor at the concentration of the corresponding sample object.

[0024] The degree of aging of the target sensor is determined based on the label sample characteristic values ​​at the concentration of each sample object.

[0025] According to a sensor signal attenuation calibration method provided by the present invention, the determination of the aging degree of the target sensor based on the tag sample characteristic value at the concentration of each sample object includes:

[0026] The target sensor is used to detect the concentration of the test object at the concentration of each sample object, and the first detection signal of the target sensor at the concentration of each sample object is obtained.

[0027] Based on the first detection signal at the concentration of each sample object, calculate the first feature value of the target sensor at the feature dimension corresponding to the label sample feature value at the concentration of each sample object;

[0028] Based on each first feature value and the corresponding feature value of the label sample under the feature dimension, determine a set of first feature values ​​and label sample feature values ​​that match the feature values.

[0029] The feature values ​​of the tag samples matched with the feature values ​​are determined as the aging degree corresponding to the target sensor.

[0030] According to the sensor signal attenuation calibration method provided by the present invention, the aforementioned multiple different feature dimensions include a signal baseline feature dimension, a peak region feature dimension, and a peak baseline feature dimension. The method for determining the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions based on each group of sample detection signals includes:

[0031] For each set of sample detection signals, the sample feature value of the sample detection signal under the signal baseline feature dimension is calculated based on the signal value of the sample detection signal at the start or end of a preset time period.

[0032] Based on the signal value of the sample detection signal in the peak region and the corresponding time, determine the sample feature value of the sample detection signal in the feature dimension of the peak region;

[0033] Based on the signal value of the sample detection signal in the peak region and the signal value of the sample detection signal during a preset time period at the start or end of the signal, the feature value of the sample detection signal under the peak baseline feature dimension is calculated.

[0034] According to a sensor signal attenuation calibration method provided by the present invention, the method further includes:

[0035] Acquire sample detection signals from sensors with different aging levels under different operating conditions;

[0036] Based on the detection signals of each group of samples, determine the sample feature value of each group of sample detection signals under each first feature dimension;

[0037] The sample feature values ​​of each group of sample detection signals under each first feature dimension are input into the initial concentration estimation model for concentration calibration to determine the predicted concentration corresponding to each group of sample detection signals.

[0038] The initial concentration estimation model is trained based on the loss between the predicted concentration of each sample detection signal and the concentration of the corresponding sample object, as well as the regularization loss corresponding to the initial concentration estimation model, to obtain the concentration estimation model.

[0039] The present invention also provides a sensor signal attenuation calibration device, comprising the following modules:

[0040] The signal acquisition module is used to acquire the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object;

[0041] The feature value determination module is used to determine the target feature value corresponding to each first feature dimension in the feature dimension set based on the real-time detection signal; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be measured and is determined based on the sample detection signal of the sample sensor with different aging degree under different operating conditions.

[0042] The signal attenuation calibration module is used to input the target feature value under each first feature dimension into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be measured. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity and the concentration of the sample object.

[0043] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the sensor signal attenuation calibration method as described above.

[0044] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the sensor signal attenuation calibration method as described above.

[0045] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the sensor signal attenuation calibration method as described above.

[0046] The sensor signal attenuation calibration method, apparatus, device, and storage medium provided by this invention acquires the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object. Based on the real-time detection signal, the target feature value corresponding to each first feature dimension in the feature dimension set is determined. Then, the target feature value under each first feature dimension is input into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the target object. Each first feature dimension in the feature dimension set is strongly correlated with the concentration of the target object and is determined based on the sample detection signals of the sample sensor with different aging degrees under different operating conditions. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimensions and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of the sample sensor with different aging degrees under different operating conditions. Different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity, and the concentration of the sample object. In this method, since the characteristic value that is strongly correlated with aging characteristics and concentration can be calculated from the actual sensor signal, and the concentration can be calibrated by inputting the characteristic value into the pre-trained concentration estimation model, the attenuation of the sensor signal can be accurately calibrated or compensated. This enables accurate concentration measurement throughout the entire life cycle of the electrochemical sensor, ensuring that the aging sensor still has stable detection performance and reliability under complex working conditions. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0048] Figure 1 is a flowchart illustrating the sensor signal attenuation calibration method provided by the present invention.

[0049] Figure 2 is a schematic diagram of the output signal of the alcohol sensor provided by the present invention under a certain operating condition.

[0050] Figure 3 is a schematic diagram of the sensor signal attenuation calibration device provided by the present invention.

[0051] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0053] Electrochemical sensors are widely used in environmental monitoring, industrial safety, automotive battery management, and health monitoring due to their advantages such as high sensitivity, good selectivity, fast response speed, and low cost. However, during long-term service, these sensors experience aging phenomena such as reduced electrode activity, electrolyte consumption, and membrane fouling, leading to a gradual attenuation of the sensor's output signal. This signal attenuation directly alters the correspondence between the sensor's response signal and the target gas concentration, distorting the concentration estimation results based on the original calibration model, and consequently affecting the sensor's lifespan and detection reliability.

[0054] Currently, some technologies utilize manually set thresholds, lookup tables, or empirical formulas to achieve a certain degree of detection and calibration of electrochemical sensors after aging. However, in complex and variable real-world applications, the aging process of sensors is influenced by a combination of factors such as temperature, humidity, and electrode decay. Relying solely on fixed thresholds or empirical formulas is insufficient to accurately characterize signal attenuation characteristics, often leading to insufficient calibration accuracy and limited applicability. Especially under long-term service conditions, current technologies lack the ability to dynamically assess the aging stage of sensors, making it difficult to guarantee the reliability and consistency of detection results. Some studies have also attempted to use black-box models for signal compensation, but these methods cannot adaptively adjust to the dynamic changes in the sensor aging process, resulting in a significant decrease in detection accuracy and a shortened effective lifespan of the sensor under long-term service conditions. Furthermore, current technologies also have the following minor drawbacks: 1. Insufficient data mining: Current technologies often rely on single variables or simplified formulas, failing to fully utilize the large amount of data accumulated during aging experiments, resulting in insufficient model generalization ability. 2. Limited scalability: Most methods are designed for specific sensors or specific gas scenarios, lacking adaptability across sensor types or multi-gas detection.

