CN113138210B - Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor - Google Patents
Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor Download PDFInfo
- Publication number
- CN113138210B CN113138210B CN202110688589.9A CN202110688589A CN113138210B CN 113138210 B CN113138210 B CN 113138210B CN 202110688589 A CN202110688589 A CN 202110688589A CN 113138210 B CN113138210 B CN 113138210B
- Authority
- CN
- China
- Prior art keywords
- regression
- concentration
- point
- data
- resistance value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/02—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance
- G01N27/04—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance by investigating resistance
- G01N27/12—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance by investigating resistance of a solid body in dependence upon absorption of a fluid; of a solid body in dependence upon reaction with a fluid, for detecting components in the fluid
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
Abstract
The invention discloses a self-adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor, and belongs to the technical field of gas sensor measurement. The method of the invention obtains enough data quantity through Gaussian process regression interpolation, then searches for a neighbor point for each data point in a multidimensional space, obtains an accurate model of the data point through local Gaussian process regression of the data point and the neighbor point, and places each data point at the center position of the neighbor point in the multidimensional space, thereby eliminating the concept of each variable edge point, avoiding the influence of the edge point on the Gaussian process regression, and realizing intelligent real-time accurate compensation of the temperature and the humidity of the gas sensor. The method improves the accuracy of the final model and the accuracy of each point prediction; by three normal distribution detection methods, automatic adaptation of the optimal neighbor point number is realized, the finally obtained model is an optimal model, and the accuracy of the model is further improved.
Description
Technical Field
The invention belongs to the technical field of gas sensor measurement, and particularly relates to a self-adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor.
Background
The gas sensor is easily influenced by temperature and humidity when measuring gas, and how to better deal with the influence caused by the temperature and humidity is a problem which is difficult to solve. The prior art researches how to compensate the influence of temperature or humidity on the gas sensor respectively, but the prior art does not compensate the influence of temperature and humidity simultaneously. In the existing method, a classical numerical algorithm is a type of algorithm which is applied more, and is mostly used for processing the influence caused by one physical quantity, when two physical quantities of temperature and humidity are processed, a plurality of items are required to be added, and the overfitting condition is easily caused when the number of items and parameters are set.
The neural network is often introduced into the compensation of temperature or humidity, a model needs to be trained in advance, so that a large amount of sample data is needed to train the model in advance, the calculated amount is large, a certain time delay is caused, the real-time compensation cannot be realized, and the requirement on hardware is higher. Moreover, due to the imperfect theoretical knowledge system, the overfitting problem is easy to occur.
The Gaussian process regression algorithm is a parameter-free estimation method, has a strict statistical learning theoretical basis, and can avoid the possibility of improper parameter estimation in a classical numerical algorithm and a complex calculation process in a neural network. The gaussian process regression algorithm is commonly used for processing high-dimensionality complex problems, but due to the limitations of the gaussian process regression algorithm, the estimation error of the data of the edge points of each variable is often large, particularly in the problem of researching the cross influence of multiple variables, the edge points of each variable occupy a large proportion in the whole data, and therefore the finally obtained model has large deviation from the rules actually contained in the data. This results in the algorithm not being able to accurately estimate the edge point data and the resulting final model is not the best model, thus also resulting in a reduced accuracy of prediction for other points.
Only when the number of samples is enough to reflect the rule of the object and the input and output variable data quantity is enough, a more accurate regression model can be obtained. However, in the actual testing process, the acquired sample data is discrete data, and the manual testing has limited data acquisition capacity, which often results in incomplete data size.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a self-adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor.
The technical problem proposed by the invention is solved as follows:
a self-adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor comprises the following steps:
s1, obtaining resistance values of the gas sensor under different temperatures, humidities and concentrations through testing, and judging physical characteristics of the gas sensor;
s2, screening and eliminating test abnormal points through the physical characteristics of the gas sensor;
s3, checking whether the resistance residual is normally distributed or not through a multiple linear regression method and a QQ-plot diagram, if the residual data are normally distributed, turning to S4, and if not, returning to S1 for testing again;
s4, establishing a first Gaussian process regression initial model with temperature, humidity and concentration as input and resistance as output, performing Gaussian process regression on the data with the test abnormal points removed, predicting the resistance of the test abnormal points by using regression results, and correcting the test abnormal points;
s5, randomly selecting 15 data points from the corrected data points as a check group, and taking the rest data points as a regression group;
s6, performing Gaussian process regression on the regression group data points by using the first Gaussian process regression initial model;
s7, predicting the resistance value of the regression group data point by using the regression result obtained in S6, and calculating the relative error between the predicted resistance value and the test resistance value of the regression group data point, wherein the error is an error abnormal point when the relative error is more than 10%; correcting the error difference constant points through iterative Gaussian process regression, and performing Gaussian process regression on the regression group data points after the error difference constant points are corrected by using a first Gaussian process regression initial model;
s8, predicting the resistance value of the data point of the test group through the regression result obtained in S7, and if the relative error between the predicted resistance value and the measured resistance value of the data point of the test group is within 10%, judging that the first Gaussian process regression initial model is suitable for the currently tested gas sensor;
s9, normalizing the temperature, humidity, concentration and resistance value data after error abnormal points are corrected, and enabling the data range to be reduced to a [0,1] interval;
s10, performing Gaussian process regression on the normalized temperature, humidity, concentration and resistance value data by using a first Gaussian process regression initial model; inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
establishing a second Gaussian process regression initial model with temperature, humidity and resistance as input and concentration as output; performing Gaussian process regression on the interpolated temperature, humidity, concentration and resistance value data by using a second Gaussian process regression initial model, and predicting a concentration value by using a regression result;
calculating a relative error between the predicted concentration value after normalization reduction and a concentration value set by testing, judging whether the percentage of data points with the relative error larger than 10% in the total data points is smaller than a set threshold value, if so, finishing interpolation, otherwise, carrying out next round of interpolation, continuously inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
s11-1, for each data point, searching 45 nearest neighbor points on a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a 46-point local data set;
s11-2, performing Gaussian process regression on the local data set by using a second Gaussian process regression initial model, and predicting the concentration prediction value of the current data point by using the regression result; traversing all the data points to obtain the concentration predicted values of all the data points;
s12, carrying out normalization reduction processing on the interpolated temperature, humidity, concentration and resistance value data;
s13, calculating the absolute error between the predicted concentration value and the concentration value set by the test, solving the confidence interval of the absolute error, and searching the best neighbor point by sequentially reducing and sequentially increasing the neighbor points participating in regression;
s13-1, calculating the absolute error between the concentration predicted value and the concentration set by the testE；
S13-2, respectively by
、
And
three test modes determine the confidence interval of the absolute error, wherein,
as an absolute errorEIs determined by the average value of (a) of (b),
as an absolute errorEStandard deviation of (d); calculating the percentage of points within the confidence interval;
s13-3, sequentially reducing and sequentially increasing the number of neighbor points, and forming a new data point for each data pointLocal data set, go to S11-2; comparing the percentage of the points in the confidence interval corresponding to all the local data sets with the standard normal distribution percentage corresponding to the three test modes, and taking the neighbor point of the local data set corresponding to the percentage of the point in the confidence interval closest to the percentage of the point in the confidence interval as the optimal neighbor point M; when the comparison is carried out, the priority of the three inspection modes is ranked as
、
And
；
s14, for each data point, searching M neighbor points with the nearest Euclidean distance in a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a local data set of M +1 points; and performing Gaussian process regression on the local data set of the M +1 point by using a second Gaussian process regression initial model, predicting the concentration predicted value of the current data point by using the regression result, and traversing all the data points to obtain the concentration predicted values of all the data points.
