Method and device for predicting concentration of mixed gas decomposition product based on feature fusion
Through feature fusion technology, long and short-term memory networks and convolutional neural networks are used to extract the characteristics of infrared spectral data, the problem of low prediction accuracy of SF6/N2 mixed gas decomposition product concentration in the prior art is solved, achieving higher prediction accuracy and better equipment maintenance support.
Patent Information
- Application Number
- CN202411731532.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-04-22
AI Technical Summary
In the prior art, the concentration prediction accuracy of SF6/N2 mixed gas decomposition products is low, which affects the early warning of GIS equipment and increases the maintenance cost and difficulty of the equipment.
Using a feature fusion method, by obtaining the one-dimensional infrared spectral data of the mixed gas, performing denoising processing, the characteristics of the one-dimensional infrared spectral data are extracted using a long and short-term memory network, and the data is converted into a two-dimensional spectral time-frequency map through wavelet transformation. The characteristics of the two-dimensional spectral time-frequency map are extracted using a convolutional neural network, and finally the two types of features are fused, and the concentration prediction values of the decomposed products SO2, NO and CS2 are output through the concentration prediction module.
Through feature fusion, the accuracy of the concentration prediction of SF6/N2 mixed gas decomposition product is improved, and the internal insulation status of high-voltage electrical equipment is effectively evaluated, providing data support for equipment maintenance strategies.
Smart Images
Figure CN119864098A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric power equipment, and in particular to a method and device for predicting the concentration of mixed gas decomposition products based on feature fusion, and a computer-readable storage medium. Background Art
[0002] With the increasing attention to environmental issues and the proposal of the "dual carbon" goal, the power industry is gradually seeking environmentally friendly gases to replace traditional SF6 gas as the insulating medium for high-voltage equipment. SF6 / N2 mixed gas has become a promising alternative to SF6 due to its low global warming potential and good insulation performance. However, SF6 / N2 mixed gas still faces the problem of partial discharge in high-voltage electrical equipment, especially in gas insulated switchgear (GIS). The partial discharge phenomenon may cause SF6 and N2 to decompose, producing decomposition products such as SO2, NO, and CS2. The presence and concentration level of these decomposition products are directly related to the insulation status and health of the equipment, and the concentration of the decomposition products needs to be predicted.
[0003] Fourier Transform Infrared Spectroscopy (FTIR) is widely used in quantitative analysis in many fields due to its advantages such as fast detection speed, high sensitivity, strong anti-interference ability and non-contact. In the field of GIS insulation fault diagnosis, FTIR technology has also been introduced for the detection of SF6 decomposition products.
[0004] However, traditional concentration prediction methods for SF6 decomposition products mostly use linear models, which have limitations when dealing with overlapping absorption peaks and nonlinear relationships of decomposition products. In addition, most existing methods only use one-dimensional spectral data and fail to fully mine the information in high-dimensional spectral data, resulting in low concentration prediction accuracy of SF6 / N2 mixed gas decomposition products, which affects the early warning of GIS equipment and increases the maintenance cost and difficulty of equipment. Summary of the invention
[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem of low accuracy in the concentration prediction of decomposition products of SF6 / N2 mixed gas in the prior art.
[0006] In order to solve the above technical problems, the present invention provides a method for predicting the concentration of mixed gas decomposition products based on feature fusion, comprising:
[0007] Obtain one-dimensional infrared spectrum data of mixed gas;
[0008] Denoising one-dimensional infrared spectral data;
[0009] Inputting the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer, and outputting a first spectrum feature;
[0010] The denoised one-dimensional infrared spectrum data is transformed by wavelet transform to obtain a two-dimensional spectrum time-frequency diagram, which is input into a second feature extraction network including a convolution layer to output a second spectrum feature;
[0011] The first spectral feature and the second spectral feature are input into the concentration prediction module for splicing, and the concentration prediction values of the decomposition products SO2, NO and CS2 are output.
[0012] Preferably, the method for obtaining the one-dimensional infrared spectrum data of the mixed gas is: using a standardized spectrum measurement operation to collect the mixed gas at 1700 cm -1 ~2700cm -1 One-dimensional infrared spectral data in the wavenumber range.
[0013] Preferably, the one-dimensional infrared spectrum data denoising includes background interference denoising and random noise suppression.
[0014] Preferably, the first feature extraction network includes a sequence input layer, a first long short-term memory layer, a first dropout layer, a second long short-term memory layer, a second dropout layer and a fully connected layer.
