CN115860056B - Sensor array neural network method for mixed gas concentration prediction - Google Patents
Sensor array neural network method for mixed gas concentration prediction Download PDFInfo
- Publication number
- CN115860056B CN115860056B CN202310129437.4A CN202310129437A CN115860056B CN 115860056 B CN115860056 B CN 115860056B CN 202310129437 A CN202310129437 A CN 202310129437A CN 115860056 B CN115860056 B CN 115860056B
- Authority
- CN
- China
- Prior art keywords
- mixed gas
- training sample
- concentration
- interpolation
- neural network
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 28
- 239000007789 gas Substances 0.000 claims abstract description 172
- 238000012549 training Methods 0.000 claims abstract description 119
- 239000013598 vector Substances 0.000 claims abstract description 88
- 230000004044 response Effects 0.000 claims abstract description 48
- 238000012880 independent component analysis Methods 0.000 claims abstract description 15
- 210000002569 neuron Anatomy 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 4
- 230000001131 transforming effect Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 10
- 230000004913 activation Effects 0.000 description 7
- 238000001514 detection method Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 5
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 4
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000003344 environmental pollutant Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 231100000719 pollutant Toxicity 0.000 description 3
- 238000009827 uniform distribution Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 239000011664 nicotinic acid Substances 0.000 description 2
- 229910052757 nitrogen Inorganic materials 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- MGWGWNFMUOTEHG-UHFFFAOYSA-N 4-(3,5-dimethylphenyl)-1,3-thiazol-2-amine Chemical compound CC1=CC(C)=CC(C=2N=C(N)SC=2)=C1 MGWGWNFMUOTEHG-UHFFFAOYSA-N 0.000 description 1
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 229910021529 ammonia Inorganic materials 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 229910002092 carbon dioxide Inorganic materials 0.000 description 1
- 239000001569 carbon dioxide Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 239000002440 industrial waste Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- JCXJVPUVTGWSNB-UHFFFAOYSA-N nitrogen dioxide Inorganic materials O=[N]=O JCXJVPUVTGWSNB-UHFFFAOYSA-N 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 231100000331 toxic Toxicity 0.000 description 1
- 230000002588 toxic effect Effects 0.000 description 1
- 230000002087 whitening effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A50/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE in human health protection, e.g. against extreme weather
- Y02A50/20—Air quality improvement or preservation, e.g. vehicle emission control or emission reduction by using catalytic converters
Landscapes
- Investigating Or Analyzing Materials By The Use Of Electric Means (AREA)
Abstract
The invention provides a sensor array neural network method for mixed gas concentration prediction, which belongs to the technical field of mixed gas concentration monitoring, and specifically comprises the following steps: acquiring each gas response value in the mixed gas by using a gas array sensor, forming training sample points with the corresponding gas concentrations, and performing nearest neighbor interpolation on a training sample set acquired based on time sequence; extracting feature vectors from the interpolated training sample set through a feature engineering model formed by an automatic encoder and an independent component analysis algorithm, taking the feature vectors as input of a fully-connected artificial neural network, taking a concentration data set of the interpolated training sample set as output, and training the fully-connected artificial neural network; and after the gas response values of the mixed gas to be tested are sequentially processed by the trained automatic encoder, the converged independent component analysis algorithm model and the trained full-connection artificial neural network, outputting the gas concentrations. The method effectively reduces the mutual interference among the mixed gases and realizes the high-accuracy prediction of the concentration of the mixed gases.
Description
Technical Field
The invention belongs to the technical field of mixed gas concentration monitoring, and particularly relates to a sensor array neural network method for mixed gas concentration prediction.
Background
With the rapid development of technology and the acceleration of urban process, the released large amount of polluted gas and toxic and harmful gas have more and more serious influence, so that the real-time and accurate concentration detection of the complex mixed gas is very necessary.