[0055] Based on this, embodiments of the present invention provide a sensor signal attenuation calibration method, apparatus, device, and storage medium, which can solve the above-mentioned technical problems. Specifically, the electrochemical sensor signal attenuation is calibrated by a data-driven approach that combines aging experimental data with machine learning algorithms. This enables online identification of the aging stage and accurate estimation of the true concentration, exhibiting stronger adaptability and robustness. Furthermore, it maximizes the use of experimental tests and real-time information, ensuring that the sensor maintains high-precision detection throughout its entire lifespan, thereby significantly improving the reliability of the sensor.

[0056] First, let's explain the proper nouns.

[0057] Baseline: Refers to the stable output signal value of an electrochemical sensor when it is not exposed to the target substance, i.e., the initial reference state of the sensor. The baseline is used to assess the signal deviation of the sensor during the detection process and is an important reference for judging the response strength and stability.

[0058] It should be noted that the execution subject of the embodiments of the present invention may be a sensor signal attenuation calibration device, or may include electronic equipment, or may include other devices, apparatuses, or systems, etc., without specific limitations. The following embodiments will use an electronic device as an example for illustration.

[0059] Figure 1 is a schematic flowchart of the sensor signal attenuation calibration method provided by the present invention. As shown in Figure 1, the method includes the following steps:

[0060] Step 102: Obtain the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object.

[0061] In this step, the target sensor can be an electrochemical sensor / gas sensor that detects the concentration of any type of gas, such as a sensor for detecting carbon dioxide concentration, a sensor for detecting oxygen concentration, or an alcohol sensor for detecting the concentration of alcohol vapors. The target sensor can be a sensor of any aging degree.

[0062] The target object refers to the object whose concentration is specifically detected by the target sensor. For example, it can be a gas to be measured. In other words, the type of the target object matches the type of object that the target sensor can detect.

[0063] Specifically, when it is necessary to detect the concentration of the target object in real time, the target sensor can be placed in the space where the target object is located. The concentration of the target object can be detected in real time by the target sensor to obtain a real-time detection signal. This real-time detection signal can be an electrical signal, such as a current signal.

[0064] Step 104: Based on the real-time detection signal, determine the target feature value corresponding to each first feature dimension in the feature dimension set; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be tested and is determined according to the sample detection signal of the sample sensor with different aging degree under different operating conditions.

[0065] This process involves pre-collecting sample sensors of the same type as the target sensor but at different aging stages, such as electrochemical sensors. These different aging stages can cover the entire lifecycle of the target sensor. After obtaining these sample sensors at different aging stages, the concentration of the sample under different operating conditions can be detected using these sensors. This yields sample detection signals (e.g., electrical signals) from the sensor at different aging stages under various operating conditions. These different operating conditions include at least one difference in environmental temperature, humidity, and sample concentration. The type of sample can be the same as the type of object being measured, such as targeting the same type of gas. Then, the characteristic information of each sample detection signal is determined across different feature dimensions, and the correlation between the characteristic information and concentration under each feature dimension is analyzed. Based on this correlation, the feature dimension with a strong correlation (i.e., a relatively strong correlation) with concentration is selected as the first feature dimension. Alternatively, the correlation between the characteristic information and concentration and aging stage under each feature dimension can be analyzed to select the feature dimension with a strong correlation (i.e., a relatively strong correlation) with both concentration and aging stage, which is then selected as the first feature dimension. All the found first feature dimensions can be combined to form a feature dimension set. It can be understood that the feature dimension set may include one or more first feature dimensions; if it includes multiple first feature dimensions, these multiple first feature dimensions can be different feature dimensions.

[0066] In addition, the above-mentioned different feature dimensions may include signal baseline feature dimension, peak region feature dimension, and peak baseline feature dimension, etc. More specifically, they may include initial baseline feature dimension, final baseline feature dimension, peak corresponding time feature dimension, peak amplitude feature dimension, feature dimension of the time before and after the peak reaching the difference between the initial baseline and the peak at 3 / 8 of the time, feature dimension of the integral area of ​​the overall response of the curve between the time before and after the peak reaching the difference between the initial baseline and the peak at 3 / 8 of the time, and peak baseline ratio feature dimension.

[0067] Specifically, after obtaining the real-time detection signal acquired by the target sensor, the feature value of the real-time detection signal under each first feature dimension in the feature dimension set can be calculated, and denoted as the target feature value of the real-time detection signal under each first feature dimension. Here, the first feature dimension can be one of the feature dimensions mentioned above, such as the initial baseline feature dimension, the integral area feature dimension of the overall response of the curve between the peak and the 3 / 8 time interval between the peak and the initial baseline and the peak, etc. The target feature value under the corresponding feature dimension can be, for example, the mean of the initial baseline, the integral area of ​​the overall response of the curve between the peak and the 3 / 8 time interval between the peak and the initial baseline and the peak, etc.

[0068] It should be noted that each of the first feature dimensions here is a feature dimension that is strongly correlated with the concentration of the object to be measured. In other words, the target feature value of each first feature dimension calculated by real-time detection signal is strongly correlated with the concentration of the object to be measured, and can reflect the true concentration of the object to be measured more accurately. Therefore, the calibrated concentration obtained by subsequent concentration calibration through the target feature value of the first feature dimension is more accurate.

[0069] Step 106: Input the target feature value under each first feature dimension into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be tested; wherein, the concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity and the concentration of the sample object.

[0070] Specifically, the concentration of sample objects under different operating conditions can be detected in advance using sample sensors with different aging levels to obtain sample detection signals of sample sensors with different aging levels under different operating conditions. The sample feature value of each sample detection signal under each first feature dimension can be determined through each sample detection signal. Then, the sample feature value of each group of sample detection signals under each first feature dimension is used as the input of the initial concentration estimation model, and the concentration of the sample object corresponding to each group of sample detection signals is used as the reference output (i.e., training label) of the initial concentration estimation model to train the initial concentration estimation model and obtain the concentration estimation model.

[0071] Furthermore, the specific architecture or type of the concentration estimation model described above can be set according to the actual situation. For example, it can be a machine learning model, such as a machine learning model based on the XGBoost regression algorithm (Extreme Gradient Boosting Regressor), which can be referred to as a regression model. The XGBoost regression algorithm is an ensemble learning algorithm based on Gradient Boosting Decision Tree (GBDT). It improves the model's fitting accuracy and generalization ability by iteratively training multiple regression trees and weighted fusion. Its basic form can be expressed as:

[0072] ;

[0073] Where, x i This represents the detection signal of the i-th sample group. This is the predicted output (i.e., predicted concentration) of the detection signal for the i-th sample group. Let K represent the k-th regression tree, where K is the total number of regression trees. It is a tree function space.