Further, in S10, the covariance function in the second gaussian process regression initial model is a square exponential covariance function, a Matern covariance function, a rational quadratic covariance function, or a linear covariance function.
Further, the specific process of S1 is as follows:
s1-1, acquiring resistance values of the gas sensor under different temperatures, humidities and concentrations;
s1-2, respectively drawing two-dimensional curves of resistance values and concentrations at different temperatures and humidity;
and S1-3, judging the physical characteristic of the gas sensor according to the two-dimensional curve of the resistance value-concentration, if the resistance value increases along with the increase of the concentration, judging the physical characteristic of the gas sensor to be a resistance-increasing sensor, and if not, judging the physical characteristic of the gas sensor to be a resistance-decreasing sensor.
Further, the specific process of S2 is as follows:
s2-1, if the physical characteristic of the gas sensor is a resistance rise type sensor, checking whether the resistance value increases with the concentration under various temperatures and humidities, and enabling a point which does not accord with the rule to be a test abnormal point;
if the physical characteristic of the gas sensor is a resistance drop type sensor, checking whether the resistance value decreases with the concentration under various temperatures and humidities, and taking a point which does not accord with the rule as a test abnormal point;
and S2-2, removing the screened test abnormal points.
Further, the specific process of S3 is as follows:
s3-1, establishing a multiple linear regression initial model according to the change relation of the resistance value along with the temperature, the humidity and the concentration;
s3-2, performing multiple linear regression on the resistance value to obtain residual error data;
and S3-3, checking whether the residual data are normally distributed by using a QQ-plot diagram of the residual data, if so, meeting the requirement of Gaussian process regression, turning to S4, and otherwise, returning to S1 for testing again.
Further, the specific process of S4 is as follows:
s4-1, setting the temperature, the humidity and the concentration as input variables of a first Gaussian process regression initial model, and setting the resistance value as an output variable;
s4-2, selecting a covariance function in the first Gaussian process regression initial model as a square exponential covariance function, a Matern covariance function, a rational secondary covariance function or a linear covariance function;
s4-3, performing Gaussian process regression on the data with the test outliers removed by using the first Gaussian process regression initial model;
and S4-4, predicting the resistance value of the rejected test abnormal point by using the regression result obtained in the S4-3, and replacing the test resistance value by using the predicted resistance value.
Further, the specific process of S7 is as follows:
s7-1, using the regression result obtained in S6 to predict the resistance value of the regression group data point;
s7-2, calculating relative errors of the predicted resistance value and the test resistance value by using the regression group data points obtained in the step S7-1:
wherein the content of the first and second substances,
in order to be a relative error,
in order to test the resistance value of the resistor,
predicting resistance values for the regression group data points;
if the relative error is larger than 10%, judging that the error difference constant points are wrong, removing the error difference constant points with the largest relative error, and performing Gaussian process regression on the regression group data points after the error difference constant points with the largest relative error are removed by using a first Gaussian process regression initial model;
s7-3, predicting the resistance value of the error abnormal point removed in the S7-2 by using the regression result obtained in the S7-2, and replacing the test resistance value by using the predicted resistance value; performing Gaussian process regression on the regression group data points after the error difference constant point with the maximum correction error by using a first Gaussian process regression initial model;
s7-4, predicting the predicted resistance value of the regression group data point after correcting the maximum error point by using the regression result obtained in S7-3, and repeating S7-2 to S7-3 until no error abnormal point with the error larger than 10% exists.
The invention has the beneficial effects that:
(1) the method of the invention obtains the accurate model of the point by searching the neighboring point for each point in the multidimensional space and performing local Gaussian process regression on the point and the neighboring point, places each point at the central position of the neighboring point in the multidimensional space, eliminates the concept of each variable edge point, thereby avoiding the influence of the edge point on the Gaussian process regression. The method of the invention improves the accuracy of the final model and the accuracy of each point prediction.
(2) The method realizes the automatic adaptation of the optimal neighbor point number through three normal distribution detection methods, so that the finally obtained model is the optimal model, and the accuracy of the model is further improved.
(3) The method solves the problem of incomplete data volume through interpolation.
(4) According to the method, temperature and humidity compensation is carried out on the gas sensor through Gaussian process regression, a training process is not needed, calculation is simpler, and the requirement of real-time correction can be met; when the gas sensor is in an extreme environment, the acquired data is affected, the data rule is poor, and the method can still make better correction.
