A method and system for diluted sampling of flue gas

Abstract

This application relates to the technical field of flue gas dilution sampling. Specifically, it discloses a flue gas dilution sampling method and system, which uses artificial intelligence technology based on deep learning to perform semantic parsing and multi-scale correlation analysis on the sampling condition data input by the user, so as to capture the semantic correlation information of sampling conditions at the first scale and the second scale, and through fine-grained semantic interaction and fusion of the two, to achieve a deep semantic understanding of the sampling conditions, thereby intelligently recommending appropriate sampling methods. In this way, it can effectively reduce the subjective judgment error of operators and improve the scientificity and accuracy of sampling method selection.

Description

Technical Field

[0001] This application relates to the technical field of flue gas dilution sampling, and more specifically, to a flue gas dilution sampling method and system. Background Art

[0002] The flue gas dilution sampling technology is a technology that dilutes the pollutants in the flue gas to reduce their concentration to a range suitable for analysis by an instrument, so as to achieve accurate measurement of the pollutants in the flue gas. In the fields of environmental protection and air quality monitoring, the flue gas dilution sampling technology is an indispensable means for evaluating atmospheric pollutant emissions, monitoring industrial emission sources, and studying the characteristics of particulate matter.

[0003] For example, the invention patent with the publication number CN115452490A discloses a method for diluting and sampling particulate matter in a flowing gas, which adopts a constant flow and constant dilution ratio sampling method or an isokinetic and constant dilution ratio sampling method. After selecting the sampling method, the flow rate of the sample gas in the sample gas pipeline is adjusted according to the set standard flow rate value of the sample gas, so that the sample gas enters the dilution mixing chamber and then enters the residence chamber. At the same time, the sample gas flow rate, pressure, and temperature signals are collected through the measurement and control system, the working condition flow rate value of the sample gas is converted into the standard flow rate value, and based on this, the gas flow rate of the bypass is adjusted to make the standard flow rate value of the sample gas equal to the determined flow rate value. This solution sets a flow sensor in the sample gas pipeline, which can conveniently, accurately, and real-time measure the flow rate value of the sample gas, improve the flow control accuracy during the sampling process, and also realize the on-site direct measurement of the particulate matter concentration, providing a more efficient and accurate technical means for environmental monitoring.

[0004] In practical applications, different sampling methods have different applicable scenarios and limiting conditions, so it is necessary to select the most suitable sampling method according to the specific sampling environment and purpose. However, in actual operation, operators usually need to rely on personal experience or simple preset standards to determine the sampling method, which not only increases the work difficulty but also may affect the accuracy and efficiency of the sampling results.

[0005] Therefore, an optimized flue gas dilution sampling method and system are expected. Summary of the Invention

[0006] In order to solve the above technical problems, this application is proposed. The embodiments of this application provide a flue gas dilution sampling method and system, which use artificial intelligence technology based on deep learning to perform semantic parsing and multi-scale correlation analysis on the sampling condition data input by the user, so as to capture the semantic correlation information of the sampling conditions at the first scale and the second scale, and through fine-grained semantic interaction and fusion of the two, to achieve a deep semantic understanding of the sampling conditions, thereby intelligently recommending a suitable sampling method. In this way, the subjective judgment error of the operator can be effectively reduced, and the scientificity and accuracy of the sampling method selection can be improved.

[0007] According to one aspect of the present application, a flue gas dilution sampling method is provided, which includes: selecting a sampling method, adopting a constant flow rate and constant dilution ratio sampling method or an isokinetic and constant dilution ratio sampling method, wherein selecting the sampling method includes:

[0008] Obtaining sampling condition data input by a user, where the sampling condition data includes a sampling environment, a sampling object, sample gas characteristics, a sampling purpose, equipment conditions, and accuracy requirements;

[0009] Performing multi-scale local correlation feature extraction on each of the sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix;

[0010] Performing feature interaction fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix;

[0011] Generating a recommended result of the sampling method based on the sampling condition multi-source multi-scale correlation feature matrix.

[0012] According to another aspect of the present application, a flue gas dilution sampling system is provided, which includes:

[0013] A sampling condition data acquisition module for obtaining sampling condition data input by a user, where the sampling condition data includes a sampling environment, a sampling object, sample gas characteristics, a sampling purpose, equipment conditions, and accuracy requirements;

[0014] A multi-scale local correlation feature extraction module for performing multi-scale local correlation feature extraction on each of the sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix;

[0015] A feature interaction fusion module for performing feature interaction fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix;

[0016] A recommended result generation module for generating a recommended result of the sampling method based on the sampling condition multi-source multi-scale correlation feature matrix.

[0017] Compared with the prior art, the flue gas dilution sampling method and system provided by the present application use artificial intelligence technology based on deep learning to perform semantic parsing and multi-scale correlation analysis on the sampling condition data input by the user, so as to capture the semantic correlation information of the sampling conditions at the first scale and the second scale, and through fine-grained semantic interaction and fusion of the two, to achieve a deep semantic understanding of the sampling conditions, thereby intelligently recommending appropriate sampling methods. In this way, the subjective judgment error of the operator can be effectively reduced, and the scientificity and accuracy of the selection of the sampling method can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By describing the embodiments of the present application in more detail with reference to the accompanying drawings, the above and other objects, features and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application, and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 It is a flowchart of the flue gas dilution sampling method according to an embodiment of the present application.

[0020] Figure 2 It is a schematic diagram of data flow of the flue gas dilution sampling method according to an embodiment of the present application.

[0021] Figure 3 It is a flowchart of sub-step S2 of the flue gas dilution sampling method according to an embodiment of the present application.

[0022] Figure 4 It is a flowchart of sub-step S3 of the flue gas dilution sampling method according to an embodiment of the present application.

[0023] Figure 5 It is a flowchart of sub-step S31 of the flue gas dilution sampling method according to an embodiment of the present application.