The mixed gas monitoring technology is widely applied to the fields of natural gas industry, electric power, gas transmission, biochemical medicine, indoor gas pollutant detection and the like. The method is characterized in that various active nitrogen/carbon trace gas pollutants are generated in industrial production, chemical synthesis, energy exploitation, industrial waste and the like, and real-time and accurate monitoring of the active nitrogen/carbon trace gas pollutants is an important guarantee for preventing and treating various pollution. The array sensor for detecting the gas, the signal processing circuit and the detection algorithm module form a bionic system together, namely an electronic nose. The traditional electronic nose smell recognition algorithm has the defects of complicated feature extraction steps and low recognition accuracy, so most researches are limited to recognition of gas types in mixed gas, and the gas concentration cannot be predicted. Therefore, the intelligent gas sensor with multifunction and integration is constructed, the concentration is effectively and accurately predicted, the key science and technology bottlenecks of the prior art are solved, and the realization of the functions of the application in multiple occasions is important.
Current solutions for detecting mixed gas by array sensors focus on methods using Machine Learning (ML) and Artificial Neural Networks (ANN). Because the feature extraction of the signals output by the array sensor is relatively complex, the complex and complicated features are difficult to fit by machine learning methods such as a principal component analysis algorithm (PCA), a K-nearest neighbor algorithm (K-Means), a Support Vector Machine (SVM) algorithm and the like, so that the fitting precision is low. Deep learning algorithms are also used in mixed gas identification, where Convolutional Neural Networks (CNNs) are trained by extracting multidimensional features of the output signals of the array sensors, which can effectively improve the accuracy of odor classification, but still have poor effect on the concentration prediction of various gases. The cyclic neural network is widely applied to various time sequences and predicts, wherein the long-short-term memory neural network (LSTM) predicts concentration values by extracting signal sequences of various sensors and fusing, but because of the mutual influence among various gases, the long-short-term memory neural network is difficult to extract key features, so that when the selectivity of the array sensor is poor, the concentration result fused by the algorithm is inaccurate, and meanwhile, the sequence with a certain time length is required to be extracted in the detection process, so that lower detection efficiency is caused. The impulse neural network (SNN) is very suitable for processing the time-space event information of the neuromorphic sensor, the algorithm is more bionic than an artificial neuron, has more advantages in the processing of spike sequences, and makes certain progress in the processing of array sensor signals. However, the calculation of the method uses asynchronous and discontinuous modes, so that the backward propagation algorithm which is successful in the neural network cannot be directly applied to the algorithm. In addition, these deep learning network models have complex structures and large model sizes, and consume a large amount of computing resources.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a sensor array neural network method for predicting the concentration of mixed gas, which realizes high-accuracy prediction of the concentration of the gas in the mixed gas.
The technical scheme adopted by the invention is as follows:
a sensor array neural network method for mixed gas concentration prediction, comprising the steps of:
and 6, introducing the mixed gas to be tested into a gas array sensor, reading response values of each gas in the mixed gas to be tested by using a signal processing circuit, extracting features by an encoder in the automatic encoder after training, converting the features into high-dimensional feature vectors by using a converged independent component analysis algorithm model, inputting the high-dimensional feature vectors into a fully-connected artificial neural network after training, and outputting the concentration of each gas in the mixed gas to be tested.
Further, the training sample set in step 1
； wherein ,Nthe total number of training sample points is 50-200; />
Is the firstiMixed gas concentration characteristic vector of each training sample point, whereinnThe number of gas types of the mixed gas is as followsn≤4，
Is the firstiThe first training sample pointmThe concentration of the seed gas;
is the firstiMixed gas response characteristic vector of each training sample point, and
correspondingly, in->
Is the firstiThe first training sample pointmA response value of the seed gas.