Drawings
FIG. 1 is a graph of concentration-resistance at 10 ℃ and 30% humidity in the examples;
FIG. 2 is a QQ-plot of residual data in an example embodiment;
FIG. 3 is a diagram illustrating the distribution of relative error points of the test set in the example;
FIG. 4 is a graph of the normalized predicted concentration value after interpolation and the normalized test set concentration value scatter versus;
FIG. 5 is a statistical chart and a normal curve chart of the absolute error frequency distribution of the predicted concentration value and the concentration value set by the test.
Detailed Description
The invention is further described below with reference to the figures and examples.
The embodiment provides a self-adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor, which comprises the following steps of:
s1, obtaining resistance values of the ammonia gas sensor under different temperatures, humidities and concentrations through testing, and judging physical characteristics of the ammonia gas sensor;
s1-1, respectively changing the ammonia concentration in a closed space where the ammonia sensor is located at different temperatures and humidities, and monitoring the resistance value of the ammonia sensor at different concentrations, wherein the temperature range is 10-30 ℃, the humidity range is 20-70%, and the concentration range is 10-50 ppm;
s1-2, respectively drawing two-dimensional curves of resistance values and concentrations at different temperatures and humidity;
and S1-3, judging the physical characteristic of the ammonia gas sensor according to the two-dimensional curve of the resistance value-concentration, if the resistance value increases along with the increase of the concentration, judging the physical characteristic of the ammonia gas sensor to be a resistance-increasing sensor, and if not, judging the physical characteristic of the ammonia gas sensor to be a resistance-decreasing sensor. The ammonia gas sensor used in this example was an up-type sensor.
S2, screening and eliminating abnormal points through the physical characteristics of the ammonia gas sensor;
s2-1, checking whether the resistance value increases with the concentration under each temperature and humidity, and taking the point which does not accord with the rule as a test abnormal point;
s2-2, removing the 13 screened test abnormal points.
S3, testing whether the resistance residual error is normally distributed through a multiple linear regression method and a QQ-plot diagram;
s3-1, a concentration-resistance curve chart at 10 ℃ and 30% humidity is shown in figure 1, and the resistance value and the concentration tend to have a linear relation; the change rule of the resistance value along with the temperature shows that the resistance value is influenced by the temperature and tends to have a cubic relation with the temperature; the change rule of the resistance value along with the humidity shows that the resistance value and the humidity tend to have a quadratic relation. Establishing a multiple linear regression initial model as follows:
wherein the content of the first and second substances,
is the model parameters of the initial model of the multiple linear regression,
as a value of the resistance, the resistance value,
it is the temperature that is set for the purpose,
in order to be the humidity level,
is a resistance;
s3-2, performing multiple linear regression on the resistance value by using the multiple linear regression initial model to obtain residual data;
s3-3, making a QQ-plot diagram of residual data, as shown in FIG. 2; and (4) checking whether the residual data are normally distributed or not by using a QQ-plot diagram of the residual data, if so, meeting the requirement of Gaussian process regression, and turning to the step S4, otherwise, turning to the step S1 for testing again. As can be seen from fig. 2, the residual data are normally distributed and satisfy the precondition of gaussian process regression.
S4, establishing a first Gaussian process regression initial model with temperature, humidity and concentration as input and resistance as output, performing Gaussian process regression on the data with the test abnormal points removed, predicting the resistance of the test abnormal points by using regression results, and correcting the test abnormal points;
s4-1, setting the temperature, the humidity and the concentration as input variables of a first Gaussian process regression initial model, and setting the resistance value as an output variable;
s4-2, selecting a covariance function in the first Gaussian process regression initial model as a square exponential covariance function;
in temperature and humidity compensation of the gas sensor, input variables comprise temperature, humidity and concentration, and different function relations are respectively formed between the input variables and the resistance value of a dependent variable, so that when high-precision requirements exist, composite operations such as addition or multiplication and the like can be carried out on different basic covariance functions to form a composite covariance function, and therefore a kernel function in a model is more fit with the data rule of the sensor;
s4-3, performing Gaussian process regression on the data with the test outliers removed by using the first Gaussian process regression initial model;
and S4-4, predicting the resistance values of the rejected 13 test abnormal points by using the regression result obtained in the S4-3, and replacing the test resistance values by using the predicted resistance values, thereby completing the correction of the test abnormal points.
And S5, randomly selecting 15 data points from the corrected data points as a check group, and using the rest data points as regression groups.
The regression set of data points were considered as discrete point data points obtained during the test, and the test set of data points were considered as unknown data points. With gaussian process regression, a finite set of discrete regression data points needs to be extended to a continuous space to obtain data points for unknown points. I.e., when the model calculated from the regression set is applied to the test set as an unknown data point, the model is considered successful.
S6, performing Gaussian process regression on the regression group data points by using the first Gaussian process regression initial model;
s7, predicting the resistance value of the regression group data point by using the regression result obtained in S6, and calculating a relative error by combining the test resistance value, wherein the error is an error abnormal point when the relative error is more than 10%; correcting the error difference constant points through iterative Gaussian process regression, and performing Gaussian process regression on the regression group data points after the error difference constant points are corrected by using a first Gaussian process regression initial model;
s7-1, using the regression result obtained in S6 to predict the resistance value of the regression group data point;
s7-2, calculating relative errors of the predicted resistance value and the test resistance value by using the regression group data points obtained in the step S7-1:
wherein the content of the first and second substances,
in order to be a relative error,
in order to test the resistance value of the resistor,
predicting resistance values for the regression group data points;
if the relative error is larger than 10%, judging that the error difference constant points are wrong, removing the error difference constant points with the largest relative error, and performing Gaussian process regression on the regression group data points after the error difference constant points with the largest relative error are removed by using a first Gaussian process regression initial model;
s7-3, predicting the resistance value of the error abnormal point removed in the S7-2 by using the regression result obtained in the S7-2, and replacing the test resistance value by using the predicted resistance value; performing Gaussian process regression on the regression group data points after the error difference constant point with the maximum correction error by using a first Gaussian process regression initial model;
s7-4, predicting the predicted resistance value of the regression group data point after correcting the maximum error point by using the regression result obtained in the S7-3, and repeating the step S7-2 to the step S7-3 until no error abnormal point with the error larger than 10% exists;
s8, verifying whether the verification group data points meet the model through the regression result obtained in S7, and if so, successfully establishing a first Gaussian process regression initial model;
and predicting the resistance value of the data point of the test group through the regression result obtained in the step S7, and if the relative errors between the predicted resistance value and the measured resistance value of the data point of the test group are within 10%, determining that the first Gaussian process regression initial model is suitable for the currently tested ammonia gas sensor, wherein the distribution of the relative error points of the test group is shown in FIG. 3. As can be seen from FIG. 3, the relative errors between the predicted resistance values and the measured resistance values of the data points in the test set are within 10%, so that the first Gaussian process regression initial model is suitable for the currently tested ammonia gas sensor.
