CN114707598A - Mixed gas identification method, system, terminal equipment and readable storage medium - Google Patents

Mixed gas identification method, system, terminal equipment and readable storage medium Download PDF

Info

Publication number
CN114707598A
CN114707598A CN202210335225.7A CN202210335225A CN114707598A CN 114707598 A CN114707598 A CN 114707598A CN 202210335225 A CN202210335225 A CN 202210335225A CN 114707598 A CN114707598 A CN 114707598A
Authority
CN
China
Prior art keywords
mixed gas
neural network
dimensional
features
convolution
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.)
Pending
Application number
CN202210335225.7A
Other languages
Chinese (zh)
Inventor
王宁
王金磊
马培元
练赛
雷绍充
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202210335225.7A priority Critical patent/CN114707598A/en
Publication of CN114707598A publication Critical patent/CN114707598A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Game Theory and Decision Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a mixed gas identification method, a system, a terminal device and a storage medium, comprising the following steps: acquiring multi-channel one-dimensional time sequence data of the mixed gas; performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation, and extracting the overall features; performing convolution and pooling operation on the overall features to obtain a feature extraction result; identifying the type of the mixed gas from the characteristic extraction result; and (4) predicting the concentration of the mixed gas in real time by using the convolutional neural network and the cyclic neural network. The method for identifying the type of the mixed gas of the one-dimensional convolutional neural network under the random condition can automatically extract the characteristics and classify the extracted characteristics to identify the type of the mixed gas, and does not need to separate the characteristic extraction and classification training. The gas type is marked in a multi-label mode, and the dimension space brought by a single-label mode is reduced.