[0074] After the concentration estimation model is trained, the target feature values ​​of the real-time detection signal under each first feature dimension can be input into the concentration estimation model. The concentration estimation model estimates the current concentration of the target object detected by the target sensor, i.e., performs concentration calibration or signal attenuation calibration of the target sensor. After the processing is completed, the calibrated concentration can be output, i.e., the current calibrated concentration of the target object detected by the target sensor is obtained.

[0075] As described above, this embodiment of the invention constructs an online concentration estimation model by mining the correlation features between the sensor output and the actual concentration during the aging process. Compared with traditional empirical methods, the method of this embodiment can effectively eliminate signal attenuation offset caused by aging, achieve dynamic calibration of the target sensor's detection accuracy, and has stronger adaptability and robustness. It can maximize the use of experimental test information and real-time electrical signals under complex operating conditions, achieving high-precision detection of gas concentration by the aged sensor throughout its entire life cycle, thereby significantly improving the sensor's reliability and application value. Simultaneously, the constructed concentration estimation model allows electrochemical sensors at any stage of their life cycle to calibrate their detected gas concentration using specific features obtained from their detection signals and the concentration estimation model, achieving calibration of the electrochemical sensor's signal attenuation. This eliminates the need for frequent replacement of electrochemical sensors, greatly increasing their lifespan and saving production costs.

[0076] In this embodiment, the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object is acquired. Based on the real-time detection signal, the target feature value corresponding to each first feature dimension in the feature dimension set is determined. Then, the target feature value under each first feature dimension is input into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the target object. Each first feature dimension in the feature dimension set is strongly correlated with the concentration of the target object and is determined according to the sample detection signal of the sample sensor with different aging degrees under different operating conditions. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined according to the sample detection signal of the sample sensor with different aging degrees under different operating conditions. Different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity and the concentration of the sample object. In this method, since the characteristic value that is strongly correlated with aging characteristics and concentration can be calculated from the actual sensor signal, and the concentration can be calibrated by inputting the characteristic value into the pre-trained concentration estimation model, the attenuation of the sensor signal can be accurately calibrated or compensated. This enables accurate concentration measurement throughout the entire life cycle of the electrochemical sensor, ensuring that the aging sensor still has stable detection performance and reliability under complex working conditions.

[0077] The above embodiments describe the process of determining the feature dimension set. The following embodiments will describe this process in detail.

[0078] In one embodiment, before determining the target feature value corresponding to the real-time detection signal under each first feature dimension in the feature dimension set based on the real-time detection signal in step 104 above, the method further includes the following steps:

[0079] Step A1: Obtain sample detection signals from sensor sensors with different aging levels under different operating conditions.

[0080] In this embodiment, the sample detection signals are the experimental results of the sample sensor under multiple controlled test conditions. First, the test environment conditions are set, i.e., the test conditions are defined, including temperature, humidity, and sample concentration. The temperature condition covers multiple (e.g., 8) uniformly measured points within the working environment range of the sample sensor; the humidity condition covers multiple (e.g., 4) uniformly measured points within the working environment range; and the concentration condition covers multiple (e.g., 8) uniformly measured points within the sensor's measurement range. Through the permutation and combination of the above conditions, multiple sets (e.g., 8×4×8=256 sets) of experimental test conditions are constructed. The measurement range corresponding to the concentration can be, for example, 0-400 BAC, and the temperature range corresponding to the temperature can be, for example, -20~40℃. The sample object can be the same as the object to be measured, for example, both being a single gas.

[0081] In addition, before the testing experiment begins, accelerated aging of the sensors can be achieved by performing different numbers of tests on the same type of sample sensors under the median concentration condition (i.e., the median concentration of the corresponding measurement range), thus obtaining sample sensors with different aging degrees. Simultaneously, under each set of test conditions, multiple (e.g., 10) sample sensors at different aging degrees are deployed for the experiment. Each sample sensor needs to complete three sets of independent tests, with a one-day interval between each set. In each set of tests, the same concentration is measured twice consecutively to ensure data stability and repeatability.

[0082] During the testing process, sample detection signals related to the training of the concentration estimation model were recorded and extracted. These sample detection signals were multidimensional data, including:

[0083] Label data for environmental conditions (temperature, humidity, concentration);

[0084] Timestamp data, denoted as t (i.e., the time series of the test process);

[0085] Electrochemical potential signal data is denoted as S (i.e., the output of the sample sensor).

[0086] The above data, after being processed, is used as input samples for training the concentration estimation model, so as to achieve modeling and calibration of the signal attenuation of the sensor at different aging stages.

[0087] Step A2: Based on the detection signals of each group of samples, determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions; the multiple different feature dimensions include the first feature dimension.

[0088] The aforementioned various feature dimensions include signal baseline feature dimensions, peak region feature dimensions, and peak baseline feature dimensions. More specifically, they may include initial baseline feature dimensions, final baseline feature dimensions, peak corresponding time feature dimensions, peak amplitude feature dimensions, the feature dimensions of the time before and after the peak when the difference between the initial baseline and the peak value is reached (3 / 8 of the time), the integral area feature dimension of the overall curve response between the time before and after the peak when the difference between the initial baseline and the peak value is reached (3 / 8 of the time), and peak baseline ratio feature dimensions. These feature dimensions are all capable of characterizing the aging characteristics and concentration response properties of the sensor.

[0089] Optionally, step A2 above, which determines the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions based on the detection signals of each group of samples, may include:

[0090] For each set of sample detection signals, the sample feature value of the sample detection signal under the signal baseline feature dimension is calculated based on the signal value of the sample detection signal at the start or end of a preset time period.

[0091] Based on the signal value of the sample detection signal in the peak region and the corresponding time, determine the sample feature value of the sample detection signal in the feature dimension of the peak region;

[0092] Based on the signal value of the sample detection signal in the peak region and the signal value of the sample detection signal during a preset time period at the start or end of the signal, the feature value of the sample detection signal under the peak baseline feature dimension is calculated.