[0024] Figure 6 It is a block diagram of the flue gas dilution sampling system according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one" and / or "the" are not specifically singular, but may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list, and the method or device may also include other steps or elements.

[0026] Although this application makes various references to certain modules in the system according to embodiments of this application, any number of different modules may be used and run on the user terminal and / or server. The modules are merely illustrative, and different aspects of the system and method may use different modules.

[0027] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the operations before or below may not necessarily be executed precisely in sequence. Instead, various steps may be processed in reverse order or simultaneously as needed. Also, other operations may be added to these processes, or one or more steps may be removed from these processes.

[0028] Next, example embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all embodiments of this application. It should be understood that this application is not limited by the example embodiments described herein.

[0029] It is worth noting that in this application, all actions of obtaining data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where it is located and obtaining the authorization given by the owner of the corresponding device.

[0030] As mentioned in the above background art, Patent CN115452490A proposes a flue gas dilution sampling method. Since in practical applications, different sampling methods have different applicable scenarios and limiting conditions, it is necessary to select the most suitable sampling method according to the specific sampling environment and purpose. However, in actual operation, operators usually need to rely on personal experience or simple preset standards to determine the sampling method, which not only increases the work difficulty but also may affect the accuracy and efficiency of the sampling results. To address this technical problem, this application proposes an optimized flue gas dilution sampling method. It uses artificial intelligence technology based on deep learning to perform semantic parsing and multi-scale correlation analysis on the sampling condition data input by the user to capture the semantic correlation information of the sampling conditions at the first scale and the second scale, and through fine-grained semantic interaction and fusion of the two, to achieve a deep semantic understanding of the sampling conditions, thereby intelligently recommending a suitable sampling method. In this way, the subjective judgment error of the operator can be effectively reduced, and the scientificity and accuracy of the sampling method selection can be improved.

[0031] Figure 1 It is a flowchart of the flue gas dilution sampling method according to an embodiment of this application. Figure 2 It is a schematic diagram of the data flow of the flue gas dilution sampling method according to an embodiment of this application. As Figure 1 and Figure 2As shown, the flue gas dilution sampling method includes the steps of: S1, obtaining sampling condition data input by a user, where the sampling condition data includes a sampling environment, a sampling object, sample gas characteristics, a sampling purpose, equipment conditions, and accuracy requirements; S2, performing multi-scale local correlation feature extraction on each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix; S3, performing feature interaction fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix; S4, generating a recommended result of a sampling method based on the sampling condition multi-source multi-scale correlation feature matrix.

[0032] In the above flue gas dilution sampling method, in step S1, sampling condition data input by a user is obtained, where the sampling condition data includes a sampling environment, a sampling object, sample gas characteristics, a sampling purpose, equipment conditions, and accuracy requirements. It should be understood that when performing flue gas dilution sampling, different sampling tasks have different requirements and constraint conditions, and selecting an appropriate sampling method is crucial for ensuring the accuracy and reliability of data. For example, the constant flow and constant dilution ratio sampling method is suitable for application scenarios that require maintaining a constant sample gas flow rate and dilution ratio. It maintains a constant dilution ratio by setting a fixed standard condition flow rate value of the sample gas and the flow rate of the dilution gas. This method is particularly suitable for monitoring fixed pollution sources with relatively stable emissions and little change in flow rate, such as the exhaust gas emissions of industrial boilers, incinerators, and other equipment. Its advantages are simple operation and easy control, which can provide consistent sampling results and facilitate long-term monitoring and trend analysis. However, since this method does not adjust according to the actual change in flow rate, when facing a dynamic environment with large flow rate fluctuations, it may lead to insufficient sampling representativeness. In contrast, the isokinetic constant dilution ratio sampling method is more flexible. It requires the flow rate at the sampling nozzle of the sample gas to be consistent with the flow rate in the sampling pipeline, thereby achieving isokinetic sampling. This means that regardless of how the flow rate in the pipeline changes, the sampling process can reflect the true gas concentration distribution. This method is very suitable for monitoring situations where the flow rate is unstable or there are significant fluctuations, such as vehicle exhaust emission tests, air quality monitoring at urban traffic intersections, etc., and can capture instantaneous concentration changes more accurately and provide more precise measurement results. However, the technical requirements for implementing isokinetic sampling are relatively high. It is necessary to monitor and adjust the flow rate of the sampling system in real time to ensure that it is always synchronized with the flow rate in the pipeline, which places higher requirements on the equipment accuracy and technical complexity. Therefore, when selecting a specific sampling method, factors such as the sampling environment, sampling object, sample gas characteristics (such as flow rate stability), sampling purpose (such as monitoring short-term peaks or long-term trends), equipment conditions, and accuracy requirements need to be comprehensively considered to ensure the accuracy and representativeness of the sampling results.

[0033] The flue gas dilution sampling method is a measurement technique used to obtain the concentration of pollutants in the emission source. To ensure the accuracy and representativeness of the sampling, a series of sampling condition data need to be collected, including but not limited to the sampling environment, sampling object, characteristics of the sampled gas, sampling purpose, equipment conditions, and accuracy requirements. This information is crucial for selecting the appropriate sampling method, preparing the sampling equipment, and conducting data analysis. The following will detail the specific implementation steps for obtaining the sampling condition data:

[0034] First, it is necessary to understand the specific sampling environment. This means visiting the location of the emission source, recording its geographical location, and evaluating the possible impacts of the surrounding geographical features on the sampling. More importantly, a series of key environmental parameters need to be measured and recorded, such as temperature, humidity, atmospheric pressure, wind speed, and wind direction. These environmental factors not only affect the behavior patterns of pollutants but also directly impact the working efficiency and accuracy of the sampling equipment.

