Mixed gas concentration detection method based on temperature modulation technology

Abstract

The invention discloses a mixed gas concentration detection method based on a temperature modulation technology, and relates to the field of semiconductor gas sensing. The problem that the concentration of the mixed gas is difficult to analyze more accurately in the prior art is solved. According to the method, the characteristics of the gas sensitive response data are expanded under the scale of the spatial domain through the temperature modulation technology, the time sequence data are processed by adopting the LSTM neural network model, the characteristic information contained in the sequence data is fully mined, and the recognition precision of the concentration of the mixed gas is remarkably improved through the combination of the temperature modulation technology and the LSTM neural network model. The device and the method are mainly used for detecting the concentration of H2 and CO gas released from the transformer oil in case of failure.

Description

Technical Field

[0001] The invention relates to the field of semiconductor gas sensing. Background Art

[0002] As a hub equipment for power transportation and distribution, the reliable operation of transformers is the key to the safety and stability of power systems. At present, dissolved gas analysis (DGA) in oil is considered to be the most effective method for transformer fault diagnosis. Monitoring the decomposition gas of transformer oil through a gas detection system helps to realize online monitoring of the operating status of the transformer. When a transformer fails, CO and H2 gases will be released successively from the transformer oil. These two gases can be regarded as two typical characteristic gases dissolved in transformer oil. Therefore, real-time detection of CO and H2 gases is of great significance to the safe and stable operation of transformer equipment. At present, gas chromatography is usually used to separate the dissolved gas in transformer oil for detection. The basic steps of dissolved gas extraction include sampling, purification, and degassing and separation of dissolved gas in oil. Subsequently, the gas chromatograph separates the mixture into single components, and then these components are sent to infrared spectroscopy or gas sensors for concentration detection. However, since CO and H2 gases move at similar speeds in the chromatographic column, the peak times of the two gases after passing through the chromatographic column are relatively close, resulting in a certain overlap of the signal peaks detected by the sensor, so there will be a certain error in actual detection.

[0003] In recent years, artificial olfactory systems composed of metal oxide (MOS) gas sensor arrays and artificial intelligence algorithms have shown significant advantages in the field of real-time gas detection. MOS gas sensors have the characteristics of fast detection speed, simple operation and low cost. Artificial intelligence algorithms have strong feature extraction and recognition capabilities, can spontaneously extract feature information from data, and complete the update of model parameters through self-learning, and have strong generalization capabilities. At present, artificial olfactory systems have played an important role in industrial pollution monitoring, food safety, and medical fields. However, many current studies on the prediction of mixed gases only focus on the prediction of species, while ignoring the key indicator of concentration, making it difficult to conduct more accurate analysis. Therefore, developing a method for predicting the concentration of mixed gases has important application value. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that it is difficult to conduct more accurate analysis on the concentration prediction of mixed gas at present. The present invention provides a mixed gas concentration detection method based on temperature modulation technology.

[0005] A mixed gas concentration detection method based on temperature modulation technology, the method comprising the following steps:

[0006] Gas preparation: prepare mixed gases of various concentrations, and the mixed gases are composed of H2 and CO;

[0007] Data acquisition: The gas-sensitive response data of the mixed gas collected by the sensor array under different temperature modulation modes in each working cycle is used as a training sample, wherein the sensor array is prepared by a plurality of metal oxide gas sensors of different types arranged in an array; the training sample is a data matrix composed of the gas-sensitive response data collected by four metal oxide gas sensors under each temperature modulation mode in each working cycle;

[0008] Data processing: pre-processing of training samples and labeling of gas types and concentrations;

[0009] Model training: Input the training samples into the LSTM neural network model to predict the concentration of each type of gas in the mixed gas and realize the training of the LSTM neural network model;

[0010] Inference stage: Use the trained LSTM neural network model to predict the concentration of mixed gas.

[0011] Preferably, each metal oxide gas sensor comprises a substrate and a gas sensing portion; the gas sensing portion is formed by mixing a doped metal target material and a SnO2 target material;

[0012] The substrate is an Al2O3 ceramic sheet, and positive and negative gold electrodes are printed on the top surface of the substrate, and positive and negative heating electrodes are printed on the bottom surface. The sensitive part is connected to the positive and negative gold electrodes, and the positive and negative gold electrodes serve as data output ends of the metal oxide gas sensor, and the positive and negative heating electrodes serve as temperature modulation ends of the metal oxide gas sensor.