[0093] Specifically, taking an alcohol sensor as an example, refer to Figure 2, which shows the output signal of the alcohol sensor under a certain operating condition. The figure illustrates the sample detection signal of the alcohol sensor under this condition. The horizontal axis represents time (in seconds), and the vertical axis represents signal amplitude. The blue curve in the figure is the sensor curve (SensorCurve), denoted as S(t). The red dashed line is the straight line corresponding to the peak value. The green dashed line is the straight line corresponding to the amplitude at 3 / 8 of the difference between the initial baseline and the peak value, i.e., the 3 / 8 Threshold. The blue curve and the purple dashed line are the straight lines corresponding to the 3 / 8 of the difference between the initial baseline and the peak value, i.e., the Fall index. The above sample detection signal curves encompass the electrochemical potential signal output by the sample sensor and the timestamp data. Based on this, the sample feature values ​​corresponding to the sample detection signal under the above eight feature dimensions can be calculated using feature engineering extraction methods. The specific calculation process is as follows:

[0094] 1. Calculate the initial baseline mean under the initial baseline feature dimension, and use this initial baseline mean as the sample feature value under this initial baseline feature dimension. Specifically, this includes: extracting the initial baseline mean F1, and letting the sampling point be S(t). i ), i=1,2,…,N b , where N b This refers to the number of sampling points within the initial baseline interval (i.e., the preset time period, such as the first 20 seconds):

[0095] ;

[0096] 2. Calculate the peak position in the feature dimension corresponding to the peak time, and use this peak position as the sample feature value in the feature dimension corresponding to the peak time. Specifically, this includes: extracting the peak position F2, i.e., the time when the signal reaches its maximum value.

[0097] ;in, ;

[0098] 3. Calculate the peak amplitude under the peak amplitude feature dimension, and use this peak amplitude as the sample feature value under this peak amplitude feature dimension. Specifically, this includes: extracting the peak amplitude F3 corresponding to the peak position.

[0099] ;

[0100] 4-5. Calculate the first 3 / 8 and last 3 / 8 moments under the feature dimension of the 3 / 8 moment difference between the initial baseline and the peak value, and use these first 3 / 8 and last 3 / 8 moments as sample feature values ​​under the feature dimension of the 3 / 8 moment difference between the initial baseline and the peak value. Specifically, this includes extracting the position F4 of the first 3 / 8 moment and the position F5 of the last 3 / 8 moment.

[0101] ;in, ;

[0102] The positions F4 (first 3 / 8 of the time) and F5 (last 3 / 8 of the time) can be used to characterize the dynamic process before and after the peak, which facilitates the accurate selection of feature dimensions that are strongly correlated with concentration.

[0103] 6. During the signal decline phase, calculate the mean of the final baseline under the final baseline feature dimension, and use this mean of the final baseline as the sample feature value under this final baseline feature dimension. Specifically, this includes: extracting the mean of the final baseline F6.

[0104] ;

[0105] Wherein, S(t) j ) represents the signal value of the j-th sampling point in the final baseline interval (i.e., a preset time period, such as the first 20 seconds), N e This represents the total number of sampling points within the final baseline interval;

[0106] 7. Calculate the integral area under the integral area feature dimension of the overall response of the curve between the initial baseline and the peak value at 3 / 8 of the time interval before and after the peak value, and use this integral area as the sample feature value under the integral area feature dimension of the overall response of the curve between the initial baseline and the peak value at 3 / 8 of the time interval before and after the peak value. Specifically, this includes calculating the integral area F7 of the overall response of the curve between F4 and F5, which is used to measure the cumulative response strength of the signal over time.

[0107] ;

[0108] Wherein, S(t) k ) represents the signal value at the k-th sampling point; Δt represents the time interval between adjacent sampling points; N represents the total number of sampling points from position F4 in the first 3 / 8 of the time interval to position F5 in the last 3 / 8 of the time interval;

[0109] 8. Calculate the peak baseline ratio under the peak baseline ratio feature dimension, and use this peak baseline ratio as the sample feature value under this peak baseline ratio feature dimension. Specifically, this includes constructing the peak baseline ratio F8, which reflects the degree of enhancement of the peak signal relative to the baseline.

[0110] ;

[0111] The above process can be used to calculate the sample feature value of each of the eight different feature dimensions for each sample detection signal.

[0112] Step A3: For each aging level of the sample sensor, based on the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging level sample sensor and the concentration of different sample objects, determine the first feature dimension under the aging level sample sensor among multiple different feature dimensions; the first feature dimension under each aging level sample sensor is strongly correlated with the concentration of the sample object under the corresponding aging level.

[0113] In this step, through the aforementioned feature value calculation process, a set of sample feature values ​​for each aging level of the sensor under different operating conditions can be calculated. This set of sample feature values ​​includes the sample feature values ​​of the sample detection signal under the corresponding operating conditions in each of the eight different feature dimensions. Then, by keeping the temperature and humidity variables constant, the correlation between all sets of sample feature values ​​and all concentrations is calculated sequentially at different aging levels / states, and the sample feature values ​​or feature dimensions that show a strong correlation with concentration under all aging levels are selected.

[0114] For each aging level of the sample sensor, optionally, in this step, based on the correlation between each sample feature value of the detection signal of each group of samples corresponding to the aging level of the sample sensor and the concentration of different sample objects, the first feature dimension under the aging level of the sample sensor is determined among multiple different feature dimensions, which may include:

[0115] Calculate the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects, and obtain multiple first correlation coefficients corresponding to the aging degree sample sensor;

[0116] Based on multiple first correlation coefficients corresponding to the aging degree of the sample sensor, determine at least one set of sample feature values ​​and the corresponding concentration of the sample object whose first correlation coefficients satisfy the strong concentration correlation condition.

[0117] The feature dimension corresponding to the feature value of the sample that meets the strong correlation condition is determined as the first feature dimension under the aging degree of the sample sensor.

[0118] For each aging level of the sample sensor, there are multiple sample detection signals under various operating conditions. The temperature and humidity variables in these conditions are kept constant. At that aging level, the correlation between each sample feature value in a set of sample feature values ​​corresponding to the sample detection signals under all concentration conditions and the concentration under all concentration conditions is calculated. Specifically, the Pearson correlation coefficient method can be used for calculation, and the formula is as follows:

[0119] ;

[0120] Where, x i This represents the i-th sample feature value in a set of sample feature values ​​corresponding to the sample detection signal. y represents the mean of the sample feature values. i This represents the concentration corresponding to the i-th sample feature value in a set of sample feature values ​​corresponding to the sample detection signal. The value represents the mean concentration, n represents the number of sample feature values ​​(or the number of sample feature values ​​in each group of sample feature values), and r represents the mean concentration. X,Y The correlation coefficient represents the correlation between sample feature value X and concentration Y. The larger the correlation coefficient, the stronger the correlation between the sample feature value and the concentration.