Further, in step 2, training sample sets are obtained
The specific steps for nearest neighbor interpolation are as follows:
step 2.1, traversing the training sample setDThe mixed gas concentration characteristic vector of all training sample points in the system is not used for repeatedly calculating the concentration distance between every two training sample points
：
Step 2.2, calculating all concentration distances
Average concentration distance>
：
wherein ,
to at the same timeNThe number of pairwise permutation and combination is carried out in the training sample points;
step 2.3 at all concentration distances
Less than the average concentration distance->
A mixed gas concentration characteristic vector interpolation point is pre-added between the two training sample points to obtain a mixed gas concentration characteristic vector interpolation point set +.>
； wherein ,numinterpolation points are added for the concentration characteristic vector of the mixed gas; />
To add topThe mixed gas concentration characteristic vector interpolation points;
step 2.4, orderq=1；
Step 2.5 according to the firstqInterpolation points of concentration characteristic vectors of mixed gas
And training sample setDIs the nearest neighbor in (3)kConcentration distance between individual training sample points +.>
Calculate->
Corresponding weighted mixed gas response eigenvector +.>
：
wherein ,
is->
Nearest neighbor of the first pairαIn the training sample points
Weights of (2); />
Is->
Nearest neighbor of the first pairαA mixed gas response feature vector of each training sample point;
step 2.6, mixing gas concentration characteristic vector interpolation points
And weighted mixed gas response eigenvector>
Constructing interpolation training sample points, adding and updating training sample setsDJudgingqWhether or not to be equal tonumIf yes, obtaining a training sample set after interpolationD’Is common toN+numTraining sample points after interpolation; otherwise, letq=q+1, switch back to step 2.5.
Further, in step 2.5kThe value is 2-8.
Further, the number of neurons of the input layer and the output layer of the automatic encoder is equal to the number of gas types of the mixed gas.
Further, the feature vector of the bottleneck layer includes a feature number interval of 15-30.
Further, the extracted feature vector includes a feature quantity interval of 5-15.
Further, the number of neurons of the hidden layer in the fully-connected artificial neural network is 5-50, and the number of neurons of the output layer is equal to the number of gas types of the mixed gas.
The beneficial effects of the invention are as follows:
the invention provides a sensor array neural network method for mixed gas concentration prediction, which is characterized in that firstly, response values of an array sensor are processed based on a nearest neighbor interpolation algorithm, more accurate interpolation results are obtained at dense sample points, response characteristics can be expressed more accurately through fewer sample points, and the subsequent prediction accuracy is ensured; then, based on the characteristic engineering formed by the automatic encoder and the independent component analysis algorithm, the characteristics of the response data are decomposed, and independent variables are extracted, so that the mutual interference between gases is reduced, and the training effect of the fully-connected artificial neural network is improved; finally, the characteristic of the concentration data is learned through a fully-connected artificial neural network, so that the high-accuracy prediction of the gas concentration in the mixed gas is realized; in addition, the method also has higher prediction precision for the scene detected by the array sensor with poor selectivity.
Drawings
Fig. 1 is a diagram showing a fusion algorithm of a sensor array neural network method for mixed gas concentration prediction according to embodiment 1;
FIG. 2 is a graph showing feature point distribution of the 12-dimensional feature vector extracted by the automatic encoder and the independent component analysis algorithm in example 1, wherein the feature point distribution is reduced to 3 dimensions at the same distance by a t-random adjacent embedding (TSNE) dimension reduction algorithm;
FIG. 3 is a graph showing the confidence intervals of the absolute error average value and 95% confidence of the prediction result of the sensor array neural network method for mixed gas concentration prediction proposed in example 1;
fig. 4 is a graph showing the confidence intervals of the predicted result relative to the mean error and 95% confidence of the sensor array neural network method for mixed gas concentration prediction proposed in example 1.
Description of the embodiments
The present invention will be described in further detail with reference to the drawings and the embodiments, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Example 1
The embodiment provides a sensor array neural network method for mixed gas concentration prediction, and the adopted fusion algorithm structure is shown in fig. 1, and specifically comprises the following steps:
step 2.1,Traversing training sample setsDThe mixed gas concentration characteristic vector of all training sample points in the system is not used for repeatedly calculating the concentration distance between every two training sample points
：/>
Step 2.2, calculating all concentration distances
Average concentration distance>
：
wherein ,
to at the same timeNThe number of pairwise permutation and combination is carried out in the training sample points;
step 2.3 at all concentration distances
Less than the average concentration distance->
A mixed gas concentration characteristic vector interpolation point is pre-added between the two training sample points to obtain a mixed gas concentration characteristic vector interpolation point set +.>
； wherein ,numthe total number of interpolation points is 679 for the added mixed gas concentration characteristic vector; />
To add topThe mixed gas concentration characteristic vector interpolation points;
step 2.4, orderq=1；
Step 2.5 according to the firstqInterpolation points of concentration characteristic vectors of mixed gas
And training sample setDIs the nearest neighbor in (3)kConcentration distance between =3 training sample points +.>
Calculate->
Corresponding weighted mixed gas response eigenvector +.>
：
wherein ,
is->
Nearest neighbor of the first pairαIn the training sample points
Weights of (2); />
Is->
Nearest neighbor of the first pairαA mixed gas response feature vector of each training sample point;
step 2.6, mixing gas concentration characteristic vector interpolation points
And weighted mixed gas response eigenvector>
Constructing interpolation training sample points, adding and updating training sample setsDJudgingqWhether or not to be equal tonumIf yes, the training sample set updated last is to be updatedDAs a post-interpolation training sample setD’Is common toN+numTraining sample points after interpolation; otherwise, letq=q+1, switch back to step 2.5;
1 layer of input layer, the number of neurons is 4;
5 hidden layers, wherein the number of neurons in each layer is 30, 20, 30 and 30 respectively;
1 output layer, with 4 neurons;
wherein, the activation functions of the input layer and the hidden layer are both linear rectification functions (ReLu), and the output layer does not use the activation functions; the activation function, the optimizer and the network structure are obtained according to the minimum loss under the condition that the automatic encoder does not have over-fitting through grid search and 3-fold cross verification;
the input layer of the automatic encoder is the first layer of the encoder, the last layer of the encoder is the bottleneck layer (the structure shown by the broken line square block in fig. 1), the number of neurons is 20, meanwhile, the bottleneck layer is also the first layer of the decoder, the last layer of the decoder is the output layer of the automatic encoder, the other layers are all hidden layers, and the number of neurons is 30;
training sample set after interpolationD’In a response data set consisting of all mixed gas response characteristic vectorsYRespectively serving as input of an encoder and output of a decoder, training the automatic encoder to obtain the trained automatic encoder, and extracting feature vectors of a bottleneck layer;
1 layer of input layer, the number of neurons is 12;
5 hidden layers, wherein the number of neurons in each layer is 30, 20 and 10 respectively;
1 output layer, with 4 neurons;
the activation functions of the input layer and the hidden layer are ReLu functions, and the output layer does not use the activation functions; initial weights are set through uniform distribution initialization; the activation function, the optimizer and the network structure are obtained according to the minimum loss under the condition that the fully-connected artificial neural network is not fitted through grid search and 3-fold cross verification;
training sample set after interpolation by taking extracted feature vector as inputD’In a concentration data set consisting of all mixed gas concentration characteristic vectorsXTraining the fully-connected artificial neural network for output to obtain the trained fully-connected artificial neural network;
FIG. 2 shows a feature point distribution diagram of 12-dimensional (high-dimensional) feature vectors extracted by a feature engineering model formed by an automatic encoder after training and a converged independent component analysis algorithm model through the response value data obtained by an array sensor, wherein the feature point distribution diagram is obtained by reducing the dimension to 3-dimensional at the same distance through a t-random adjacent embedding (TSNE) dimension reduction algorithm, so that the data distribution of the 12-dimensional extracted feature vectors extracted by the feature engineering model is more uniform, the extracted feature vectors with larger deviation do not exist, and the extracted feature vectors with more uniform distribution have higher accuracy in the training of a subsequent fully-connected artificial neural network.
According to the 12-dimensional feature vector extracted by the feature engineering, the 12-dimensional feature vector is sent into a full-connection artificial neural network for prediction, and the optimal parameters of the full-connection artificial neural network training are obtained through grid search and 3-fold cross verification, as shown in Table 1:
TABLE 1 optimal parameters for fully connected artificial neural network training
| Super parameter | Value of |
| Training batch | 128 |
| Training times | 3000 |
| Optimizer | Adaptive moment estimation |
| Initial weight hidden layer activation function | Uniform distribution initialization ReLu |
In this embodiment, a mixed gas formed by mixing carbon dioxide, nitrogen dioxide, methane and ammonia is taken as an example, an adaptive moment estimation (Adam) optimizer is adopted to optimize the back propagation of a fully-connected artificial neural network, and finally, the prediction result of the fully-connected artificial neural network after training is represented by a confidence interval of a relative error and an absolute error under a 95% confidence, and the results are respectively shown in fig. 3 and fig. 4, so that four gases have good prediction accuracy, wherein the relative error is within 4% of the confidence interval, the accuracy is high, and the prediction effect is good.
