Flue gas emission continuous monitoring system and device

Abstract

The invention belongs to the technical field of environmental monitoring, and provides a flue gas emission continuous monitoring system and device. The system comprises a data acquisition module, a flue gas pretreatment module, an ultraviolet spectrum acquisition module, a multi-mode spectral analysis module, a self-adaptive interference compensation module, a particulate matter fingerprint and pollution source tracing module, a multi-source collaborative data fusion module, a carbon flow rate measurement and calculation module, a digital twinborn maintenance module, a block chain evidence storage module and a system control and communication module. According to the system, the monitoring precision is improved through multi-mode fusion, synchronous detection of gas concentration and particulate matter element composition is achieved through collaborative analysis of a quantum dot enhanced DOAS technology, a Stern-Volmer fluorescence quenching technology and an LIBS technology, the detection effect of NH3 is enhanced by introducing a CdSe / ZnS quantum dot coating, and meanwhile, the precision of gas concentration and particulate matter element analysis is remarkably improved through the means of humidity correction and the like; particularly, the detection precision of NH3 in a complex flue gas environment is improved, and the problem that the monitoring error of a traditional single technology is large is solved.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring technology, and in particular to a continuous emission monitoring system and device for flue gas. Background Technology

[0002] A continuous emission monitoring system (CEMS) is an automated system used to monitor the concentrations and related parameters of various pollutants in industrial flue gas in real time. Through a series of operations such as sampling, pretreatment, and analysis of the flue gas, it provides accurate pollutant emission data for environmental regulatory departments and enterprises, and is an important means to control industrial pollution and ensure environmental quality.

[0003] Currently, existing flue gas emission monitoring technologies mainly employ single spectral analysis or chemical analysis methods. Spectroscopic analysis methods are mostly based on infrared or ultraviolet absorption principles; for example, DOAS technology is commonly used to measure gas concentrations. Chemical analysis methods detect pollutants through chemical reactions. Additionally, some systems incorporate data processing and communication functions to achieve basic monitoring and data transmission.

[0004] However, existing technologies have some shortcomings, such as: insufficient monitoring accuracy, susceptibility to interference from complex components, humidity, temperature, and other factors in flue gas, especially for the detection of gases such as NH3, where errors are significant in high-humidity environments; weak anti-interference capability, making it difficult to identify and eliminate the influence of unknown interfering gases, leading to reduced reliability of measurement results; relatively limited functionality, lacking the ability to analyze the elemental composition of particulate matter and trace pollution sources, thus failing to provide accurate basis for pollution control; low accuracy in carbon flow rate calculation, failing to adequately meet the current needs of carbon emission management; and reliance on manual inspections for equipment maintenance, lacking predictive maintenance capabilities, which may result in excessively long system downtime and affect monitoring continuity. Summary of the Invention

[0005] The purpose of this application is to provide a continuous emission monitoring system and device for flue gas to solve at least one of the technical problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this application provides a continuous emission monitoring system for flue gas, employing the following technical solution: A continuous emission monitoring system for flue gas includes: The data acquisition module is used to collect raw flue gas in the flue and simultaneously monitor the sampling flow rate. It adopts an in-situ extraction sampling probe with a CdSe / ZnS quantum dot coating to specifically adsorb NH3. The flue gas pretreatment module is used to remove dust and moisture from the raw flue gas, stabilize its temperature, pressure and flow rate, and output clean and stable test gas and fluorescence signal excited by quantum dot coating. The ultraviolet spectral acquisition module is used to emit ultraviolet light to the gas to be tested and receive the transmitted light signal to obtain raw spectral data in the range of 200-400nm; simultaneously, the quantum dot coating is irradiated with ultraviolet excitation light and the fluorescence signal is collected by a photomultiplier tube. The multimodal spectral analysis module combines DOAS technology, Stern-Volmer fluorescence quenching principle and LIBS technology to calculate gas concentration and particulate matter elemental composition; The adaptive interference compensation module identifies interference features by improving the convolutional neural network and corrects the gas concentration based on a dynamic compensation algorithm. The particulate matter fingerprinting and pollution source tracing module analyzes the elemental composition of particulate matter based on LIBS technology and matches the pollution source fingerprint database through a feature element ratio decision tree. The multi-source collaborative data fusion module uses the Kriging algorithm to perform spatial interpolation to generate a spatial heat map of pollution distribution, and uses the ARIMA model to predict pollutant concentration and diffusion trend. The carbon flow rate calculation module calculates real-time carbon flow rate and cumulative emissions based on a dual-source fusion model of coal consumption inversion and measured CO2. The digital twin maintenance module combines physical wear and tear models with LSTM neural networks to predict the remaining lifespan of equipment and output maintenance priorities. The blockchain evidence storage module is used to ensure the immutability of data. It achieves encrypted evidence storage of monitoring data by improving the PBFT consensus algorithm and data signature technology. The system control and communication module adjusts the opening of the ammonia injection valve based on the PID algorithm and triggers multi-level alarms according to the level of abnormality.