S9, normalizing the temperature, humidity, concentration and resistance value data after error abnormal points are corrected, and enabling the data range to be reduced to a [0,1] interval;
s10, performing Gaussian process regression on the normalized temperature, humidity, concentration and resistance value data by using a first Gaussian process regression initial model; inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
establishing a second Gaussian process regression initial model with temperature, humidity and resistance value as input and concentration as output, and selecting a covariance function in the second Gaussian process regression initial model as a square exponential covariance function, a Matern covariance function, a rational quadratic covariance function or a linear covariance function; performing Gaussian process regression on the interpolated temperature, humidity, concentration and resistance value data by using a second Gaussian process regression initial model, and predicting a concentration value by using a regression result;
calculating a relative error between the predicted concentration value after normalization reduction and a concentration value set by a test, judging whether the percentage of data points with the relative error larger than 10% to the total data points is smaller than 1%, if so, finishing interpolation, otherwise, carrying out next round of interpolation, continuously inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
in this example, three rounds of interpolation are performed on the data. Firstly, inserting concentration values of 15ppm, 25ppm, 35ppm and 45ppm on the basis of original data points of 10ppm, 20ppm, 30ppm, 40ppm and 50ppm, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result; then, concentration values of 12.5ppm, 17.5ppm, 22.5ppm, 27.5ppm, 32.5ppm, 37.5ppm, 42.5ppm and 47.5ppm are inserted in sequence, and a predicted resistance value corresponding to the intermediate concentration value after the normalization processing is predicted by using a regression result; the values of concentration 11.25ppm, 13.75ppm, 16.25ppm, 18.75ppm, 21.25ppm, 23.75ppm, 26.25ppm, 28.75ppm, 31.25ppm, 33.75ppm, 36.25ppm, 38.75ppm, 41.25ppm, 43.75ppm, 46.25ppm and 48.75ppm were again inserted in this order, and the regression results were used to predict the predicted resistance values corresponding to the intermediate concentration values after the normalization processing. At this time, the percentage of the data points with the relative error between the predicted concentration value and the concentration value set by the test larger than 10% to the total data points is smaller than 1%, and the interpolation is completed. Fig. 4 shows a graph comparing the normalized predicted concentration value after interpolation and the normalized concentration value set by the test.
S11-1, for each data point, searching 45 nearest neighbor points on a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a 46-point local data set;
s11-2, performing Gaussian process regression on the local data set by using a second Gaussian process regression initial model, and predicting the concentration prediction value of the current data point by using the regression result; traversing all the data points to obtain the concentration predicted values of all the data points;
after the data is normalized, the distance between a calculation point and a point is ensured to ensure that the weight occupied by each variable is the same. When the concentration of a certain point is predicted, local Gaussian process regression is carried out by searching neighbor points of the point on a four-dimensional space, so that the estimation abnormality of the Gaussian process regression on edge points of each variable can be avoided, and accurate prediction of each point can be realized by training a model by using neighbor points with similar rules.
S12, carrying out normalization reduction processing on the interpolated temperature, humidity, concentration and resistance value data;
s13, calculating the absolute error between the predicted concentration value and the concentration value set by the test, solving the confidence interval of the absolute error, and searching the best neighbor point by sequentially reducing and sequentially increasing the neighbor points participating in regression;
s13-1, calculating the absolute error between the concentration predicted value and the concentration set by the testE；
S13-2, respectively by
、
And
three test modes determine the confidence interval of the absolute error, wherein,
as an absolute errorEIs determined by the average value of (a) of (b),
as an absolute errorEStandard deviation of (d); calculating the percentage of points within the confidence interval;
s13-3, sequentially reducing and sequentially increasing the number of neighbor points, and turning to S11-2 for a new local data set formed by each data point; comparing the percentage of the points in the confidence interval corresponding to all the local data sets with the standard normal distribution percentage corresponding to the three test modes, and taking the neighbor point of the local data set corresponding to the percentage of the point in the confidence interval closest to the percentage of the point in the confidence interval as the optimal neighbor point M; when the comparison is carried out, the priority of the three inspection modes is ranked as
、
And
；
when the number of neighbor points participating in the regression is 45,
confidence interval of [ -1.56,1.57]The percentage of the points in the confidence interval is (594-10)/594 ≈ 98.3%;
the confidence interval is [ -1.04,1.05]The percentage of the points in the confidence interval is (594-24)/594 ≈ 96.0%;
confidence interval of [ -0.51,0.53]The percentage of the points in the confidence interval is (594-59)/594 ≈ 90.0%;
respectively increasing or decreasing the number of neighbor points, and correspondingly setting confidence of local data setThe point number in the interval is 30 neighbor point numbers which are closest to the standard normal distribution percentage corresponding to the three test modes in percentage
Confidence interval is [ -1.47,1.47]The percentage of the points in the confidence interval is (594-10)/594 ≈ 98.3%;
confidence interval of [ -0.98,0.98 [)]The percentage of the points in the confidence interval is (594-22)/594 ≈ 96.3%;
confidence interval of [ -0.49,0.49 [)]The percentage of points within the confidence interval is (594-57)/594 ≈ 90.4%. The normal distribution of the absolute error can be observed intuitively through the frequency distribution statistical chart of the absolute error and the normal curve, and the frequency distribution statistical chart of the absolute error and the normal curve of the predicted concentration value and the concentration value set by the test are shown in fig. 5.