[0015] Preferably, the second feature extraction network includes an image input layer, a first convolutional layer, a second convolutional layer and a fully connected layer.
[0016] Preferably, the concentration prediction module comprises a concatenation layer, a first fully connected layer, a hyperbolic tangent function activation layer, a second fully connected layer, a leaky rectified linear unit activation layer and a third fully connected layer.
[0017] Preferably, the process of constructing a spectral database to train the first feature extraction network, the second feature extraction network and the concentration prediction module includes:
[0018] A dynamic gas distributor is used to configure mixed gases of different concentration ratios; the mixed gases include SF6 / N2 mixed gas, SO2 gas, NO gas and CS2 gas, wherein the SF6 / N2 mixed gas is the background gas, and the SO2 gas, NO gas and CS2 gas are the gases to be tested;
[0019] Standardized spectral measurement operations are used to collect spectral data of mixed gases with different concentration ratios within a preset wavenumber range to build a spectral database.
[0020] Preferably, a dynamic gas distributor is used to configure mixed gases of different concentration ratios, and the formula is:
[0021] Samplesn ~N(2 n ,(n+0.5) 2 ),n=0,1,2,…,7
[0022] Among them, Samples n Indicates that the random concentrations of the three gases to be tested follow a mean of 2 n , a normal distribution with a standard deviation of n+0.5.
[0023] The present invention also provides a concentration prediction device for mixed gas decomposition products based on feature fusion, comprising:
[0024] Spectral data acquisition module, used to obtain one-dimensional infrared spectrum data of mixed gas;
[0025] Denoising module, used to denoise one-dimensional infrared spectrum data;
[0026] A feature extraction module is used to input the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer, and output a first spectrum feature; obtain a two-dimensional spectrum time-frequency graph by wavelet transforming the denoised one-dimensional infrared spectrum data, and input it into a second feature extraction network including a convolution layer, and output a second spectrum feature;
[0027] The prediction module is used to input the first spectral feature and the second spectral feature into the concentration prediction module for splicing, and output the concentration prediction values of the decomposition products SO2, NO and CS2.
[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for predicting the concentration of mixed gas decomposition products based on feature fusion are implemented.
[0029] The above technical solution of the present invention has the following beneficial effects compared with the prior art:
[0030] The method for predicting the concentration of mixed gas decomposition products based on feature fusion described in the present invention first obtains one-dimensional infrared spectrum data of the mixed gas and performs denoising, then uses a long short-term memory network to extract the features of the one-dimensional infrared spectrum data to obtain a first spectrum feature; uses wavelet transform to convert the one-dimensional infrared spectrum data into a two-dimensional spectrum time-frequency graph, and extracts the features of the two-dimensional spectrum time-frequency graph through a convolutional neural network to obtain a second spectrum feature. Finally, the first spectrum feature and the second spectrum feature are fused, and the concentration prediction values of three gases, SO2, NO and CS2, are output through a concentration prediction module network. The present invention combines the features of one-dimensional spectrum data and the feature extraction of two-dimensional spectrum data, and can obtain richer feature information in two dimensions. On the one hand, the static information of the spectrum features is retained, and on the other hand, the dynamic changes in the time domain-frequency domain are captured. Finally, the accuracy of the concentration prediction of the decomposition products of the SF6 / N2 mixed gas is improved through feature fusion, and then the internal insulation state of high-voltage electrical equipment can be effectively evaluated, providing data support for equipment maintenance strategies. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:
[0032] Figure 1 It is a flow chart of a concentration prediction method of mixed gas decomposition products based on feature fusion of the present invention;
[0033] Figure 2 The gas mixture is at 1700cm -1 ~2700cm -1 Absorption spectra over a range of wavenumbers;
[0034] Figure 3 It is a schematic diagram of the structure of the LSTM unit;
[0035] Figure 4 It is a schematic diagram of the structure of the concentration prediction module;
[0036] Figure 5 is a schematic diagram of the spectral data collection device. DETAILED DESCRIPTION
[0037] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0038] Embodiment 1
[0039] Reference Figure 1 As shown, the present invention provides a method for predicting the concentration of mixed gas decomposition products based on feature fusion, comprising:
[0040] S1. Obtain one-dimensional infrared spectrum data of the mixed gas.
[0041] Preferably, the method for obtaining the one-dimensional infrared spectrum data of the mixed gas is: using a standardized spectrum measurement operation to collect the mixed gas at 1700 cm -1 ~2700cm -1 One-dimensional infrared spectrum data in the wavenumber range. Figure 2 As shown in the figure, the decomposition products SO2, NO and CS2 all have obvious absorption peaks in this wave number range, while the background gases SF6 and N2 do not. Therefore, the spectral data in this band has a high signal-to-noise ratio.
[0042] S2. De-noising one-dimensional infrared spectrum data.
[0043] Methods for denoising one-dimensional infrared spectral data include but are not limited to background interference denoising and random noise suppression.
[0044] S3, inputting the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer (Long Short-Term Memory, LSTM), and outputting a first spectrum feature;
[0045] The denoised one-dimensional infrared spectrum data is transformed by wavelet transform to obtain a two-dimensional spectrum time-frequency diagram, which is input into a second feature extraction network including a convolution layer to output a second spectrum feature.
[0046] Fully extracting effective features from infrared spectral data is the key to ensuring the accuracy of gas concentration prediction. The present invention adopts a deep feature extraction and fusion network. The first branch directly takes the denoised one-dimensional infrared spectral data as input, and passes through a first feature extraction network including a long short-term memory layer to obtain a first spectral feature containing contextual correlation information in the one-dimensional data; the second branch uses wavelet transform to map the denoised one-dimensional infrared spectral data to a two-dimensional spectral time-frequency spectrum diagram, and then passes through a second feature extraction network including a convolutional layer to extract cross-scale features and correlation pattern features in the time-frequency domain in the two-dimensional spectral time-frequency spectrum diagram to obtain a second spectral feature.
[0047] The first feature extraction network includes a sequence input layer, a first long short-term memory layer, a first dropout layer, a second long short-term memory layer, a second dropout layer and a fully connected layer. Specific parameter settings are shown in Table 1.
[0048] Table 1. First characteristic network parameter settings
[0049]
[0050]
[0051] The Recurrent Neural Network (RNN) model is designed to process sequence data. The key feature of RNN is that it achieves memory by using neurons with self-feedback connections. represents the input of the network at time t, represents the state of the hidden layer, then h t It is not only related to the input at time t, but also to the hidden layer state h at time t-1. t-1 The main problem with RNNs is that they have difficulty modeling dependencies between states at long intervals, the so-called long-term dependency problem, due to the exploding or vanishing gradient problem.
[0052] A reasonable solution is to introduce a gating mechanism to control the speed of information accumulation, including selectively adding new information and forgetting previously accumulated information. LSTM is a gated RNN. The LSTM layer includes several LSTM units, each of which uses an internal state and three gates:
[0053] Internal State Information is transferred linearly between internal states and to hidden states Nonlinear output information;
[0054] Forget Gate t ∈[0,1] D : Control the internal state c t-1 How much information needs to be forgotten?
[0055] Input Gate i t ∈[0,1] D : Control temporary internal state How much information needs to be retained?
[0056] Output gate o t ∈[0,1] D : Control the internal state c t In the hidden state h t How much information to output.
[0057] The structure diagram of the standard LSTM unit is as follows Figure 3 As shown, the calculation formula is as follows:
[0058] i t =σ(W i x t +U i h t-1 +b i )
[0059] f t =σ(Wf x t +U f h t-1 +b f )
[0060] o t =σ(W o x t +U o h t-1 +b o )
[0061]
[0062] h t =o t ⊙tanh(c t )
[0063] Among them, W i , W f , W o , W c , U i , U f , U o , U c 、b i 、b f 、b o and b c are the parameters of the LSTM network, ⊙ represents point-by-point multiplication, σ represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function.
[0064] Compared with one-dimensional infrared spectral data, high-dimensional spectrograms often contain richer spatiotemporal information. By performing time-frequency domain analysis on spectral data, the changes in the spectral characteristics of the signal over time can be captured from a dynamic perspective. The time-frequency diagram can reveal the time-varying characteristics of the gas concentration change process, especially in the case of mixed gases. These time-frequency characteristics help to distinguish the dynamic changes of different gases.
[0065] Therefore, the present invention utilizes wavelet transform to convert the denoised one-dimensional infrared spectrum data into a two-dimensional spectrum time-frequency diagram.
[0066] The two-dimensional spectral time-frequency graph is unified into a grayscale image with a size of 64*64, and then a convolutional neural network is used as the second feature extraction network to extract the cross-scale features of the two-dimensional spectral time-frequency graph and the correlation pattern features in the time-frequency domain, and output the second spectral features.
[0067] The denoised one-dimensional infrared spectrum data is transformed by wavelet transform to obtain a two-dimensional spectrum time-frequency diagram, which is input into a second feature extraction network including a convolution layer to output a second spectrum feature.
[0068] The second feature extraction network includes an image input layer, a first convolutional layer, a second convolutional layer, and a fully connected layer.
[0069] The grayscale image of size 64*64 is input into the second feature extraction network, and after two convolutional layers, the fully connected layer outputs the second spectral feature of size 20*1. The network parameters of the second feature extraction network are shown in Table 2.
[0070] Table 2 Second feature extraction network parameter settings
[0071]
[0072] S4. Input the first spectral feature and the second spectral feature into a concentration prediction module, and output the concentration prediction values of the decomposition products SO2, NO and CS2.
[0073] Reference Figure 4 As shown, the concentration prediction module fuses the two extracted features to perform quantitative analysis of several characteristic gases, including a splicing layer, a first fully connected layer, a hyperbolic tangent function activation layer, a second fully connected layer, a leakage corrected linear unit activation layer, and a third fully connected layer. The first spectral feature and the second spectral feature are input into the concentration prediction module.
[0074] Combining spectral data with time-frequency plots allows the model to better learn complex, nonlinear, time-varying patterns.
[0075] The concentration prediction module connects the first spectral feature and the second spectral feature in the concatenation layer, and then passes through three fully connected layers and two activation layers to finally output the concentration prediction values of the decomposition products SO2, NO and CS2. The parameter settings of the concentration prediction module are shown in Table 3.
[0076] Table 3 Concentration prediction module parameter settings
[0077]
[0078] Preferably, this embodiment constructs a spectral database to train the first feature extraction network, the second feature extraction network and the concentration prediction module.
[0079] In order to collect one-dimensional infrared spectrum data of mixed gases at different concentration ratios for training the calibration model, the following Figure 5 The spectral data collection device shown mainly includes: a gas tank (including SF6 / N2 mixed gas, SO2 gas, NO gas, CS2 gas), a dynamic gas distributor, an infrared light source, a gas pool, an infrared spectrometer, and a data storage / display PC. Among them, the SF6 / N2 mixed gas is used as the background gas, and the SO2 gas, NO gas, and CS2 gas are the gases to be tested.
[0080] The process of building a spectral database includes:
[0081] The dynamic gas distributor is used to configure mixed gases with different concentration ratios. The formula is:
[0082] Samples n ~N(2 n ,(n+0.5) 2 ),n=0,1,2,…,7
[0083] Among them, Samples n Indicates that the random concentrations of the three gases to be tested follow a mean of 2 n , with a normal distribution with a standard deviation of n+0.5. For example, when n is 0, the prepared concentration follows N(1,0.5 2 ); when n is 7, the preparation concentration follows N(128,7.5 2 ).
[0084] Obviously, the above formula can roughly cover the gas preparation concentration range of 0 to 150 ppm, and the lower the concentration, the denser the concentration levels that can be obtained, which is more conducive to obtaining more spectral data of mixed gases at low concentration ratios.
[0085] Standardized spectral measurement operation was used to collect mixed gas at different concentration ratios at 1700 cm -1 ~2700cm -1 Spectral data within a certain wavenumber range are used to construct a spectral database.
[0086] The total number of samples in the spectral database constructed in this embodiment is 220, which is roughly distributed according to the rule that the lower the concentration, the more samples there are. The final spectral database is shown in Table 4. For example, when n = 0, 50 groups of SO2, NO and CS2 mixed gases are randomly configured, and the concentration of each gas in each group of mixed gases follows N (1, 0.5 2 )distributed.
[0087] Table 4 Example of structured spectral data
[0088]
[0089] In summary, the concentration prediction method of the mixed gas decomposition product based on feature fusion described in the present invention first obtains the one-dimensional infrared spectrum data of the mixed gas and performs denoising, then uses the long short-term memory network to extract the features of the one-dimensional infrared spectrum data to obtain the first spectrum feature; uses wavelet transform to convert the one-dimensional infrared spectrum data into a two-dimensional spectrum time-frequency graph, and extracts the features of the two-dimensional spectrum time-frequency graph through a convolutional neural network to obtain the second spectrum feature. Finally, the first spectrum feature and the second spectrum feature are fused, and the concentration prediction values of the three gases SO2, NO and CS2 are output through the concentration prediction module network. The present invention combines the features of one-dimensional spectrum data and the feature extraction of two-dimensional spectrum data, and can obtain richer feature information in two dimensions. On the one hand, the static information of the spectrum features is retained, and on the other hand, the dynamic changes in the time domain-frequency domain are captured. Finally, the accuracy of the concentration prediction of the decomposition products of the SF6 / N2 mixed gas is improved through feature fusion, which has important practical significance for evaluating the internal insulation state of high-voltage electrical equipment.