[0007] Preferably, the multimodal spectral analysis module includes: The gas concentration analysis unit calculates SO2 and NO based on DOAS technology and Beer-Lambert's law. x Original concentrations of O2 and CO2; NH3 calibration unit, combined with humidity correction Formula for calculating the corrected concentration of NH3; The first LIBS analysis unit uses a laser to bombard particulate matter and collects plasma emission spectra to calculate the elemental composition of the particulate matter.

[0008] Preferably, the adaptive interference compensation module includes: The interference identification unit uses a CNN+DNN combined model to extract interference features and outputs the interference confidence score. The concentration correction unit dynamically corrects gas concentration based on the extended Beer-Lambert law. The transfer learning unit invokes the pre-trained model for transfer learning and triggers online incremental learning when the interference recognition success rate is less than a preset threshold.

[0009] Preferably, the particulate matter fingerprint and pollution source tracing module includes: The second LIBS analysis unit measures the plasma emission spectrum from 200 to 800 nm by bombarding particulate matter with a 50 mJ pulsed laser. The fingerprint matching unit compares the measured element ratio with the pollution source fingerprint database based on the feature element ratio decision tree, and calculates the matching degree using cosine similarity.

[0010] Preferably, the multi-source collaborative data fusion module includes: Spatial interpolation units are used to generate a global pollution distribution heat map using the Kriging algorithm; The time series prediction unit uses the ARIMA model to predict the concentration in the next 1 hour and the diffusion trend in the next 72 hours.

[0011] Preferably, the carbon flow rate calculation module includes: The coal consumption inversion unit calculates the coal consumption and inverts the carbon flow rate based on boiler operating parameters. The actual measurement calculation unit calculates the actual carbon flow rate based on the flue gas flow rate and CO2 concentration; The fusion unit uses a dynamic weighted fusion algorithm to output the final carbon flow rate.

[0012] Preferably, the digital twin maintenance module includes: The life prediction unit combines a physical loss model and an LSTM neural network to predict the remaining life of a component. The health assessment unit outputs equipment maintenance priorities based on health indicator formulas.

[0013] Preferably, the blockchain evidence storage module includes: The signature unit generates a data digest using the SHA3-256 hash algorithm and a data signature using ECDSA signature technology. The consensus unit employs an improved PBFT algorithm to achieve rapid consensus within a 5-node regional consensus group. The system control and communication module includes: The intelligent control unit adjusts the opening degree of the ammonia injection valve based on the PID algorithm; The alarm unit triggers audible and visual alarms, SMS notifications, or platform push notifications based on the level of abnormality.

[0014] To address the aforementioned technical problems, this application also provides a continuous emission monitoring device for flue gas, comprising a processor and a memory. The processor is used to process instructions stored in the memory to implement the functions of the continuous emission monitoring system for flue gas described above. The memory stores program instructions for performing data acquisition, flue gas pretreatment, ultraviolet spectral acquisition, multimodal spectral analysis, adaptive interference compensation, particulate matter fingerprinting and pollution source tracing, multi-source collaborative data fusion, carbon flow rate calculation, digital twin maintenance, blockchain evidence storage, and system control and communication.

[0015] Preferably, the continuous emission monitoring device further includes a sampling component, a preprocessing component, a spectral detection component, a data transmission component, and an execution component connected to the processor. The sampling component is used to collect raw flue gas in the flue. The preprocessing component is used to perform dust removal and dehumidification treatment on the raw flue gas. The spectral detection component is used to acquire the absorption spectrum and fluorescence intensity data of the flue gas. The data transmission component is used to realize data interaction between the device and external equipment. The execution component is used to perform ammonia injection valve adjustment and alarm operations according to the instructions of the processor.

[0016] The beneficial effects of this invention are as follows: This application provides a continuous emission monitoring system for flue gas. Multimodal fusion improves monitoring accuracy. By using quantum dot-enhanced DOAS technology, Stern-Volmer fluorescence quenching, and LIBS technology for synergistic analysis, it achieves simultaneous detection of gas concentration and particulate matter elemental composition. The introduction of a CdSe / ZnS quantum dot coating enhances the detection effect of NH3. At the same time, through humidity correction and other means, it significantly improves the accuracy of gas concentration and particulate matter elemental analysis, especially the detection accuracy of NH3 in complex flue gas environments, and solves the problem of large monitoring errors in traditional single technology.

