Industrial exhaust gas component detection method, device and equipment

Abstract

The invention relates to the technical field of gas component detection, and discloses an industrial exhaust gas component detection method, device and equipment, the method comprises the following steps: driving an induction probe in an industrial exhaust gas detection device to determine an exhaust gas sampling area, and controlling angle time control parameters based on the exhaust gas sampling area; driving the inductive probe to perform concentration sampling and reliability evaluation on the industrial waste gas at a preset angle position, and determining an optimal sampling angle set; performing multispectral sampling on the industrial waste gas to obtain multispectral waste gas sampling data, and performing feature integration on the multispectral waste gas sampling data in combination with the multidimensional environment parameters to obtain an environment-multispectral waste gas feature vector; multi-component feature recognition is executed through the CNN-GasNet deep learning model, the pollutant components in the industrial waste gas and the concentration values of the pollutant components are obtained, the detection accuracy of the industrial waste gas is improved, and high-precision recognition and concentration prediction of the multiple pollutant components are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of gas composition detection, and in particular to a method, device and equipment for detecting the composition of industrial exhaust gas. Background Art

[0002] Traditional industrial waste gas detection systems primarily use fixed-position probes for sampling and analysis, but this approach faces significant challenges in real-world industrial environments. Due to complex airflow, uneven temperatures, and pressure fluctuations at industrial sites, waste gas distribution exhibits significant spatial non-uniformity. Fixed probes often only capture pollutant information in a localized area, leading to significant discrepancies between test results and actual emissions, impacting regulatory decisions and pollution control effectiveness.

[0003] Existing industrial waste gas detection devices commonly suffer from limited sensor probe detection range and fixed angles, making them unable to cope with the dynamic changes in gas composition distribution under complex operating conditions. In particular, under conditions of high and uneven smoke concentration, fixed-position sampling results in incomplete data, affecting detection accuracy. Furthermore, industrial waste gas is often a complex mixture of multiple gases, subject to cross-interference. Traditional single-wavelength or single-point detection technologies struggle to effectively distinguish the exact concentrations of each component, especially for low-concentration pollutants, which lacks sensitivity, further exacerbating the uncertainty of monitoring data. Summary of the Invention

[0004] The present invention provides a method, device and equipment for detecting the components of industrial exhaust gases, which improves the detection accuracy of industrial waste gas and realizes high-precision identification and concentration prediction of multiple pollutant components.

[0005] In a first aspect, the present invention provides a method for detecting components of industrial exhaust gas, the method comprising: Determine the exhaust gas sampling area by driving the induction probe in the industrial exhaust gas detection device, and control the angle and time control parameters based on the exhaust gas sampling area; According to the angle time control parameters, the sensing probe is driven to perform concentration sampling and reliability evaluation on the industrial waste gas at a preset angle position to determine the optimal sampling angle set; Performing multispectral sampling on industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and integrating the multispectral waste gas sampling data with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector; The environmental-multispectral exhaust gas feature vector is input into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and concentration values ​​in the industrial exhaust gas.

[0006] In a second aspect, the present invention provides an industrial exhaust gas component detection device, the industrial exhaust gas component detection device comprising: A control module, configured to determine an exhaust gas sampling area by driving a sensing probe in an industrial exhaust gas detection device, and to control angle and time control parameters based on the exhaust gas sampling area; A concentration sampling module is used to drive the sensing probe to perform concentration sampling and reliability evaluation on industrial waste gas at a preset angle position according to the angle time control parameter, and determine the optimal sampling angle set; a feature integration module, configured to perform multispectral sampling on the industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and perform feature integration on the multispectral waste gas sampling data in combination with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector; The feature recognition module is used to input the environmental-multispectral exhaust gas feature vector into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and their concentration values ​​in the industrial exhaust gas.

[0007] A third aspect of the present invention provides a computer device comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned industrial exhaust gas composition detection method.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to execute the above-mentioned industrial exhaust gas composition detection method.

[0009] In the technical solution provided by the present invention, a rotatable multi-angle sensing probe technology is adopted. The probe is driven by a rotating block to perform a semicircular trajectory movement in a rotating slide, which expands the sampling space range to 120°, overcomes the defect of limited detection range of traditional fixed-position probes, and makes the detection of exhaust gas components more comprehensive and accurate. The dynamic sampling strategy based on the angle-time control parameters enables the sensing probe to adaptively adjust the residence time at each sampling point according to the exhaust gas flow characteristics and concentration change rate, improves the pertinence and representativeness of data collection, and avoids the problem of information loss caused by fixed-frequency sampling. Through multiple data preprocessing technologies such as adaptive median filtering, drift correction and outlier detection, the noise interference, time drift and outlier effects in the sampling process are effectively eliminated, and the reliability and accuracy of the exhaust gas concentration data are greatly improved. The multiple cross-validation sampling algorithm MRCS is used to perform reliability evaluation and cross-validation on the exhaust gas concentration data in different angle intervals, accurately lock the optimal sampling angle set, and solve the sampling position optimization problem caused by uneven exhaust gas diffusion. An array of ultraviolet, visible, and infrared multispectral sensors is deployed at optimal angles to perform high-precision scanning of the characteristic absorption wavelengths of different pollutants. Combining environmental parameters with probe motion parameters to construct multidimensional feature vectors significantly enhances the ability to distinguish complex gas mixtures. The CNN-GasNet deep learning model, through its shared encoder and dual-branch architecture (component identification branch and concentration calculation branch), achieves high-precision identification and concentration prediction of multiple pollutant components. This model is particularly effective in distinguishing low-concentration and spectrally overlapping components, overcoming the limitations of traditional detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 Schematic diagram of the steps of the method for detecting the composition of industrial exhaust gas according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of an industrial exhaust gas composition detection device according to an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Embodiments of the present invention provide a method, device and equipment for detecting the composition of industrial exhaust gases. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe the order or precedence of the objects. It should be understood that the numbers used in this way are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products or apparatus.