[0035] After having a preliminary understanding of the sampling environment, the next step is to gain an in-depth understanding of the sampling object, namely the flue gas itself. This step involves identifying the source of the flue gas, such as whether it comes from a thermal power plant, a chemical plant, or other types of industrial facilities. Flue gases from different sources have different emission characteristics, such as continuous or intermittent emissions, the magnitude and variation trend of the emission volume, and the emission cycle. Understanding these basic information helps to select the appropriate sampling strategies and technical means.

[0036] Then, it enters the analysis stage of the flue gas characteristics. This can be achieved by using advanced portable or fixed-installed gas analysis instruments, aiming to obtain the presence and concentration levels of common air pollutants, such as sulfur dioxide (SO2), nitrogen oxides (NOx), carbon monoxide (CO), volatile organic compounds (VOCs), and other potential harmful substances. In addition to chemical components, physical properties are also an aspect that cannot be ignored, including the temperature, pressure, flow rate, and moisture content of the flue gas. These parameters also have a significant impact on the sampling process.

[0037] Next, clarify the sampling purpose. Before planning any sampling activity, it is crucial to clarify the sampling purpose. Sampling purposes can be classified into several major categories according to different needs. For example, regulatory compliance monitoring, environmental impact assessment, scientific research, accident emergency response, etc., as well as equipment performance evaluation. Among them, regulatory compliance monitoring usually requires regular and continuous sampling to ensure that the emission levels meet the relevant environmental protection standards. Environmental impact assessment may require a detailed analysis of pollutant concentrations in a specific area or time period to evaluate their potential impact on the surrounding environment. Scientific research may focus on the generation mechanism, transmission process, or long-term trends of pollutants, while accident emergency response requires rapid and accurate acquisition of pollutant concentration information in order to take timely response measures. Each category of sampling purpose corresponds to different sampling strategies and methods. For example, regulatory compliance monitoring may focus more on the monitoring of long-term trends, while accident emergency response focuses more on rapid response and high-precision measurement. Therefore, the sampling purpose will directly affect the formulation of the sampling plan and the selection of sampling methods.

[0038] After sufficient background information is obtained, consideration can be given to how to use existing resources for effective sampling. This involves checking the performance parameters of all available sampling equipment, including but not limited to the flow range of the sampling pump, the material and inner diameter of the sampling tube, the type and efficiency of the filter, and the accuracy and stability of the data recording and processing equipment.

[0039] Through the above steps, sampling condition data can be systematically obtained, thus laying a solid foundation for subsequent data analysis and processing.

[0040] In the above flue gas dilution sampling method, in step S2, multi-scale local correlation feature extraction is performed on each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix. Among them, Figure 3 It is a flowchart of sub-step S2 of the flue gas dilution sampling method according to an embodiment of the present application. As Figure 3As shown, step S2 includes steps: S21, performing structured processing on each sampling condition data in the sampling condition data to obtain a sampling environment embedded coding vector, a sampling object embedded coding vector, a sampled gas characteristic embedded coding vector, a device condition embedded coding vector, and a precision requirement embedded coding vector; S22, arranging the sampling environment embedded coding vector, the sampling object embedded coding vector, the sampled gas characteristic embedded coding vector, the device condition embedded coding vector, and the precision requirement embedded coding vector along the sample dimension to form a sampling condition multi-source low-dimensional aggregation matrix; S23, inputting the sampling condition multi-source low-dimensional aggregation matrix into a multi-scale sequence dual-stream encoder based on an RNN-LSTM hybrid model to obtain the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix.

[0041] Specifically, in step S21, structured processing is performed on each sampling condition data in the sampling condition data to obtain a sampling environment embedded coding vector, a sampling object embedded coding vector, a sampled gas characteristic embedded coding vector, a device condition embedded coding vector, and a precision requirement embedded coding vector. It should be understood that considering that the sampling condition data is unstructured text information, therefore, in order to convert it into a form recognizable and processable by a computer to ensure that it can be effectively processed by a deep learning model, the present application further performs structured processing on each sampling condition data in the sampling condition data respectively, maps the multi-source sampling condition data to a high-dimensional semantic feature space through text embedding coding technology, captures the context semantic information of each sampling condition data, and converts it into a structured numerical representation. In a specific example of the present application, step S21 includes: respectively passing each sampling condition data in the sampling condition data through a semantic embedding encoder based on the Bert model to obtain the sampling environment embedded coding vector, the sampling object embedded coding vector, the sampled gas characteristic embedded coding vector, the device condition embedded coding vector, and the precision requirement embedded coding vector. Those of ordinary skill in the art should know that the Bert model is a pre-trained language model that can be used for various natural language processing tasks. It uses the encoder part in the Transformer architecture and performs deep bidirectional encoding on the text through a bidirectional training method to capture the context information. In the present application, the Bert model is used as the semantic embedding encoder, which can convert different sampling condition data into embedded coding vectors in vector form. The specific implementation steps of this method will be introduced in detail below:

[0042] For each sampling condition data, such as the sampling environment description, it is first necessary to convert it into a format suitable for processing by the Bert model. This usually involves tokenization, that is, breaking the sentence into word or sub-word units. Then, these words will form the input sequence of the model together with special tokens, such as [CLS] (used to indicate the start of the sequence) and [SEP] (a special token used to separate different sentences). In addition, positional embeddings are added to preserve the position information of the words. The core of the Bert model is a multi-layer bidirectional Transformer encoder, and each layer of it receives information from all positions of the previous layer, allowing the model to learn more complex features in all layers. This means that for a given word, the Bert model considers not only the context on the left but also the context on the right. After encoding, each word corresponds to a hidden state vector, which contains rich semantic information of the word. Finally, the overall context semantic feature representation of the sampling environment description can be obtained by aggregating the feature outputs of multiple layers, and the sampling environment embedding encoding vector is obtained.

[0043] In this way, the Bert model can capture key information in the text description of the sampling environment, such as conditions like temperature and humidity, so as to express the specific characteristics of the sampling target object, helping to understand the nature of the sample. This is helpful for representing the unique properties of gas samples, such as composition and pressure, and providing rich semantic information for subsequent data analysis.