[0090] Embodiment 2
[0091] Based on the method for predicting the concentration of mixed gas decomposition products based on feature fusion described in Example 1, this embodiment provides a device for predicting the concentration of mixed gas decomposition products based on feature fusion, including:
[0092] Spectral data acquisition module, used to obtain one-dimensional infrared spectrum data of mixed gas;
[0093] Denoising module, used to denoise one-dimensional infrared spectrum data;
[0094] A feature extraction module is used to input the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer, and output a first spectrum feature; obtain a two-dimensional spectrum time-frequency graph by wavelet transforming the denoised one-dimensional infrared spectrum data, and input it into a second feature extraction network including a convolution layer, and output a second spectrum feature;
[0095] The prediction module is used to input the first spectral feature and the second spectral feature into the concentration prediction module for splicing, and output the concentration prediction values of the decomposition products SO2, NO and CS2.
[0096] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for predicting the concentration of mixed gas decomposition products based on feature fusion are implemented.
[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0101] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.
Claims
1. A method for predicting the concentration of mixed gas decomposition products based on feature fusion, characterized in that: include: Obtain one-dimensional infrared spectrum data of mixed gas; Denoising one-dimensional infrared spectral data; Inputting the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer, and outputting a first spectrum feature; The denoised one-dimensional infrared spectrum data is transformed by wavelet transform to obtain a two-dimensional spectrum time-frequency diagram, which is input into a second feature extraction network including a convolution layer to output a second spectrum feature; The first spectral feature and the second spectral feature are input into the concentration prediction module for splicing, and the concentration prediction values of the decomposition products SO2, NO and CS2 are output.
2. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: The method for obtaining one-dimensional infrared spectrum data of mixed gas is: using standardized spectrum measurement operation to collect mixed gas at 1700cm -1 ~2700cm -1 One-dimensional infrared spectral data in the wavenumber range.
3. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: The one-dimensional infrared spectrum data denoising includes background interference denoising and random noise suppression.
4. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: The first feature extraction network includes a sequence input layer, a first long short-term memory layer, a first discard layer, a second long short-term memory layer, a second discard layer and a fully connected layer.
5. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: The second feature extraction network includes an image input layer, a first convolutional layer, a second convolutional layer and a fully connected layer.
6. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: The concentration prediction module includes a concatenation layer, a first fully connected layer, a hyperbolic tangent function activation layer, a second fully connected layer, a leaky rectified linear unit activation layer and a third fully connected layer.
7. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 1, characterized in that: Constructing a spectral database to train the first feature extraction network, the second feature extraction network and the concentration prediction module, the process of constructing the spectral database includes: A dynamic gas distributor is used to configure mixed gases of different concentration ratios; the mixed gases include SF6 / N2 mixed gas, SO2 gas, NO gas and CS2 gas, wherein the SF6 / N2 mixed gas is the background gas. SO2 gas, NO gas and CS2 gas are the gases to be tested; Standardized spectral measurement operations are used to collect spectral data of mixed gases with different concentration ratios within a preset wavenumber range to build a spectral database.
8. The method for predicting the concentration of mixed gas decomposition products based on feature fusion according to claim 7, characterized in that: The dynamic gas distributor is used to configure mixed gases with different concentration ratios. The formula is: Samples n ~N(2 n ,(n+0.5) 2 ),n=0,1,2,…,7 Among them, Samples n Indicates that the random concentrations of the three gases to be tested follow a mean of 2 n , a normal distribution with a standard deviation of n+0.
5.
9. A concentration prediction device for mixed gas decomposition products based on feature fusion, characterized in that: include: Spectral data acquisition module, used to obtain one-dimensional infrared spectrum data of mixed gas; Denoising module, used to denoise one-dimensional infrared spectrum data; A feature extraction module is used to input the denoised one-dimensional infrared spectrum data into a first feature extraction network including a long short-term memory layer, and output a first spectrum feature; obtain a two-dimensional spectrum time-frequency graph by wavelet transforming the denoised one-dimensional infrared spectrum data, and input it into a second feature extraction network including a convolution layer, and output a second spectrum feature; The prediction module is used to input the first spectral feature and the second spectral feature into the concentration prediction module for splicing, and output the concentration prediction values of the decomposition products SO2, NO and CS2.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for predicting the concentration of mixed gas decomposition products based on feature fusion as claimed in any one of claims 1 to 8 are implemented.