Description

Mixed gas identification method, system, terminal device and readable storage medium
Technical Field
The invention belongs to the technical field of gas identification, and relates to a mixed gas identification method, a mixed gas identification system, terminal equipment and a storage medium.
Background
The intelligent gas identification system is a system for realizing qualitative and quantitative analysis of detected gas by referring to an olfactory perception mechanism of animals or human beings. With the rapid development of society, the industrial field is rapidly developed, the gas identification technology is rapidly developed, the gas can be rapidly and accurately detected, and the method plays an important role in a plurality of fields of society, including industrial pollution monitoring, food safety, medical fields and the like. The system adsorbs gas molecules through the sensor to generate electric signals, and the AD conversion circuit is used for carrying out quantitative collection on the electric signals to finish the extraction of features, and finally the type identification and concentration judgment of the gas are finished. Natural gas is used as an important clean energy source and has wide application in the fields of chemical industry and fuels. In an effort to provide efficient, clean, and diverse use of natural gas, the quality of the natural gas affects every aspect of this subsequent processing, and thus analysis of the gas composition content in the natural gas facilitates natural gas development and transportation and further processing and energy utilization. Natural gas is a mixture of most of methane, a small amount of alkane gases such as ethane and propane, and a small amount of impurity gases.
In recent years, many studies on identification of mixed gas have been made, and these studies have also achieved a good effect in identification of the kind of mixed gas. However, these studies are generally performed under a relatively stable condition, and most of the input data adopts stable characteristics such as response time, recovery time or sensitivity in the sensor array, and after simple data processing, some mainstream classification methods are adopted, so that a better recognition effect can be finally achieved. However, to apply these research results to actual production life, there are two problems: 1) in practical application, the form and concentration of the mixed gas are complex, random and irregular; 2) the previous work has relatively few studies on the real-time prediction of the concentration of the mixed gas, and more, the concentration prediction of the complete gas reaction process. 3) In recent years, research on prediction of the concentration of a mixed gas has never been stopped. However, the work is regular periodic continuous reaction under laboratory conditions, and it is also based on ideal results, the concentration values of gases are random in real life many times, and the reaction may not be completely started to recover when an intelligent gas identification system is used for identification, and under such a condition, the prediction of the concentration values is more challenging and practical.
Disclosure of Invention
The present invention is directed to solving the problems of the prior art, and provides a mixed gas identification method, a system, a terminal device, and a storage medium, which can identify the type and detect the concentration of the mixed gas at random.
In order to achieve the purpose, the invention adopts the following technical scheme to realize the purpose:
a mixed gas identification method comprises the following steps:
acquiring multi-channel one-dimensional time sequence data of the mixed gas;
performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation, and extracting the overall features; performing convolution and pooling operation on the overall features to obtain a feature extraction result;
identifying the type of the mixed gas from the result of the feature extraction by using a multi-label classification method;
and establishing a regression model, and predicting the concentration of the mixed gas in real time by using a convolutional neural network and a cyclic neural network according to the type of the mixed gas and combining the result of the feature extraction.
The method is further improved in that:
the acquiring of the multi-channel one-dimensional time sequence data of the mixed gas by the gas sensor array comprises:
and acquiring multi-channel one-dimensional time sequence data of the mixed gas through the gas sensor array.
The one-dimensional convolution operation is specifically as follows:
the one-dimensional convolution equation is as follows:
Figure BDA0003576549340000031
wherein A is an input sequence, and the dimension of the input sequence A is (1, N)a) B is a convolution kernel with dimension of (1, N)b) And C (j) is the convolution result of the j-th neuron, and j is more than or equal to 1 and less than or equal to Na-Nb+ 1; n represents the size of each layer of convolution kernel, the initialization of the weight adopts Xavier initialization, and the uniform distribution of the Xavier initialization is as follows:
Figure BDA0003576549340000032
wherein, W represents uniform distribution, U represents sampling range, and m and n are the number of input units and output units of each layer.
The multi-label classification method specifically comprises the following steps:
prediction of target gas ypreThe following were used:
ypre＝sigmoid(WTh+b) (3)
wherein sigmoid () represents an activation function, WTRepresenting the weight of a hidden layer of a second layer, T representing a transposed matrix, h representing a network input, and b representing a bias item;
adopting the output of the sigmoid function as a classification result, using a threshold value of 0.5 as a judgment criterion for the output result, setting a predicted value label with an output value higher than 0.5 as 1, and setting the predicted value label as 0 if the output value is lower than 0.5; the Log loss for the second class of the model is as follows:
Figure BDA0003576549340000033
wherein, ciIs a class, y (c)i) Is a true tag, N is the number of classifiers, ypre(ci) And q (c)i) Are respectively class ciThe average value of the Log loss values of the classifier is finally used as the final loss value.
The identification of the kind of the mixed gas comprises the following steps:
inputting the result obtained by automatic feature extraction into a multilayer sensor, and identifying the type of the mixed gas; wherein the multilayer perceptron consists of full connections.
The real-time prediction of the concentration of the mixed gas comprises the following steps:
combining original time sequence data, selecting meaningful high-level features, inputting the high-level features output by a Convolutional Neural Network (CNN) into a Recurrent Neural Network (RNN) layer, analyzing the high-level features in a time dimension by using the RNN, and screening out the features which are most suitable for prediction in time; and splicing the screened feature vectors into a full-connection mode for calculating linear regression, and finally obtaining a predicted concentration value.
A mixed gas identification system comprising:
the data acquisition module is used for acquiring multi-channel one-dimensional time sequence data of the mixed gas;