[0044] Specifically, in step S22, the sampling environment embedding encoding vector, the sampling object embedding encoding vector, the sample gas characteristic embedding encoding vector, the device condition embedding encoding vector, and the accuracy requirement embedding encoding vector are arranged along the sample dimension to form a multi-source low-dimensional aggregation matrix of sampling conditions. It should be understood that considering that in practical applications, there are usually certain correlation relationships between various sampling conditions. For example, a specific type of sampling object usually determines its sample gas characteristics, and the sample gas characteristics in turn affect the selection of sampling equipment and accuracy requirements. Therefore, in order to accurately capture the semantic associations between various sampling conditions for better recommendation of sampling methods, it is necessary to further perform semantic association analysis on various sampling conditions. Based on this, in this application, the sampling environment embedding encoding vector, the sampling object embedding encoding vector, the sample gas characteristic embedding encoding vector, the device condition embedding encoding vector, and the accuracy requirement embedding encoding vector are arranged along the sample dimension to form a multi-source low-dimensional aggregation matrix of sampling conditions, so as to perform data integration, establish an association structure between various sampling conditions, and thus facilitate subsequent extraction of association features.

[0045] Specifically, in step S23, the sampled-condition multi-source low-dimensional aggregation matrix is input into a multi-scale sequence dual-stream encoder based on an RNN-LSTM hybrid model to obtain the first-scale sampled-condition multi-source local correlation feature matrix and the second-scale sampled-condition multi-source local correlation feature matrix. Specifically, in order to further explore the interaction and influence of various sampled conditions in different scale ranges, the present application uses an RNN-LSTM hybrid model to construct a multi-scale sequence dual-stream encoder, and captures local correlation information and broader correlation information in the sampled-condition multi-source low-dimensional aggregation matrix through a dual-channel structure, so as to obtain the first-scale sampled-condition multi-source local correlation feature matrix and the second-scale sampled-condition multi-source local correlation feature matrix. Those of ordinary skill in the art should be aware that the RNN model is a neural network model capable of processing sequence data. Through a cyclic connection method, the network can remember previous information and use this information to affect subsequent outputs. The LSTM (Long Short-Term Memory network) is a special type of RNN. By introducing a gating mechanism, it solves the problem of gradient vanishing or gradient explosion encountered by traditional RNN models when processing long-sequence data, and thus can effectively capture longer-term dependencies. Based on this, in the present application, the RNN model and the LSTM model are respectively used to process the sampled-condition multi-source low-dimensional aggregation matrix, and semantic information transmission between the feature of each sampled condition in the matrix is carried out through multiple hidden layers in the network model, so as to simultaneously extract relatively direct, local or short-term sampled-condition correlation information, as well as more global, long-term or indirect sampled-condition correlation information, thus achieving a comprehensive understanding of the sampled conditions.

[0046] In the above flue gas dilution sampling method, in step S3, the first-scale sampled-condition multi-source local correlation feature matrix and the second-scale sampled-condition multi-source local correlation feature matrix are subjected to feature interaction fusion based on external knowledge modulation to obtain a sampled-condition multi-source multi-scale correlation feature matrix. It should be understood that in order to comprehensively utilize multi-scale sampled-condition correlation information, it is necessary to further fuse the first-scale sampled-condition multi-source local correlation feature matrix and the second-scale sampled-condition multi-source local correlation feature matrix. In particular, in order to improve the accuracy and effect of feature fusion, the present application proposes a feature interaction fusion method based on external knowledge modulation, which guides the fusion process of multi-scale sampled-condition correlation features by introducing external knowledge related to sampled conditions, thereby enhancing the ability to identify key features and improving the richness and accuracy of feature representation. Among them, Figure 4 is a flowchart of sub-step S3 of the flue gas dilution sampling method according to an embodiment of the present application. As Figure 4As shown in the figure, step S3 includes the steps of: S31, based on external knowledge, performing fine-grained feature interaction optimization on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix to obtain an external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix; S32, based on the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix, performing feature modulation optimization on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix respectively to obtain an optimized first-scale sampling condition multi-source local association feature matrix and an optimized second-scale sampling condition multi-source local association feature matrix; S33, performing per-position semantic interaction fusion on the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix to obtain the sampling condition multi-source multi-scale association feature matrix.

[0047] Figure 5 It is a flowchart of sub-step S31 of the flue gas dilution sampling method according to an embodiment of the present application. As Figure 5 shown, step S31 includes the steps of: S311, inputting the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix into a fine-grained feature interaction network to obtain a sampling condition multi-source multi-scale local association interaction feature matrix; S312, inputting the sampling condition multi-source multi-scale local association interaction feature matrix into an attention unit based on external knowledge to obtain the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix

[0048] More specifically, S311 is expressed by the formula:

[0049]

[0050] where represents the first-scale sampling condition multi-source local association feature matrix, represents the second-scale sampling condition multi-source local association feature matrix, represents the transpose of the matrix, represents matrix multiplication operation, represents the sampling condition multi-source multi-scale local association interaction feature matrix.

[0051] That is, through the fine-grained feature interaction network, fine-grained feature interaction processing is performed on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix, so as to understand the correlation between the two from the micro level, capture the fine-grained interaction information between the two, and thus generate the sampling condition multi-source multi-scale local association interaction feature matrix.

[0052] More specifically, the S312 can be expressed by the formula:

[0053]

[0054] where and represent the learnable memory parameter matrices of the attention unit based on external knowledge, represents the normalization function, represents the multi-source multi-scale local association interaction feature matrix of the external knowledge optimization sampling condition.