[0013] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the method for detecting components of industrial exhaust gas in an embodiment of the present invention includes: Step S1: determining an exhaust gas sampling area by driving a sensing probe in an industrial exhaust gas detection device, and controlling angle and time control parameters based on the exhaust gas sampling area; It is understandable that the execution subject of the present invention may be an industrial exhaust gas component detection device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0014] Specifically, by driving the induction probe in the industrial waste gas detection device to perform a semicircular trajectory along the rotating slide in the set structure, the space is covered within a rotation range of 120°. The induction probe works in coordination with the first rotating shaft, gear system, connecting rod and second rotating shaft tube in the rotating device to achieve accurate arrival at each rotation position in the rotating slide, and establish a perceptible area around the detector body, namely the waste gas sampling area. The waste gas sampling area is divided into non-uniform angles to avoid the problem of spatial information redundancy or insufficient density caused by equal-interval sampling. The non-uniform division principle is formulated based on the gradient characteristics of waste gas concentration at different angles in historical data. The angle division is encrypted in sections with drastic concentration changes, and the points are sparsely distributed in sections with gentle concentration changes, to obtain a discrete sampling angle set with spatial distribution characteristics. The residence time of each sampling point in the discrete sampling angle set is calculated according to the waste gas flow characteristics. During actual operation, the concentration change rate at adjacent angular positions is continuously estimated in real time. This rate, denoted as ΔC / Δθ, is then modeled using the concentration value C(θi-1) at the previous angular point and the position information at the current point θi. The specific dwell time calculation formula, t=α·C(θi-1)+β, is constructed by combining the adjustment coefficients α and β. Using this formula, the sampling system automatically adjusts the dwell time at each sampling point based on actual exhaust gas concentration fluctuations. In areas of high fluctuation, the sampling time is extended to improve data reliability, while in areas of stable concentration, the dwell time is appropriately shortened to enhance overall scanning efficiency. This creates an angle-dwell time correspondence table. The rotational speed control curve of the rotating block is calculated based on this angle-dwell time correspondence table. The rotational speed of the rotating block is controlled based on the sampling sequence and time distribution. Based on the angle-time requirements, the control unit uses a PID control algorithm to generate the rotating block speed control curve. This control curve also establishes a mapping relationship between the sampling angle, dwell time, and sampling frequency, achieving parameter coordination throughout the entire process, from physical motion to sampling behavior, and ultimately determining the angle-time control parameters.

[0015] Step S2: According to the angle-time control parameters, the sensing probe is driven to perform concentration sampling and reliability evaluation of the industrial waste gas at a preset angle position to determine the optimal sampling angle set; Specifically, according to the angle time control parameter set, the sensing probe is driven along the rotating slide to reach the set N angle position points in sequence, and a resident sampling operation is performed at each angle point to obtain the exhaust gas concentration value corresponding to the point. nThe continuous positioning and stay of the probe constitute the original angle-concentration discrete data set. The original angle-concentration discrete data set is subjected to noise elimination, and the adaptive median filtering algorithm is used to dynamically adjust the filter window size according to the local signal-to-noise ratio of the sampling point to ensure a balance between smoothing ability and edge retention ability. After the filtering process is completed, in order to avoid the influence of sensitivity drift caused by long-term operation of the probe on the concentration data and cause systematic errors, a time-related drift correction algorithm is introduced. The algorithm constructs a drift function through regular detection of reference standard gas, applies it to the concentration value at each angle, and obtains the true concentration data after eliminating the influence of time drift. The concentration data that eliminates the influence of time drift are subjected to outlier detection and local interpolation, and the modified Z-score method is used to identify non-physical deviation points caused by local concentration mutations, instrument anomalies or instantaneous data loss. When the absolute value of the Z-score of a certain sampling point is greater than 3.5, it is regarded as an abnormal point, and is compensated and repaired by local interpolation of the previous and next valid data to obtain the first exhaust gas concentration distribution data. The entire sampling angle interval is divided into several angle intervals, each of which covers a certain angle range, and a subset within the interval is extracted from the first concentration distribution data to form the second exhaust gas concentration distribution data. An independent reliability assessment is performed on the second concentration distribution data of each angle interval. The assessment includes data consistency verification, local fluctuation amplitude statistics, and variance calculation of multiple measurements. At the same time, a cross-validation mechanism based on different angle intervals is combined to verify the stability, representativeness, and redundancy of the data in each angle interval. Compare the consistency of concentration trends between adjacent intervals, and whether the current interval data is in a change-sensitive area in the global fitting curve. Based on the evaluation scores of each interval, the angle points with significant concentration changes, strong representativeness, and minimal noise are selected to form the optimal sampling angle set.

[0016] The full angle range covered by the first exhaust gas concentration distribution data is structured, that is, the complete angle interval is divided into several angle segments with independent interval attributes according to the set interval or concentration change trend. Each angle segment constitutes a local statistical unit, forming an angle interval set. For each sub-interval in the angle interval set, the concentration sample data corresponding to the interval is extracted, and two core indicators are calculated based on this data, namely the mean concentration level of the interval and the standard deviation of the concentration fluctuation degree, to construct the second exhaust gas concentration distribution data. The interval weight coefficient is calculated based on the second exhaust gas concentration distribution data of each angle interval. The interval weight coefficient comprehensively considers the pollutant intensity performance of the data collected in the angle interval, the recognizability of the concentration change, the stability of the data, and the spatial representativeness. The weight of each angle interval is calculated through a predefined evaluation model to form an interval weight coefficient set. The angle intervals are sorted from large to small according to the interval weight coefficient set, and the top M intervals are selected to form a candidate optimal interval set. A cross-validation operation is performed on each candidate angle interval. The cross-validation process includes data partitioning, fitting verification, and error analysis. Through multiple rounds of training and testing of data within the candidate interval, a systematic score is assigned to the data in terms of fitting curve accuracy, fluctuation consistency, and sampling responsiveness, and a comprehensive score is assigned to each interval. The comprehensive scoring results are sorted, and the interval with the highest score is selected, which is considered to have the strongest data reliability and pollutant response characteristics and is confirmed as the final optimal angle interval. Based on the range of this optimal angle interval, its starting and ending angle boundaries are determined. Within this range, multiple sampling angle points are set, combining factors such as concentration variation characteristics, spectral response characteristics, and the physical limitations of the sampling mechanism, to construct the optimal sampling angle set.

[0017] Step S3: Perform multispectral sampling on the industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and integrate the multispectral waste gas sampling data with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector; Specifically, based on the optimal sampling angle set, a multispectral sensor array is installed within the optimal range of the rotating slide. This sensor array consists of ultraviolet, visible, and infrared spectrometers covering different electromagnetic bands, ensuring that spectral response information from shortwave to longwave is captured at each sampling angle. Based on this structure, the sensor probe is precisely driven by a rotating block, rotating it sequentially along a set path to each designated position within the optimal angle set. Each time the sensor array reaches the target position, it automatically activates and initiates the collection of multi-band spectral absorption data for the exhaust gas sample at that angle, forming a complete set of raw spectral data covering the ultraviolet, visible, and infrared bands. To highlight the spectral response characteristics of different pollutant components in specific wavelength regions, multiple pollutant characteristic wavelength windows are set in the data processing. For example, a specific absorption window in the ultraviolet band is set for sulfur dioxide, and windows in the visible and infrared bands are set for nitrogen oxides or volatile organic compounds. A high-resolution scanning algorithm is used to perform high-precision scanning and curve analysis of the characteristic absorption peak region in each window. The effective absorption response of each pollutant in its representative wavelength region is extracted from the raw spectral data, generating a set of multispectral exhaust gas sampling data with identification significance. While performing spectral feature extraction, multi-dimensional environmental parameters are simultaneously collected, including on-site operating environment information such as temperature, pressure, humidity, wind speed and direction, as well as dynamic structural state data such as the current rotational angular velocity, motion acceleration and rotation radius of the sensing probe. The effective absorption characteristics are vector-integrated with the synchronously collected multi-dimensional environmental parameters to form multi-dimensional coupled input features. To improve feature representativeness and compress dimensionality, a differential optical absorption spectroscopy algorithm is used to perform feature enhancement and background stripping on the multispectral signal, thereby accurately extracting the differential absorption characteristics of pollutants in each wavelength channel. Combined with the normalized environmental parameters, this generates an environmental-multispectral exhaust gas feature vector.