The method for automatically searching the optimal neighbor point number, combining the neighbor point number and the point to form a local data set and performing local Gaussian process regression for predicting the point concentration is called as an adaptive local Gaussian regression algorithm, so that the problem of large deviation of Gaussian process regression on edge point prediction is solved, and the prediction accuracy of all the points is improved.
S14, for each data point, searching 30 neighbor points with the nearest Euclidean distance in a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a local data set with 31 points; and performing Gaussian process regression on the 31-point local data set by using a second Gaussian process regression initial model, predicting the concentration predicted value of the current data point by using a regression result, and traversing all the data points to obtain the concentration predicted values of all the data points.
The above is an embodiment of the present invention. The specific parameters in the above embodiments and examples are only for the purpose of clearly illustrating the invention verification process of the inventor and are not intended to limit the scope of the invention, which is defined by the claims, and all equivalent structural changes made by using the contents of the specification and the drawings of the present invention should be covered by the scope of the present invention.
Claims (6)
1. A self-adaptive local Gaussian temperature and humidity compensation method of an intelligent gas sensor is characterized by comprising the following steps:
s1, obtaining resistance values of the gas sensor under different temperatures, humidities and concentrations through testing, and judging physical characteristics of the gas sensor;
the specific process of S1 is as follows:
s1-1, acquiring resistance values of the gas sensor under different temperatures, humidities and concentrations;
s1-2, respectively drawing two-dimensional curves of resistance values and concentrations at different temperatures and humidity;
s1-3, judging the physical characteristic of the gas sensor according to the two-dimensional curve of the resistance value-concentration, if the resistance value increases along with the increase of the concentration, judging the physical characteristic of the gas sensor to be a resistance-increasing sensor, otherwise, judging the physical characteristic of the gas sensor to be a resistance-decreasing sensor;
s2, screening and eliminating test abnormal points through the physical characteristics of the gas sensor;
s3, checking whether the resistance residual is normally distributed or not through a multiple linear regression method and a QQ-plot diagram, if the residual data are normally distributed, turning to S4, and if not, returning to S1 for testing again;
s4, establishing a first Gaussian process regression initial model with temperature, humidity and concentration as input and resistance as output, performing Gaussian process regression on the data with the test abnormal points removed, predicting the resistance of the test abnormal points by using regression results, and correcting the test abnormal points;
s5, randomly selecting 15 data points from the corrected data points as a check group, and taking the rest data points as a regression group;
s6, performing Gaussian process regression on the regression group data points by using the first Gaussian process regression initial model;
s7, predicting the resistance value of the regression group data point by using the regression result obtained in S6, and calculating the relative error between the predicted resistance value and the test resistance value of the regression group data point, wherein the error is an error abnormal point when the relative error is more than 10%; correcting the error difference constant points through iterative Gaussian process regression, and performing Gaussian process regression on the regression group data points after the error difference constant points are corrected by using a first Gaussian process regression initial model;
s8, predicting the resistance value of the data point of the test group through the regression result obtained in S7, and if the relative error between the predicted resistance value and the measured resistance value of the data point of the test group is within 10%, judging that the first Gaussian process regression initial model is suitable for the currently tested gas sensor;
s9, normalizing the temperature, humidity, concentration and resistance value data after error abnormal points are corrected, and enabling the data range to be reduced to a [0,1] interval;
s10, performing Gaussian process regression on the normalized temperature, humidity, concentration and resistance value data by using a first Gaussian process regression initial model; inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
establishing a second Gaussian process regression initial model with temperature, humidity and resistance as input and concentration as output; performing Gaussian process regression on the interpolated temperature, humidity, concentration and resistance value data by using a second Gaussian process regression initial model, and predicting a concentration value by using a regression result;
calculating a relative error between the predicted concentration value after normalization reduction and a concentration value set by testing, judging whether the percentage of data points with the relative error larger than 10% in the total data points is smaller than a set threshold value, if so, finishing interpolation, otherwise, carrying out next round of interpolation, continuously inserting an intermediate concentration value between every two concentration values, and predicting a predicted resistance value corresponding to the intermediate concentration value after normalization processing by using a regression result;
s11-1, for each data point, searching 45 nearest neighbor points on a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a 46-point local data set;
s11-2, performing Gaussian process regression on the local data set by using a second Gaussian process regression initial model, and predicting the concentration prediction value of the current data point by using the regression result; traversing all the data points to obtain the concentration predicted values of all the data points;
s12, carrying out normalization reduction processing on the interpolated temperature, humidity, concentration and resistance value data;
s13, calculating the absolute error between the predicted concentration value and the concentration value set by the test, solving the confidence interval of the absolute error, and searching the best neighbor point by sequentially reducing and sequentially increasing the neighbor points participating in regression;
s13-1, calculating the absolute error between the concentration predicted value and the concentration set by the testE；
S13-2, respectively by
、
And
three test modes determine the confidence interval of the absolute error, wherein,
as an absolute errorEIs determined by the average value of (a) of (b),
as an absolute errorEStandard deviation of (d); calculating the percentage of points within the confidence interval;
s13-3, sequentially reducing and sequentially increasing the number of neighbor points, forming a new local data set for each data point, and turning to S11-2; comparing the percentage of the points in the confidence interval corresponding to all the local data sets with the standard normal distribution percentage corresponding to the three test modes, and taking the point closest to the point in the confidence intervalThe neighbor point number of the local data set corresponding to the percentage of the point number of (a) is taken as the optimal neighbor point number M; when the comparison is carried out, the priority of the three inspection modes is ranked as
、
And
；
s14, for each data point, searching M neighbor points with the nearest Euclidean distance in a four-dimensional space consisting of temperature, humidity, concentration and resistance value to form a local data set of M +1 points; and performing Gaussian process regression on the local data set of the M +1 point by using a second Gaussian process regression initial model, predicting the concentration predicted value of the current data point by using the regression result, and traversing all the data points to obtain the concentration predicted values of all the data points.