the feature extraction module is used for performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation and extracting the overall features; performing convolution and pooling operation on the overall features to obtain a feature extraction result;
the type identification module is used for identifying the type of the mixed gas from the result of the characteristic extraction by using a multi-label classification method;
and the concentration prediction module is used for establishing a regression model, and predicting the concentration of the mixed gas in real time by utilizing the convolutional neural network and the cyclic neural network according to the type of the mixed gas and combining the result of the characteristic extraction.
A terminal device comprising a memory, a processor and a computer program stored in said memory and executable on said processor, said processor implementing the steps of the method as described above when executing said computer program.
A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method as described above.
Compared with the prior art, the invention has the following beneficial effects:
the method for identifying the type of the one-dimensional convolution neural network 1D-DCNN mixed gas under the random condition is realized based on a multi-label mode of the one-dimensional convolution neural network, can automatically extract the characteristics and classify the extracted characteristics to identify the type of the mixed gas, does not need to separate the characteristic extraction and classification training, and is an end-to-end classifier. Secondly, the gas type is marked in a multi-label mode, and the dimension space brought by a single-label mode is reduced. The convolutional neural network and the cyclic neural network (convolutional neural network CNN-cyclic neural network RNN) firstly combine original response data into more meaningful and higher-level features by using the convolutional neural network, and then establish a regression prediction model for the output features of the convolutional neural network CNN by using the cyclic neural network RNN, so that the species identification and the concentration real-time prediction of mixed gas species under random conditions are realized. Finally, the convolutional neural network CNN-lstm can achieve better prediction effect than ESN and LR on the real-time prediction of the concentration of the mixed natural gas.
Drawings
In order to more clearly explain the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of the mixed gas identification method of the present invention.
FIG. 2 is a schematic diagram of the mixed gas identification system of the present invention.
FIG. 3 is a diagram of the overall network framework of the one-dimensional convolutional neural network 1D-DCNN of the present invention.
FIG. 4 is a diagram of a one-dimensional convolution according to the present invention.
FIG. 5 is an overall framework diagram of the convolutional neural network CNN-recurrent neural network RNN of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined or explained in subsequent figures.
In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. are used for indicating the orientation or positional relationship based on the orientation or positional relationship shown in the drawings or the orientation or positional relationship which is usually arranged when the product of the present invention is used, the description is merely for convenience and simplicity, and the indication or suggestion that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and thus, cannot be understood as limiting the present invention. Furthermore, the terms "first," "second," and the like are used merely to distinguish one description from another, and are not to be construed as indicating or implying relative importance.
Furthermore, the term "horizontal", if present, does not mean that the component is required to be absolutely horizontal, but may be slightly inclined. For example, "horizontal" merely means that the direction is more horizontal than "vertical" and does not mean that the structure must be perfectly horizontal, but may be slightly inclined.
In the description of the embodiments of the present invention, it should be further noted that unless otherwise explicitly stated or limited, the terms "disposed," "mounted," "connected," and "connected" should be interpreted broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
The invention is described in further detail below with reference to the accompanying drawings:
referring to fig. 1, an embodiment of the invention discloses a mixed gas identification method, which includes the following steps:
s1, acquiring multi-channel one-dimensional time sequence data of the mixed gas through the gas sensor array;
s2, using the one-dimensional time sequence data obtained by the sensors as the input of an automatic feature extraction part, using the one-dimensional convolution operation of the automatic feature part on each sensor, splicing the features after convolution together to be used as the overall feature, and finally using the features after convolution and pooling as the input of the third step;
and S3, taking the result obtained by automatic feature extraction as the input of a classification part, wherein the classification part mainly comprises a multi-layer perceptron MLP formed by fully connecting the residual layers, and finally the features extracted in the convolution stage enter the multi-layer perceptron MLP to identify the types of the mixed gas one by one.
S4, original response data are combined into more meaningful and higher-level features through the convolutional neural networks S2 and S3, and then a regression model is built for the output features of the convolutional neural network CNN through the recurrent neural network RNN, so that real-time concentration prediction is achieved. Firstly, the convolutional neural network CNN mainly combines original time series data and selects significant high-level features at the same time, inputs the high-level features output by the convolutional neural network CNN to a recurrent neural network RNN layer, analyzes the high-level features in a time dimension by using the recurrent neural network RNN, and screens out features most suitable for prediction in the time dimension. And finally, splicing the screened feature vectors into a full-connection mode for calculation of linear regression to finally obtain a predicted concentration value.
As shown in fig. 2, an embodiment of the present invention discloses a mixed gas identification system, including:
the data acquisition module is used for acquiring multi-channel one-dimensional time sequence data of the mixed gas;
the feature extraction module is used for performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation and extracting the overall features; performing convolution and pooling operation on the overall features to obtain a feature extraction result;
the type identification module is used for identifying the type of the mixed gas from the result of the characteristic extraction by using a multi-label classification method;
and the concentration prediction module is used for establishing a regression model, and predicting the concentration of the mixed gas in real time by utilizing the convolutional neural network and the cyclic neural network according to the type of the mixed gas and combining the result of the characteristic extraction.
The principle of the invention is as follows:
the invention provides a method for identifying the type of a one-dimensional convolution neural network 1D-DCNN mixed gas under a random condition, which is realized based on a multi-label mode of one-dimensional convolution neural channels and collaterals. The invention can automatically extract the characteristics and classify the extracted characteristics to identify the types of the mixed gas, does not need to separate the characteristic extraction and classification training, and is an end-to-end classifier. The specific principle of the proposed one-dimensional convolutional neural network 1D-DCNN and the verification comparison result of the experiment will be explained in detail below.
As shown in fig. 3, fig. 3 is a deep learning framework based on a one-dimensional convolutional neural network 1D-DCNN, which is integrally composed of two parts: a feature extraction part and a classification part. In the feature extraction part, convolution operation, batch standardization operation and an activation function are taken as a convolution stage, and the automatic feature extraction part consists of a plurality of convolution stages; and the classification part is a multi-layer perceptron MLP network.
The present invention is described in further detail below:
(1) feature extraction section
In contrast to conventional convolutional neural networks CNN, the response of the gas mixture at the gas sensor is a one-dimensional time series rather than a two-dimensional image pixel, while unlike image recognition, the response of the gas sensor array is horizontally correlated rather than vertically and horizontally on a two-dimensional image. Thus, a one-dimensional convolution operation can be used on the one-dimensional response of each sensor. Subsequently, the features obtained after the convolution are spliced together to further extract the overall features of the different sensors, which can increase the selectivity of the gas sensor array. In turn, a relatively small number of convolution kernels (64 or 128) are applied and the number of layers of the model (i.e., the number of convolution stages) is increased, thereby improving the generalization performance of the model.
In addition, a Dropout layer is inserted between the third convolution stage and the fourth convolution stage, so that the operation of randomly reducing the number of neurons in the third convolution stage can be reduced, the occurrence of overfitting can be reduced in the stage of training the model, and the calculation efficiency can be improved. In the overall network framework, only one pooling operation (averaging pooling) is employed after the last convolution stage, which can downsample the number of extracted features to 1/3 of the original number and can remove redundant features of the fourth convolution stage. The response of the gas sensor is continuous and correlated in the time domain, and the operation of averaging pooling can be used to process values at different nearby locations on different feature maps and can ensure continuity of the downsampled feature.
(2) Sorting section
And the residual layers of the one-dimensional convolutional neural network 1D-DCNN are fully connected to form a multilayer sensor MLP, and finally the features extracted in the convolution stage enter the multilayer sensor MLP to identify the types of the mixed gas one by one. Because the computing complexity of the full connection of the multilayer perceptron MLP occupies the main computing amount of the whole one-dimensional convolutional neural network 1D-DCNN, the computing complexity is kept at a proper level, and the number of the neurons of the multilayer perceptron MLP is reduced to 64 and 32. In addition, ReLU functions are used as activation functions of the ReLU functions, and a dropout operation is inserted before the first fully connected layer so as to improve the calculation efficiency and reduce overfitting. Finally, in the output layer, 3 neurons and sigmoid functions are used to output the probabilities of 3 multi-label classes, and the solution of multi-label classification will be described in detail later.
(3) Convolution stage
The convolution phase of the one-dimensional convolutional neural network 1D-DCNN is composed of operations such as convolution operations, batch normalization operations, and nonlinear function activation. As described above, the convolution operation is comprised of a series of one-dimensional convolution kernels. Wherein the sizes of the convolution kernels of the 4 convolution stages are 1 × 16 × 8, 1 × 3 × 64, and 1 × 3 × 128, respectively, where 8, 64, 64, and 128 are the number of convolution kernels. Details of the one-dimensional convolution operation are shown in FIG. 4, which is similar to a conventional two-dimensional convolution, assuming that the input sequence A has dimensions (1, N)a) And convolution kernel B has dimension (1, N)b) The one-dimensional convolution and equation can be expressed as in equation (1):
Figure BDA0003576549340000091
wherein j is more than or equal to 1 and less than or equal to Na-Nb+1, and C (j) as the result of the convolution of the j-th neuron. Since weight initialization strongly affects the performance of the model and its generalization capability, xavier initialization is used as a weight initialization method, which is an effective neural network initialization method because it can make the variance of each layer nearly equal and optimize the information flow. The uniform distribution of Xavier initialization can be expressed as in equation (2):
Figure BDA0003576549340000092
wherein m and n are the number of input units and output units of each layer.
Batch normalization and activation function: batch normalization can effectively improve the performance of each layer with a nonlinear sigmoid activation function, so batch normalization processing is performed in each layer structure. Meanwhile, the addition of the batch normalization layer can not only accelerate the training process, but also reduce the occurrence of overfitting especially in a relatively small data set. As for the activation function, similar to the conventional convolutional neural network CNN, the ReLU equation is adopted as the activation function at each convolution stage.
(4) Multi-label classification method
As the number of the mixed gas components increases, the kinds of gas mixture will increase exponentially, and if the label is expressed using the form of a one-hot, the dimension of the label will become abnormally large. For example, assuming a mixed gas containing n components, the label dimension using the form of one-hot is n, whereas the label dimension value using the multi-label approach is n. Finally, a multi-label form is adopted in the label form, so that the overall calculation amount caused by large label dimension is reduced. The problem transformation method is adopted to solve the multi-label problem, and multi-label classification is transformed into a plurality of second classifications. As shown in Table 1, the data set has a mix of 6 multi-labels, where a "1" indicates the corresponding gas occurrence and [0,0]Indicating that no measured gas is present and only background gas (e.g., air) is present. The multi-label classification is divided into three classes containing only single labels:
Figure BDA0003576549340000103
and
Figure BDA0003576549340000104
because the labels of the mixed gas data sets are independent, the conversion of the three binary problems can be realized by a sigmoid activation function, and the method can be better combined with a Convolutional Neural Network (CNN) expansion method.
TABLE 1 examples of mixed gas multi-label data
Figure BDA0003576549340000101
The output of the one-dimensional convolutional neural network 1D-DCNN is shown as formula (3):
ypre＝sigmoid(WTh+b) (3)
wherein y ispreDenotes the prediction of the target gas, WTRepresenting the weight of the second layer hidden layer. In the output layer, the output of the sigmoid function is adopted as the results of the three classifiers, meanwhile, the threshold value of 0.5 is used as a judgment criterion for the output result, the label of a predicted value with the output value higher than 0.5 is set as 1, and if the label is lower than 0.5, the label is set as 0.
Loss values of the model, Log losses of the three second classifications are modified to formula (4):
Figure BDA0003576549340000102
wherein, y (c)i) Is a true tag, N is the number of classifiers, ypre(ci) And q (c)i) Are respectively class ciAnd (4) the probabilities of the positive prediction and the negative prediction are finally used as the final loss value by taking the average value of the Log loss values of the three classifiers.
The invention relates to a method for predicting and analyzing concentration values of components in mixed gas under the condition of random distribution of the concentration values of the components in the mixed gas, which mainly studies on the real-time prediction of the concentration of the mixed gas. Because deep learning obtains good results in multivariate time series prediction, a model is established by using an observed time series to predict a new time series trend, and deep neural network methods such as a model combining a recurrent neural network RNN, a recurrent neural network and a dynamic Boltzmann machine, a model combining an ARIMA and a multilayer perceptron MLP and the like obtain good results on predicting the time series. Because the response curve of the mixed gas also belongs to the time series, a regression model of the response value and the concentration label can be established by adopting a recurrent neural network when the concentration of the mixed gas is predicted in real time. The concentration of the mixed gas is linear with the response value of the gas at a lower condition, but the concentration of the mixed gas and the response value of the gas are nonlinear with the increase of the concentration. Therefore, for a data set with nonlinearity, the data set can be mapped into a high-dimensional space, and then a regression model is established through a regression method to perform prediction estimation on new data. As can be known from the one-dimensional convolutional neural network, the convolutional neural network can be used for firstly combining the original response data into more meaningful and higher-level features, and then the recurrent neural network RNN is used for establishing a regression prediction model for the output features of the convolutional neural network CNN, so that the real-time prediction of the concentration is realized.
The invention designs a novel convolutional neural network and cyclic neural network (CNN-RNN) framework, respectively designs the frameworks and loss functions of the convolutional neural network and the cyclic neural network, and sets a comparison method and analyzes the experimental result. The structure provided by the invention is shown in fig. 5 and mainly comprises a convolutional neural network CNN part, a cyclic neural network RNN part and a fully-connected linear regression layer. First, the convolutional neural network CNN mainly uses the original time series data xtAnd combining and simultaneously selecting meaningful high-level features, inputting the high-level features output by the convolutional neural network CNN into a recurrent neural network RNN layer, analyzing the high-level features on a time dimension by using the recurrent neural network RNN, and screening out the features which are most suitable for prediction on the time. Finally, the screened feature vectors are spliced into a full-connection mode and used for calculating linear regression, and finally a prediction output value y is obtainedpre。
The network structure of the present invention is explained in detail below:
(1) convolutional neural network CNN
Considering that the cyclic neural network RNN is adopted for processing on a time axis subsequently, a two-dimensional convolution mode is adopted for multivariable mixed gas response, and original multi-sensor response data are converted into one-dimensional vectors after being convolved, so that the cyclic neural network RNN can be processed subsequently. Assuming that the dimensionality of an input response sequence A is (M, N), wherein M is the number of sampling points in a time domain, and N is the number of sensor arrays; the size of the convolution kernel B is (H, N), the width of the convolution kernel B is equal to that of the convolution kernel A, so that the result after convolution is a one-dimensional vector, and a specific calculation formula is as follows:
Figure BDA0003576549340000121
in the above formula, i is 1 ≦ M — H +1, j is 1, S is a one-dimensional vector, and S (i, j) is expressed as a result after the ith convolution.
The convolutional neural network part does not refer to pooling operations, but only to extract temporal features in the time domain, and new temporally continuous features are extracted after convolution. In this configuration, the result after convolution is nonlinearly transformed using the ReLU function as shown below, followed by stitching together the vectors after each time step convolution extraction to provide a temporally continuous input to the recurrent neural network RNN layer.
ht＝ReLU(Wt*Xt+bt)
In the above formula, htFor the output result after convolution as the input feature in the subsequent time domain, WtIs a convolution kernel in a time segment, XtThe multivariate response value of the mixed gas in the time domain is obtained.
(2) Recurrent neural network portion
As shown in fig. 5, the convolutional neural network CNN concatenates the output features (one-dimensional vectors) after the convolution of the mixed gas response to form temporally continuous features, and then directly inputs these concatenated features to the recurrent neural network RNN network. In practical application, the recurrent neural network RNN has two most effective sequence models, namely a long-short term memory (LSTM) and a Gated Recurrent Unit (GRU). Long and short term memory is used as a processing unit of the recurrent neural network RNN. Structurally, the gating circulation unit GRU is simpler than the long-short term memory LSTM, but in the later experiment, due to the fact that the data set is large, the expression performance of the long-short term memory LSTM is better, and the result of the subsequent comparison experiment is verified. The long-short term memory LSTM introduces three gates: an input gate, a forgetting gate and an output gate, while introducing memory cells in the same state as hidden.
(3) Objective function
The real-time hidden state output from the recurrent neural network RNN finally enters a fully-connected linear regression layer, the hiding at each time step is developed into a fully-connected mode, and the output value corresponding to each time is predicted by solving a linear regression value, as shown in the following formula:
Figure BDA0003576549340000131
wherein, ytFor the prediction outputs corresponding to different time instants,
Figure BDA0003576549340000132
is a full connection weight parameter.
For the regression problem, the square error is generally adopted as the optimization target, as shown in the following formula:
Figure BDA0003576549340000133
wherein, theta is expressed as a parameter set of the model needing optimization updating, T is a set of all time steps, and Y istpFor predicting concentration values at different time steps, YtThe real concentration value of the corresponding time step. The optimization objective is to minimize the sum of the errors by updating the set of weight parameters of the optimization model, i.e. the predicted values are closest to the true values.
The invention also has the following advantages:
1. in the aspect of the type identification problem, the invention provides a novel multi-label one-dimensional convolutional neural network 1D-DCNN algorithm for classifying binary mixed gas of ethylene, carbon monoxide and methane. The one-dimensional convolutional neural network 1D-DCNN is capable of automatically extracting features from the raw data set and classifying while processing the mixed gas data in a multi-label manner. In addition, the projection of the feature automatically extracted by the one-dimensional convolutional neural network 1D-DCNN on PCA is easier to classify than the original response, the experiment is expanded to 10-fold cross validation, the identification accuracy reaches 96.30 percent, and the method is far superior to the performance of a Support Vector Machine (SVM), an Artificial Neural Network (ANN), k-nearest neighbor (KNN) and a Random Forest (RF).
2. On the aspect of real-time concentration prediction, the invention provides a novel mixed gas concentration real-time prediction structure combining a convolutional neural network and a cyclic neural network (CNN-RNN), and the real-time prediction of the mixed gas concentration of ethylene, carbon monoxide and methane every 0.5s is realized. In the experimental process, a long-short term memory unit (LSTM) and a Root Mean Square Error (RMSE) are respectively used as a processing unit and a performance evaluation standard of the RNN part of the recurrent neural network. Errors in the predictions of methane, ethane and nitrogen all come below 1%. In addition, the experiment is further expanded into three comparison methods, and the proposed convolutional neural network CNN-lstm is superior to the concentration prediction performance of a structure of the convolutional neural network and a gate cycle control unit (convolutional neural network CNN-gru), an Echo State Network (ESN) and a Linear Regression (LR).
The invention provides a one-dimensional convolution neural network 1D-DCNN mixed gas type identification method under random conditions, which is realized based on a multi-label mode of one-dimensional convolution neural channels and collaterals. The invention can automatically extract the characteristics and classify the extracted characteristics to identify the types of the mixed gas, does not need to separate the characteristic extraction and classification training, and is an end-to-end classifier. The gas type is marked in a multi-label mode, and the dimension space brought by a single-label mode is reduced.
Real-time prediction of the mixed gas concentration and analysis of each component is more challenging than pattern classification of the mixed gas. The invention predicts and analyzes the concentration value of each component in the mixed gas under the condition of random distribution of the concentration value of each component in the mixed gas. Meanwhile, most of previous researches on the component analysis of the mixed gas begin to predict and analyze the concentration value after the gas reaction is complete, and the researches on the real-time prediction of the concentration value of the mixed gas are relatively few, so that the invention mainly researches on the real-time prediction of the concentration of the mixed gas. The mainstream regression prediction analysis method uses more stable characteristic parameters of gas response, such as sensitivity, response time, recovery time and the like, as input, and in the mixed gas analysis of random concentration, the stable characteristics lack applicability and use more characteristics of transient response. Because deep learning obtains good results in multivariate time series prediction, a model is established by using an observed time series to predict a new time series trend, and deep neural network methods such as a circulating neural network (RNN), a model combining the circulating neural network and a dynamic Boltzmann machine, a model combining ARIMA and a multilayer perceptron MLP and the like obtain good results on predicting the time series. Because the response curve of the mixed gas also belongs to the time series, a regression model of the response value and the concentration label can be established by adopting a recurrent neural network when the concentration of the mixed gas is predicted in real time. The concentration of the mixed gas is linear with the response value of the gas at a low level, but the concentration of the mixed gas is non-linear with the response value of the gas. Therefore, for a data set with nonlinearity, the data set can be mapped into a high-dimensional space, and then a regression model is established through a regression method to perform prediction estimation on new data. As can be known from the one-dimensional convolutional neural network, the convolutional neural network can be used for firstly combining the original response data into more meaningful and higher-level features, and then the recurrent neural network RNN is used for establishing a regression prediction model for the output features of the convolutional neural network CNN, so that the real-time prediction of the concentration is realized.
The terminal device provided by the embodiment of the invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor realizes the steps of the above-mentioned method embodiments when executing the computer program. Alternatively, the processor implements the functions of the modules/units in the above device embodiments when executing the computer program.
The computer program may be partitioned into one or more modules/units that are stored in the memory and executed by the processor to implement the invention.
The terminal device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The terminal device may include, but is not limited to, a processor, a memory.
The Processor may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component, etc.
The memory may be used to store the computer programs and/or modules, and the processor may implement various functions of the terminal device by executing or executing the computer programs and/or modules stored in the memory and calling data stored in the memory.
The terminal device integrated modules/units, if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments may be implemented. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like. It should be noted that the computer readable medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable media does not include electrical carrier signals and telecommunications signals as is required by legislation and patent practice.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes will occur to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (9)

1. A mixed gas identification method is characterized by comprising the following steps:
acquiring multi-channel one-dimensional time sequence data of the mixed gas;
performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation, and extracting the overall features; performing convolution and pooling operation on the overall characteristics to obtain a characteristic extraction result;
identifying the type of the mixed gas from the result of the feature extraction by using a multi-label classification method;
and establishing a regression model, and predicting the concentration of the mixed gas in real time by using a convolutional neural network and a cyclic neural network according to the type of the mixed gas and combining the result of the feature extraction.
2. The mixed gas identification method according to claim 1, wherein the acquiring of the multi-channel one-dimensional time-series data of the mixed gas by the gas sensor array comprises:
and acquiring multi-channel one-dimensional time sequence data of the mixed gas through the gas sensor array.
3. The mixed gas identification method according to claim 1 or 2, wherein the one-dimensional convolution operation is specifically as follows:
the one-dimensional convolution equation is as follows:
Figure FDA0003576549330000011
wherein A is an input sequence, and the dimension of the input sequence A is (1, N)a) B is a convolution kernel with dimension of (1, N)b) And C (j) is the convolution result of the j-th neuron, and j is more than or equal to 1 and less than or equal to Na-Nb+ 1; n represents the size of each layer of convolution kernel, the initialization of the weight adopts Xavier initialization, and the uniform distribution of the Xavier initialization is as follows:
Figure FDA0003576549330000012
wherein, W represents uniform distribution, U represents sampling range, and m and n are the number of input units and output units of each layer.
4. The mixed gas identification method according to claim 1, wherein the multi-label classification method specifically comprises:
prediction of target gas ypreThe following were used:
ypre＝sigmoid(WTh+b) (3)
wherein sigmoid () represents an activation function, WTRepresenting the weight of a hidden layer of a second layer, T representing a transposed matrix, h representing a network input, and b representing a bias item;
adopting the output of the sigmoid function as a classification result, using a threshold value of 0.5 as a judgment criterion for the output result, setting a predicted value label with an output value higher than 0.5 as 1, and setting the predicted value label as 0 if the output value is lower than 0.5; the Log loss for the second class of the model is as follows:
Figure FDA0003576549330000021
wherein, ciAs class, y (c)i) Is a true tag, N is the number of classifiers, ypre(ci) And q (c)i) Are respectively class ciThe average value of the Log loss values of the classifier is finally used as the final loss value.
5. The mixed gas identification method according to claim 1, wherein the identifying a kind of the mixed gas includes:
inputting the result obtained by automatic feature extraction into a multilayer sensor, and identifying the type of the mixed gas; wherein the multilayer perceptron consists of full connections.
6. The mixed gas identification method according to claim 1, wherein the predicting the concentration of the mixed gas in real time comprises:
combining original time sequence data, selecting meaningful high-level features, inputting the high-level features output by the convolutional neural network CNN into a recurrent neural network RNN layer, analyzing the high-level features on a time dimension by using the recurrent neural network RNN, and screening out the features which are most suitable for prediction on the time; and splicing the screened feature vectors into a full-connection mode for calculating linear regression, and finally obtaining a predicted concentration value.
7. A mixed gas identification system, comprising:
the data acquisition module is used for acquiring multi-channel one-dimensional time sequence data of the mixed gas;
the feature extraction module is used for performing one-dimensional convolution operation on the multi-channel one-dimensional time sequence data, splicing the features after the one-dimensional convolution operation and extracting the overall features; performing convolution and pooling operation on the overall features to obtain a feature extraction result;
the type identification module is used for identifying the type of the mixed gas from the result of the characteristic extraction by using a multi-label classification method;
and the concentration prediction module is used for establishing a regression model, and predicting the concentration of the mixed gas in real time by utilizing the convolutional neural network and the cyclic neural network according to the type of the mixed gas and combining the result of the characteristic extraction.
8. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1-6 when executing the computer program.
9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 6.
CN202210335225.7A 2022-03-31 2022-03-31 Mixed gas identification method, system, terminal equipment and readable storage medium Pending CN114707598A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210335225.7A CN114707598A (en) 2022-03-31 2022-03-31 Mixed gas identification method, system, terminal equipment and readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210335225.7A CN114707598A (en) 2022-03-31 2022-03-31 Mixed gas identification method, system, terminal equipment and readable storage medium

Publications (1)

Publication Number Publication Date
CN114707598A true CN114707598A (en) 2022-07-05

Family

ID=82171608

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210335225.7A Pending CN114707598A (en) 2022-03-31 2022-03-31 Mixed gas identification method, system, terminal equipment and readable storage medium

Country Status (1)

Country Link
CN (1) CN114707598A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114997251A (en) * 2022-08-02 2022-09-02 中国农业科学院农业信息研究所 Mixed gas identification method and system based on gas sensor array
CN115618927A (en) * 2022-11-17 2023-01-17 中国人民解放军陆军防化学院 Gas type identification method based on time sequence-graph fusion neural network
CN115860056A (en) * 2023-02-17 2023-03-28 电子科技大学 Sensor array neural network method for mixed gas concentration prediction
CN116759014A (en) * 2023-08-21 2023-09-15 启思半导体(杭州)有限责任公司 Random forest-based gas type and concentration prediction method, system and device
CN117216659A (en) * 2023-09-12 2023-12-12 暨南大学 Atmospheric particulate source analysis method and system based on single-particle aerosol mass spectrum

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114997251A (en) * 2022-08-02 2022-09-02 中国农业科学院农业信息研究所 Mixed gas identification method and system based on gas sensor array
CN114997251B (en) * 2022-08-02 2022-11-08 中国农业科学院农业信息研究所 Mixed gas identification method and system based on gas sensor array
CN115618927A (en) * 2022-11-17 2023-01-17 中国人民解放军陆军防化学院 Gas type identification method based on time sequence-graph fusion neural network
CN115860056A (en) * 2023-02-17 2023-03-28 电子科技大学 Sensor array neural network method for mixed gas concentration prediction
CN115860056B (en) * 2023-02-17 2023-05-26 电子科技大学 Sensor array neural network method for mixed gas concentration prediction
CN116759014A (en) * 2023-08-21 2023-09-15 启思半导体(杭州)有限责任公司 Random forest-based gas type and concentration prediction method, system and device
CN116759014B (en) * 2023-08-21 2023-11-03 启思半导体(杭州)有限责任公司 Random forest-based gas type and concentration prediction method, system and device
CN117216659A (en) * 2023-09-12 2023-12-12 暨南大学 Atmospheric particulate source analysis method and system based on single-particle aerosol mass spectrum

Similar Documents

Publication Publication Date Title
CN114707598A (en) Mixed gas identification method, system, terminal equipment and readable storage medium
Al‐gaashani et al. Tomato leaf disease classification by exploiting transfer learning and feature concatenation
Wang et al. Time series classification from scratch with deep neural networks: A strong baseline
KOKLU et al. The classification of eye state by using kNN and MLP classification models according to the EEG signals
Dharshana et al. A Novel Approach for Detection and Classification of Fish Species
Rajasekar et al. Performance analysis of soft computing techniques for the automatic classification of fruits dataset
Li et al. SeedSortNet: a rapid and highly effificient lightweight CNN based on visual attention for seed sorting
Banerjee et al. A regularized cnn-svm framework for improved diagnosis of citrus fruit diseases, both common and exotic
Joy et al. A comprehensive study on the performance of different Multi-class Classification Algorithms and Hyperparameter Tuning Techniques using Optuna
Albu An approach for predicting protein-protein interactions using supervised autoencoders
Meng et al. Biological image temporal stage classification via multi-layer model collaboration
Zhang et al. Accurate Inference of Unseen Combinations of Multiple Rootcauses with Classifier Ensemble
Toda et al. Learning from synthetic dataset for crop seed instance segmentation
CN109444360B (en) Fruit juice storage period detection algorithm based on cellular neural network and electronic nose feature extraction
Ren et al. Video-based emotion recognition using multi-dichotomy RNN-DNN
Dani et al. Survey on the use of CNN and Deep Learning in Image Classification
Mamatha et al. Visual Sentiment Classification of Restaurant Review Images using Deep Convolutional Neural Networks
Kang et al. Object recognition under occlusion revisited: elucidating algorithmic advantages of recurrent computation
Kavitha et al. Deep Learning and Internet of Things based Detection of Diseases and Prediction of Pesticides in Fruits
Koloski et al. Latent graph powered semi-supervised learning on biomedical tabular data
Marinoni et al. Enhancing ensemble learning and transfer learning in multimodal data analysis by adaptive dimensionality reduction
Ramadhani et al. Performance Comparison of Convolutional Neural Network and MobileNetV2 for Chili Diseases Classification
Huang et al. Image feature selection based on orthogonal ℓ2, 0 norms
Yang et al. Prediction effect of ADMET properties based on machine learning
AABIDI et al. A NEW APPROACH OF LEAF DISEASE DETECTION USING BAG OF VISUAL WORDS

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