[0121] The correlation coefficient between each sample feature value and concentration at this aging level can be calculated using the above method, and is denoted as the first correlation coefficient. Then, among these multiple first correlation coefficients, at least one first correlation coefficient that satisfies the strong concentration correlation condition can be found. This strong concentration correlation condition could be, for example, finding the largest or most significant first correlation coefficient. This method yields the largest or most significant first correlation coefficient, denoted as the target first correlation coefficient. Subsequently, the sample feature value and concentration used in its calculation can be obtained from the target first correlation coefficient, denoted as the target sample feature value and target concentration. Simultaneously, the feature dimension corresponding to this target sample feature value can be obtained, i.e., the feature dimension corresponding to the sample feature value that satisfies the strong concentration correlation condition, which serves as the first feature dimension after strong concentration correlation screening for samples at this aging level.

[0122] This method allows us to identify the primary feature dimension that is strongly correlated with concentration at each aging stage. It is understandable that the primary feature dimension strongly correlated with concentration may be the same or different at different aging stages.

[0123] Step A4: Determine the set of feature dimensions based on the first feature dimension of the sample sensors for various aging degrees.

[0124] In this step, after obtaining the first feature dimensions that are strongly correlated with concentration under different aging degrees, these first feature concentrations can be combined and deduplicated to obtain a feature dimension set consisting of one or more first feature dimensions.

[0125] In this embodiment, by collecting the detection signals of each sample from sensors at each aging level, a set of multidimensional sample feature values ​​corresponding to each sample detection signal is determined. Then, by analyzing the correlation between each set of multidimensional sample feature values ​​and the concentration, the feature dimensions strongly correlated with the concentration at each aging level are determined, and a set of feature dimensions is obtained accordingly. This allows for the rapid and accurate identification of feature dimensions strongly correlated with both concentration and aging level, improving the accuracy of features calculated for subsequent concentration estimation. Furthermore, by calculating the correlation coefficient between each set of multidimensional sample feature values ​​and the concentration at each aging level, sample feature values ​​that satisfy the concentration-strong correlation condition are identified, and their corresponding feature dimensions are used as the first feature dimension at the corresponding aging level. This further improves the efficiency and accuracy of determining feature dimensions strongly correlated with both concentration and aging level.

[0126] In actual testing, it is sometimes necessary to understand the aging degree of the target sensor. Based on this, this invention presents a scheme for quickly determining the aging degree of the target sensor based on sample feature values. The following embodiments illustrate this.

[0127] In one embodiment, the above method further includes the following steps:

[0128] Step B1: Obtain sample detection signals from sensor sensors with different aging levels under different operating conditions.

[0129] For an explanation of this step, please refer to the explanation of step A1 above, which will not be repeated here.

[0130] Step B2: Based on the detection signals of each group of samples, determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions.

[0131] For an explanation of this step, please refer to the explanation of step A2 above, which will not be repeated here.

[0132] Step B3: Based on the correlation between the sample feature values ​​at the concentration of each sample object and different degrees of aging, determine a set of sample feature values ​​that satisfy the strong correlation condition at the concentration of each sample object.

[0133] Step B4: Take a set of sample feature values ​​that meet the strong correlation condition at the concentration of each sample object as the label sample feature value at the concentration of the corresponding sample object; the above label sample feature value is used to characterize the aging degree of the sample sensor at the concentration of the corresponding sample object.

[0134] In steps B3 and B4, the temperature and humidity variables can be kept constant. The correlation between the feature values ​​of all groups of samples and the degree of aging is calculated at different concentrations and at all aging levels. The feature values ​​of samples that have a strong correlation with the degree of sensor aging at different concentrations (i.e., high correlation) are selected.

[0135] Specifically, for each concentration, multiple sets of sample detection signals from sensors with different aging levels under different temperature and humidity conditions at the same concentration can be generated. With temperature and humidity variables kept constant, at that concentration, the correlation between each sample feature value in a set of sample feature values ​​corresponding to each sample detection signal under all aging levels and the aging level is calculated. This can be done using the Pearson correlation coefficient method, and the calculation formula can also be the aforementioned r... X,Y The calculation formula, but x in it i This represents the i-th sample feature value in a set of sample feature values ​​corresponding to the sample detection signal. y represents the mean of the sample feature values. i This represents the aging degree of the sample sensor corresponding to the i-th aging degree of the sample detection signal (i.e., the number of tests corresponding to each aging degree of the sensor in the accelerated aging test). The mean value representing the degree of aging, r X,Y The correlation coefficient represents the relationship between the sample feature value X and the degree of aging Y. The larger the correlation coefficient, the more correlated the sample feature value is with the degree of aging.

[0136] The correlation coefficient between each sample feature value and the degree of aging at each concentration can be calculated using the above method. This correlation coefficient is denoted as the second correlation coefficient. Then, based on multiple second correlation coefficients at each concentration, at least one set of target sample feature values ​​that satisfy the strong correlation condition is determined and used as the label sample feature value for that concentration. The strong correlation condition could be, for example, finding the largest or highest first correlation coefficient among the second correlation coefficients at each concentration. For instance, first find the second correlation coefficient with r ≥ 0.5, then find the largest r for calibration experiments and obtain its corresponding sample feature value, which is then used as the label sample feature value for that concentration.

[0137] The above method can be used to obtain the characteristic values ​​of the label samples at each concentration, forming an aging label system. The aging label, i.e., the size of the characteristic value of the label sample, can reflect / characterize the aging degree of the sensor at the corresponding concentration.

[0138] Step B5: Determine the aging degree of the target sensor based on the label sample characteristic value at the concentration of each sample object.

[0139] In this step, after obtaining the feature value of the label sample at each concentration, the feature value at the corresponding concentration can be calculated by the real-time detection signal of the target sensor at one or more concentrations, and matched with the feature value of the label sample at the corresponding concentration. The feature value of the label sample that is successfully matched is the aging degree of the target sensor.

[0140] Optionally, determining the aging degree of the target sensor based on the tag sample characteristic values ​​at the concentration of each sample object in this step may include:

[0141] The target sensor is used to detect the concentration of the test object at the concentration of each sample object, and the first detection signal of the target sensor at the concentration of each sample object is obtained.