[0017] The system uses an improved CNN+DNN combined model to identify interference features. By combining transfer learning and online incremental learning, it can quickly adapt to new types of interference, dynamically update the model, effectively eliminate interference from unknown gases, and achieve more accurate measurement errors. It solves the problem of interference from complex flue gas components and ensures the reliability of monitoring data.

[0018] This system integrates multiple functions, including pollution source tracing, multi-source data fusion prediction, carbon flow rate calculation, digital twin maintenance, blockchain evidence storage, and intelligent control alarms. It combines physical models with LSTM neural networks to predict the remaining lifespan of equipment and generate maintenance plans, reducing downtime. It generates pollution distribution heat maps through Kriging interpolation and combines ARIMA models to predict 72-hour diffusion trends, providing decision support for precise pollution control. The dual-source carbon flow rate fusion model dynamically adapts to boiler load, improving the accuracy of carbon emission calculations. It achieves full-process automation and intelligence from flue gas monitoring to data processing, source tracing analysis, carbon emission management, equipment maintenance, and data evidence storage. Attached Figure Description

[0019] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of the continuous monitoring method for flue gas emissions provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the continuous emission monitoring device for flue gas provided in the embodiments of this application. Detailed Implementation

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0024] A continuous emission monitoring system for flue gas includes a data acquisition module, a flue gas pretreatment module, an ultraviolet spectral acquisition module, a multimodal spectral analysis module, an adaptive interference compensation module, a particulate matter fingerprinting and pollution source tracing module, a multi-source collaborative data fusion module, a carbon flow rate calculation module, a digital twin maintenance module, a blockchain evidence storage module, and a system control and communication module. Among them: 100. Data acquisition module, used to collect raw flue gas in the flue and simultaneously monitor the sampling flow rate, adopts an in-situ extraction sampling probe with CdSe / ZnS quantum dot coating for specific adsorption of NH3.

[0025] In yet another embodiment, the flue gas collection module includes: The sampling probe unit adopts in-situ extraction sampling and integrates a CdSe / ZnS quantum dot coating for specific adsorption of NH3, preparing for fluorescence detection; CdSe / ZnS is a core-shell semiconductor nanocrystal, with CdSe (cadmium selenide) as the core and ZnS (zinc sulfide) as the outer shell. CdSe determines the optical properties of the quantum dots, such as the fluorescence emission wavelength. The fluorescence wavelength can be changed by adjusting its particle size; in this embodiment, it is adapted to 365nm ultraviolet excitation. ZnS serves as a protective layer, which can improve the chemical stability and fluorescence quantum yield of the quantum dots, reduce nonradiative transition losses, and extend their lifespan in high-temperature and high-humidity flue gas environments. CdSe / ZnS quantum dots are chemically modified, such as by grafting amino functional groups onto the surface, and fixed onto the inner surface of the sampling probe to form a uniform thin film. This modification endows them with the ability to specifically adsorb NH3, a polar molecule that can form hydrogen bonds with the functional groups on the quantum dot surface.

[0026] The fluorescence quantum yield of quantum dots is far higher than that of traditional Rhodane lamp organic fluorescent dyes. They still produce detectable fluorescence changes even at low concentrations of NH3, exhibiting high sensitivity and meeting the trace monitoring requirements for NH3 emissions in environmental standards. Through surface modification, they selectively adsorb only NH3, reducing interference from other gases in the flue gas and demonstrating high specificity, thus solving the problem of cross-interference in traditional spectroscopic methods. The ZnS shell protects the CdSe core from corrosion by acidic gases and particulate matter in the flue gas, and it remains stable even under heating conditions at 180℃, demonstrating good stability.

[0027] The flow monitoring unit is used to measure the sampled flow rate in real time with an accuracy of ±2%.

[0028] 200. Flue gas pretreatment module, used to remove dust and moisture from raw flue gas, stabilize its temperature, pressure and flow rate, and output clean and stable test gas and fluorescence signal excited by quantum dot coating.

[0029] In yet another embodiment, the flue gas pretreatment module includes: The heating and heat tracing unit is based on the flue gas dew point T. dp Calculate the heat tracing temperature T h T h =T dp +40℃, where T dpThe dew point of the flue gas is measured in real time by a dew point sensor, and a constant temperature control is adopted using a 180℃ heating and heat tracing pipeline to prevent acid gas condensation from corroding the equipment. The graded filtration unit uses a gradient filter element from 5μm to 1μm to remove dust particles of different sizes in stages. The backflush unit regularly cleans the dust accumulated in the graded filter unit. When the filter element pressure difference is >2000Pa, pulse backflush is performed at a pressure of 0.6MPa and an interval of 0.2s.

[0030] 300. Ultraviolet spectral acquisition module, used to emit ultraviolet light to the gas to be tested and receive the transmitted light signal to acquire raw spectral data in the range of 200-400nm.