[0018] Step S4: Input the environment-multispectral exhaust gas feature vector into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and their concentration values ​​in the industrial exhaust gas.

[0019] Specifically, the ambient-multispectral exhaust gas feature vector is input into the deep learning model. This feature vector contains the absorption characteristics of pollutants in multiple bands, including ultraviolet, visible, and infrared, and incorporates environmental factors such as temperature, humidity, and air pressure at the time of detection, as well as the sensor's rotational state parameters. This vector is then fed into the feature enhancement preprocessing unit within the CNN-GasNet deep learning architecture for spatial and channel normalization. Spatial normalization normalizes the spatial scale differences introduced by sampling points at different angles, while channel normalization performs isoscalization on the intensity ranges and environmental parameter dimensions of different spectral bands, forming an enhanced feature matrix with a balanced numerical distribution and no scale bias. This enhanced feature matrix is ​​then fed into the shared encoder of the CNN-GasNet architecture for multi-scale feature extraction. The shared encoder consists of multiple layers of convolutional units, each with a different receptive field size and step size. It simultaneously extracts potential correlations between the pollutant's spectral response and environmental disturbances at both the local detail and global trend scales. After multiple convolutions and pooling steps, the encoder outputs a set of high-dimensional shared feature codes. The shared feature codes are fed into two functional branches within the model structure, one for pollutant component identification and the other for concentration estimation. In the component identification branch, component-specific features are extracted based on the response patterns of the different pollutants in the feature codes. Classification is performed through a nonlinear mapping layer, outputting a set of probability distributions representing the likelihood of each pollutant being present in the current sample. In the concentration calculation branch, the shared features are mapped to a continuous output space, and a regression model is constructed based on concentration information from historically annotated data. This yields a set of predicted concentration values ​​for each pollutant component. To enhance the confidence of the prediction results, a concentration confidence interval assessment mechanism is implemented at the output stage, based on the matching relationship between the type probability distribution and the predicted concentration values. This generates confidence intervals with upper and lower bounds for each pollutant component, indicating the stability and uncertainty of the model output. The resulting output is a complete set of identification results, including the specific name of each pollutant component in the current industrial waste gas, the basis for determining its probability of presence, its corresponding predicted concentration value, and the upper and lower confidence intervals for the concentration.

[0020] The shared feature encodings are fed into the pollutant component identification branch of the CNN-GasNet deep learning model. This identification branch is designed to handle the specific discrimination of complex multi-component pollutants. Its core mechanism is to perform a goal-oriented deconstruction and identification of the multi-band spectral features, angular position information, and environmental response characteristics contained in the overall feature space. To achieve feature decoupling and identification, a channel separation operation is performed within the identification branch. Based on predefined component class labels or clustering results, the shared feature encodings are partitioned into multiple sub-matrices based on wavelength, angular, or channel response strength. This results in a set of pollutant target feature matrices, each corresponding to the response of a potential pollutant component in the input data. These pollutant target feature matrices are then fed into k regression sub-networks within the pollutant component identification branch, each of which is responsible for identifying a specific pollutant or class. To enhance the sub-network's ability to extract effective information from high-dimensional features, an attention-guided mechanism is implemented within each sub-network to dynamically highlight key dimensional features in the input matrix and suppress irrelevant background interference. By adjusting the attention weight distribution, each regression subnetwork performs a weighted operation on the feature submatrix for which it is responsible, outputting a set of target-guided weighted feature vectors. These vectors represent the significant response characteristics of pollutants under specific environmental and wavelength conditions, and are highly representative and discriminative. The weighted feature vectors are input into the three fully connected layers of the pollutant component identification branch, which sequentially perform linear mapping and nonlinear activation operations. This process aims to achieve a high-level abstraction of the original feature vectors, transforming the spectral response patterns, spatial distribution characteristics, and dynamic state signals perceived at the lower level into discriminable latent feature representations. Each network layer continuously enhances the discriminability of the input signal in the category space by adjusting the weight coefficients and activation function form, and outputs the target latent feature vector. The target latent feature vector is then corrected for environmental conditions and compensated for cross-interference. Environmental correction uses parameters such as temperature, humidity, and air pressure collected at the time of detection to correct for dimensions in the feature space that are significantly affected by environmental disturbances, thereby reducing errors caused by environmental fluctuations in classification. Cross-interference compensation, based on prior knowledge and data-driven models, detects classification deviations between pollutant components caused by spectral absorption overlap, mutual interference in physical sensor responses, or proximity of bands. By adjusting or reconstructing the feature space, it reduces the negative impact of inter-component interference on classification accuracy, thereby obtaining a probability distribution of the type of each pollutant component.

[0021] In an embodiment of the present invention, a rotatable multi-angle sensing probe technology is adopted. The rotating block drives the probe to perform a semicircular trajectory movement in the rotating slide, thereby expanding the sampling space range to 120°, overcoming the defect of the limited detection range of traditional fixed-position probes, and making the detection of exhaust gas components more comprehensive and accurate. The dynamic sampling strategy based on the angle-time control parameters enables the sensing probe to adaptively adjust the residence time at each sampling point according to the exhaust gas flow characteristics and concentration change rate, thereby improving the pertinence and representativeness of data collection and avoiding the problem of information loss caused by fixed-frequency sampling. Through multiple data preprocessing technologies such as adaptive median filtering, drift correction and outlier detection, the noise interference, time drift and outlier effects in the sampling process are effectively eliminated, greatly improving the reliability and accuracy of the exhaust gas concentration data. The multiple cross-validation sampling algorithm MRCS is adopted to perform reliability evaluation and cross-validation on the exhaust gas concentration data in different angle intervals, accurately lock the optimal sampling angle set, and solve the sampling position optimization problem caused by uneven exhaust gas diffusion. An array of ultraviolet, visible, and infrared multispectral sensors is deployed at optimal angles to perform high-precision scanning of the characteristic absorption wavelengths of different pollutants. Combining environmental parameters with probe motion parameters to construct multidimensional feature vectors significantly enhances the ability to distinguish complex gas mixtures. The CNN-GasNet deep learning model, through its shared encoder and dual-branch architecture (component identification branch and concentration calculation branch), achieves high-precision identification and concentration prediction of multiple pollutant components. This model is particularly effective in distinguishing low-concentration and spectrally overlapping components, overcoming the limitations of traditional detection methods.