2. The adaptive local Gaussian temperature and humidity compensation method for an intelligent gas sensor according to claim 1, wherein in S10, the covariance function in the regression initial model of the second Gaussian process is a square exponential covariance function, a Matern covariance function, a rational quadratic covariance function or a linear covariance function.
3. The adaptive local Gaussian temperature and humidity compensation method for the intelligent gas sensor according to claim 2, wherein the specific process of S2 is as follows:
s2-1, if the physical characteristic of the gas sensor is a resistance rise type sensor, checking whether the resistance value increases with the concentration under various temperatures and humidities, and enabling a point which does not accord with the rule to be a test abnormal point;
if the physical characteristic of the gas sensor is a resistance drop type sensor, checking whether the resistance value decreases with the concentration under various temperatures and humidities, and taking a point which does not accord with the rule as a test abnormal point;
and S2-2, removing the screened test abnormal points.
4. The adaptive local Gaussian temperature and humidity compensation method for the intelligent gas sensor according to claim 3, wherein the specific process of S3 is as follows:
s3-1, establishing a multiple linear regression initial model according to the change relation of the resistance value along with the temperature, the humidity and the concentration;
s3-2, performing multiple linear regression on the resistance value to obtain residual error data;
and S3-3, checking whether the residual data are normally distributed by using a QQ-plot diagram of the residual data, if so, meeting the requirement of Gaussian process regression, turning to S4, and otherwise, returning to S1 for testing again.
5. The adaptive local Gaussian temperature and humidity compensation method for the intelligent gas sensor according to claim 4, wherein the specific process of S4 is as follows:
s4-1, setting the temperature, the humidity and the concentration as input variables of a first Gaussian process regression initial model, and setting the resistance value as an output variable;
s4-2, selecting a covariance function in the first Gaussian process regression initial model as a square exponential covariance function, a Matern covariance function, a rational secondary covariance function or a linear covariance function;
s4-3, performing Gaussian process regression on the data with the test outliers removed by using the first Gaussian process regression initial model;
and S4-4, predicting the resistance value of the rejected test abnormal point by using the regression result obtained in the S4-3, and replacing the test resistance value by using the predicted resistance value.
6. The adaptive local Gaussian temperature and humidity compensation method for the intelligent gas sensor according to claim 5, wherein the specific process of S7 is as follows:
s7-1, using the regression result obtained in S6 to predict the resistance value of the regression group data point;
s7-2, calculating relative errors of the predicted resistance value and the test resistance value by using the regression group data points obtained in the step S7-1:
wherein the content of the first and second substances,
in order to be a relative error,
in order to test the resistance value of the resistor,
predicting resistance values for the regression group data points;
if the relative error is larger than 10%, judging that the error difference constant points are wrong, removing the error difference constant points with the largest relative error, and performing Gaussian process regression on the regression group data points after the error difference constant points with the largest relative error are removed by using a first Gaussian process regression initial model;
s7-3, predicting the resistance value of the error abnormal point removed in the S7-2 by using the regression result obtained in the S7-2, and replacing the test resistance value by using the predicted resistance value; performing Gaussian process regression on the regression group data points after the error difference constant point with the maximum correction error by using a first Gaussian process regression initial model;
s7-4, predicting the predicted resistance value of the regression group data point after correcting the maximum error point by using the regression result obtained in S7-3, and repeating S7-2 to S7-3 until no error abnormal point with the error larger than 10% exists.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110688589.9A CN113138210B (en) | 2021-06-22 | 2021-06-22 | Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110688589.9A CN113138210B (en) | 2021-06-22 | 2021-06-22 | Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113138210A CN113138210A (en) | 2021-07-20 |
| CN113138210B true CN113138210B (en) | 2021-09-24 |
Family
ID=76815849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110688589.9A Active CN113138210B (en) | 2021-06-22 | 2021-06-22 | Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113138210B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114021454B (en) * | 2021-11-03 | 2022-10-21 | 国网山东省电力公司营销服务中心(计量中心) | Electric energy meter comprehensive verification test error decoupling method and system |
| CN114543711B (en) * | 2022-01-23 | 2023-08-25 | 中南大学 | Sensing data processing method of gas distance measurement system |
Citations (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1459876A (en) * | 2002-12-19 | 2003-12-03 | 李扬 | Orgaic polymer-inorganic nano composite resistance type thin film humidity sensitive element and its manufacturing method |
| CN1779460A (en) * | 2004-11-22 | 2006-05-31 | 山东理工大学 | Intelligent antistaling experimental box for fruits and vegetables |
| CN1800842A (en) * | 2006-01-13 | 2006-07-12 | 南京埃森环境技术有限公司 | Online resistance-capacitance type method and apparatus for analyzing water content in high temperature smoke |