[0055] That is, the generated multi-source multi-scale local association interaction feature matrix of the sampling condition is sent into the attention unit based on external knowledge, and the external knowledge is used to further optimize the multi-source multi-scale local association interaction feature matrix of the sampling condition, so as to enhance the model's understanding ability of the knowledge in the field of flue gas dilution sampling.

[0056] In a specific example of the present application, relevant knowledge of flue gas dilution sampling, such as the applicable scenarios, advantages and disadvantages of different sampling methods (such as isokinetic sampling, constant flow sampling, etc.), environmental protection rules, standards and requirements related to flue gas dilution sampling, etc., are used to train the parameter matrix of the attention unit, and guide the attention unit to weight the multi-source multi-scale local association interaction feature matrix of the sampling condition, so as to effectively identify and strengthen the important association interaction features between multi-source sampling conditions, and capture the potential connections between multi-source sampling conditions within a larger range.

[0057] Specifically, in a specific example of the present application, the step S32 includes: performing a linear transformation on the multi-source local association feature matrix of the first-scale sampling condition to obtain a first query feature matrix and a first value feature matrix, and using the multi-source multi-scale local association interaction feature matrix of the external knowledge optimization sampling condition as a key matrix, and inputting the first query feature matrix, the first value feature matrix and the key matrix into the fine-grained modulation module based on the Transformer structure to obtain the optimized multi-source local association feature matrix of the first-scale sampling condition; performing a linear transformation on the multi-source local association feature matrix of the second-scale sampling condition to obtain a second query feature matrix and a second value feature matrix, and using the multi-source multi-scale local association interaction feature matrix of the external knowledge optimization sampling condition as a key matrix, and inputting the second query feature matrix, the second value feature matrix and the key matrix into the fine-grained modulation module based on the Transformer structure to obtain the optimized multi-source local association feature matrix of the second-scale sampling condition.

[0058]

[0059] Among them, and respectively represent the first query embedding matrix and the first value embedding matrix, and respectively represent the first query feature matrix and the first value feature matrix, and respectively represent the second query embedding matrix and the second value embedding matrix, and respectively represent the second query feature matrix and the second value feature matrix, , , and respectively represent different bias terms, represents the normalized exponential function, represents the feature scale value of the external knowledge optimized fine-grained feature interaction matrix, and respectively represent the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix.

[0060] Here, this application uses the sampling condition multi-source multi-scale local association interaction feature matrix optimized by external knowledge as the key matrix. At the same time, based on the first-scale sampling condition multi-source local association feature matrix, the first query feature matrix and the first value feature matrix are constructed, and based on the second-scale sampling condition multi-source local association feature matrix, the second query feature matrix and the second value feature matrix are constructed. The information exchange and integration between the features inside and the external knowledge are realized through the attention mechanism of the Transformer structure, so as to ensure that the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix can benefit from the external knowledge to improve the quality of their feature expressions.

[0061] Specifically, the step S33 includes: calculating the sampling condition multi-source multi-scale association feature matrix by dividing the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix at the corresponding positions.

[0062]

[0063] Among them, represents the sampling condition multi-source multi-scale association feature matrix.

[0064] That is, the optimized first-scale sampling condition multi-source local correlation feature matrix and the optimized second-scale sampling condition multi-source local correlation feature matrix are subjected to position-by-position semantic interaction fusion to integrate the multi-scale multi-source sampling condition semantic correlation information, obtaining a sampling condition multi-source multi-scale correlation feature matrix, thereby achieving a comprehensive understanding of the sampling conditions.

[0065] In the above flue gas dilution sampling method, in step S4, based on the sampling condition multi-source multi-scale correlation feature matrix, a recommended result of the sampling method is generated. In a specific example of the present application, step S4 includes: inputting the sampling condition multi-source multi-scale correlation feature matrix into a sampling method intelligent recommendation module based on a classifier to obtain the recommended result, and the recommended result is used to represent the recommended type label of the sampling method. Here, the classifier is used to predict the most suitable sampling method according to the input sampling condition multi-source multi-scale correlation feature matrix and output the corresponding recommended type label. Specifically, the training of the classifier is based on a large amount of sampling condition data and corresponding sampling method labels. Through supervised learning, the model can learn the mapping relationship between different sampling conditions and sampling methods. After training, the classifier can, according to the real-time input sampling condition multi-source multi-scale correlation feature matrix, by learning and identifying the feature patterns and semantic information contained therein, combined with the classification mapping relationship learned during the training process, quickly and accurately recommend the sampling method that best suits the current sampling conditions, thereby providing decision-making support for users.

[0066] Particularly, considering that the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix respectively represent the sampling condition time-series semantic correlation coding features under different time-series correlation coding scales. In this way, when performing feature fine-grained interaction based on external knowledge modulation, the introduction of external prior knowledge will optimize the semantic interaction quality between the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix. However, the introduction of external knowledge may also introduce interference information at the same time. This makes the overall feature manifold representation of the sampling condition multi-source multi-scale correlation feature matrix obtained through feature fine-grained interaction have fine-grained feature structure holes in the high-dimensional feature space. The existence of fine-grained feature structure holes will not only cause insufficient semantic coverage of the features of the sampling condition multi-source multi-scale correlation feature matrix corresponding to class probability labels, but also cause off-target outlier class regression inference mapping, affecting the accuracy of the recommended result obtained by inputting the sampling condition multi-source multi-scale correlation feature matrix into the sampling method intelligent recommendation module based on a classifier.

[0067] Based on this, in a preferred example of the present application, when the multi-source multi-scale associated feature matrix of the sampling conditions is input into the intelligent recommendation module of the sampling method based on the classifier to obtain the recommendation result, the multi-source multi-scale associated feature matrix of the sampling conditions is first optimized, and the specific optimization process includes:

[0068] Expand the multi-source multi-scale associated feature matrix of the sampling conditions into a multi-source multi-scale associated feature vector of the sampling conditions;

[0069] Input the multi-source multi-scale associated feature vector of the sampling conditions into a pre-classifier based on the Softmax function to obtain a multi-source multi-scale fine-grained class probability label vector, which is expressed by the formula:

[0070]

[0071] Wherein, is the eigenvalue at the th position in the multi-source multi-scale fine-grained class probability label vector, is the eigenvalue at the th position in the multi-source multi-scale associated feature vector, is the scale value of the multi-source multi-scale associated feature vector, represents the natural constant.

[0072] Calculate the product between the multi-source multi-scale fine-grained class probability label vector and its transposed vector to obtain a multi-source multi-scale fine-grained full label domain modulation matrix; divide the eigenvalues at each position in the multi-source multi-scale fine-grained full label domain modulation matrix by the square root of the scale value of the multi-source multi-scale associated feature vector to obtain a multi-source multi-scale fine-grained full label domain scale modulation matrix, which is expressed by the formula:

[0073]

[0074] Wherein, is the multi-source multi-scale fine-grained full label domain scale modulation matrix, is the multi-source multi-scale fine-grained class probability label vector, is the transpose of the vector, is the matrix multiplication operation.

[0075] Input the multi-source multi-scale fine-grained full label domain scale modulation matrix into a class probability domain attention sparsity module based on a multi-level gating function to obtain a sparsified multi-source multi-scale fine-grained full label domain scale modulation matrix, which is expressed by the formula:

[0076]

[0077]

[0078] Among them, is the sparse sampling condition multi-source multi-scale fine-grained full label domain scale modulation matrix, is the multi-level gating function, is the eigenvalue at the th position in the sampling condition multi-source multi-scale fine-grained full label domain scale modulation matrix, is the scaling factor, is the gating threshold.

[0079] Taking the sampling condition multi-source multi-scale correlation feature vector as the query feature vector, multiplying the sparse sampling condition multi-source multi-scale fine-grained full label domain scale modulation matrix by the sampling condition multi-source multi-scale correlation feature vector to obtain an optimized sampling condition multi-source multi-scale correlation feature vector, which is expressed by the formula:

[0080]

[0081] Among them, is the optimized sampling condition multi-source multi-scale correlation feature vector.

[0082] Inputting the optimized sampling condition multi-source multi-scale correlation feature vector into the classifier-based sampling method intelligent recommendation module to obtain the recommendation result.

[0083] Thus, by adding class probability domain-level optimizable perturbations to the original feature distribution to strengthen the entanglement of the dependencies based on label domain modulation among different variables in the original feature distribution, the scale modulation optimization of the source domain feature vector based on class probability query is realized. In this way, it is possible to more effectively retain the significant inherent modal information in the source domain features and effectively mask the interfering components in the feature distribution, thereby improving the adversarial robustness of the manifold expression of the feature distribution. In this way, the accuracy of the recommendation result obtained by inputting it into the classifier-based sampling method intelligent recommendation module is improved.

[0084] In the technical solution of this application, after determining the sampling method, the sampling process can be started. The constant flow rate and constant dilution ratio sampling method and the isokinetic and constant dilution ratio sampling method are two common sampling methods. Among them, the constant flow rate and constant dilution ratio sampling method emphasizes keeping the flow rates of the dilution gas and the sampled flue gas constant. This means that during the entire sampling process, the flow rates of the dilution gas and the flue gas are preset and will not change with the change of the flue gas flow rate in the flue. This method is applicable to the situation where the flue gas flow rate is relatively stable or when precise measurement is not required. The constant flow rate and constant dilution ratio sampling method uses a precise flow controller to ensure that the flow rates of the dilution gas and the flue gas are consistent, thereby ensuring a fixed dilution ratio. The mixed gas is then transported to the analytical instrument for further analysis. The specific implementation steps of this method will be introduced in detail below:

[0085] First is the preparation stage. It is necessary to ensure that all used flow meters, thermometers, pressure sensors, and other related equipment are calibrated and in good working condition, and select appropriate dilution air. Usually, filtered ambient air is used as the dilution air to ensure that it does not contain the pollutants to be measured. It is also necessary to set up the sampling system. According to the type of pollutants to be measured and the expected concentration, design and install appropriate sampling pipelines, probes, filter membranes, or other collection devices.

[0086] Next is the determination of the dilution ratio. Determine the appropriate dilution ratio according to the expected flue gas concentration and the maximum concentration that the analytical instrument can handle. The dilution ratio is the ratio of the dilution air flow rate to the flue gas flow rate. By adjusting the valve or using an electronic flow controller, the required dilution ratio can be achieved.

[0087] Then, start sampling. Turn on the pump or compressor that provides clean air to allow the dilution air to enter the mixing chamber, and then insert the sampling probe into the chimney or emission outlet to start extracting the flue gas sample. It is necessary to ensure that the probe is in the correct position to obtain a representative sample. Continuously monitor the flow rates of the dilution air and the flue gas to maintain a stable dilution ratio. At the same time, record key parameters such as pressure and temperature changes during sampling.

[0088] Secondly is the sample collection stage. Pass the diluted flue gas through the pre-prepared filter membrane or other trapping media to intercept particulate matter or adsorb the target gas compounds. Take samples at predetermined time intervals for subsequent laboratory analysis. For long-term continuous sampling, it is necessary to consider replacing the filter membrane or container.

[0089] Finally, end the sampling. After reaching the predetermined sampling time, first close the flue gas inlet, and then gradually reduce the dilution air flow rate until it stops completely. Then immediately seal the container or packaging material containing the collected sample to prevent external interference. Then label each sample with relevant information such as date, location, dilution ratio, etc., and store it properly for transportation to the laboratory.

[0090] In subsequent data processing and reporting, use the known dilution ratio to back-calculate the actual concentration of pollutants in the original flue gas, and compare the results of parallel samples, blank samples, etc. to evaluate the quality of the entire sampling process. Organize all data, including sampling conditions, operation procedures, final measured values, and uncertainty estimates, to form a formal report.