[0013] Preferably, the heating electrode is implemented using ruthenium oxide.

[0014] Preferably, the sensor array includes four metal oxide gas sensors, and the metals doped into the SnO2 target materials of the four metal oxide gas sensors are In, Ni, Pd and Co respectively.

[0015] Preferably, the LSTM neural network model includes 1 input layer, 2 LSTM layers, 1 fully connected layer and 1 output layer arranged in sequence;

[0016] The received gas-sensitive response data is sent to two LSTM layers in sequence through the input layer for feature advance, and then the extracted features are integrated through the fully connected layer and output through the output layer.

[0017] Preferably, the fully connected layer is formed by a combination of a linear transformation and an activation function.

[0018] Preferably, the output layer is implemented using a linear regression activation function.

[0019] Preferably, the preprocessing of the training samples is implemented by: performing smoothing and normalization processing in sequence.

[0020] A mixed gas concentration detection device based on temperature modulation technology includes a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement the mixed gas concentration detection method based on temperature modulation technology.

[0021] A computer-readable storage device stores a computer program, and when the computer program is executed, the mixed gas concentration detection method based on the temperature modulation technology is implemented.

[0022] Advantages of the present invention:

[0023] The present invention provides a mixed gas concentration detection method based on temperature modulation technology. Compared with the existing mixed gas concentration prediction method, the present invention significantly improves the recognition accuracy of the mixed gas concentration by expanding the characteristics of the gas-sensitive response data at the scale of the spatial domain through temperature modulation technology. At the same time, the LSTM neural network model is used to process the time series data to fully mine the characteristic information contained in the sequence data. The recognition method has the advantages of high recognition accuracy, high prediction accuracy, etc. Through the method of the present invention, the gas concentration prediction task under complex background can be completed, providing an effective method for detecting the released gas in transformer oil.

[0024] The gas concentration detection method of the present invention has the advantages of being fast, efficient, and convenient, and is of great significance to the development of the gas identification field. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the mixed gas concentration detection method based on temperature modulation technology of the present invention;

[0026] Figure 2 The scanning electron microscope images of SnO2 target materials doped with four different metals are shown below. Figure 2 a is an electron microscope image of In-doped SnO2. Figure 2 b is the electron microscope image of Ni-doped SnO2. Figure 2 c is the electron microscope image of Pd-doped SnO2. Figure 2 d is an electron microscope image of Co-doped SnO2; and 100 nm in the figure represents the scale size;

[0027] Figure 3 It is a temperature variation curve diagram of the sensor array in the temperature modulation mode;

[0028] Figure 4This is a comparison chart of the predicted concentration and actual concentration of CO gas in the mixed gas;

[0029] Figure 5 This is a comparison chart of the predicted concentration and actual concentration of H2 gas in the mixed gas.

[0030] Figure 6 This is a graph of the gas-sensitive response data obtained by detecting CO and H2 mixed gases using four metal oxide gas sensors in temperature modulation mode. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0033] Specific implementation method 1. Figure 1 The present embodiment describes a mixed gas concentration detection method based on temperature modulation technology, the method comprising the following steps:

[0034] Gas preparation: prepare mixed gases of various concentrations, and the mixed gases are composed of H2 and CO;

[0035] Data acquisition: The gas-sensitive response data of the mixed gas collected by the sensor array under different temperature modulation modes in each working cycle is used as a training sample, wherein the sensor array is prepared by a plurality of metal oxide gas sensors of different types arranged in an array; the training sample is a data matrix composed of the data collected by four metal oxide gas sensors under each temperature modulation mode in each working cycle;

[0036] Data processing: pre-processing of training samples and labeling of gas types and concentrations;

[0037] Model training: Input the training samples into the LSTM neural network model to predict the concentration of each type of gas in the mixed gas and realize the training of the LSTM neural network model;

[0038] Inference stage: Use the trained LSTM neural network model to predict the concentration of mixed gas.

[0039] In this embodiment, temperature modulation technology is used to expand the feature dimension of gas-sensitive response data. Each training sample uses a working cycle of a sensor array as the data capture period, and each training sample includes data output by four metal oxide gas sensors under each temperature modulation. The data collected under each temperature modulation mode within the working cycle is used as a feature dimension. At the same time, a long short-term memory neural network model (LSTM neural network model) is used to extract the features of the gas-sensitive response data to fully mine the feature information contained in the sequence data, thereby improving the prediction accuracy of the mixed gas concentration.

[0040] Figure 3 The temperature modulation of the sensor array is shown in Figure 5 for a duty cycle of 500 seconds.

[0041] Furthermore, each metal oxide gas sensor includes a substrate and a gas sensing portion;

[0042] The gas sensing part is prepared by mixing the doped metal target and the SnO2 target;

[0043] The substrate is an Al2O3 ceramic sheet, and positive and negative gold electrodes are printed on the top surface of the substrate, and positive and negative heating electrodes are printed on the bottom surface. The sensitive part is connected to the positive and negative gold electrodes, and the positive and negative gold electrodes serve as data output ends of the metal oxide gas sensor, and the positive and negative heating electrodes serve as temperature modulation ends of the metal oxide gas sensor.

[0044] The preferred embodiment provides a specific structure of a metal oxide gas sensor, which is simple to prepare and easy to implement.

[0045] A further heating electrode is realized using ruthenium oxide.

[0046] Furthermore, the sensor array includes four metal oxide gas sensors, and the metals doped in the SnO2 target materials of the four metal oxide gas sensors are In, Ni, Pd and Co respectively. Figure 2 , scanning electron microscope images of four metal-doped SnO2 target materials are given. From the images, it can be seen that the morphology of the four thin film materials is basically the same, they are uniform thin films, and the doping elements have no obvious effect on the morphology of the film.

[0047] See also Figure 6 The response value curve of the gas-sensitive response data collected by the metal oxide gas sensor with doped metals being In, Ni, Pd and Co is given. The gas-sensitive response data is the resistance change of the metal oxide gas sensor.

[0048] Furthermore, the LSTM neural network model includes 1 input layer, 2 LSTM layers, 1 fully connected layer and 1 output layer which are arranged in sequence;

[0049] The received gas-sensitive response data is sent to two LSTM layers in sequence through the input layer for feature advance, and then the extracted features are integrated through the fully connected layer and output through the output layer.

[0050] Specifically, the fully connected layer is composed of a combination of linear transformation and activation function; the output layer is implemented using a linear regression activation function, and the gas concentration is predicted by establishing a mapping relationship between the input data and the concentration.

[0051] LSTM controls the flow of information through a special "gate" mechanism (forget gate, input gate, output gate), thereby solving the problem of gradient vanishing or gradient exploding when traditional RNN processes long sequences, and can more effectively capture long-term dependencies.

[0052] Furthermore, the preprocessing of the training samples is implemented by: smoothing and normalizing in sequence. Specifically, the data is scaled to the range of [0, 1] by the Min-Max normalization method to eliminate the error caused by the dimension of the data.

[0053] Specific implementation method 2: A mixed gas concentration detection device based on temperature modulation technology includes a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement the mixed gas concentration detection method based on temperature modulation technology.

[0054] Specific implementation method 3: A computer-readable storage device stores a computer program, and when the computer program is executed, the mixed gas concentration detection method based on temperature modulation technology is implemented.

[0055] Verification test:

[0056] 1. Preparation of sensor array. Magnetron sputtering technology was used to co-sputter multiple targets to prepare doped metal oxide gas-sensitive materials. The base material target used was SnO2 target, and the doped metal targets included In, Ni, Pd, and Co targets. The sputtering power of SnO2 target was 100W, the sputtering power of doped metal target was 10W, the sputtering pressure was 3.4Pa, the sputtering time was 20min, the sputtering temperature was 100℃, the Ar / O2 ratio was 60 / 30, and the bias voltage was 50V. A sensor array with four metal oxide gas sensors was obtained.

[0057] 2. Use the prepared sensor array to collect the gas-sensing properties of the mixed gas formed by CO and H2;

[0058] 3. Perform pre-processing such as smoothing and normalization on the collected gas-sensitive response data, assign type and concentration labels, and divide them into training sets and test sets;

[0059] 4. The training set is used to train the LSTM neural network model, and the output is the concentration value of each type of gas; the data in the test set is input into the trained LSTM neural network model to predict the H2 gas and CO gas concentrations.

[0060] See also Figure 4 and Figure 5 The comparison charts between the predicted results and the actual concentrations of CO gas and H2 gas concentrations are given respectively; Figure 4 and Figure 5 Most of the predicted concentrations are the same as the actual concentrations, and the predicted concentrations of the different parts are also infinitely close to the actual concentrations, which proves that the prediction accuracy of the present invention is high and the prediction method of the present invention is effective.

[0061] Compared with the prior art, the positive effects of the present invention are: the temperature modulation technology is used to expand the characteristic dimension of gas-sensitive data, enrich the characteristics of gas-sensitive performance, and thus improve the prediction accuracy of mixed gases. The LSTM neural network model is used to process time series data, which fully mines the characteristic information contained in the sequence data, and further improves the accuracy of the prediction method.

[0062] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A mixed gas concentration detection method based on temperature modulation technology, characterized in that: The method comprises the following steps: Gas preparation: prepare mixed gases of various concentrations, and the mixed gases are composed of H2 and CO; Data acquisition: The gas-sensitive response data of the mixed gas collected by the sensor array under different temperature modulation modes in each working cycle is used as a training sample, wherein the sensor array is prepared by a plurality of metal oxide gas sensors of different types arranged in an array; the training sample is a data matrix composed of the gas-sensitive response data collected by four metal oxide gas sensors under each temperature modulation mode in each working cycle; Data processing: pre-processing of training samples and labeling of gas types and concentrations; Model training: Input the training samples into the LSTM neural network model to predict the concentration of each type of gas in the mixed gas and realize the training of the LSTM neural network model; Inference stage: Use the trained LSTM neural network model to predict the concentration of mixed gas.

2. The mixed gas concentration detection method based on temperature modulation technology according to claim 1 is characterized in that: Each metal oxide gas sensor includes a substrate and a gas-sensitive portion; the gas-sensitive portion is prepared by mixing a doped metal target material and a SnO2 target material; The substrate is an Al2O3 ceramic sheet, and positive and negative gold electrodes are printed on the top surface of the substrate, and positive and negative heating electrodes are printed on the bottom surface. The sensitive part is connected to the positive and negative gold electrodes, and the positive and negative gold electrodes serve as data output ends of the metal oxide gas sensor, and the positive and negative heating electrodes serve as temperature modulation ends of the metal oxide gas sensor.

3. The mixed gas concentration detection method based on temperature modulation technology according to claim 2 is characterized in that: The heating electrode is made of ruthenium oxide.

4. The mixed gas concentration detection method based on temperature modulation technology according to claim 2 is characterized in that: The sensor array includes four metal oxide gas sensors, and the metals doped into the SnO2 target materials of the four metal oxide gas sensors are In, Ni, Pd and Co respectively.

5. The mixed gas concentration detection method based on temperature modulation technology according to claim 1 is characterized in that: The LSTM neural network model includes 1 input layer, 2 LSTM layers, 1 fully connected layer and 1 output layer, which are set in sequence; The received gas-sensitive response data is sent to two LSTM layers in sequence through the input layer for feature advance, and then the extracted features are integrated through the fully connected layer and output through the output layer.

6. The mixed gas concentration detection method based on temperature modulation technology according to claim 5 is characterized in that: The fully connected layer is composed of a combination of linear transformation and activation function.

7. The mixed gas concentration detection method based on temperature modulation technology according to claim 5 is characterized in that: The output layer is implemented using a linear regression activation function.

8. The mixed gas concentration detection method based on temperature modulation technology according to claim 1 is characterized in that: The implementation method of preprocessing the training samples is: smoothing and normalization processing in sequence.

9. A mixed gas concentration detection device based on temperature modulation technology, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that: The processor executes a computer program to implement the mixed gas concentration detection method based on temperature modulation technology as claimed in any one of claims 1 to 8.

10. A computer-readable storage device storing a computer program, characterized in that: When the computer program is executed, the mixed gas concentration detection method based on temperature modulation technology as claimed in any one of claims 1 to 8 is implemented.

Images