| CN101135670A (en) * | 2007-10-12 | 2008-03-05 | 电子科技大学 | Two-channel sonic surface wave gas sensors having humidity compensate function |
| CN102192927A (en) * | 2010-11-05 | 2011-09-21 | 重庆大学 | Air-quality monitoring system based on electronic nose technique, and monitoring method thereof |
| WO2012178071A2 (en) * | 2011-06-23 | 2012-12-27 | Brown University | Device and methods for temperature and humidity measurements using a nanocomposite film sensor |
| CN102944583A (en) * | 2012-11-30 | 2013-02-27 | 重庆大学 | Metal-oxide gas sensor array concentration detecting method based on drift compensation |
| CN103398744A (en) * | 2013-08-09 | 2013-11-20 | 山西埃尔气体系统工程有限公司 | Monitor for integrated parameters of medical molecular sieve oxygen generation system |
| CN204330663U (en) * | 2014-12-24 | 2015-05-13 | 广州兰泰仪器有限公司 | A kind of fast tester for water content |
| CN104903732A (en) * | 2013-01-11 | 2015-09-09 | 诺基亚技术有限公司 | Sensor array with fixed output |
| CN105181754A (en) * | 2015-06-29 | 2015-12-23 | 电子科技大学 | Compensation type resistor type integrated gas sensor array and preparation method thereof |
| CN105301064A (en) * | 2015-12-10 | 2016-02-03 | 郑州大学 | In2O3-based hotline type semiconductor gas sensor with environment temperature and humidity self-compensation capabilities |
| CN105388187A (en) * | 2015-11-13 | 2016-03-09 | 大连民族大学 | Measuring method of humidity-controllable semiconductor gas sensitive element |
| DE102016124831A1 (en) * | 2016-12-19 | 2018-06-21 | Martin Donath | Method for permanent monitoring of a drying process of moistened components and apparatus for carrying out the method |
| US10119927B2 (en) * | 2010-12-23 | 2018-11-06 | Crc Care Pty Ltd | Analyte ion detection method and device |
| CN109470743A (en) * | 2018-11-08 | 2019-03-15 | 西南大学 | A kind of gas sensor characteristic test system of multiple physical field coupling environment |
| CN109781800A (en) * | 2019-01-25 | 2019-05-21 | 青岛大学 | A kind of gas sensor and preparation method thereof based on metal molybdate nanocomposite |
| CN109799772A (en) * | 2019-01-07 | 2019-05-24 | 重庆水利电力职业技术学院 | A kind of red kiwi fruit canker pre-alarming control system and method |
| US10610111B1 (en) * | 2006-06-30 | 2020-04-07 | Bao Tran | Smart watch |
| CN111693579A (en) * | 2020-07-16 | 2020-09-22 | 长沙理工大学 | Hydrogen sulfide gas detection method and sensor based on nanosheet composite membrane |
| CN111830110A (en) * | 2020-08-17 | 2020-10-27 | 江苏利得智能监测科技有限公司 | Oxygen sensor self-calibration method, device and system |
| CN112162019A (en) * | 2020-09-22 | 2021-01-01 | 上海摇橹仪器设备有限公司 | Temperature and humidity compensation blood glucose detection chip, detector, blood glucose detection method based on chip and detector, and electronic equipment |
| CN112585453A (en) * | 2018-09-05 | 2021-03-30 | 大阪瓦斯株式会社 | Gas detection device |
| CN112858394A (en) * | 2020-12-31 | 2021-05-28 | 广州奥松电子有限公司 | Detection system, detection device, measurement method thereof and temperature and humidity compensation method |
| CN112903758A (en) * | 2021-01-21 | 2021-06-04 | 电子科技大学 | Intelligent temperature and humidity self-calibration method for overlapped blocks of gas sensor |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7243945B2 (en) * | 1992-05-05 | 2007-07-17 | Automotive Technologies International, Inc. | Weight measuring systems and methods for vehicles |
| US6843100B2 (en) * | 2002-01-14 | 2005-01-18 | Kendro Laboratory Products, L.P. | Thermal conductivity measurement of carbon dioxide gas with relative humidity and temperature compensation |
| US7842243B2 (en) * | 2006-02-21 | 2010-11-30 | Sergey Sergeyevich Sarkisov | Chemical sensor with an indicator dye |
| US8884382B2 (en) * | 2007-07-17 | 2014-11-11 | Kwj Engineering, Inc. | Multi-Dimensional sensors and sensing systems |
| RU2649366C1 (en) * | 2008-04-30 | 2018-04-02 | Ксилеко, Инк. | Biomass treatment |
| CN102663219B (en) * | 2011-12-21 | 2015-03-04 | 北京理工大学 | Fuel cell output prediction method and system based on mixing model |
| US9696272B2 (en) * | 2015-08-07 | 2017-07-04 | Silicon Laboratories Inc. | Systems and methods for humidity measurement using dielectric material-based relative humidity sensors |
| US10238131B2 (en) * | 2016-08-08 | 2019-03-26 | Premier Innovations, LLC | System, device, and method for preservative concentration control in baking production |
| US10775395B2 (en) * | 2018-10-18 | 2020-09-15 | Arctoris Limited | System and method of performing a biological experiment with adaptive cybernetic control of procedural conditions |
| CN213148673U (en) * | 2020-07-22 | 2021-05-07 | 南京逸然电子科技有限公司 | Atmospheric particulate concentration measuring device based on laser scattering method |
-
2021
- 2021-06-22 CN CN202110688589.9A patent/CN113138210B/en active Active
Patent Citations (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1459876A (en) * | 2002-12-19 | 2003-12-03 | 李扬 | Orgaic polymer-inorganic nano composite resistance type thin film humidity sensitive element and its manufacturing method |
| CN1779460A (en) * | 2004-11-22 | 2006-05-31 | 山东理工大学 | Intelligent antistaling experimental box for fruits and vegetables |
| CN1800842A (en) * | 2006-01-13 | 2006-07-12 | 南京埃森环境技术有限公司 | Online resistance-capacitance type method and apparatus for analyzing water content in high temperature smoke |
| US10610111B1 (en) * | 2006-06-30 | 2020-04-07 | Bao Tran | Smart watch |
| CN101135670A (en) * | 2007-10-12 | 2008-03-05 | 电子科技大学 | Two-channel sonic surface wave gas sensors having humidity compensate function |
| CN102192927A (en) * | 2010-11-05 | 2011-09-21 | 重庆大学 | Air-quality monitoring system based on electronic nose technique, and monitoring method thereof |
| US10119927B2 (en) * | 2010-12-23 | 2018-11-06 | Crc Care Pty Ltd | Analyte ion detection method and device |
| WO2012178071A2 (en) * | 2011-06-23 | 2012-12-27 | Brown University | Device and methods for temperature and humidity measurements using a nanocomposite film sensor |