While the invention has been described in terms of specific embodiments, any feature disclosed in this specification may be replaced by alternative features serving the equivalent or similar purpose, unless expressly stated otherwise; all of the features disclosed, or all of the steps in a method or process, except for mutually exclusive features and/or steps, may be combined in any manner.
Claims (7)
1. A sensor array neural network method for mixed gas concentration prediction, comprising the steps of:
step 1, introducing mixed gas with known gas concentrations into a gas array sensor, reading response values of the gases in real time by using a signal processing circuit, wherein the gas concentrations form mixed gas concentration feature vectors, the response values of the gases form mixed gas response feature vectors, and taking the mixed gas concentration feature vectors and the corresponding mixed gas response feature vectors as a group of training sample points to further obtain a training sample set based on time sequence acquisitionD；
Step 2, training sample setDNearest neighbor interpolation is carried out, specifically: calculating the concentration distance between every two training sample points, performing mixed gas concentration feature vector interpolation between two training sample points smaller than the average concentration distance, sequentially calculating weighted mixed gas response feature vectors of a plurality of nearest neighbor training sample points corresponding to each interpolation position, obtaining all interpolation training sample points, and further obtaining a training sample set after interpolationD’；
The specific process of calculating the weighted mixed gas response characteristic vector of a plurality of nearest neighbor training sample points corresponding to each interpolation position is as follows:
step a, orderq=1；
Step b, according to the firstqInterpolation points of concentration characteristic vectors of mixed gas
And training sample setDIs the nearest neighbor in (3)kConcentration distance between individual training sample points +.>
Calculate->
Corresponding weighted mixed gas response eigenvector +.>
：
wherein ,
is->
Nearest neighbor of the first pairαMixed gas concentration characteristic vector of individual training sample points +.>
Weights of (2); />
Is->
Nearest neighbor of the first pairαA mixed gas response feature vector of each training sample point;
step c, mixed gas concentration characteristic vector interpolation point
And weighted mixed gas response eigenvector>
Constructing interpolation training sample points, adding and updating training sample setsDJudgingqWhether or not the total number of interpolation points of the characteristic vector of the added mixed gas concentration is equal tonumIf yes, obtaining a training sample set after interpolationD’Is common toN+numTraining sample points after interpolation, whereinNTraining the total number of sample points; otherwise, letq=q+1, switching back to step b;
step 3, constructing an automatic encoder comprising an encoder and a decoder, and training a sample set after interpolationD’In a response data set consisting of all mixed gas response characteristic vectorsYRespectively serving as input of an encoder and output of a decoder, training the automatic encoder to obtain the trained automatic encoder, and extracting feature vectors of a bottleneck layer in the automatic encoder;
step 4, transforming the feature vector of the bottleneck layer into an extracted feature vector through an independent component analysis algorithm, and obtaining a converged independent component analysis algorithm model;
step 5, constructing a fully-connected artificial neural network, taking the extracted feature vector as input, and training a sample set after interpolationD’In a concentration data set consisting of all mixed gas concentration characteristic vectorsXTraining the fully-connected artificial neural network for output to obtain the trained fully-connected artificial neural network;
and 6, introducing the mixed gas to be tested into a gas array sensor, reading response values of each gas in the mixed gas to be tested by using a signal processing circuit, extracting features by an encoder in the automatic encoder after training, converting the features into high-dimensional feature vectors by using a converged independent component analysis algorithm model, inputting the high-dimensional feature vectors into a fully-connected artificial neural network after training, and outputting the concentration of each gas in the mixed gas to be tested.
2. The sensor array neural network method for mixed gas concentration prediction according to claim 1,characterized in that the training sample set in step 1
； wherein ,Nthe total number of training sample points is 50-200; />
Is the firstiMixed gas concentration characteristic vector of each training sample point, whereinnThe number of gas types of the mixed gas is as followsn≤4，/>
Is the firstiThe first training sample pointmThe concentration of the seed gas; />
Is the firstiMixed gas response characteristic vector of each training sample point, and +.>
Correspondingly, in->
Is the firstiThe first training sample pointmA response value of the seed gas.
3. The sensor array neural network method for mixed gas concentration prediction of claim 2, wherein the training sample set is performed in step 2
The specific steps for nearest neighbor interpolation are as follows:
step 2.1, traversing the training sample setDThe mixed gas concentration characteristic vector of all training sample points in the system is not used for repeatedly calculating the concentration distance between every two training sample points
：
Step 2.2, calculating all concentration distances
Average concentration distance of (c):
wherein ,
to at the same timeNThe number of pairwise permutation and combination is carried out in the training sample points;
step 2.3 at all concentration distances
Distance less than average concentration
A mixed gas concentration characteristic vector interpolation point is pre-added between the two training sample points to obtain a mixed gas concentration characteristic vector interpolation point set +.>
； wherein ,numinterpolation points are added for the concentration characteristic vector of the mixed gas; />
To add topThe mixed gas concentration characteristic vector interpolation points;
step 2.4, orderq=1；
Step 2.5 according to the firstqInterpolation points of concentration characteristic vectors of mixed gas
And training sampleBook setDIs the nearest neighbor in (3)kConcentration distance between individual training sample points +.>
Calculate->
Corresponding weighted mixed gas response eigenvector +.>
：
wherein ,
is->
Nearest neighbor of the first pairαIn training sample points->
Weights of (2); />
Is->
Nearest neighbor of the first pairαA mixed gas response feature vector of each training sample point;
step 2.6, mixing gas concentration characteristic vector interpolation points
And weighted mixed gas response eigenvector>
Constructing interpolation training sample points, adding and updating training sample setsDJudgingqWhether or not to be equal tonumIf yes, obtaining a training sample set after interpolationD’Is common toN+numTraining sample points after interpolation; otherwise, letq=q+1, switch back to step 2.5.
4. The sensor array neural network method for mixed gas concentration prediction according to claim 3, wherein in step 2.5kThe value is 2-8.
5. The sensor array neural network method for mixed gas concentration prediction according to claim 1, wherein the feature vector of the bottleneck layer comprises a feature number interval of 15-30.
6. The sensor array neural network method for mixed gas concentration prediction according to claim 1, wherein the extracted feature vector comprises a feature quantity interval of 5-15.
7. The sensor array neural network method for mixed gas concentration prediction according to claim 1, wherein the number of neurons in the hidden layer in the fully connected artificial neural network is 5-50.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310129437.4A CN115860056B (en) | 2023-02-17 | 2023-02-17 | Sensor array neural network method for mixed gas concentration prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310129437.4A CN115860056B (en) | 2023-02-17 | 2023-02-17 | Sensor array neural network method for mixed gas concentration prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115860056A CN115860056A (en) | 2023-03-28 |
| CN115860056B true CN115860056B (en) | 2023-05-26 |
Family
ID=85658316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310129437.4A Active CN115860056B (en) | 2023-02-17 | 2023-02-17 | Sensor array neural network method for mixed gas concentration prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115860056B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117929655B (en) * | 2024-03-25 | 2024-06-21 | 理想创新实验室科技(江苏)有限公司 | Smell intensity detection method, device, equipment, storage medium and vehicle |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110161181A (en) * | 2018-02-13 | 2019-08-23 | 中国石油化工股份有限公司 | The concentration of component recognition methods of mixed gas and system |
| CN111126575A (en) * | 2020-01-09 | 2020-05-08 | 同济大学 | Gas sensor array mixed gas detection method and device based on machine learning |
| CN114707598A (en) * | 2022-03-31 | 2022-07-05 | 西安交通大学 | Mixed gas identification method, system, terminal equipment and readable storage medium |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101949826B (en) * | 2010-09-02 | 2012-09-05 | 西安交通大学 | Positive model and inverse model-based quantitative spectrometric analysis and calibration method of multi-component gas |
| CN104820072B (en) * | 2015-04-30 | 2016-03-09 | 重庆大学 | Based on the monitoring method of the Electronic Nose air-quality monitoring system of cloud computing |
| CN105938116B (en) * | 2016-06-20 | 2019-09-10 | 吉林大学 | Gas sensor array concentration detection method based on fuzzy division and model integrated |
| US10948467B2 (en) * | 2018-05-17 | 2021-03-16 | East China University Of Science And Technology | Online centralized monitoring and analysis method for multi-point malodorous gases using electronic nose instrument |
| CN108732316A (en) * | 2018-07-06 | 2018-11-02 | 天津澜禹笙环保科技有限公司 | A kind of stench intelligent monitor system based on cloud computing platform |
| EP3736564A1 (en) * | 2019-05-09 | 2020-11-11 | Infineon Technologies AG | Gas sensing device and method for operating a gas sensing device |
| EP3770583B1 (en) * | 2019-07-26 | 2023-07-19 | Infineon Technologies AG | Gas sensing device for sensing a gas in a mixture of gases and method for operating a gas sensing device |
| EP3879266A1 (en) * | 2020-03-11 | 2021-09-15 | Infineon Technologies AG | Gas sensing device and method for operating a gas sensing device |
| CN112464544B (en) * | 2020-11-17 | 2024-06-18 | 北京工业大学 | Method for constructing prediction model of dioxin emission concentration in urban solid waste incineration process |
| CN115324698B (en) * | 2022-08-09 | 2023-06-16 | 上海交通大学 | Diesel engine tail gas analysis sensing method based on neural network |
-
2023
- 2023-02-17 CN CN202310129437.4A patent/CN115860056B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110161181A (en) * | 2018-02-13 | 2019-08-23 | 中国石油化工股份有限公司 | The concentration of component recognition methods of mixed gas and system |
| CN111126575A (en) * | 2020-01-09 | 2020-05-08 | 同济大学 | Gas sensor array mixed gas detection method and device based on machine learning |
| CN114707598A (en) * | 2022-03-31 | 2022-07-05 | 西安交通大学 | Mixed gas identification method, system, terminal equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115860056A (en) | 2023-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | A surrogate-assisted multiswarm optimization algorithm for high-dimensional computationally expensive problems | |
| CN109142171B (en) | Urban PM10 concentration prediction method based on feature expansion and fusing with neural network | |
| CN111967343A (en) | Detection method based on simple neural network and extreme gradient lifting model fusion | |
| CN112884056A (en) | Optimized LSTM neural network-based sewage quality prediction method | |
| CN115860056B (en) | Sensor array neural network method for mixed gas concentration prediction | |
| CN110880369A (en) | Gas marker detection method based on radial basis function neural network and application | |
| CN114530258A (en) | Deep learning drug interaction prediction method, device, medium and equipment | |
| CN111008224A (en) | Time sequence classification and retrieval method based on deep multitask representation learning | |
| Pei et al. | Genetic algorithms for classification and feature extraction | |
| CN114399066A (en) | Mechanical equipment predictability maintenance system and maintenance method based on weak supervision learning | |
| CN115099296A (en) | Sea wave height prediction method based on deep learning algorithm | |
| CN115659254A (en) | Power quality disturbance analysis method for power distribution network with bimodal feature fusion | |
| CN115327041A (en) | Air pollutant concentration prediction method based on correlation analysis | |
| CN113505477A (en) | Process industry soft measurement data supplementing method based on SVAE-WGAN | |
| Song et al. | Importance weighted expectation-maximization for protein sequence design | |
| CN111737294A (en) | Data flow classification method based on dynamic increment integration fuzzy | |
| CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
| CN111797979A (en) | Vibration transmission system based on LSTM model | |
| CN117313516A (en) | Fermentation product prediction method based on space-time diagram embedding | |
| CN116910049A (en) | MDAN-based power load data missing value filling model and construction method thereof | |
| CN115396198B (en) | Power grid intrusion detection system and method based on CNN architecture and parameter parallel optimization | |
| CN111310974A (en) | Short-term water demand prediction method based on GA-ELM | |
| CN114358058A (en) | Wireless communication signal open set identification method and system based on deep neural network | |
| Tian et al. | Incremental learning of Bayesian networks with hidden variables | |
| Ma et al. | TPAD: temporal pattern based neural network model for anomaly detection in multivariate time series |
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 |