[0142] Based on the first detection signal at the concentration of each sample object, calculate the first feature value of the target sensor at the feature dimension corresponding to the label sample feature value at the concentration of each sample object;

[0143] Based on each first feature value and the corresponding feature value of the label sample under the feature dimension, determine a set of first feature values ​​and label sample feature values ​​that match the feature values.

[0144] The feature values ​​of the tag samples matched with the feature values ​​are determined as the aging degree corresponding to the target sensor.

[0145] This process involves setting up multiple concentration environments according to the concentration parameters described in the experiment. The target sensor is then placed in each concentration environment to acquire signals, obtaining the first detection signal acquired by the target sensor at each concentration. The feature value of the target sensor at each concentration is then calculated for the corresponding feature dimension, and denoted as the first feature value. Next, the feature values ​​of the label samples at the same concentration are compared with the first feature value to determine if they match. If a set of label sample feature values ​​matches the first feature value, then the match between the label sample feature values ​​and the first feature value is successful. The successfully matched label sample feature value can be used as the aging degree of the target sensor, and its magnitude reflects / characterizes the aging degree of the target sensor.

[0146] In this embodiment, by collecting the detection signals of each sample from the sensor at each aging level, a set of multi-dimensional sample feature values ​​corresponding to each sample detection signal is determined. Then, by analyzing the correlation between each set of multi-dimensional sample feature values ​​and the aging level, the feature values ​​strongly correlated with the aging level at each concentration are determined. This allows for the rapid and accurate identification of feature values ​​strongly correlated with the aging level, thereby quickly calculating the aging level of the target sensor. Furthermore, by using the real-time detection signals of the target sensor at one or more concentrations to calculate the feature values ​​at the corresponding concentrations, and matching them with the feature values ​​of the label samples at the corresponding concentrations, the successfully matched label sample feature values ​​represent the aging level of the target sensor. This further improves the efficiency and accuracy of calculating the aging level of the target sensor.

[0147] The above embodiments mentioned that the concentration estimation model can be trained. The following embodiments will explain the specific training process.

[0148] In one embodiment, the above method further includes the following steps:

[0149] Acquire sample detection signals from sensors with different aging levels under different operating conditions;

[0150] Based on the detection signals of each group of samples, determine the sample feature value of each group of sample detection signals under each first feature dimension;

[0151] The sample feature values ​​of each group of sample detection signals under each first feature dimension are input into the initial concentration estimation model for concentration calibration to determine the predicted concentration corresponding to each group of sample detection signals.

[0152] The initial concentration estimation model is trained based on the loss between the predicted concentration of each sample detection signal and the concentration of the corresponding sample object, as well as the regularization loss corresponding to the initial concentration estimation model, to obtain the concentration estimation model.

[0153] The process of acquiring the detection signals of each sample and calculating the sample feature values ​​of each group of sample detection signals under each first feature dimension can be found in the explanations of steps A1 and A2 above, and will not be repeated here.

[0154] After obtaining the sample feature values ​​of each group of sample detection signals under each first feature dimension, i.e., after obtaining a set of sample feature values ​​corresponding to each group of sample detection signals, these values ​​can be input into the initial concentration estimation model for concentration calibration processing to obtain the calibration concentration predicted by the model for each group of sample detection signals, denoted as the predicted concentration. Then, the loss (e.g., mean square error) between the predicted concentration of each group of sample detection signals and the concentration under the corresponding operating conditions is calculated, as well as the regularization loss corresponding to the initial concentration estimation model is calculated. These two losses are then summed to obtain the total loss L( The total loss can be expressed as follows:

[0155] ;

[0156] in, Let y be the loss function between the predicted concentration of the detection signal for each group of samples and the concentration under the corresponding operating conditions. i Let be the concentration of the detection signal of the i-th sample under the corresponding operating conditions. This is the regularization term for the loss function of the model complexity.

[0157] In the implementation, model training uses Python and the XGBoost open-source library. Input features are standardized and then divided into training, validation, and test sets. The validation set is used for selecting model hyperparameters and evaluating performance, while the test set is used to verify the model's generalization ability on different sensors. The example uses the following parameter configurations:

[0158] Learning rate: 0.1;

[0159] Maximum depth: 6;

[0160] Subsampling ratio: 0.8;

[0161] Number of iteration rounds (n_estimators): 500;

[0162] Random seed (random state): 42.

[0163] After model training, the trained concentration estimation model is used to predict the input features, obtaining the estimated concentration value of the sensor under aging conditions. R is calculated by comparing this estimated value with the actual concentrations on the validation and test sets. 2 The coefficient of determination and RMSE (root mean square error) are used to evaluate the accuracy and stability of the model.

[0164] In this embodiment, the initial concentration estimation model is trained by calculating the loss between the predicted concentration of the detection signal of each group of samples and the concentration under the corresponding working condition, as well as the regularization loss corresponding to the initial concentration estimation model, to obtain the concentration estimation model. This can improve the accuracy of the trained model, thereby improving the accuracy of the concentration predicted by the trained model.

[0165] In summary, this invention employs label selection and machine learning model training based on aging experimental test data. First, a multi-dimensional database is constructed through offline experiments, collecting operational data from a number of sensors at different aging levels under varying temperatures, humidity, and test gas concentrations. Key features are extracted through feature engineering, and based on these features, aging characteristics and concentration-related features are extracted and selected through correlation analysis to establish an aging label system and train a concentration estimation model. Subsequently, using the concentration features as input, a signal attenuation calibration model is constructed using machine learning algorithms to assess the sensor's aging level and estimate the true concentration. Finally, based on the trained model, during online operation, inputting real-time sensor data outputs the current aging level and the corrected concentration value, thus achieving adaptive adjustment as aging dynamically changes. This effectively improves detection accuracy and sensor lifespan, overcomes the signal attenuation caused by aging, and ensures the sensor's detection accuracy and stability throughout its entire lifespan. Furthermore, the method of this invention maximizes the use of experimental and real-time information, achieving data fusion and feature selection, overcoming the limitations of current methods that rely on single variables or simplified formulas, and enhancing the model's generalization ability. Finally, the method and process of this invention can be extended to different types of electrochemical sensors and various gas detection scenarios, exhibiting good scalability.

[0166] The sensor signal attenuation calibration device provided by the present invention is described below. The sensor signal attenuation calibration device described below can be referred to in correspondence with the sensor signal attenuation calibration method described above.

[0167] Figure 3 is a schematic diagram of the sensor signal attenuation calibration device provided by the present invention. As shown in Figure 3, the device may include:

[0168] The signal acquisition module 310 is used to acquire the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object;

[0169] The feature value determination module 320 is used to determine the target feature value corresponding to each first feature dimension in the feature dimension set based on the real-time detection signal; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be measured and is determined based on the sample detection signal of the sample sensor with different aging degree under different operating conditions.

[0170] The signal attenuation calibration module 330 is used to input the target feature value under each first feature dimension into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be measured. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity and the concentration of the sample object.

[0171] In one embodiment, before the feature value determination module 320 determines the target feature value corresponding to the real-time detection signal under each first feature dimension in the feature dimension set based on the real-time detection signal, the apparatus further includes:

[0172] The sample signal acquisition module is used to acquire sample detection signals from sample sensors with different aging levels under different operating conditions.

[0173] The sample feature determination module is used to determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions based on the detection signals of each group of samples; the multiple different feature dimensions include the first feature dimension.

[0174] The feature dimension filtering module is used to determine the first feature dimension of the sample sensor under each aging level based on the correlation between each sample feature value of the detection signal of each group of samples corresponding to the aging level sample sensor and the concentration of different sample objects among multiple different feature dimensions; the first feature dimension of the sample sensor under each aging level is strongly correlated with the concentration of the sample object under the corresponding aging level.

[0175] The feature dimension set determination module is used to determine the feature dimension set based on the first feature dimension of the sample sensors under various aging degrees.

[0176] Optionally, the aforementioned feature dimension filtering module is specifically used to calculate the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects, to obtain multiple first correlation coefficients corresponding to the aging degree sample sensor; based on the multiple first correlation coefficients corresponding to the aging degree sample sensor, to determine at least one set of sample feature values ​​and the corresponding concentration of sample objects whose first correlation coefficients satisfy the concentration strong correlation condition; and to determine the feature dimension corresponding to the sample feature values ​​that satisfy the concentration strong correlation condition as the first feature dimension under the aging degree sample sensor.

[0177] In one embodiment, the above-mentioned apparatus further includes:

[0178] The sample signal acquisition module is used to acquire sample detection signals from sample sensors with different aging levels under different operating conditions.

[0179] The sample feature determination module is used to determine the sample feature values ​​of each group of sample detection signals under multiple different feature dimensions based on the detection signals of each group of samples.

[0180] The sample feature value screening module is used to determine a set of sample feature values ​​that meet the strong correlation condition at the concentration of each sample object, based on the correlation between the sample feature values ​​at the concentration of each sample object and different aging degrees.

[0181] The concentration-related feature value determination module is used to take a set of sample feature values ​​that meet the strong feature correlation condition at the concentration of each sample object as the label sample feature value at the concentration of the corresponding sample object; the above label sample feature value is used to characterize the aging degree of the sample sensor at the concentration of the corresponding sample object.

[0182] The aging degree determination module is used to determine the aging degree of the target sensor based on the label sample feature values ​​at the concentration of each sample object.

[0183] Optionally, the aforementioned aging degree determination module is specifically used to perform concentration detection of the test object using the target sensor at the concentration of each sample object, and obtain a first detection signal of the target sensor at the concentration of each sample object; calculate a first feature value of the target sensor in the feature dimension corresponding to the label sample feature value at the concentration of each sample object based on the first detection signal of each sample object; determine a set of first feature values ​​and label sample feature values ​​that match the feature values ​​based on each first feature value and the label sample feature value in the corresponding feature dimension; and determine the aging degree of the target sensor as the label sample feature value that matches the feature values.

[0184] In one embodiment, the aforementioned multiple different feature dimensions include a signal baseline feature dimension, a peak region feature dimension, and a peak baseline feature dimension. The aforementioned sample feature determination module is specifically used to, for each group of sample detection signals, calculate the sample feature value of the sample detection signal under the signal baseline feature dimension based on the signal value of the sample detection signal during a preset time period at the start or end of the signal; determine the sample feature value of the sample detection signal under the peak region feature dimension based on the signal value of the sample detection signal in the peak region and the corresponding time; and calculate the feature value of the sample detection signal under the peak baseline feature dimension based on the signal value of the sample detection signal in the peak region and the signal value of the sample detection signal during the preset time period at the start or end of the signal.

[0185] In one embodiment, the above-mentioned apparatus further includes:

[0186] The training module is used to acquire sample detection signals from sensor samples with different aging levels under different operating conditions; based on the sample detection signals of each group, the sample feature value of each group of sample detection signals under each first feature dimension is determined; the sample feature value of each group of sample detection signals under each first feature dimension is input into the initial concentration estimation model for concentration calibration processing to determine the predicted concentration corresponding to each group of sample detection signals; based on the loss between the predicted concentration of each group of sample detection signals and the concentration of the corresponding sample object, and the regularization loss corresponding to the initial concentration estimation model, the initial concentration estimation model is trained to obtain the concentration estimation model.

[0187] It should be noted that the apparatus provided in this embodiment of the invention can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0188] Figure 4 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 4, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. The processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logic instructions in the memory 430 to execute a sensor signal attenuation calibration method. The method includes: acquiring a real-time detection signal obtained by a target sensor after real-time concentration detection of the target object; determining a target feature value corresponding to each first feature dimension in a feature dimension set based on the real-time detection signal; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the target object and is determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions; inputting the target feature value under each first feature dimension into a concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the target object; wherein the concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimensions and the concentration of the sample object corresponding to each sample feature value, and each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions, wherein the different operating conditions include at least one parameter being different among ambient temperature, ambient humidity, and the concentration of the sample object.

[0189] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0190] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the sensor signal attenuation calibration method provided by the above methods. The method includes: acquiring a real-time detection signal obtained by a target sensor after real-time concentration detection of the object to be measured; determining a target feature value corresponding to each first feature dimension in a feature dimension set based on the real-time detection signal; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be measured and is determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions; inputting the target feature value under each first feature dimension into a concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be measured; wherein the concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimensions and the concentration of the sample object corresponding to each sample feature value, and each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions, wherein the different operating conditions include at least one parameter being different among ambient temperature, ambient humidity, and the concentration of the sample object.

[0191] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the sensor signal attenuation calibration method provided by the above methods. The method includes: acquiring a real-time detection signal obtained by a target sensor after real-time concentration detection of a test object; determining a target feature value corresponding to each first feature dimension in a feature dimension set based on the real-time detection signal; each first feature dimension in the feature dimension set is strongly correlated with the concentration of the test object and is determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions; inputting the target feature value under each first feature dimension into a concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the test object; wherein the concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimensions and the concentration of the sample object corresponding to each sample feature value, and each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions, wherein the different operating conditions include at least one parameter being different among ambient temperature, ambient humidity, and the concentration of the sample object.

[0192] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0193] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0194] 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 of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for calibrating sensor signal attenuation, characterized in that, include: Acquire the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object; determine the target feature value corresponding to the real-time detection signal under each first feature dimension in the feature dimension set based on the real-time detection signal; Each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be tested and is determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The target feature value under each first feature dimension is input into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be tested. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and the concentration of each sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity, and the concentration of the sample object.

2. The sensor signal attenuation calibration method according to claim 1, characterized in that, Before determining the target feature value corresponding to each first feature dimension in the feature dimension set based on the real-time detection signal, the method further includes: acquiring sample detection signals of sample sensors with different aging degrees under different operating conditions; determining the sample feature value of each group of sample detection signals under multiple different feature dimensions based on each group of sample detection signals; the multiple different feature dimensions include the first feature dimension; for each aging degree of sample sensor, determining the first feature dimension under the aging degree of sample sensor among the multiple different feature dimensions based on the correlation between each sample feature value of each group of sample detection signals corresponding to the aging degree of sample sensor and the concentration of different sample objects; the first feature dimension under each aging degree of sample sensor is strongly correlated with the concentration of the sample object under the corresponding aging degree; determining the feature dimension set based on the first feature dimensions under various aging degrees of sample sensors.

3. The sensor signal attenuation calibration method according to claim 2, characterized in that, The step of determining the first feature dimension under the sample sensor at the aging degree based on the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects from multiple different feature dimensions includes: calculating the correlation between each sample feature value of the sample detection signal corresponding to each group of samples of the aging degree sample sensor and the concentration of different sample objects to obtain multiple first correlation coefficients corresponding to the aging degree sample sensor; determining at least one set of sample feature values ​​and the corresponding concentration of sample objects whose first correlation coefficients satisfy the strong concentration correlation condition based on the multiple first correlation coefficients corresponding to the aging degree sample sensor; and determining the feature dimension corresponding to the sample feature values ​​that satisfy the strong concentration correlation condition as the first feature dimension under the sample sensor at the aging degree.

4. The sensor signal attenuation calibration method according to any one of claims 1 to 3, characterized in that, The method further includes: acquiring sample detection signals of sample sensors with different aging degrees under different operating conditions; determining sample feature values ​​of each group of sample detection signals in multiple different feature dimensions based on the correlation between the sample feature values ​​at the concentration of each sample object and different aging degrees; determining a set of sample feature values ​​that satisfy the feature strong correlation condition at the concentration of each sample object; using the set of sample feature values ​​that satisfy the feature strong correlation condition at the concentration of each sample object as the label sample feature value at the concentration of the corresponding sample object; the label sample feature value is used to characterize the aging degree of the sample sensor at the concentration of the corresponding sample object; and determining the aging degree corresponding to the target sensor based on the label sample feature value at the concentration of each sample object.

5. The sensor signal attenuation calibration method according to claim 4, characterized in that, The step of determining the aging degree of the target sensor based on the label sample feature value at the concentration of each sample object includes: using the target sensor to detect the concentration of the test object at the concentration of each sample object to obtain a first detection signal of the target sensor at the concentration of each sample object; calculating a first feature value of the target sensor at the concentration of each sample object based on the first detection signal at the concentration of each sample object, corresponding to the feature dimension of the label sample feature value; determining a set of first feature values ​​and label sample feature values ​​that match the feature values ​​based on each first feature value and the label sample feature value at the corresponding feature dimension; and determining the label sample feature value that matches the feature values ​​as the aging degree of the target sensor.

6. The sensor signal attenuation calibration method according to claim 2 or 3, characterized in that, The multiple different feature dimensions include a signal baseline feature dimension, a peak region feature dimension, and a peak baseline feature dimension. Determining the sample feature value of each group of sample detection signals under multiple different feature dimensions based on each group of sample detection signals includes: for each group of sample detection signals, calculating the sample feature value of the sample detection signal under the signal baseline feature dimension based on the signal value of the sample detection signal during a preset time period at the start or end of the signal; determining the sample feature value of the sample detection signal under the peak region feature dimension based on the signal value of the sample detection signal in the peak region and the corresponding time; and calculating the feature value of the sample detection signal under the peak baseline feature dimension based on the signal value of the sample detection signal in the peak region and the signal value of the sample detection signal during the preset time period at the start or end of the signal.

7. The sensor signal attenuation calibration method according to any one of claims 1 to 3, characterized in that, The method further includes: acquiring sample detection signals from sample sensors with different aging degrees under different operating conditions; determining sample feature values ​​for each group of sample detection signals in each first feature dimension based on each group of sample detection signals; inputting the sample feature values ​​for each group of sample detection signals in each first feature dimension into an initial concentration estimation model for concentration calibration processing to determine the predicted concentration corresponding to each group of sample detection signals; and training the initial concentration estimation model based on the loss between the predicted concentration of each group of sample detection signals and the concentration of the corresponding sample object, as well as the regularization loss corresponding to the initial concentration estimation model, to obtain the concentration estimation model.

8. A sensor signal attenuation calibration device, characterized in that, include: The signal acquisition module is used to acquire the real-time detection signal obtained by the target sensor after real-time concentration detection of the target object; The feature value determination module is used to determine the target feature value corresponding to the real-time detection signal under each first feature dimension in the feature dimension set based on the real-time detection signal. Each first feature dimension in the feature dimension set is strongly correlated with the concentration of the object to be measured and is determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The signal attenuation calibration module is used to input the target feature value under each first feature dimension into the concentration estimation model for concentration calibration processing to determine the calibration concentration corresponding to the object to be measured. The concentration estimation model is pre-trained based on multiple sample feature values ​​under the first feature dimension and the concentration of the sample object corresponding to each sample feature value. Each sample feature value and each concentration of the sample object are determined based on the sample detection signals of sample sensors with different aging degrees under different operating conditions. The different operating conditions include at least one parameter that is different among ambient temperature, ambient humidity, and the concentration of the sample object.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the sensor signal attenuation calibration method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the sensor signal attenuation calibration method as described in any one of claims 1 to 7.

Images