| CN102944583A (en) * | 2012-11-30 | 2013-02-27 | 重庆大学 | Metal-oxide gas sensor array concentration detecting method based on drift compensation |
| CN104903732A (en) * | 2013-01-11 | 2015-09-09 | 诺基亚技术有限公司 | Sensor array with fixed output |
| CN103398744A (en) * | 2013-08-09 | 2013-11-20 | 山西埃尔气体系统工程有限公司 | Monitor for integrated parameters of medical molecular sieve oxygen generation system |
| CN204330663U (en) * | 2014-12-24 | 2015-05-13 | 广州兰泰仪器有限公司 | A kind of fast tester for water content |
| CN105181754A (en) * | 2015-06-29 | 2015-12-23 | 电子科技大学 | Compensation type resistor type integrated gas sensor array and preparation method thereof |
| CN105388187A (en) * | 2015-11-13 | 2016-03-09 | 大连民族大学 | Measuring method of humidity-controllable semiconductor gas sensitive element |
| CN107677709A (en) * | 2015-11-13 | 2018-02-09 | 大连民族大学 | With SnO2The method of testing of the semiconductor gas sensor of the controlled humidity of gas sensitive preparation process |
| CN105301064A (en) * | 2015-12-10 | 2016-02-03 | 郑州大学 | In2O3-based hotline type semiconductor gas sensor with environment temperature and humidity self-compensation capabilities |
| DE102016124831A1 (en) * | 2016-12-19 | 2018-06-21 | Martin Donath | Method for permanent monitoring of a drying process of moistened components and apparatus for carrying out the method |
| CN112585453A (en) * | 2018-09-05 | 2021-03-30 | 大阪瓦斯株式会社 | Gas detection device |
| CN109470743A (en) * | 2018-11-08 | 2019-03-15 | 西南大学 | A kind of gas sensor characteristic test system of multiple physical field coupling environment |
| CN109799772A (en) * | 2019-01-07 | 2019-05-24 | 重庆水利电力职业技术学院 | A kind of red kiwi fruit canker pre-alarming control system and method |
| CN109781800A (en) * | 2019-01-25 | 2019-05-21 | 青岛大学 | A kind of gas sensor and preparation method thereof based on metal molybdate nanocomposite |
| CN111693579A (en) * | 2020-07-16 | 2020-09-22 | 长沙理工大学 | Hydrogen sulfide gas detection method and sensor based on nanosheet composite membrane |
| CN111830110A (en) * | 2020-08-17 | 2020-10-27 | 江苏利得智能监测科技有限公司 | Oxygen sensor self-calibration method, device and system |
| CN112162019A (en) * | 2020-09-22 | 2021-01-01 | 上海摇橹仪器设备有限公司 | Temperature and humidity compensation blood glucose detection chip, detector, blood glucose detection method based on chip and detector, and electronic equipment |
| CN112858394A (en) * | 2020-12-31 | 2021-05-28 | 广州奥松电子有限公司 | Detection system, detection device, measurement method thereof and temperature and humidity compensation method |
| CN112903758A (en) * | 2021-01-21 | 2021-06-04 | 电子科技大学 | Intelligent temperature and humidity self-calibration method for overlapped blocks of gas sensor |
Non-Patent Citations (3)
| Title |
|---|
| Electron compensation in p-type 3DOM NiO by Sn doping for enhanced formaldehyde sensing performance;Z Wang 等;《Journal of Materials Chemistry C》;20170301(第5期);第3254-3263页 * |
| Evolution of breath analysis based on humidity and gas sensors: Potential and challenges;Tai HL 等;《SENSORS AND ACTUATORS B-CHEMICAL》;20200901;第318卷;第128104页 * |
| 基于RGO-PEO复合薄膜的QCM湿度传感器研究;王斯 等;《电子元件与材料》;20170511;第36卷(第5期);第67-71页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113138210A (en) | 2021-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113138210B (en) | Self-adaptive local Gaussian temperature and humidity compensation method for intelligent gas sensor | |
| CN109799269B (en) | Electronic nose gas sensor array optimization method based on dynamic feature importance | |
| CN107803350B (en) | A kind of method of lithium battery automatic sorting, storage medium and battery sorting device | |
| CN110687072B (en) | Calibration set and verification set selection and modeling method based on spectral similarity | |
| CN109816031B (en) | Transformer state evaluation clustering analysis method based on data imbalance measurement | |
| CN112903758B (en) | Intelligent temperature and humidity self-calibration method for overlapped blocks of gas sensor | |
| CN116075733A (en) | Battery management system for classifying battery modules | |
| CN114114039A (en) | Method and device for evaluating consistency of single battery cells of battery system | |
| CN112817842B (en) | Imperfect debug SRGM decision method, system, medium, equipment and application | |
| CN114859249B (en) | Method and device for detecting battery pack capacity | |
| CN114636827B (en) | Method, device, equipment and medium for eliminating abnormal points of reaction curve | |
| TWI774291B (en) | Secondary battery inspection method and secondary battery inspection device | |
| CN116256402A (en) | Valve side sleeve defect detection method based on frequency domain dielectric spectrum data | |
| CN113987240B (en) | Customs inspection sample tracing method and system based on knowledge graph | |
| CN115855133A (en) | Calibration method and device of sensor, computer equipment and readable storage medium | |
| Buschmann et al. | Data-driven decision support for process quality improvements | |
| CN111062118B (en) | Multilayer soft measurement modeling system and method based on neural network prediction layering | |
| CN112397404B (en) | Semiconductor process detection method and system | |
| CN107506824B (en) | Method and device for detecting bad observation data of power distribution network | |
| CN113688506A (en) | Potential atmospheric pollution source identification method based on multidimensional data such as micro-station | |
| CN116990450B (en) | Defect detection method and system for cornea shaping mirror | |
| CN116992308B (en) | Data and knowledge fusion process fluctuation analysis and optimization method and device | |
| CN114660095B (en) | Sintered corundum porosity measuring method and system based on optics | |
| Nishimura et al. | Anomaly detection and analysis by a gradient boosting trees and neural network ensemble model | |
| CN116467557A (en) | Accurate comparison method for load data of user power failure event |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |