CN115656437B - Soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data - Google Patents
Soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data Download PDFInfo
- Publication number
- CN115656437B CN115656437B CN202211314862.2A CN202211314862A CN115656437B CN 115656437 B CN115656437 B CN 115656437B CN 202211314862 A CN202211314862 A CN 202211314862A CN 115656437 B CN115656437 B CN 115656437B
- Authority
- CN
- China
- Prior art keywords
- data
- signal
- concentration
- image
- gas
- 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
- 238000000691 measurement method Methods 0.000 title claims abstract description 10
- 239000000126 substance Substances 0.000 claims abstract description 20
- 239000002574 poison Substances 0.000 claims abstract description 16
- 238000003556 assay Methods 0.000 claims abstract description 15
- 239000007789 gas Substances 0.000 claims description 52
- 230000004044 response Effects 0.000 claims description 46
- 238000013527 convolutional neural network Methods 0.000 claims description 38
- 238000012360 testing method Methods 0.000 claims description 29
- 238000000034 method Methods 0.000 claims description 20
- 239000002341 toxic gas Substances 0.000 claims description 19
- 238000000354 decomposition reaction Methods 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 13
- 238000005259 measurement Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 239000003440 toxic substance Substances 0.000 description 12
- 239000000463 material Substances 0.000 description 9
- 239000001301 oxygen Substances 0.000 description 7
- 229910052760 oxygen Inorganic materials 0.000 description 7
- -1 oxygen ion Chemical class 0.000 description 6
- 238000001179 sorption measurement Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000004888 barrier function Effects 0.000 description 3
- 231100000481 chemical toxicant Toxicity 0.000 description 3
- 238000013075 data extraction Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 150000002500 ions Chemical class 0.000 description 2
- 239000002086 nanomaterial Substances 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003795 desorption Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Abstract
The invention provides a soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data, and relates to the technical field of gas concentration analysis. According to the invention, by establishing a gradient image feature and signal feature and gas concentration sample library and fitting the relation between the gradient image feature and signal feature and the gas concentration by using SVR (support vector regression) algorithm, the function of judging the gas concentration of the easy-to-poison chemical is realized. The current situation that the concentration of the gas of the easy-to-make chemical is estimated by manual assay is changed, and the working efficiency of security and protection detection personnel is improved.
Description
Technical Field
The invention relates to the technical field of gas concentration analysis, in particular to a soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data.
Background
In the concentration judgment of the easily-toxic gas, the adsorption and desorption phenomena of the gas occur on the surface of the sensitive material on the surface of the sensor, and the conductivity of the nano material also changes along with the exchange of electrons in the process, and the change size of the nano material is related to the factors such as the temperature of the sensor, the type of the gas and the like.
On the surface of the semiconductor material sensor, three adsorption-state ions of O2-, O2-and O-are mainly existed. When the room temperature or the temperature is lower, the adsorbed oxygen mainly takes O2-as a main component, and the oxygen ion adsorption on the surface of the material is in dynamic balance; when the sensitive material is at a higher temperature (i.e. in a heating working state), the adsorbed O2-abstracts electrons to O2-and O-. The positive charge layer is formed on the surface of the material, so that the width of the grain boundary barrier is narrowed, the conductivity of the sensitive material is increased, and the resistance is reduced. When the surface of the sensitive material is contacted with the easy-to-poison gas in the heating working state, the gas reacts with oxygen ions in the adsorption state and releases electrons to further narrow the barrier width of the surface of the material, so that the conductivity is rapidly increased. When the adsorption process reaches an equilibrium state, the surface resistance of the sensitive material also reaches a stable state. The temperature modulation breaks the dynamic balance of oxygen ion species at a constant temperature in a static test, so that the width of a grain boundary barrier is periodically changed, and the form of surface adsorption ions is also periodically changed. In this process, the oxygen ion reaction of the test gas and the oxygen ion reaction of different adsorption states are carried out, when the oxygen ion state activity is matched with the test gas, the peak value change of the response curve is generated, and the higher the gas concentration is, the larger the amplitude of the response curve is.
However, the easy-to-poison chemicals are rare in life, so that the amplitude difference of the current response curves of some sensors under the condition of temperature adjustment is very small, and the concentration of the easy-to-poison chemicals is difficult to be directly and quickly judged by artificial experience and assay technology, so that an effective artificial intelligent method is not available, and an operator is assisted in effectively judging the concentration of the easy-to-poison chemicals.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a soft measurement method for the concentration of the easy-to-poison gas based on signal multi-feature data.
A soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data comprises the following steps:
step 1: collecting original response signal data under dynamic temperature modulation of a gas sensor, carrying out signal decomposition on the original response signal to obtain a modal component signal, and carrying out characteristic data extraction on the modal component signal, wherein the characteristic data comprises a continuous mean Slope of the modal component signal, the modal component signal and a coordinate axis Area characteristic Area;
step 1.1: the original response signal data is subjected to signal decomposition through empirical mode decomposition, and the formula is as follows:
where f (t) is the original response signal, IMF i (t) is the decomposed modal component signal at the i-th time, and r (t) is the residual component signal. Where n represents the number of modal component signals and t represents the response time.
Step 1.2: the continuous mean slope IMF' (t) of the modal component signal is calculated as follows:
where h represents consecutive sampling intervals.
Step 1.3: the area S of the modal component signal and the x-axis is calculated as follows:
Step 2: calculating continuous mean gradient of an original response signal, drawing a gradient curve characteristic image according to the numerical value of the continuous mean gradient, binarizing to obtain a binarized gradient curve image, and extracting the characteristic of the binarized gradient curve image through a convolutional neural network CNN.
Step 2.1: the continuous mean gradient is calculated for the original response signal as follows:
Step 2.2: extracting pixel values of all pixels of an image, performing gray level processing, and adopting a gray level conversion formula as follows:
r∈[0,1]
Where c is a constant, r is an input gray level, and x i is a pixel value of each pixel;
step 2.3: binarizing the obtained gray level image, and outputting a continuous gradient change curve binary image, wherein a binarization transformation formula is as follows:
R＝w1z1+w2z2+...+w9z9
where z is the gray scale of the pixel and w is the filter template coefficient. R is the gray average value.
Where g (x, y) is the output image and T is a non-negative threshold.
Step 2.4: designing a convolutional neural network CNN image feature extractor to extract gradient image features;
The convolutional neural network CNN image feature extractor comprises an input layer, an output layer, three CONV convolutional layers and two Dense full-connection layers; the input layer is input_shape, the size of the input image is 240x1, the dimension of the CONV layer is CONV1:60 x 11, CONV2:60 x 5, CONV3:20 x1, and the dimension of the Dense full connection layer is Dense1:180×1 and Dense2:4 x 1; training a CNN image feature extractor of the convolutional neural network, and outputting the Dense2 serving as a gradient image feature Binary by an output layer;
step 3: sequentially repeating the steps 1-2 for all the collected original response signals with different concentrations;
Step 4: testing the gas concentration of the existing sample of the easy-to-poison chemical to obtain actual gas concentration test data;
Step 5: establishing a sample library database by taking the multi-characteristic data in the steps 1 and 2, namely Slope, area, binary and the gas concentration assay data in the step 4 as samples;
Step 6: fitting the samples in the sample library database to obtain a functional relation between multi-characteristic data and gas concentration test data through a support vector regression algorithm, and storing the functional relation in a functional form;
Step 6.1: carrying out standardized processing on multi-characteristic data and gas concentration assay data in a sample library;
the normalization process is a normalization process, and the processing formula is as follows:
Pi＝2(P-Pmin)/(Pmax-Pmin)-1
Wherein: p i is the processed data, P is the input data, P max is the maximum value in the input data, and P min is the minimum value in the input data;
Step 6.2: the method comprises the steps of scrambling all samples, randomly sequencing, taking part of samples as training data, and taking the rest samples as test data;
Step 6.3: fitting training data through a support vector regression algorithm SVR, testing a fitting result, and reserving the data when the testing error is less than 0.1%, wherein the function relation S is obtained as follows:
Grade＝S(Slope,Area,Binary)
Wherein Grade represents the soft measurement concentration of the easy-to-poison gas, slope represents the continuous mean Slope characteristic, area represents the modal component signal and the coordinate axis Area characteristic, and Binary represents the gradient image characteristic.
Step 7: acquiring response signal data to be detected under dynamic temperature modulation of a gas sensor, and extracting multi-feature data of the response signal according to the steps 1 and 2;
Step 8: and (3) taking the multi-characteristic data obtained in the step (7) as input, calculating the concentration of the gas through the function obtained in the step (4) and outputting the gas to finish the soft measurement of the concentration of the easily-toxic gas.
The beneficial effects of adopting above-mentioned technical scheme to produce lie in:
The invention provides a soft measurement method for concentration of easily-made toxic gas based on signal multi-feature fusion. And extracting signal features of the modal component signal slope, the modal component signal and the coordinate axis area aiming at differences of the modal component signal slope, the decomposition signal and the coordinate axis area. Meanwhile, aiming at the difference of the original response curve forms of the semiconductor sensor, the thought of calculating continuous gradient values and carrying out binary graphics on the continuous gradient values is provided. The binarized gradient curve characteristic image is used as input, and gradient image characteristics are extracted based on a convolutional neural network algorithm CNN. The gradient (Slope) of the modal component signal, the Area characteristic (Area) of the coordinate axis and the gradient image characteristic (Binary) together form multi-characteristic data to be used as input, and concentration prediction modeling is carried out based on a support vector regression algorithm SVR. The function of soft measurement of the concentration of the easily-made chemical is realized.
Compared with the existing method, the soft measurement method for the concentration of the easily-toxic gas based on signal multi-feature fusion has the following advantages: and through a signal decomposition technology and an image processing technology, a multi-feature extraction technology is adopted to extract the multiple features of the decomposed signals, and an easily-made toxic gas concentration judgment model is established. The method realizes the judgment of the concentration of the easy-to-poison gas, changes the current situation that the concentration of the easy-to-poison chemical is evaluated by manual assay, and improves the working efficiency of security and protection detection personnel. Meanwhile, manual intervention is reduced, and unnecessary errors are avoided.
Drawings
FIG. 1 is a flow chart of determining concentration of a toxic chemical according to an embodiment of the present invention.
FIG. 2 is an illustration of empirical mode decomposition of an easily-toxic chemical 1 in an embodiment of the present invention.
FIG. 3 is an exemplary diagram of empirical mode decomposition of newly collected toxic chemicals 1 in an embodiment of the present invention.
FIG. 4 is a graph showing a continuous gradient of an easily-toxic chemical 1 according to an embodiment of the present invention
FIG. 5 is a gray scale view of an easily-made chemical 1 in an embodiment of the present invention;
FIG. 6 is a binary diagram of an easily-toxic chemical 1 in an embodiment of the present invention;
FIG. 7 is a graph of a continuous gradient of response curves for an easily-toxic chemical 1 in an embodiment of the present invention.
FIG. 8 is an illustration of empirical mode decomposition of a toxic chemical 2 susceptible to the embodiment of the present invention.
FIG. 9 is an exemplary diagram of empirical mode decomposition of newly collected easy-to-poison chemicals 2 in an embodiment of the present invention.
FIG. 10 is a graph showing a continuous gradient of an easily-toxic chemical 2 in accordance with an embodiment of the present invention
FIG. 11 is a gray scale view of an easy-to-poison chemical 2 in an embodiment of the present invention;
FIG. 12 is a binary diagram of an easily-toxic chemical 2 in an embodiment of the present invention;
FIG. 13 is a graph showing a continuous gradient of response curve of an easily-toxic chemical 2 in an embodiment of the present invention
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
A soft measurement method for concentration of easily-toxic gas based on signal multi-characteristic data, as shown in figure 1, comprises the following steps:
example 1: take the easy-to-poison chemical 1 as an example;
step 1: collecting original response signal data under dynamic temperature modulation of a gas sensor, carrying out signal decomposition on the original response signal to obtain a modal component signal, and carrying out characteristic data extraction on the modal component signal, wherein the characteristic data comprises a continuous mean Slope of the modal component signal, the modal component signal and a coordinate axis Area characteristic Area;
Step 1.1: the original response signal data is subjected to signal decomposition by empirical mode decomposition, as shown in fig. 2, and the formula is as follows:
where f (t) is the original response signal, IMF i (t) is the decomposed modal component signal at the i-th time, and r (t) is the residual component signal. Where n represents the number of modal component signals and t represents the response time.
Step 1.2: the continuous mean slope IMF' (t) of the modal component signal is calculated as follows:
where h represents consecutive sampling intervals.
The continuous mean slope values in this example are shown in table 1:
table 1 continuous mean slope values
Step 1.3: the area S of the modal component signal and the x-axis is calculated as follows:
The mode component signals and the coordinate axis area values in this embodiment are shown in table 2:
TABLE 2 Modal component Signal and coordinate axis Area value Area
Step 2: calculating continuous mean gradient of an original response signal, drawing a gradient curve characteristic image according to the numerical value of the continuous mean gradient, binarizing to obtain a binarized gradient curve image, and extracting the characteristic of the binarized gradient curve image through a convolutional neural network CNN.
Step 2.1: calculating a continuous mean gradient for the original response signal, drawing a continuous gradient curve as shown in fig. 4, and the formula is as follows:
Step 2.2: the pixel values of each pixel point of the image are extracted and gray scale processing is performed, and as shown in fig. 5, the gray scale conversion formula is as follows:
r∈[0,1]
Where c is a constant, r is an input gray level, and x i is a pixel value of each pixel;
Step 2.3: binarizing the obtained gray level image, and outputting a continuous gradient change curve binary image, wherein the binarization conversion formula is as follows, as shown in fig. 6:
R＝w1z1+w2z2+...+w9z9
where z is the gray scale of the pixel and w is the filter template coefficient. R is the gray average value.
Where g (x, y) is the output image and T is a non-negative threshold.
Step 2.4: designing a convolutional neural network CNN image feature extractor to extract gradient image features;
The convolutional neural network CNN image feature extractor comprises an input layer, an output layer, three CONV convolutional layers and two Dense full-connection layers; the input layer is input_shape, the size of the input image is 240x1, the dimension of the CONV layer is CONV1:60 x 11, CONV2:60 x 5, CONV3:20 x1, and the dimension of the Dense full connection layer is Dense1:180×1 and Dense2:4 x 1; training a CNN image feature extractor of the convolutional neural network, and outputting the Dense2 serving as a gradient image feature Binary by an output layer;
in this embodiment, the characteristics of the gradient image extracted by the convolutional neural network CNN are shown in table 3;
TABLE 3 gradient image feature Binary
Step 3: sequentially repeating the steps 1-2 for all the collected original response signals with different concentrations;
Step 4: testing the gas concentration of the existing sample of the easy-to-poison chemical to obtain actual gas concentration test data;
the results of the gas concentration assay data in this example are shown in table 4;
TABLE 4 easy-to-toxicity chemical gas concentration value Grade
Step 5: establishing a sample library database by taking the multi-characteristic data in the steps 1 and 2, namely Slope, area, binary and the gas concentration assay data in the step 4 as samples;
Step 6: fitting the samples in the sample library database to obtain a functional relation between multi-characteristic data and gas concentration test data through a support vector regression algorithm, and storing the functional relation in a functional form;
Step 6.1: carrying out standardized processing on multi-characteristic data and gas concentration assay data in a sample library;
the normalization process is a normalization process, and the processing formula is as follows:
Pi＝2(P-Pmin)/(Pmax-Pmin)-1
Wherein: p i is the processed data, P is the input data, P max is the maximum value in the input data, and P min is the minimum value in the input data;
Step 6.2: the method comprises the steps of scrambling all samples, randomly sequencing, taking part of samples as training data, and taking the rest samples as test data;
in the embodiment, a sample of the fourth fifth is taken as training data, and the fifth is taken as test data;
Step 6.3: and fitting training data through a support vector regression algorithm SVR, testing the fitting result, and when the testing error is less than 0.1%, considering that the training result is satisfactory, and storing the set of functional relations. The functional relation S is obtained as follows:
Grade＝S(Slope,Area,Binary)
Wherein Grade represents the soft measurement concentration of the easy-to-poison gas, slope represents the continuous mean Slope characteristic, area represents the modal component signal and the coordinate axis Area characteristic, and Binary represents the gradient image characteristic.
Step 7: acquiring response signal data to be detected under dynamic temperature modulation of a gas sensor, and extracting multi-feature data of the response signal according to the steps 1 and 2;
The gradient image features and the signal features in this embodiment are:
TABLE 5 gradient image features and Signal features
Step 8: and (3) taking the multi-characteristic data obtained in the step (7) as input, and calculating the concentration of the gas through the function obtained in the step (4) and outputting the gas, so as to finish the soft measurement of the concentration of the easily-toxic gas.
The final results of the concentrations of the easy-to-poison chemicals and the actual assay results in this example are shown in the following table:
TABLE 6 concentration values of easily-toxic chemicals gases
Example 2: take the easy-to-poison chemical 2 as an example;
step 1: collecting original response signal data under dynamic temperature modulation of a gas sensor, carrying out signal decomposition on the original response signal to obtain a modal component signal, and carrying out characteristic data extraction on the modal component signal, wherein the characteristic data comprises a continuous mean Slope of the modal component signal, the modal component signal and a coordinate axis Area characteristic Area;
step 1.1: the original response signal data is subjected to signal decomposition by empirical mode decomposition, as shown in fig. 8, the formula is as follows:
where f (t) is the original response signal, IMF i (t) is the decomposed modal component signal at the i-th time, and r (t) is the residual component signal. Where n represents the number of modal component signals and t represents the response time.
Step 1.2: the continuous mean slope IMF' (t) of the modal component signal is calculated as follows:
where h represents consecutive sampling intervals.
The continuous mean slope values in this example are shown in table 7:
TABLE 7 continuous mean slope values
Step 1.3: the area S of the modal component signal and the x-axis is calculated as follows:
the mode component signals and the coordinate axis area values in this embodiment are shown in table 8:
table 8 modal component signals and coordinate axis Area values Area
Step 2: calculating continuous mean gradient of an original response signal, drawing a gradient curve characteristic image according to the numerical value of the continuous mean gradient, binarizing to obtain a binarized gradient curve image, and extracting the characteristic of the binarized gradient curve image through a convolutional neural network CNN.
Step 2.1: the continuous mean gradient is calculated for the original response signal, and a continuous gradient curve is drawn as shown in fig. 10, with the following formula:
Step 2.2: the pixel values of each pixel point of the image are extracted and gray scale processing is performed, and as shown in fig. 11, the gray scale conversion formula is as follows:
r∈[0,1]
Where c is a constant, r is an input gray level, and x i is a pixel value of each pixel;
Step 2.3: binarizing the obtained gray-scale image, and outputting a continuous gradient change curve binary image, as shown in fig. 12, the binarization conversion formula is as follows:
R＝w1z1+w2z2+...+w9z9
where z is the gray scale of the pixel and w is the filter template coefficient. R is the gray average value.
Where g (x, y) is the output image and T is a non-negative threshold.
Step 2.4: designing a convolutional neural network CNN image feature extractor to extract gradient image features;
The convolutional neural network CNN image feature extractor comprises an input layer, an output layer, three CONV convolutional layers and two Dense full-connection layers; the input layer is input_shape, the size of the input image is 240x1, the dimension of the CONV layer is CONV1:60 x 11, CONV2:60 x 5, CONV3:20 x1, and the dimension of the Dense full connection layer is Dense1:180×1 and Dense2:4 x 1; training a CNN image feature extractor of the convolutional neural network, and outputting the Dense2 serving as a gradient image feature Binary by an output layer;
The characteristics of the gradient image extracted by the convolutional neural network CNN in the embodiment are shown in table 9;
TABLE 9 gradient image feature Binary
Step 3: sequentially repeating the steps 1-2 for all the collected original response signals with different concentrations;
Step 4: testing the gas concentration of the existing sample of the easy-to-poison chemical to obtain actual gas concentration test data;
the results of the gas concentration assay data in this example are shown in table 10;
TABLE 10 easy-to-poison chemical gas concentration value Grade
Step 5: establishing a sample library database by taking the multi-characteristic data in the steps 1 and 2, namely Slope, area, binary and the gas concentration assay data in the step 4 as samples;
Step 6: fitting the samples in the sample library database to obtain a functional relation between multi-characteristic data and gas concentration test data through a support vector regression algorithm, and storing the functional relation in a functional form;
Step 6.1: carrying out standardized processing on multi-characteristic data and gas concentration assay data in a sample library;
the normalization process is a normalization process, and the processing formula is as follows:
Pi＝2(P-Pmin)/(Pmax-Pmin)-1
Wherein: p i is the processed data, P is the input data, P max is the maximum value in the input data, and P min is the minimum value in the input data;
Step 6.2: the method comprises the steps of scrambling all samples, randomly sequencing, taking part of samples as training data, and taking the rest samples as test data;
in the embodiment, a sample of the fourth fifth is taken as training data, and the fifth is taken as test data;
Step 6.3: and fitting training data through a support vector regression algorithm SVR, testing the fitting result, and when the testing error is less than 0.1%, considering that the training result is satisfactory, and storing the set of functional relations. The functional relation S is obtained as follows:
Grade＝S(Slope,Area,Binary)
wherein Grade represents soft measurement concentration of the easy-to-poison gas, slope represents continuous mean Slope characteristic, modal component signal and coordinate axis area characteristic, and Binary represents gradient image characteristic.
Step 7: acquiring response signal data to be detected under dynamic temperature modulation of a gas sensor, and extracting multi-feature data of the response signal according to the steps 1 and 2;
The gradient image features and the signal features in this embodiment are:
TABLE 11 gradient image features and Signal features
Step 8: and (3) taking the multi-characteristic data obtained in the step (7) as input, and calculating the concentration of the gas through the function obtained in the step (4) and outputting the gas, so as to finish the soft measurement of the concentration of the easily-toxic gas, as shown in fig. 9 and 13.
The final results of the concentrations of the easy-to-poison chemicals and the actual assay results in this example are shown in the following table:
TABLE 12 values for concentration of easily-toxic chemicals gases
The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above technical features, but encompasses other technical features formed by any combination of the above technical features or their equivalents without departing from the spirit of the invention. Such as the above-described features, are mutually substituted with (but not limited to) the features having similar functions disclosed in the embodiments of the present disclosure.
Claims (3)
1. The soft measurement method for the concentration of the easily-toxic gas based on the signal multi-characteristic data is characterized by comprising the following steps of:
Step 1: collecting original response signal data under dynamic temperature modulation of a gas sensor, carrying out signal decomposition on the original response signal to obtain a modal component signal, and extracting characteristic data of the modal component signal; the characteristic data specifically comprises a continuous mean Slope of the modal component signal, the modal component signal and a coordinate axis Area characteristic Area;
step 1.1: the original response signal data is subjected to signal decomposition through empirical mode decomposition, and the formula is as follows:
Wherein f (t) is an original response signal, IMF i (t) is a decomposed modal component signal at the ith moment, r (t) is a residual component signal, n represents the number of modal component signals, and t represents the response moment;
Step 1.2: the continuous mean slope IMF' (t) of the modal component signal is calculated as follows:
wherein h represents a continuous sampling interval;
step 1.3: the area S of the modal component signal and the x-axis is calculated as follows:
Step 2: calculating a continuous mean gradient of an original response signal, drawing a gradient curve characteristic image according to the numerical value of the continuous mean gradient, binarizing to obtain a binarized gradient curve image, and extracting the characteristic of the binarized gradient curve image through a convolutional neural network CNN;
Step 2.1: the continuous mean gradient is calculated for the original response signal as follows:
Step 2.2: extracting pixel values of all pixels of an image, performing gray level processing, and adopting a gray level conversion formula as follows:
r∈[0,1]
Where c is a constant, r is an input gray level, and x i is a pixel value of each pixel;
step 2.3: binarizing the obtained gray level image, and outputting a continuous gradient change curve binary image, wherein a binarization transformation formula is as follows:
R＝w1z1+w2z2+...+w9z9
wherein z is the gray scale of the pixel, and w is the filter template coefficient; r is the average gray level;
wherein g (x, y) is the output image and T is a non-negative threshold;
Step 2.4: designing a convolutional neural network CNN image feature extractor to extract gradient image features;
The convolutional neural network CNN image feature extractor comprises an input layer, an output layer, three CONV convolutional layers and two Dense full-connection layers; the input layer is input_shape, the size of the input image is 240x1, the dimension of the CONV layer is CONV1:60 x 11, CONV2:60 x 5, CONV3:20 x1, and the dimension of the Dense full connection layer is Dense1:180×1 and Dense2:4 x 1; training a CNN image feature extractor of the convolutional neural network, and outputting the Dense2 serving as a gradient image feature Binary by an output layer;
step 3: sequentially repeating the steps 1-2 for all the collected original response signals with different concentrations;
Step 4: testing the gas concentration of the existing sample of the easy-to-poison chemical to obtain actual gas concentration test data;
Step 5: establishing a sample library database by taking the multi-characteristic data in the steps 1 and 2, namely Slope, area, binary and the gas concentration assay data in the step 4 as samples;
Step 6: fitting the samples in the sample library database to obtain a functional relation between multi-characteristic data and gas concentration test data through a support vector regression algorithm, and storing the functional relation in a functional form;
Step 7: acquiring response signal data to be detected under dynamic temperature modulation of a gas sensor, and extracting multi-feature data of the response signal according to the steps 1 and 2;
Step 8: and (3) taking the multi-characteristic data obtained in the step (7) as input, calculating the concentration of the gas through the function obtained in the step (4) and outputting the gas to finish the soft measurement of the concentration of the easily-toxic gas.
2. The method for soft measurement of concentration of toxic gas based on signal multi-feature data according to claim 1, wherein the step 6 specifically comprises the following steps:
Step 6.1: carrying out standardized processing on multi-characteristic data and gas concentration assay data in a sample library;
the normalization process is a normalization process, and the processing formula is as follows:
Pi＝2(P-Pmin)/(Pmax-Pmin)-1
Wherein: p i is the processed data, P is the input data, P max is the maximum value in the input data, and P min is the minimum value in the input data;
Step 6.2: the method comprises the steps of scrambling all samples, randomly sequencing, taking part of samples as training data, and taking the rest samples as test data;
Step 6.3: and fitting training data through a support vector regression algorithm SVR, testing a fitting result, and retaining the data when the testing error is less than 0.1%, so as to obtain a functional relation S.
3. The method for soft measurement of concentration of toxic gas based on signal multi-feature data of claim 2, wherein the functional relation S of step 6.3 is as follows:
Grade＝S(Slope,Area,Binary)
Wherein Grade represents the soft measurement concentration of the easy-to-poison gas, slope represents the continuous mean Slope characteristic, area represents the modal component signal and the coordinate axis Area characteristic, and Binary represents the gradient image characteristic.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211314862.2A CN115656437B (en) | 2022-10-26 | Soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211314862.2A CN115656437B (en) | 2022-10-26 | Soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115656437A CN115656437A (en) | 2023-01-31 |
| CN115656437B true CN115656437B (en) | 2024-06-11 |
Family
ID=
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108226148A (en) * | 2018-01-05 | 2018-06-29 | 湖南理工学院 | A kind of poisonous mushroom identification classification and detection device and method |
| CN109583566A (en) * | 2018-11-20 | 2019-04-05 | 东北农业大学 | Ammonia concentration prediction technique in a kind of cold ground pig house based on empirical mode decomposition and Elman neural network |
| CN109724941A (en) * | 2019-02-27 | 2019-05-07 | 大唐长山热电厂 | A kind of CO based on radial base neural net2High-temperature gas concentration detection method |
| CN111610232A (en) * | 2019-02-25 | 2020-09-01 | 英飞凌科技股份有限公司 | Gas sensing device and method for operating a gas sensing device |
| CN112100924A (en) * | 2020-09-17 | 2020-12-18 | 云南电力技术有限责任公司 | Gas concentration prediction method and device based on extreme learning machine model |
| CN112307950A (en) * | 2020-10-29 | 2021-02-02 | 重庆大学 | Method for extracting and intelligently analyzing detail characteristic quantity of GIS vibration defect identification |
| CN113988146A (en) * | 2020-07-10 | 2022-01-28 | 英飞凌科技股份有限公司 | Gas sensing device for sensing one or more gases in a gas mixture |
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108226148A (en) * | 2018-01-05 | 2018-06-29 | 湖南理工学院 | A kind of poisonous mushroom identification classification and detection device and method |
| CN109583566A (en) * | 2018-11-20 | 2019-04-05 | 东北农业大学 | Ammonia concentration prediction technique in a kind of cold ground pig house based on empirical mode decomposition and Elman neural network |
| CN111610232A (en) * | 2019-02-25 | 2020-09-01 | 英飞凌科技股份有限公司 | Gas sensing device and method for operating a gas sensing device |
| CN109724941A (en) * | 2019-02-27 | 2019-05-07 | 大唐长山热电厂 | A kind of CO based on radial base neural net2High-temperature gas concentration detection method |
| CN113988146A (en) * | 2020-07-10 | 2022-01-28 | 英飞凌科技股份有限公司 | Gas sensing device for sensing one or more gases in a gas mixture |
| CN112100924A (en) * | 2020-09-17 | 2020-12-18 | 云南电力技术有限责任公司 | Gas concentration prediction method and device based on extreme learning machine model |
| CN112307950A (en) * | 2020-10-29 | 2021-02-02 | 重庆大学 | Method for extracting and intelligently analyzing detail characteristic quantity of GIS vibration defect identification |
Non-Patent Citations (2)
| Title |
|---|
| EEMD结合改进PCNN模型的气体泄漏信号降噪;孙烨辰;《计算机仿真》;20200915(第09期);全文 * |
| 氧化铟传感器对VOCs的动态测试方法研究;孟凡利;《电子测量技术》;20200908(第17期);全文 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109444232B (en) | Multichannel intelligent polluted gas monitoring device and diffusion tracing method | |
| CN109614851B (en) | Image detection method for morphological parameters of air pollution particles | |
| CN115876258B (en) | Livestock and poultry breeding environment abnormity monitoring and alarming system based on multi-source data | |
| CN111652853A (en) | Magnetic powder flaw detection method based on deep convolutional neural network | |
| CN114677362B (en) | Surface defect detection method based on improved YOLOv5 | |
| CN114544894A (en) | PMF model-based source analysis uncertainty analysis method and system | |
| CN115656437B (en) | Soft measurement method for concentration of easily-poisoned gas based on signal multi-feature data | |
| CN111881974B (en) | Water inrush source identification method based on Piper-PCA-FCL discrimination model | |
| CN113340902A (en) | Method and device for detecting surface defects of catalyst carrier, and electronic device | |
| JP3002325B2 (en) | Surface inspection equipment | |
| CN114974452B (en) | Method and device for determining control target of secondary conversion source | |
| CN116933084A (en) | Pollutant emission prediction method and device | |
| CN114487284B (en) | Method and system for measuring concentration of heavy metal in air | |
| CN115656437A (en) | Soft measurement method for concentration of gas easy to prepare poison based on signal multi-feature data | |
| CN109241590B (en) | Method for judging heavy metal pollution sediment pollution type in environment-friendly dredging engineering | |
| CN115965796A (en) | Metal corrosion detection method and system based on image recognition | |
| CN115170822A (en) | Combined method for analyzing qualitative source of heavy metal in farmland soil | |
| CN113077002B (en) | Machine olfaction visual sensing data analysis method based on space heterodyne Raman spectrum | |
| CN112259172B (en) | Reverse identification method for pollution source with air pollutant chemical reaction condition | |
| CN112986497A (en) | Pollution gas tracing method based on gas sensor array fingerprint identification | |
| CN111007220B (en) | Method for agile monitoring of water quality of live pig breeding sewage | |
| CN114487014A (en) | Defect geometric feature extraction method for infrared nondestructive testing | |
| López et al. | Change-point method applied to the detection of temporal variations in seafloor bacterial mat coverage | |
| CN117761263B (en) | Formaldehyde gas detection system and method | |
| CN115639354B (en) | Marine plastic identification method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |