Method, device and equipment for training gas concentration monitoring model and storage medium

Abstract

The application provides a training method and device of a gas concentration monitoring model, equipment and a storage medium, wherein the method comprises: obtaining a training set, performing layer-by-layer down-sampling on the training set to obtain feature sequences at different time scales as an initial set; inputting the initial set into a first layer of a stacked model, performing information enhancement on each feature sequence, taking all the enhanced feature sequences as an initial set input into a next layer of the stacked model, performing layer-by-layer iterative enhancement, and taking a feature sequence output from a last layer of the stacked model as a final enhanced sequence; obtaining potential information of the final enhanced sequence through a sequence analysis model, fusing the potential information and the corresponding final enhanced sequence into a key sequence; obtaining a gas concentration sequence corresponding to each key sequence through a fusion prediction model, fusing and activating results of all the gas concentration sequences to obtain a final monitoring result, taking a prediction error as a loss to correct and fuse parameters of the fusion prediction model, and obtaining a gas concentration monitoring model.

Description

Technical Field

[0001] This invention relates to the field of industrial forecasting technology, and in particular to a training method, apparatus, equipment, and storage medium for a gas concentration monitoring model. Background Technology

[0002] In modern industrial production, especially in large industrial facilities such as thermal power plants, gas emissions are a significant environmental issue, particularly carbon dioxide. Controlling and monitoring gas emissions is crucial for environmental protection and is one of the key factors in ensuring compliance with environmental regulations.

[0003] In related technologies, taking the monitoring of carbon dioxide gas as an example, traditional monitoring schemes mainly rely on periodic sampling and analysis, which often cannot provide real-time monitoring functions. In recent years, carbon dioxide sensors based on non-dispersive infrared technology have become a widely used monitoring tool, which can realize real-time monitoring of carbon dioxide in flues and can operate reliably under harsh environmental conditions. However, non-dispersive infrared technology uses a linear model to convert the sensor output, making the sensor output a carbon dioxide concentration value. This model has a simple structure and easily ignores complex influencing factors, thus affecting the monitoring effect of carbon dioxide concentration values.

[0004] Based on the above analysis of the development status of this technology, the existing technologies lack integrated gas-related characteristics and more complex gas concentration monitoring models. Summary of the Invention

[0005] The purpose of this invention is to provide a training method, apparatus, device, and storage medium for a gas concentration monitoring model, aiming to solve the above-mentioned problems in the prior art.

[0006] According to a first aspect of the present invention, a method for training a gas concentration monitoring model is provided, comprising:

[0007] A training set of gas-related features changing over time is obtained, and the training set is downsampled layer by layer to obtain feature sequences at different time scales, which serve as the initial set.

[0008] The initial set is input into the first layer of the stacked model, and information enhancement is performed on each feature sequence. All the enhanced feature sequences are used as the initial set of inputs for the next layer of the stacked model, and layer-by-layer iterative enhancement is performed. The feature sequence output by the last layer of the stacked model is used as the final enhanced sequence.

[0009] The potential information of the final augmented sequence is obtained through sequence analysis model, and the potential information is fused with the corresponding final augmented sequence to form the key sequence.

[0010] The gas concentration sequences corresponding to each key sequence are obtained by fusing the prediction model. The results of all gas concentration sequences are fused and activated to obtain the final monitoring results. The prediction error is used as the loss to correct the parameters of the fusing prediction model, thus obtaining the gas concentration monitoring model.

[0011] According to a second aspect of the present invention, a training apparatus for a gas concentration monitoring model is provided, comprising:

[0012] The downsampling module is used to acquire a training set of gas-related features that change over time. The training set is downsampled layer by layer to obtain feature sequences at different time scales, which serve as the initial set.

[0013] The multi-scale time decomposition hybrid module is used to input the initial set into the first layer of the stacked model, enhance the information of each feature sequence, use all the enhanced feature sequences as the initial set of input for the next layer of the stacked model, perform iterative enhancement layer by layer, and use the feature sequence output by the last layer of the stacked model as the final enhanced sequence.

[0014] The collaborative attention module is used to obtain the latent information of the final augmented sequence through the sequence analysis model, and to fuse the latent information with the corresponding final augmented sequence into a key sequence.

[0015] The fusion prediction module is used to obtain the gas concentration sequences corresponding to each key sequence through the fusion prediction model, fuse the results of all gas concentration sequences and activate them to obtain the final monitoring results, and use the prediction error as a loss to correct the parameters of the fusion prediction model to obtain the gas concentration monitoring model.

[0016] According to a third aspect of the present invention, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of a training method for a gas concentration monitoring model as provided in the first aspect of the present disclosure.

[0017] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which an information transmission implementation program is stored, which, when executed by a processor, implements the steps of the training method for the gas concentration monitoring model provided in the first aspect of the present disclosure.

[0018] The technical solution provided by the embodiments of the present invention includes the following beneficial effects: the model is trained by incorporating gas-related features to consider the influence of factors on gas concentration, thereby enhancing the environmental adaptability of the model; feature sequences at different time scales are obtained through layer-by-layer downsampling technology, and sequence information from different time zones is utilized to train multiple predictors based on key sequences at different time scales, with data complementarity between scales improving prediction accuracy; a stacked model is used to enhance sequence information, and then a sequence analysis model is used to mine potential important information, making full use of the feature information incorporating gas-related features to train the model.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the training method for the gas concentration monitoring model according to an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the sequence analysis model according to an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the gas concentration monitoring model framework according to an embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of the training device for the gas concentration monitoring model according to an embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0027] Method Implementation Examples

[0028] According to embodiments of the present invention, a training method for a gas concentration monitoring model is provided. Figure 1 This is a flowchart of the training method for the gas concentration monitoring model according to an embodiment of the present invention, such as... Figure 1 As shown, the training method for the gas concentration monitoring model according to an embodiment of the present invention specifically includes:

[0029] In step S110, a training set of gas-related features changing over time is obtained. The training set is downsampled layer by layer to obtain feature sequences at different time scales, which serve as the initial set. Specifically, this includes:

[0030] In this embodiment of the invention, the gas-related features, i.e., the input variables, include but are not limited to ambient temperature, ambient humidity, atmospheric pressure, and light intensity absorption values ​​of multiple bands, and the output variable is carbon dioxide concentration. Both the input and output variables are time series.

[0031] Preferably, the training set is a dataset that has undergone preprocessing such as filling missing values, smoothing outlier removal, and normalization to ensure data reliability and standardization. Alternatively, a test set containing the same features can be obtained to verify the model's performance. The time series training set and monitoring labels are in the form of... , It is a training set based on historical reality. Represents all historical data. Represents the dataset Total time, Dimensions representing gas-related characteristics Indicates the first Moment The values ​​of the input variables, , This represents the time series of the output variables;

[0032] For a given training set, proceed sequentially. N Each downsampling layer, according to Formula 1 and Formula 2, applies a downsampling layer to the training set. Average pooling N Layer downsampling yields the initial set:

[0033] Formula 1;

[0034] Formula 2;

[0035] in, Indicates the time dimension as The number of gas-related features is The training set, Indicates the first Feature sequences of layer downsampling, N represents the total number of downsampling layers. Indicates the average pooling layer; This represents the feature sequence corresponding to the lowest-level original training set, containing the finest temporal variations. The highest level N The feature sequence of order X contains the most macroscopic time changes, and the original initial set is... ;

[0036] The feature sequence represents the changes of gas-related features over time. Since there are multiple gas-related features, the feature sequence is represented as a two-dimensional matrix, with each time point containing multiple features.

[0037] In step S120, the initial set is input into the first layer of the stacked model, and information enhancement is performed on each feature sequence. All enhanced feature sequences are then used as the initial set for the input of the next layer of the stacked model, and layer-by-layer iterative enhancement is performed. The feature sequence output from the last layer of the stacked model is used as the final enhanced sequence. Specifically, this includes:

[0038] In each layer of the stacked model, seasonal and trend sequences are extracted from the various feature sequences of the initial set using time series analysis tools. For the ... The decomposition and mixing module, based on the time series decomposition tool of Formula 5, extracts seasonal and trend series from the initial set:

[0039] Formula 5;

[0040] in, Represents the initial set. Represents the set of seasonal items. Represents a set of trend item sequences. , , Indicates the first l The first decomposition module n The characteristic sequence of order, and These are the corresponding seasonal sequence and trend sequence, respectively. This represents the decomposition function of a time series analysis tool, which decomposes feature sequences at different scales. Decompose sequentially. This indicates the order of downsampling. The internal calculation process of the decomposition function is shown in Formulas 6 and 7:

[0041] Formula 6;

[0042] Formula 7;

[0043] in, Represents a seasonal sequence. Represents the trend sequence. Represents the characteristic sequence of the initial set. Indicates the average pooling layer. This indicates element padding, typically using 0 to ensure input integrity. and output Having the same height and width, subscript In the stacked model, the first In the layer, , and These represent the time dimension and the feature dimension of the input variables, respectively.

[0044] Multi-scale fusion of seasonal data series yields a mixed seasonal data series, and multi-scale fusion of trend data series yields a mixed basic data series, specifically including:

[0045] Multi-scale fusion of the seasonal series is performed using Formula 3, and multi-scale fusion of the trend series is performed using Formula 4:

[0046] Formula 3;

[0047] Formula 4;

[0048] in, Indicates a mixed seasonal sequence. Represents a seasonal sequence. Represents a mixed basic sequence. Indicates the trend sequence, superscript and superscript Indicates the downsampling order, subscript In the stacked model, the first In layer +1, This indicates two linear layers in between, including the GELU activation function. Reduce the size of the input features This indicates two linear layers in between, including the GELU activation function. Enlarge the size of the input features;

[0049] although and They are all linear layers, but for In other words, it needs to be downsampled again so that its feature size can be compared with... Same; but for In other words, its feature size needs to be enlarged so that its feature size can be compared with that of other features. same;

[0050] In the process of calculating mixed seasonal series The values ​​range from 0 to N -1, , Along the time dimension, its input dimension is Its output dimension is Along the input variable dimension, its input dimension is Its output dimension is That is, starting from the original scale and going downwards, the mixed seasonal sequence of the previous larger time scale is merged with the seasonal sequence of the current time scale to promote the flow of information between scales.

[0051] In the process of calculating the mixed basic sequence The values ​​are taken sequentially from N Up to 1, , Along the time dimension, its input dimension is Its output dimension is Along the input variable dimension, its input dimension is Its output dimension is That is, starting from the smallest scale, the mixed basic sequence of the previous smaller time scale is merged with the trend term sequence of the current time scale to promote the flow of information between scales.

[0052] The set of mixed seasonal sequences is The set of mixed basic sequences is .

[0053] The mixed seasonal sequence and the mixed basic sequence at the corresponding time scale are added together to obtain the comprehensive fluctuation sequence. The comprehensive fluctuation sequence is then input into a feedforward neural network for activation. The feedforward neural network is a forward linear layer. The activated sequence is added to the corresponding feature sequence in the initial set of the current layer for information enhancement, according to formulas 8 and 9.

[0054] Formula 8;

[0055] Formula 9;

[0056] in, This represents a feedforward neural network consisting of two linear layers, with a GELU activation function placed between them. All the enhanced feature sequences serve as the initial set of inputs for the next layer of the stacked model. Therefore, the subscripts on the left-hand side of equation 8 are... , It is the first +1 input to the decompose-mix module;

[0057] The final set of enhanced sequences is That is, the output set of the last layer of the stacked model.

[0058] A stacking model is set up to enhance the depth of the network. In this embodiment of the invention, L layers are stacked, and each layer is decomposed and mixed at multiple time scales, where L is greater than or equal to 1.

[0059] In step S130, the latent information of the final enhanced sequence is obtained through a sequence analysis model, and the latent information is fused with the corresponding final enhanced sequence to form a key sequence, specifically including:

[0060] The Res-TFA time series analysis model, which includes the temporal attention module TAM and the factor attention module FAM, is used to obtain potential information.

[0061] The temporal attention module assigns attention weights to the temporal dimension of the final augmented sequence to obtain the temporal latent sequence. Specifically, for an augmented sequence at scale m... , TAM is for all Attention intensity is calculated at each time step. First, max pooling and average pooling are performed on each row vector along the variable factor dimension to obtain two corresponding description vectors. The two description vectors are fed into the linear layer respectively to obtain , ,Will and After adding them, inputting them into the sigmoid activation function, we get... column vectors of the form Column vector Each element corresponds to the attention intensity at a different time point. Multiplying the column vector of attention intensity with the value of each row of the final augmented sequence yields the temporal latent sequence. ;

[0062] The gas-related feature dimension of the final enhanced sequence is assigned attention weights by the factor attention module to obtain the feature latent sequence. The specific process is as follows: Enter FAM, FAM applies to all The attention intensity is calculated for each input variable. First, max pooling and average pooling are performed on each row vector along the time dimension to obtain two corresponding description vectors. The two description vectors are fed into the linear layer respectively to obtain , ,Will and After adding them, inputting them into the sigmoid activation function, we get... row vectors of the form row vector Each element corresponds to the attention intensity of a different input variable. The row vector of attention intensity is multiplied by the value of each column of the final augmented sequence to obtain the feature latent sequence. ;

[0063] Figure 2 This is a schematic diagram of the sequence analysis model according to an embodiment of the present invention, as shown below. Figure 2 As shown, the custom processing logic is demonstrated. For a certain final enhanced sequence, the Temporal Attention Module (TAM) is used to focus on important time points in the sequence, and the Factor Attention Module (FAM) is used to focus on important attribute features in the sequence, capturing potential key information points in different dimensions to highlight information that is effective for prediction.

[0064] Time latent sequence and characteristic latent sequences Added as the final enhancement sequence The corresponding latent information is then used to connect the sequence of latent information with the final enhanced sequence via residual connections. Adding them together yields the key sequence of order m. For the set as Latent information fusion is performed on all the final enhanced sequences to obtain a set of key sequences. .

[0065] In step S140, the gas concentration sequences corresponding to each key sequence are obtained through the fusion prediction model. The results of all gas concentration sequences are fused and activated to obtain the final monitoring result. The prediction error is used as a loss to correct the parameters of the fusion prediction model, resulting in a gas concentration monitoring model, which specifically includes:

[0066] Key sequences are regressed using a predictor consisting of two linear layers. Each key sequence's corresponding predictor yields a predictor of size [value missing]. The monitoring results, namely the gas concentration sequence, can yield a corresponding gas concentration prediction value for each time step based on the input variable characteristics. The set includes N At a time scale, one can obtain N The size is The monitoring results are summed and then fed into two linear layers with the GELU activation function set in between to obtain the final prediction result. ;

[0067] Preferably, since the final result is obtained by adding up the monitoring results, the prediction result... The prediction results at certain time points may not be accurate enough. If the monitoring results of a key sequence are inaccurate, it may lead to a large error in the overall prediction results. Therefore, after obtaining the prediction results... After the sequence is processed, an adaptive filter is used to correct the sequence. Errors in the process.

[0068] Calculate the prediction result according to formula 10. Compared with the actual results MAE loss:

[0069] Formula 10;

[0070] in, Indicates MAE loss, Represents a sequence exist The value of time, Represents a sequence exist The value at time is used to perform backpropagation training and correct model parameters based on the loss value. After correction, the gas concentration monitoring model is completed. The gas concentration monitoring model includes downsampling processing, stacking model and fusion prediction model in steps S110 to S140. Figure 3 This is a schematic diagram of the gas concentration monitoring model framework according to an embodiment of the present invention, as shown below. Figure 3 As shown, the order of downsampling on the time scale N Taking =4 as an example, the design framework of the gas concentration monitoring model, including sampling processing, stacking model and fusion prediction model, is shown. Processing is carried out at multiple time scales within each sub-module.

[0071] The method further includes:

[0072] In step S150, the gas concentration monitoring model is used to acquire gas-related features in real time, and the monitoring results of gas concentration values ​​are obtained based on the gas-related features.

[0073] In summary, addressing the existing problems, this invention's gas concentration monitoring model training method incorporates gas-related features to train the model, considering the impact of influencing factors on gas concentration and enhancing the model's environmental adaptability. It obtains feature sequences at different time scales through layer-by-layer downsampling technology, utilizing sequence information from different time zones. Multiple predictors are trained based on key sequences at different time scales, with data complementarity between scales improving prediction accuracy. A stacked model is used to enhance sequence information; in each layer of the stacked model, seasonal sequences are extracted to capture periodic fluctuations, and trend sequences are extracted to capture long-term overall fluctuations, fully understanding the fluctuation patterns of data over time. Furthermore, multi-scale fusion of seasonal and trend sequences is performed separately, combining mixed sequences from adjacent scales with fluctuation sequences at the current time scale, effectively promoting information flow between scales. A Res-TFA sequence analysis model is used to mine potentially important information, enhancing the model's performance when processing time series and multi-factor data. Overall, the training process of the gas concentration monitoring model fully utilizes feature information incorporating gas-related features and differentiates the training model across different time scales.

[0074] Device Examples

[0075] According to an embodiment of the present invention, a training device for a gas concentration monitoring model is provided. Figure 4 This is a schematic diagram of the training device for the gas concentration monitoring model according to an embodiment of the present invention, as shown below. Figure 4 As shown, the training device for the gas concentration monitoring model according to an embodiment of the present invention specifically includes:

[0076] Downsampling module 40 is used to acquire a training set of gas-related features that change over time. The training set is downsampled layer by layer to obtain feature sequences at different time scales, which serve as the initial set. Specifically, it is used for:

[0077] Based on Formulas 1 and 2, the training set The initial set is obtained by using N layers of downsampling with average pooling:

[0078] Formula 1;

[0079] Formula 2;

[0080] in, Indicates the time dimension as The number of gas-related features is The training set, Indicates the first The feature sequence of the downsampling layers, where N represents the total number of downsampling layers. This represents the average pooling layer.

[0081] The multi-scale time decomposition mixing module 42 is used to input the initial set into the first layer of the stacked model, enhance the information of each feature sequence, use all the enhanced feature sequences as the initial set of input for the next layer of the stacked model, perform iterative enhancement layer by layer, and use the feature sequence output from the last layer of the stacked model as the final enhanced sequence. Specifically, it is used for:

[0082] In each layer of the stacked model, seasonal and trend sequences are extracted from the feature sequences of the initial set using time series analysis tools. The seasonal sequences are fused at multiple scales to obtain a mixed seasonal sequence, and the trend sequences are fused at multiple scales to obtain a mixed basic sequence.

[0083] Multi-scale fusion of the seasonal series is performed using Formula 3, and multi-scale fusion of the trend series is performed using Formula 4:

[0084] Formula 3;

[0085] Formula 4;

[0086] in, Indicates a mixed seasonal sequence. Represents a seasonal sequence. Represents a mixed basic sequence. Indicates the trend sequence, superscript and superscript Indicates the downsampling order, subscript In the stacked model, the first In layer +1, This indicates two linear layers in between, including the GELU activation function. Reduce the size of the input features This indicates two linear layers in between, including the GELU activation function. Enlarge the size of the input features.

[0087] The mixed seasonal sequence and the mixed basic sequence are added together to obtain the comprehensive fluctuation sequence. The comprehensive fluctuation sequence is then input into the feedforward neural network for activation. The activated sequence is then added to the corresponding feature sequence in the initial set of input to the current layer for information enhancement.

[0088] Collaborative attention module 44 is used to obtain latent information of the final augmented sequence through sequence analysis model, and to fuse the latent information with the corresponding final augmented sequence into a key sequence. Specifically, it is used for:

[0089] The Res-TFA time series analysis model, which includes a temporal attention module and a factor attention module, is used to obtain latent information. The temporal attention module is used to assign attention weights to the time dimension of the final augmented sequence to obtain the temporal latent sequence. The factor attention module is used to assign attention weights to the gas-related feature dimension of the final augmented sequence to obtain the feature latent sequence.

[0090] The temporal latent sequence and the feature latent sequence are added together to obtain the latent information corresponding to the final enhanced sequence.

[0091] The fusion prediction module 46 is used to obtain the gas concentration sequences corresponding to each key sequence through the fusion prediction model, fuse and activate the results of all gas concentration sequences to obtain the final monitoring result, and use the prediction error as a loss to correct the parameters of the fusion prediction model to obtain the gas concentration monitoring model. Specifically, it is used for:

[0092] The device further includes:

[0093] Application module 48 is used to acquire gas-related features in real time using a gas concentration monitoring model, and obtain monitoring results of gas concentration values ​​based on the gas-related features.

[0094] In summary, addressing the existing problems, this invention provides a training device for a gas concentration monitoring model. It incorporates gas-related features to train the model, considering the impact of various factors on gas concentration and enhancing the model's environmental adaptability. Through layer-by-layer downsampling, feature sequences at different time scales are obtained, utilizing sequence information from different time zones. Multiple predictors are trained based on key sequences at different time scales, with data complementarity between scales improving prediction accuracy. A stacked model is used to enhance sequence information. In each layer of the stacked model, seasonal sequences are extracted to capture periodic fluctuations, and trend sequences are extracted to capture long-term overall fluctuations, enabling a full understanding of the fluctuation patterns of data over time. Furthermore, multi-scale fusion of seasonal and trend sequences is performed separately, combining mixed sequences from adjacent scales with fluctuation sequences at the current time scale, effectively promoting information flow between scales. A Res-TFA sequence analysis model is used to mine potentially important information, enhancing the model's performance when processing time series and multi-factor data. Overall, the training process of the gas concentration monitoring model fully utilizes feature information incorporating gas-related characteristics and differentiates between different time scales for model training.

[0095] Electronic device examples

[0096] Figure 5This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device 500 may include at least one processor 510 and a memory 520. The processor 510 can execute instructions stored in the memory 520. The processor 510 is communicatively connected to the memory 520 via a data bus. In addition to the memory 520, the processor 510 can also be communicatively connected to an input device 530, an output device 540, and a communication device 550 via the data bus.

[0097] Processor 510 can be any conventional processor, such as a commercially available CPU. Processors may also include graphics processing units (GPUs), field-programmable gate arrays (FPGAs), systems-on-chips (SoCs), application-specific integrated circuits (ASICs), or combinations thereof.

[0098] The memory 520 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0099] In this embodiment of the present disclosure, the memory 520 stores executable instructions, and the processor 510 can read the executable instructions from the memory 520 and execute the instructions to implement all or part of the steps of the training method for any of the gas concentration monitoring models in the exemplary embodiments described above.

[0100] Computer-readable storage medium embodiments

[0101] In addition to the methods and apparatus described above, exemplary embodiments of this disclosure may also be a computer program product or a computer-readable storage medium storing the computer program product, the computer product including computer program instructions that can be executed by a processor to implement all or part of the steps described in the training method for any of the gas concentration monitoring models in the exemplary embodiments described above.

[0102] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. Programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages, and scripting languages ​​(e.g., Python). The program code can be executed entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0103] Computer-readable storage media may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: static random access memory (SRAM) having one or more electrically connected wires; electrically erasable programmable read-only memory (EEPROM); erasable programmable read-only memory (EPROM); programmable read-only memory (PROM); read-only memory (ROM); magnetic storage; flash memory; magnetic disk or optical disk; or any suitable combination thereof.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for a gas concentration monitoring model, characterized in that... include: A training set of gas-related features changing over time is obtained, and the training set is downsampled layer by layer to obtain feature sequences at different time scales, which are used as the initial set. The initial set is input into the first layer of the stacked model to enhance the information of each feature sequence. All the enhanced feature sequences are used as the initial set of inputs to the next layer of the stacked model. Iterative enhancement is performed layer by layer. The feature sequence output by the last layer of the stacked model is used as the final enhanced sequence. The potential information of the final augmented sequence is obtained through a sequence analysis model, and the potential information is fused with the corresponding final augmented sequence to form a key sequence. The gas concentration sequences corresponding to each key sequence are obtained by fusing the prediction model. The results of all gas concentration sequences are fused and activated to obtain the final monitoring results. The prediction error is used as the loss to correct the parameters of the fusing prediction model, thus obtaining the gas concentration monitoring model.