[0022] In a specific embodiment, the process of executing step S1 may specifically include the following steps: The exhaust gas sampling area is obtained by driving the induction probe in the industrial exhaust gas detection device to rotate in a semicircular trajectory in the rotating slide; The exhaust gas sampling area is divided into non-uniform angles to obtain a discrete sampling angle set; The residence time of each sampling point in the discrete sampling angle set is calculated based on the exhaust gas flow characteristics, and the residence time of the current sampling point is determined based on the concentration change rate of adjacent angles to obtain an angle-residence time correspondence table; The rotation block speed control curve is calculated according to the angle-dwelling time correspondence table, and the mapping relationship between the sampling angle, dwelling time and sampling frequency is established according to the rotation block speed control curve to obtain the angle time control parameters.

[0023] Specifically, the exhaust gas detection device's sensor probe is activated by a rotation control mechanism on a rotating block within a drive unit, causing it to rotate within a pre-set semicircular slideway. Structurally, the slideway forms a 120-degree sector-shaped scanning area centered on the industrial detection instrument body. Driven by gear linkage and shaft guidance, the sensor probe executes a stable and controllable arc-shaped trajectory, forming an exhaust gas sampling area with a specific coverage range. Within this area, airflow exhibits significant spatial unevenness due to influences such as pipeline structure, pressure fluctuations, and obstruction interference. Therefore, a spatial sampling scheme was developed that accurately reflects actual concentration trends. Using a non-uniform angular partitioning strategy, combined with previous testing experience and preliminary real-time data scan results, the entire semicircular trajectory area was adaptively partitioned using density. Sampling points were densely spaced in areas of significant concentration fluctuations or sudden increases in pollutants, while they were relatively sparsely spaced in areas of gentle fluctuations. This resulted in a discrete sampling angle set with unequal angular spacing. Each angle in this set serves as a target position for dynamic control. The probe's dwell time at that position is calculated based on the flow behavior of industrial exhaust gas at that angle and the pollutant transport characteristics. A dynamic dwell mechanism based on concentration gradients is introduced. After the sensor probe completes concentration acquisition at the previous angle point, the rate of change of concentration between two adjacent points is estimated. This rate of change serves as a direct indicator of the intensity of airflow fluctuations. In areas with high rates of change, indicating significant pollutant concentration fluctuations, the probe dwell time should be extended to collect more instantaneous samples and enhance data representativeness and statistical stability. In areas with lower rates of change, the dwell time should be shortened to improve overall scanning cadence and resource efficiency. Based on this strategy, the optimal dwell time corresponding to each angle point is calculated, forming an angle-dwell time correspondence table. This table, indexed by angle and valued by time, provides the desired dwell time for each position along the scanning path. To map this angle-dwell time logic to the specific execution logic of the rotation mechanism, a speed control curve for the rotation block is calculated. This dynamic speed adjustment mechanism ensures that the rotation block not only maintains accurate angles but also maintains reasonable speed control at each angle point, ensuring that the dwell time matches the desired duration. To achieve this, a control logic module is introduced, integrating a feedback control mechanism with the angle sensor and motor controller into the transmission system to adjust the motor speed in stages. When the rotating block approaches a preset sampling angle, the controller slows down the speed based on the dwell time information at that angle in the angle-time correspondence table, allowing it to achieve precise soft landing alignment at the target angle, remain at the specified position for a constant time, and then reaccelerate to enter the next target angle segment. The entire speed control curve exhibits periodic acceleration and deceleration fluctuations, and its trajectory is positively correlated with the rate of change of concentration. Based on this speed control curve, a mapping relationship between sampling angle, dwell time, and sampling frequency is established.In the sampling frequency dimension, the control system achieves global scheduling control of the sampling frequency by adjusting the ratio between the number of angles covered in each cycle and the sum of the residence time of each sampling point. When the flow rate of industrial wastewater accelerates or the fluctuation range of pollutant concentration increases, the system, while ensuring that the angle coverage remains unchanged, increases the number of samples per unit time by shortening the residence time of low-priority angle points or slightly compressing the overall scanning cycle, thereby ensuring a balance between the system response rate and data timeliness. Conversely, when the exhaust gas flow is relatively stable, the system appropriately reduces the sampling frequency to reduce hardware power consumption and extend the service life of the equipment. The above-mentioned angle index, residence time, corresponding sampling frequency and speed curve parameters are integrated to generate a set of angle time control parameter sets.

[0024] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Based on the angle-time control parameters, the sensing probe is driven to the preset N angle positions in sequence, and dwell sampling is performed at each angle position to obtain the original angle-concentration discrete data set; Denoising the original angle-concentration discrete data set to obtain filtered concentration data, and performing drift correction on the filtered concentration data to obtain concentration data with the effect of time drift eliminated; Performing outlier detection and local interpolation on the concentration data from which the time drift effect is eliminated to obtain first exhaust gas concentration distribution data; The first exhaust gas concentration distribution data is divided into multiple angle intervals to obtain second exhaust gas concentration distribution data for each angle interval. Reliability evaluation and cross-validation are performed on the second exhaust gas concentration distribution data for each angle interval to determine the optimal sampling angle set.

[0025] Specifically, the rotational path of the sensing probe is controlled based on the angle-time control parameters—namely, the corresponding relationship between the sampling angle, dwell time, and rotation control speed. Using N preset angular positions as key sampling points, the sensing probe is driven sequentially and stabilized on the rotating slide, with a preset dwell operation performed at each angular point. During this process, whenever the probe rotates to a certain sampling angle position, the control module temporarily locks its posture and maintains a sufficient dwell time to allow the probe to collect a stable pollutant concentration signal at that position, forming a pair of raw data points indexed by angle and measured by concentration. As N sampling points are completed continuously, a raw angle-concentration discrete dataset is formed, structurally presenting a series of concentration observations at non-uniform angular intervals. Noise removal is performed on the raw data. Using a filtering algorithm, an adaptive median filtering mechanism based on local statistical features is employed to automatically identify concentration change trends before and after each sampling point. This method effectively suppresses random noise while retaining concentration mutation information, and incorporates an edge preservation strategy to prevent over-smoothing of the data. After filtering, a standard calibration process is introduced. The sensor response baseline is periodically calibrated using standard gases. The calibration results are then mapped to the measured data to form a drift-corrected concentration data series. Outlier detection and missing data repair are then performed on the drift-corrected concentration data. An anomaly recognition model is constructed based on central tendency and dispersion indicators. An improved deviation detection mechanism is used to locate suspected outliers. A local interpolation strategy is used to replace missing values ​​or outliers by spatially referencing normal data from adjacent angles, generating continuous, anomaly-free first exhaust gas concentration distribution data. The first exhaust gas concentration distribution data is then partitioned into multiple angular intervals to generate second exhaust gas concentration distribution data corresponding to these angular intervals. The concentration subset corresponding to each angular interval serves as the basic unit for reliability assessment. Evaluation calculations are performed on these subsets, including the mean, coefficient of variation, and stability distribution form of the concentration values ​​within the interval, to quantify the reliability of the concentration observations within that interval. Furthermore, a cross-validation mechanism is used between angular intervals to compare the data within one interval with those from other intervals in terms of trend, magnitude of change, or fitting residuals, to assess their consistency and synergy within the overall spatial distribution. By performing the aforementioned reliability assessment and cross-validation process on all angle intervals, a comprehensive score was ultimately obtained for each angle interval, which was then sorted from high to low based on the score. Intervals with higher scores indicate more stable concentration variation characteristics, more representative data, and a significant spatial response. Several of the angle intervals with the highest scores were selected as candidate optimal angle segments, and key angle points were extracted from these segments to ultimately determine the optimal sampling angle set with high recognition value, strong response capability, and excellent spatial expression characteristics.