[0091] The core of the isokinetic constant dilution ratio sampling method lies in maintaining the consistency between the sampling rate and the flue gas flow rate in the flue. This method requires the sampling probe to be able to adjust the pumping rate in real time to match the change in the flue gas flow rate. Usually, the sampling system is equipped with a flow rate sensor to monitor the flue gas flow rate in the flue, and the pumping speed of the sampling probe is dynamically adjusted through a feedback control system to ensure that the two always remain synchronized. In addition, in order to maintain a fixed dilution ratio, it is also necessary to adjust the supply amount of the dilution gas according to actual needs, so that the ratio of the dilution gas to the flue gas remains unchanged throughout the sampling process. The following will introduce the specific implementation steps of this method in detail:

[0092] First is the preparation stage. It is necessary to select appropriate dilution air. Usually, clean and dry air is used as the dilution medium. The dilution air should not contain the pollutants to be measured or its content is extremely low to avoid affecting the results. It is also necessary to prepare and calibrate instrument equipment, including flow meters, temperature and pressure sensors, etc., to ensure that all measurement tools have been calibrated and are within the validity period, and prepare sampling containers, such as porous plate glass bottles or other containers suitable for storing samples, and ensure that they are clean and pollution-free.

[0093] Then determine the sampling points and sampling frequency. Determine the optimal sampling location according to factors such as the flue size, shape, and internal flow field distribution. For flues with rectangular or circular cross-sections, the sampling points can be selected according to the methods specified in relevant standards. The sampling frequency, that is, the sampling time interval, is generally at least one sample per hour.

[0094] Then set the dilution system parameters. Adjust the dilution system to make the dilution air flow rate match the flue gas flow rate, that is, keep the flow rates of the two the same. This goal can be achieved by adjusting the supply amount of the dilution air. It is also necessary to set a fixed dilution ratio, such as 1:10 or adjust it according to the actual situation. This ratio determines the final total volume after mixing and the proportion of the original flue gas in it.

[0095] After the above preparations are completed, sampling can begin. Before starting sampling, let the dilution system run for a period of time to stabilize the conditions. When the conditions reach the preset values, start the sampling program, and at the same time record the timestamp and environmental parameters at this time, such as temperature, humidity, atmospheric pressure, etc. These information are very important for subsequent data analysis. Then draw a certain amount of undiluted raw flue gas from the selected sampling point into the sampling pipeline and immediately mix it with the dilution air provided in a set ratio.

[0096] Introduce the diluted flue gas obtained from the above steps into a pre-prepared sampling container for storage. Pay attention to controlling the sampling time and volume to ensure that the requirements of laboratory analysis can be met. For some special components, such as volatile organic compounds, special sampling methods may be required, such as adsorbent sampling method, to ensure its integrity.

[0097] After sampling, bring the collected samples back to the laboratory for further chemical analysis to determine the specific concentrations of various pollutants. The actual concentrations of various pollutants in the flue gas in the undiluted state can be calculated using the known dilution ratio. During the analysis process, attention should be paid to considering possible interference factors and taking appropriate measures to eliminate them.

[0098] Finally, summarize all the experimental data to form a complete report document. The report content should include but not be limited to sampling location, date, time, adopted method, model specifications of main instruments and equipment, name of the operator, original data record form, calculation formula and its application description, final test results, and uncertainty evaluation, etc.

[0099] It should be noted that when performing each of the above steps, relevant national standards or industry specifications and other applicable standard guidelines must be strictly adhered to. In addition, in order to ensure the accuracy and representativeness of sampling, it is also necessary to consider the specific on-site conditions, such as the influence of meteorological conditions, emission characteristics and other factors.

[0100] In summary, the flue gas dilution sampling method based on the embodiments of the present application is clarified. It uses artificial intelligence technology based on deep learning to perform semantic parsing and multi-scale correlation analysis on the sampling condition data input by the user to capture the semantic association information of the sampling conditions at the first scale and the second scale, and through fine-grained semantic interaction and fusion of the two, to achieve a deep semantic understanding of the sampling conditions, so as to intelligently recommend appropriate sampling methods. In this way, the subjective judgment error of the operator can be effectively reduced, and the scientificity and accuracy of the sampling method selection can be improved.

[0101] Furthermore, a flue gas dilution sampling system is also provided.

[0102] Figure 6 Is a block diagram of the flue gas dilution sampling system according to the embodiments of the present application. AsFigure 6 As shown in Figure 6 , the flue gas dilution sampling system 100 according to an embodiment of the present application includes: a sampling condition data acquisition module 110, configured to acquire sampling condition data input by a user, where the sampling condition data includes a sampling environment, a sampling object, a sample gas characteristic, a sampling purpose, equipment conditions, and accuracy requirements; a multi-scale local correlation feature extraction module 120, configured to perform multi-scale local correlation feature extraction on each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix; a feature interaction and fusion module 130, configured to perform feature interaction and fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix; and a recommendation result generation module 140, configured to generate a recommendation result of a sampling method based on the sampling condition multi-source multi-scale correlation feature matrix.

[0103] Here, those skilled in the art can understand that the specific operations of each module in the above flue gas dilution sampling system have been described in detail in the description of the flue gas dilution sampling method above with reference to Figures 1 to 5 and thus, the repeated description thereof will be omitted.

[0104] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, advantages, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, advantages, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purposes of illustration and easy understanding, rather than limitations, and the above details do not limit the present invention to necessarily adopt the above specific details for implementation.

[0105] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

[0107] In addition, it is obvious that the term "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The multiple elements stated in the system claims can also be implemented by one element through software or hardware.

[0108] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, 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 preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A flue gas dilution sampling method, comprising: Select a sampling method, adopting a constant flow rate and constant dilution ratio sampling method or an isokinetic and constant dilution ratio sampling method, characterized in that selecting a sampling method includes: Obtain the sampling condition data input by the user, where the sampling condition data includes the sampling environment, sampling object, sample gas characteristics, sampling purpose, equipment conditions, and accuracy requirements; Extract multi-scale local correlation features from each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix; Perform feature interaction fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix; Generate a recommended result of the sampling method based on the sampling condition multi-source multi-scale correlation feature matrix.

2. The flue gas dilution sampling method according to claim 1, wherein Extract multi-scale local correlation features from each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix, including: Perform structured processing on each sampling condition data in the sampling condition data to obtain a sampling environment embedding encoding vector, a sampling object embedding encoding vector, a sample gas characteristic embedding encoding vector, an equipment condition embedding encoding vector, and an accuracy requirement embedding encoding vector; Arrange the sampling environment embedding encoding vector, the sampling object embedding encoding vector, the sample gas characteristic embedding encoding vector, the equipment condition embedding encoding vector, and the accuracy requirement embedding encoding vector along the sample dimension into a sampling condition multi-source low-dimensional aggregation matrix; Input the sampling condition multi-source low-dimensional aggregation matrix into a multi-scale sequence dual-stream encoder based on an RNN-LSTM hybrid model to obtain the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix.

3. The flue gas dilution sampling method according to claim 2, wherein, Perform structured processing on each sampling condition data in the sampling condition data to obtain a sampling environment embedding encoding vector, a sampling object embedding encoding vector, a sample gas characteristic embedding encoding vector, an equipment condition embedding encoding vector, and an accuracy requirement embedding encoding vector, including: Pass each sampling condition data in the sampling condition data through a semantic embedding encoder based on the Bert model to obtain the sampling environment embedding encoding vector, the sampling object embedding encoding vector, the sample gas characteristic embedding encoding vector, the equipment condition embedding encoding vector, and the accuracy requirement embedding encoding vector.

4. The flue gas dilution sampling method according to claim 3, characterized in that, Perform feature interaction fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a sampling condition multi-source multi-scale correlation feature matrix, including: Based on external knowledge, perform fine-grained feature interaction optimization on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain an external knowledge optimized sampling condition multi-source multi-scale local correlation interaction feature matrix; Optimize the sampling condition multi-source multi-scale local association interaction feature matrix based on the external knowledge, and perform feature modulation optimization on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix respectively to obtain the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix; Perform per-position semantic interaction fusion on the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix to obtain the sampling condition multi-source multi-scale association feature matrix.

5. The flue gas dilution sampling method according to claim 4, wherein Based on external knowledge, perform fine-grained feature interaction optimization on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix to obtain the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix, including: Input the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix into a fine-grained feature interaction network to obtain the sampling condition multi-source multi-scale local association interaction feature matrix; Input the sampling condition multi-source multi-scale local association interaction feature matrix into an attention unit based on external knowledge to obtain the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix.

6. The flue gas dilution sampling method according to claim 5, characterized in that, Optimize the sampling condition multi-source multi-scale local association interaction feature matrix based on the external knowledge, and perform feature modulation optimization on the first-scale sampling condition multi-source local association feature matrix and the second-scale sampling condition multi-source local association feature matrix respectively to obtain the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix, including: Perform a linear transformation on the first-scale sampling condition multi-source local association feature matrix to obtain a first query feature matrix and a first value feature matrix, and use the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix as the key matrix. Input the first query feature matrix, the first value feature matrix, and the key matrix into a fine-grained modulation module based on the Transformer structure to obtain the optimized first-scale sampling condition multi-source local association feature matrix; Perform a linear transformation on the second-scale sampling condition multi-source local association feature matrix to obtain a second query feature matrix and a second value feature matrix, and use the external knowledge optimized sampling condition multi-source multi-scale local association interaction feature matrix as the key matrix. Input the second query feature matrix, the second value feature matrix, and the key matrix into the fine-grained modulation module based on the Transformer structure to obtain the optimized second-scale sampling condition multi-source local association feature matrix.

7. The flue gas dilution sampling method according to claim 6, wherein, Perform per-position semantic interaction fusion on the optimized first-scale sampling condition multi-source local association feature matrix and the optimized second-scale sampling condition multi-source local association feature matrix to obtain the sampling condition multi-source multi-scale association feature matrix, including: Calculate the multi-source multi-scale correlation feature matrix of sampling conditions by dividing the optimized first-scale sampling condition multi-source local correlation feature matrix and the optimized second-scale sampling condition multi-source local correlation feature matrix at each position point.

8. The flue gas dilution sampling method according to claim 7, wherein Based on the multi-source multi-scale correlation feature matrix of sampling conditions, generate a recommended result for the sampling method, including: Input the multi-source multi-scale correlation feature matrix of sampling conditions into the intelligent recommendation module for sampling methods based on a classifier to obtain the recommended result, and the recommended result is used to represent the recommended type label of the sampling method.

9. A flue gas dilution sampling system, characterized in that, Including: A sampling condition data acquisition module, which is used to acquire the sampling condition data input by the user, and the sampling condition data includes sampling environment, sampling object, sample gas characteristics, sampling purpose, equipment conditions, and accuracy requirements; A multi-scale local correlation feature extraction module, which is used to extract multi-scale local correlation features from each sampling condition data in the sampling condition data to obtain a first-scale sampling condition multi-source local correlation feature matrix and a second-scale sampling condition multi-source local correlation feature matrix; A feature interaction and fusion module, which is used to perform feature interaction and fusion based on external knowledge modulation on the first-scale sampling condition multi-source local correlation feature matrix and the second-scale sampling condition multi-source local correlation feature matrix to obtain a multi-source multi-scale correlation feature matrix of sampling conditions; A recommended result generation module, which is used to generate a recommended result for the sampling method based on the multi-source multi-scale correlation feature matrix of sampling conditions.

Images