2. The method according to claim 1, characterized in that, The method further includes: The gas concentration monitoring model is used to acquire gas-related features in real time, and the monitoring results of gas concentration values ​​are obtained based on the gas-related features.

3. The method according to claim 1, characterized in that, The step of downsampling the training set layer by layer to obtain feature sequences at different time scales, which serve as the initial set, specifically includes: Based on Formulas 1 and 2, the training set The initial set is obtained by using N layers of downsampling with average pooling: Formula 1: Formula 2: in, Indicates the time dimension as The number of gas-related features is The training set, Indicates the first The feature sequence of the downsampling layers, where N represents the total number of downsampling layers. This represents the average pooling layer.

4. The method according to claim 1, characterized in that, The process of enhancing information in each feature sequence, using all enhanced feature sequences as the initial set of inputs for the next layer of the stacked model, and performing iterative enhancement layer by layer specifically includes: In each layer of the stacked model, seasonal and trend sequences are extracted from the feature sequences of the initial set using time series analysis tools. The seasonal sequences are then fused at multiple scales to obtain a mixed seasonal sequence, and the trend sequences are fused at multiple scales to obtain a mixed basic sequence. The mixed seasonal sequence and the mixed basic sequence are added together to obtain a comprehensive fluctuation sequence. The comprehensive fluctuation sequence is then input into a feedforward neural network for activation. The activated sequence is then added to the corresponding feature sequence in the initial set of the current layer for information enhancement.

5. The method according to claim 4, characterized in that, The process of multi-scale fusion of the seasonal term series to obtain a mixed seasonal series, and multi-scale fusion of the trend term series to obtain a mixed basic series, specifically includes: Multi-scale fusion of the seasonal series is performed using Formula 3, and multi-scale fusion of the trend series is performed using Formula 4: Formula 3: Formula 4: in, Indicates a mixed seasonal sequence. Represents a seasonal sequence. Represents a mixed basic sequence. Indicates the trend sequence, superscript and superscript Indicates the downsampling order, subscript In the stacked model, the first In layer +1, This indicates two linear layers in between, including the GELU activation function. Reduce the size of the input features This indicates two linear layers in between, including the GELU activation function. Enlarge the size of the input features.

6. The method according to claim 1, characterized in that, The specific steps of obtaining the potential information of the final enhanced sequence through sequence analysis model include: The Res-TFA time series analysis model, which includes a temporal attention module and a factor attention module, is used to obtain latent information. The temporal attention module is used to assign attention weights to the time dimension of the final augmented sequence to obtain the temporal latent sequence. The factor attention module is used to assign attention weights to the gas-related feature dimension of the final augmented sequence to obtain the feature latent sequence. The temporal latent sequence and the feature latent sequence are added together to obtain the latent information corresponding to the final enhanced sequence.

7. A training device for a gas concentration monitoring model, characterized in that... include: The downsampling module is used to acquire a training set of gas-related features that change over time. The training set is downsampled layer by layer to obtain feature sequences at different time scales, which serve as the initial set. The multi-scale time decomposition hybrid module is used to input the initial set into the first layer of the stacked model, enhance the information of each feature sequence, use all the enhanced feature sequences as the initial set of input to the next layer of the stacked model, perform iterative enhancement layer by layer, and use the feature sequence output by the last layer of the stacked model as the final enhanced sequence. The collaborative attention module is used to obtain potential information of the final augmented sequence through a sequence analysis model, and to fuse the potential information with the corresponding final augmented sequence into a key sequence. The fusion prediction module is used to obtain the gas concentration sequences corresponding to each key sequence through the fusion prediction model, fuse the results of all gas concentration sequences and activate them to obtain the final monitoring results, and use the prediction error as a loss to correct the parameters of the fusion prediction model to obtain the gas concentration monitoring model.

8. The apparatus according to claim 7, characterized in that, The device further includes: The application module is used to acquire gas-related features in real time using the gas concentration monitoring model, and to obtain the monitoring results of gas concentration values ​​based on the gas-related features.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the training method for the gas concentration monitoring model as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an information transmission implementation program, which, when executed by a processor, implements the steps of the training method for the gas concentration monitoring model as described in any one of claims 1 to 6.

Images