[0026] In a specific embodiment, the execution step divides the first exhaust gas concentration distribution data into multiple angle intervals to obtain second exhaust gas concentration distribution data for each angle interval, and the reliability evaluation and cross-validation of the second exhaust gas concentration distribution data for each angle interval are performed to determine the optimal sampling angle set. The process may specifically include the following steps: Performing angle interval division on the first exhaust gas concentration distribution data to obtain an angle interval set; Calculating the mean and standard deviation of the concentration data in each angle interval in the angle interval set, and determining the second exhaust gas concentration distribution data of each angle interval according to the mean and standard deviation; Calculating interval weight coefficients based on the second exhaust gas concentration distribution data of each angle interval to obtain an interval weight coefficient set; Sort the angle intervals from large to small according to the interval weight coefficient set, and select the first M intervals to form the candidate optimal interval set; Perform a cross-validation process on each angle interval in the candidate optimal interval set to obtain a comprehensive interval score; According to the comprehensive score of the interval, the interval with the highest score is selected as the optimal interval, the angle range corresponding to the optimal interval is determined, and the optimal sampling angle set is set within the angle range.

[0027] Specifically, the first exhaust gas concentration distribution data is segmented into regions along the angular dimension. Adaptive segmentation is employed, dynamically adjusting the segmentation density based on the localized behavior of the pollutant concentration rate. This achieves finer spatial resolution within angular ranges where pollutant concentrations fluctuate dramatically, while maintaining wider coverage in regions with less fluctuation. This results in a set of non-overlapping, angularly continuous sub-intervals, which form the basic segmentation unit for pollutant spatial distribution analysis. Statistical characteristics are calculated for the concentration data within each interval. For each angular interval, the mean and standard deviation of the concentrations for all sampling points are calculated. The mean concentration reflects the average response level of the pollutant in that region, while the standard deviation reflects the severity of concentration fluctuations within that interval. Intervals with higher mean concentrations indicate pollutant distribution centers or high-emission areas, while lower standard deviations indicate stable concentration distributions in those regions that are less affected by random interference. Based on these two metrics, the original concentration data is transformed to generate a new concentration representation, which forms the second exhaust gas concentration distribution data. Weight coefficients are calculated for each interval based on the second exhaust gas concentration distribution data. When evaluating weights, the system tends to assign higher weights to angle intervals with strong pollutant responses and stable concentrations. Therefore, in the comprehensive scoring model, the mean concentration and standard deviation are weighted in a positive and negative manner, meaning that intervals with higher concentration values ​​and smaller fluctuations have greater weights. The same weight calculation logic is applied to each interval to obtain a set of interval weight coefficients. All angle intervals are sorted from high to low according to the weight coefficient set, and the angle intervals with the top M weights are extracted to form a set of candidate optimal intervals. A cross-validation process is performed on each angle interval in the candidate optimal interval set. During the cross-validation process, internal data is partitioned for each angle interval in the candidate optimal interval set, and the samples within the interval are divided into several subsets, which serve as training subsets and validation subsets, respectively. Stability is verified by reconstructing pollutant concentration distribution trends, comparing fitting residuals, and observing the consistency of concentration changes. At the same time, the concentration pattern of the interval is compared with the data structure of adjacent areas to identify whether it has good boundary transition properties and independent information contribution. The cross-validation results ultimately form a comprehensive interval scoring system. The angle interval with the highest overall score among all candidate intervals is selected as the final optimal interval, and its angular coverage is defined as the optimal sampling area for the current detection task. Within this optimal interval, multiple representative angle points are set based on the concentration distribution gradient or signal response amplitude. These points are distributed both in the concentration peak area and in the transition edge area, ensuring that the sampled data has both global expressiveness and local specificity, ultimately forming the optimal sampling angle set.

[0028] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Based on the optimal sampling angle set, a multispectral sensor array consisting of ultraviolet spectrum sensors, visible spectrum sensors, and infrared spectrum sensors is installed in the optimal range of the rotating slide. The rotating block is controlled to drive the sensing probe to accurately rotate to various angle positions and activate the multispectral sensor array to collect multi-band spectral absorption data. The pollutant characteristic wavelength window is set and the characteristic absorption peak is scanned for the multi-band spectral absorption data to obtain multi-spectral exhaust gas sampling data; The differential optical absorption spectroscopy algorithm is used to extract the effective absorption characteristics of the multi-spectral exhaust gas sampling data, and the effective absorption characteristics are vector-integrated with the synchronously collected multi-dimensional environmental parameters to obtain the environmental-multi-spectral exhaust gas feature vector.

[0029] Specifically, based on the optimal sampling angle set, a spectral sensor array is installed within the optimal range of the rotating slide. This sensor array includes three types of structural units: ultraviolet spectrum sensors, visible spectrum sensors, and infrared spectrum sensors. The three together form a cross-band, wide-response, high-resolution composite spectral acquisition system. These sensors correspond to the absorption characteristics of pollutants in the shortwave, mediumwave, and longwave regions according to the band coverage. The ultraviolet band is used to detect components with strong shortwave absorption, such as sulfur dioxide and benzene series. The visible light band is used to detect substances with significant absorption between 400 and 500 nanometers, such as nitrogen oxides. The infrared band covers gases such as volatile organic compounds and carbon monoxide with characteristic absorption peaks in the mid- and far-infrared regions. In terms of structural layout, the multi-spectral sensor array is embedded in specific angle node areas of the rotating slide along the distribution trajectory of the optimal sampling angle set, ensuring that when the sensing probe passes through these angle points, it can form a corresponding spectral acquisition channel with the array. By finely adjusting the control precision of the rotating block and employing a closed-loop feedback control mechanism, the sensor probe accurately reaches each optimal angular position while operating on the rotating slide. Upon reaching that angle, the corresponding multispectral sensor unit is automatically activated, initiating the spectral data acquisition process. During each stop, the three spectral sensors are activated synchronously, simultaneously collecting the spectral absorption behavior of the exhaust gas sample corresponding to that angular position in different wavelength bands with extremely high temporal synchronization and spatial overlap accuracy, generating a multi-band spectral raw signal set. The multi-band spectral absorption data is then windowed with pollutant-specific wavelengths. Based on a knowledge base of pollutant physical spectral fingerprints, key absorption wavelength ranges for each pollutant are predefined across different wavelengths. For example, sulfur dioxide exhibits stable absorption between 290 and 310 nanometers, nitrogen oxides exhibit a significant response between 400 and 450 nanometers, and some volatile organic compounds exhibit strong absorption peaks in the infrared region of 3.3 to 3.5 microns, or carbon monoxide at around 4.6 microns. Based on these window settings, each characteristic window in the original spectral data is focused and scanned, and a high-precision absorption peak search is performed within these characteristic intervals. By constructing a local fitting curve and analyzing the absorption peak displacement, waveform width, and absorption intensity, the actual response results of each pollutant component in its respective band are extracted, thereby compressing the original multi-band data into a set of effective characteristic data with a corresponding relationship with the pollutants, forming a multi-spectral exhaust gas sampling data set. Each element in this data set corresponds to the absorption performance of a specific band, a specific angle, and a specific component, and has a clear physical meaning and mapping relationship with the pollutant. The differential optical absorption spectroscopy algorithm is introduced to enhance the aforementioned multi-spectral sampling data. The algorithm constructs the differential characteristics between the background reference spectrum and the measured spectrum, removes the influence of non-target components such as the system response function, background scattering, and reflection interference, thereby effectively enhancing the true absorption characteristics of the pollutant itself in the spectrum and extracting key differential absorption modes in the frequency domain.The effective absorption features output by this algorithm constitute the dominant representation vector of the pollutant in spectral space. Simultaneously, considering the indirect interference of external environmental variables on the spectral response, multi-dimensional environmental parameters at the current moment are synchronously collected and recorded in parallel with the spectral feature extraction process. These include real-time operating conditions such as temperature, pressure, humidity, wind speed, and direction, as well as the dynamic behavior parameters of the sensing probe during sampling, such as rotational angular velocity, acceleration, and radius. These environmental parameters have a significant impact on gas absorption characteristics in certain bands, particularly in the infrared region. For example, temperature changes can affect the energy level transition probability of infrared gas molecules, while humidity changes can introduce interference from the water vapor absorption band. Therefore, the aforementioned environmental data and multispectral absorption features are uniformly normalized and integrated into a composite feature input. By constructing a multi-source data structure model, the spectral response information and environmental disturbance information are jointly encoded into a set of environmental-multispectral exhaust gas feature vectors.

[0030] In a specific embodiment, the process of executing step S4 may specifically include the following steps: The ambient-multispectral exhaust gas feature vector is input into the feature enhancement preprocessing unit in the CNN-GasNet deep learning model for spatial normalization and channel normalization to obtain an enhanced feature matrix. The enhanced feature matrix is ​​input into the shared encoder in the CNN-GasNet deep learning model for multi-scale feature extraction to obtain the shared feature encoding; The shared feature encoding is input into the pollutant component identification branch of the CNN-GasNet deep learning model to perform component-specific feature extraction and classification, and obtain the type probability distribution of each pollutant component; The shared feature encoding is input into the concentration calculation branch of the CNN-GasNet deep learning model to perform concentration value mapping calculation to obtain a set of concentration prediction values ​​for each pollutant component; Based on the type probability distribution and concentration prediction value set, the concentration confidence interval of each pollutant component is calculated, and the pollutant components and their concentration values ​​in industrial waste gas are output.

[0031] Specifically, the ambient-multispectral exhaust gas feature vector is input into the feature enhancement preprocessing unit in the CNN-GasNet deep learning model for spatial normalization and channel normalization. Spatial normalization is used to solve the problem of differences in sampling time, response amplitude, and signal continuity between sampling points at different angles, so that the data of all sampling angles are mapped to a unified scale interval in the spatial dimension, ensuring the equal response ability of the model to the angle distribution; while channel normalization is aimed at the differences in response intensity of multi-band input signals, especially the orders of magnitude difference in the original energy response between ultraviolet signals and infrared signals. Normalization can prevent a certain channel data from being misjudged as the dominant feature by the model due to scale deviation during training and inference, thereby weakening the response weights of other pollutant components. After the two-level normalization processing is completed, the enhanced feature matrix is ​​output. The enhanced feature matrix is ​​input into the shared encoder module in the CNN-GasNet model for multi-scale feature extraction. The shared encoder consists of several layers of convolutional network units, with different levels corresponding to different receptive fields and step-size designs. The shallow structure captures short-wave behaviors such as local strong absorption peaks, while the deep structure is suitable for extracting overall trends, cross-absorption patterns, and spectral drift features. Through layer-by-layer convolution, activation, and pooling operations, the system extracts a high-dimensional representation that combines local and global information while maintaining the integrity of the input structure, namely, the shared feature code. The shared feature code is input in parallel into the two functional branches of the CNN-GasNet model, namely the pollutant component identification branch and the concentration calculation branch. In the identification branch, component-specific feature extraction is performed in the feature space according to the pollutant component categories marked in the training phase, that is, the channel response related to the absorption spectrum characteristics of specific pollutants is selectively strengthened in the high-dimensional code, and the band signal or background interference characteristics that are not related to the current task are weakened. After component feature extraction, these features are mapped to a higher order through a set of fully connected layers and nonlinear mapping functions, forming a set of probability distributions for pollutant component types. The probability values ​​reflect the model's confidence in the presence or absence of each pollutant. The output is a probability vector whose length equals the number of component categories supported by the model, with each element representing the probability of the corresponding pollutant being identified under the current sampling state. Simultaneously, the shared feature encodings are fed into the concentration calculation branch, where, through convolution or regression mapping mechanisms, the input encodings are projected from feature space onto a continuous domain, establishing a functional relationship between the pollutant response signature and the actual concentration. The concentration calculation module is trained using a large amount of annotated concentration data as a supervisory signal, enabling it to not only distinguish the presence or absence of a pollutant but also predict its magnitude under each sampling state. The output is a set of concentration predictions, where each element represents an estimated current concentration of a particular pollutant component. These estimates are used for emission assessment, early warning analysis, and compliance assessment.Based on the type probability distribution and the set of concentration prediction values, the concentration confidence interval is evaluated. By comprehensively considering the model's confidence in the existence of a certain pollutant and the numerical distribution stability of its corresponding concentration prediction value, the concentration confidence interval of each pollutant component is calculated. The upper and lower boundaries of the confidence interval are estimated by the fluctuation range of the historical model residual, the confidence propagation mechanism of the current prediction value, or the distribution modeling method based on probability regression. For pollutant components with higher confidence, a narrower confidence interval is output, indicating that the concentration estimate has a strong reliability; for components with lower confidence or larger fluctuations in concentration estimates, the system will expand the confidence interval range, prompting the user that the identification result is uncertain, and recommends repeated sampling or calibration. The existence probability, predicted concentration value and corresponding concentration confidence interval of each pollutant component are structured and output to form a set of industrial waste gas multi-component identification and concentration estimation results.

[0032] In a specific embodiment, the step of inputting the shared feature encoding into the pollutant component identification branch of the CNN-GasNet deep learning model to perform component-specific feature extraction and classification to obtain the type probability distribution of each pollutant component can specifically include the following steps: The shared feature encoding is input into the pollutant component identification branch in the CNN-GasNet deep learning model, and the channel separation operation is performed through the pollutant component identification branch to obtain the pollutant target feature matrix group; The pollutant target feature matrix group is input into the k regression sub-networks of the pollutant component identification branch to perform attention-guided feature selection and obtain a weighted feature vector; The weighted feature vector is input into the three-layer fully connected layer of the pollutant component identification branch to perform linear transformation and nonlinear mapping to obtain the target hidden feature vector; The target hidden feature vector is corrected for environmental conditions and compensated for cross-interference to obtain the type probability distribution of each pollutant component.

[0033] Specifically, the shared feature code is fed into the pollutant component recognition branch of the CNN-GasNet deep learning model, where it performs a channel separation operation. During the channel separation process, the recognition branch slices and reorganizes the shared code along the feature dimension based on predefined pollutant component channel indices or channel attention patterns learned during training. This results in a structured set of pollutant target feature matrices, each corresponding to a feature subspace of a pollutant or pollutant category. The pollutant target feature matrices are then fed into k parallel regression subnetworks constructed within the recognition branch. Each regression subnetwork corresponds to a pollutant component type or subcategory within the same type. Its core task is to perform attention-guided feature selection. Within each regression subnetwork, an attention module is deployed. This module automatically identifies the most relevant feature dimensions for the current task by calculating attention weights for each channel, spatial location, or spectral dimension of the input feature matrix and dynamically adjusts their weight coefficients. This process does not rely on external rules or static masks, but rather automatically learns the attention distribution based on the response characteristics of the current input data and gradient feedback from the network's historical training, resulting in significant adaptability. Each attention module outputs a set of weight coefficients, which, after element-wise weighted multiplication with the original matrix, form a set of weighted feature vectors. The weighted feature vectors are sequentially input into the three-layer fully connected structure of the pollutant component identification branch. This structure, consisting of alternating linear mapping layers and nonlinear activation units, gradually maps low- and medium-level feature vectors to a higher-order feature space. In the first layer, linear transformations are used to transform the input features into the unified computational space of the neural network. In the second layer, nonlinear functions such as ReLU are introduced to activate the features to enhance their expressiveness in the category dimension. In the third layer, the network further focuses on the category distribution boundary, compressing the feature distribution, ultimately outputting a set of target hidden feature vectors. Environmental condition correction and cross-interference compensation are performed on the target hidden feature vectors. Environmental condition correction is based on previously collected parameter data such as temperature, humidity, air pressure, and wind speed and direction. Environmentally dependent portions of the hidden features are reconstructed and adjusted based on the current operating conditions. In the infrared band, gas absorption intensity is particularly sensitive to temperature and humidity. This correction effectively suppresses response shifts that occur under different detection conditions for the same pollutant. Cross-interference compensation primarily addresses spectral overlap and structural coupling between multiple pollutants in feature space. Because some pollutants have similar absorption characteristics in certain bands, a preset inter-spectral interference matrix or interference correlation weights automatically learned during training are used to redistribute or penalize the portions of the latent vector that belong to common absorption regions, thereby enhancing distinction between categories and reducing the risk of misclassification.After the above corrections and compensations, the adjusted target hidden feature vector is finally input into the classifier structure, and a set of standardized type probability distribution vectors is output, in which each element represents the probability of existence of a certain pollutant under the current sampling state. The vector as a whole constitutes the probabilistic expression of the pollutant component identification result.

[0034] The above describes the method for detecting the components of industrial exhaust gas in the embodiment of the present invention. The following describes the device for detecting the components of industrial exhaust gas in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an industrial exhaust gas component detection device includes: A control module is used to determine the exhaust gas sampling area by driving the sensing probe in the industrial exhaust gas detection device, and control the angle and time control parameters based on the exhaust gas sampling area; The concentration sampling module is used to drive the sensing probe to perform concentration sampling and reliability assessment of industrial waste gas at a preset angle position according to the angle-time control parameters, and determine the optimal sampling angle set; The feature integration module is used to perform multispectral sampling on industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and integrate the multispectral waste gas sampling data with multi-dimensional environmental parameters to obtain the environment-multispectral waste gas feature vector; The feature recognition module is used to input the environmental-multispectral exhaust gas feature vector into the CNN-GasNet deep learning model to perform multi-component feature recognition and obtain the pollutant components and their concentration values ​​in the industrial exhaust gas.

[0035] Through the coordinated cooperation of the various components mentioned above, a rotatable multi-angle sensing probe technology is adopted. The rotating block drives the probe to perform a semicircular trajectory movement within the rotating slide, expanding the sampling space range to 120°. This overcomes the limited detection range of traditional fixed-position probes and makes exhaust gas composition detection more comprehensive and accurate. The dynamic sampling strategy based on angle-time control parameters enables the sensing probe to adaptively adjust its residence time at each sampling point according to the exhaust gas flow characteristics and concentration change rate, improving the targetedness and representativeness of data collection and avoiding the information loss problem caused by fixed-frequency sampling. Through multiple data preprocessing technologies such as adaptive median filtering, drift correction, and outlier detection, the noise interference, time drift, and outlier effects in the sampling process are effectively eliminated, significantly improving the reliability and accuracy of exhaust gas concentration data. The multiple cross-validation sampling algorithm MRCS is used to perform reliability assessment and cross-validation on exhaust gas concentration data in different angle intervals, accurately lock the optimal sampling angle set, and solve the sampling position optimization problem caused by uneven exhaust gas diffusion. An array of ultraviolet, visible, and infrared multispectral sensors is deployed at optimal angles to perform high-precision scanning of the characteristic absorption wavelengths of different pollutants. Combining environmental parameters with probe motion parameters to construct multidimensional feature vectors significantly enhances the ability to distinguish complex gas mixtures. The CNN-GasNet deep learning model, through its shared encoder and dual-branch architecture (component identification branch and concentration calculation branch), achieves high-precision identification and concentration prediction of multiple pollutant components. This model is particularly effective in distinguishing low-concentration and spectrally overlapping components, overcoming the limitations of traditional detection methods.

[0036] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0037] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0038] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0039] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0040] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0041] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0042] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting components of industrial exhaust gas, characterized in that: include: Determine the exhaust gas sampling area by driving the induction probe in the industrial exhaust gas detection device, and control the angle and time control parameters based on the exhaust gas sampling area; According to the angle time control parameters, the sensing probe is driven to perform concentration sampling and reliability evaluation on the industrial waste gas at a preset angle position to determine the optimal sampling angle set; Performing multispectral sampling on industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and integrating the multispectral waste gas sampling data with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector; The environmental-multispectral exhaust gas feature vector is input into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and concentration values ​​in the industrial exhaust gas.

2. The method for detecting components of industrial exhaust gas according to claim 1, characterized in that: The method of determining the exhaust gas sampling area by driving the sensing probe in the industrial exhaust gas detection device and controlling the angle and time control parameters based on the exhaust gas sampling area includes: The exhaust gas sampling area is obtained by driving the induction probe in the industrial exhaust gas detection device to rotate in a semicircular trajectory in the rotating slide; Dividing the exhaust gas sampling area into non-uniform angles to obtain a discrete sampling angle set; Calculating the residence time of each sampling point in the discrete sampling angle set according to the exhaust gas flow characteristics, determining the residence time of the current sampling point based on the concentration change rate of adjacent angles, and obtaining an angle-residence time correspondence table; The rotation block speed control curve is calculated according to the angle-dwelling time correspondence table, and a mapping relationship among the sampling angle, the dwelling time and the sampling frequency is established according to the rotation block speed control curve to obtain the angle time control parameters.

3. The method for detecting components of industrial exhaust gas according to claim 1, characterized in that: The method of driving the sensing probe to perform concentration sampling and reliability evaluation on the industrial waste gas at a preset angle position according to the angle time control parameter and determining the optimal sampling angle set includes: Based on the angle time control parameters, the sensing probe is driven to reach N preset angle positions in sequence, and dwell sampling is performed at each angle position to obtain an original angle-concentration discrete data set; performing noise elimination on the original angle-concentration discrete data set to obtain filtered concentration data, and performing drift correction processing on the filtered concentration data to obtain concentration data with the effect of time drift eliminated; Performing outlier detection and local interpolation on the concentration data from which the time drift effect is eliminated to obtain first exhaust gas concentration distribution data; The first exhaust gas concentration distribution data is divided into multiple angle intervals to obtain second exhaust gas concentration distribution data for each angle interval, and reliability evaluation and cross-validation are performed on the second exhaust gas concentration distribution data for each angle interval to determine the optimal sampling angle set.

4. The method for detecting components of industrial exhaust gas according to claim 3, characterized in that: The step of dividing the first exhaust gas concentration distribution data into a plurality of angle intervals to obtain second exhaust gas concentration distribution data for each angle interval, performing reliability assessment and cross-validation on the second exhaust gas concentration distribution data for each angle interval, and determining an optimal sampling angle set includes: Performing angle interval division on the first exhaust gas concentration distribution data to obtain an angle interval set; Calculating a mean value and a standard deviation for the concentration data in each angle interval in the angle interval set, and determining second exhaust gas concentration distribution data for each angle interval based on the mean value and the standard deviation; Calculating interval weight coefficients based on the second exhaust gas concentration distribution data of each angle interval to obtain an interval weight coefficient set; Sort the angle intervals from largest to smallest according to the interval weight coefficient set, and select the first M intervals to form a candidate optimal interval set; Performing a cross-validation process on each angle interval in the candidate optimal interval set to obtain a comprehensive interval score; The interval with the highest score is selected as the optimal interval according to the comprehensive score of the intervals, the angle range corresponding to the optimal interval is determined, and the optimal sampling angle set is set within the angle range.

5. The method for detecting components of industrial exhaust gas according to claim 1, characterized in that: According to the optimal sampling angle set, multispectral sampling is performed on the industrial waste gas to obtain multispectral waste gas sampling data, and the multispectral waste gas sampling data is feature integrated in combination with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector, including: Based on the optimal sampling angle set, a multispectral sensor array consisting of ultraviolet spectrum sensors, visible spectrum sensors and infrared spectrum sensors is installed in the optimal range of the rotating slide, and the rotating block is controlled to drive the sensing probe to accurately rotate to each angle position and activate the multispectral sensor array to collect multi-band spectral absorption data; Perform pollutant characteristic wavelength window setting and characteristic absorption peak scanning on the multi-band spectral absorption data to obtain multi-spectral exhaust gas sampling data; A differential optical absorption spectroscopy algorithm is used to extract the effective absorption features of the multi-spectral exhaust gas sampling data, and the effective absorption features are vector-integrated with the synchronously collected multi-dimensional environmental parameters to obtain an environment-multi-spectral exhaust gas feature vector.

6. The method for detecting components of industrial exhaust gas according to claim 1, characterized in that: The environmental-multispectral exhaust gas feature vector is input into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and concentration values ​​in the industrial exhaust gas, including: Inputting the environment-multispectral exhaust gas feature vector into the feature enhancement preprocessing unit in the CNN-GasNet deep learning model for spatial normalization and channel normalization to obtain an enhanced feature matrix; Inputting the enhanced feature matrix into the shared encoder in the CNN-GasNet deep learning model to perform multi-scale feature extraction to obtain a shared feature code; Inputting the shared feature code into the pollutant component identification branch of the CNN-GasNet deep learning model to perform component-specific feature extraction and classification to obtain a type probability distribution of each pollutant component; Inputting the shared feature code into the concentration calculation branch of the CNN-GasNet deep learning model to perform concentration value mapping calculation to obtain a set of concentration prediction values ​​for each pollutant component; Based on the type probability distribution and the concentration prediction value set, the concentration confidence interval of each pollutant component is calculated, and the pollutant components and their concentration values ​​in the industrial waste gas are output.

7. The method for detecting components of industrial exhaust gas according to claim 6, characterized in that: The shared feature encoding is input into the pollutant component identification branch of the CNN-GasNet deep learning model to perform component-specific feature extraction and classification to obtain a type probability distribution of each pollutant component, including: Inputting the shared feature code into the pollutant component identification branch in the CNN-GasNet deep learning model, performing a channel separation operation through the pollutant component identification branch, and obtaining a pollutant target feature matrix group; Inputting the pollutant target feature matrix group into the k regression sub-networks of the pollutant component identification branch to perform attention-guided feature selection to obtain a weighted feature vector; Inputting the weighted feature vector into the three-layer fully connected layer of the pollutant component identification branch to perform linear transformation and nonlinear mapping to obtain a target hidden feature vector; Environmental condition correction and cross-interference compensation are performed on the target hidden feature vector to obtain the type probability distribution of each pollutant component.

8. An industrial exhaust gas component detection device, characterized in that: For executing the industrial exhaust gas component detection method according to any one of claims 1 to 7, the industrial exhaust gas component detection device comprises: A control module, configured to determine an exhaust gas sampling area by driving a sensing probe in an industrial exhaust gas detection device, and to control angle and time control parameters based on the exhaust gas sampling area; A concentration sampling module is used to drive the sensing probe to perform concentration sampling and reliability evaluation on industrial waste gas at a preset angle position according to the angle time control parameter, and determine the optimal sampling angle set; a feature integration module, configured to perform multispectral sampling on the industrial waste gas according to the optimal sampling angle set to obtain multispectral waste gas sampling data, and perform feature integration on the multispectral waste gas sampling data in combination with multi-dimensional environmental parameters to obtain an environment-multispectral waste gas feature vector; The feature recognition module is used to input the environmental-multispectral exhaust gas feature vector into the CNN-GasNet deep learning model to perform multi-component feature recognition to obtain the pollutant components and their concentration values ​​in the industrial exhaust gas.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for detecting components of industrial exhaust gas according to any one of claims 1 to 7 is implemented.

Images