[0031] In yet another embodiment, the ultraviolet spectral acquisition module includes: The ultraviolet light source unit emits 200-400nm ultraviolet light to irradiate the gas to be tested; The spectral acquisition unit acquires the absorption spectrum through a 2048-pixel CCD spectrometer, simultaneously irradiates the quantum dot coating with 365nm ultraviolet excitation light, and collects the fluorescence signal by a photomultiplier tube.

[0032] The gas to be tested was irradiated with a 200-400 nm ultraviolet light source, and its absorption spectrum was acquired using a 2048-pixel CCD spectrometer. Simultaneously, a 365 nm ultraviolet excitation light was applied to the CdSe / ZnS quantum dot coating, which enhanced the characteristic absorption of NH3. Fluorescence signals were acquired using a photomultiplier tube for NH3 concentration calibration. The raw spectral data obtained after acquisition included absorption spectra and fluorescence intensity data in the 200-400 nm range, spaced 0.1 nm apart.

[0033] 400. Multimodal spectral analysis module, which combines DOAS technology, Stern-Volmer fluorescence quenching principle and LIBS technology to calculate gas concentration and particulate matter elemental composition, and outputs the original gas concentration, NH3 corrected concentration and particulate matter elemental percentage.

[0034] As another embodiment, the multimodal spectral analysis module includes: The gas concentration analysis unit calculates SO2 and NO based on DOAS technology and Beer-Lambert's law. x Original concentrations of O2 and CO2; The NH3 calibration unit calculates the corrected NH3 concentration using the humidity-corrected Stern-Volmer formula and outputs a high-precision NH3 concentration. The first LIBS analysis unit uses an Nd:YAG laser to bombard particulate matter and collects plasma emission spectra to calculate the proportions of elements such as Fe, Al, Ca, Mg, Zn, and Pb.

[0035] The processing method of the multimodal spectral analysis module is as follows: 401. Quantum dot enhanced DOAS.

[0036] Using DOAS technology, the initial concentration of the gas was calculated based on Beer-Lambert's law. The formula is:

[0037] A(λ) is the absorbance at wavelength λ, measured by a spectrometer; ε(λ) is the molar absorptivity of the target gas, in units of... The values ​​are derived from the NIST standard spectral library and constructed through pure gas calibration experiments; C is the target gas concentration in mol / L; L is the optical path length in cm, a fixed hardware value; A0(λ) is the background absorption, obtained through zero-gas calibration. SO2 and NO can be calculated based on the above formula. x The initial concentrations C of O2 and CO2 gases. 402. Quantum dot-enhanced NH3 detection.

[0038] Based on the Stern-Volmer fluorescence quenching principle, quantum dots are excited by 365nm ultraviolet light to produce fluorescence, which is quenched by NH3. A high humidity correction factor K is introduced. x Modified Stern-Volmer equation, formula:

[0039] Where I0 is the fluorescence intensity without NH3, in μW, factory calibrated; I is the fluorescence intensity with NH3, in μW, measured in real time by the photomultiplier tube; K s =0.118 L / mol is the quenching constant, obtained through calibration with pure NH3 standard gas; K x As a high humidity correction factor, RH stands for relative humidity, which is measured by a temperature and humidity sensor.

[0040] The corrected NH3 concentration was calculated based on the above formula. Unit: ppm, accuracy: ±0.1ppm.

[0041] 403.LIBS particulate matter analysis.

[0042] Particulate matter was bombarded with an Nd:YAG laser, and plasma emission spectra in the 200-800 nm range were collected by an ICCD spectrometer to calculate the proportions of elements such as Fe, Al, and Ca.

[0043] Quantitative formula for elements:

[0044] Where I is the intensity of the elemental characteristic peak; C0 is the elemental concentration; k and b are constants, which are determined by establishing a calibration curve using NIST standard dust samples and then by multiple linear regression.

[0045] The concentrations C0 of particulate matter elements such as Fe, Al, Ca, Mg, Zn, and Pb can be calculated using the above formula.

[0046] 500. Adaptive Interference Compensation Module: This module identifies interference features by improving convolutional neural networks, eliminates interference from unknown gases, and corrects gas concentrations based on a dynamic compensation algorithm.

[0047] In yet another embodiment, the interference compensation module includes: The interference identification unit uses a CNN+DNN combined model to extract interference features and outputs the interference confidence score. The concentration correction unit, based on the extended Beer-Lambert law, corrects the gas concentration through a dynamic compensation algorithm to ensure an error of less than ±2%. The transfer learning unit calls the pre-trained model for transfer learning and triggers online incremental learning when the interference recognition success rate is less than the preset success rate threshold. The preset success rate threshold is 95% by default.

[0048] The adaptive interference compensation module processes interference as follows: 501. Interference Feature Extraction.

[0049] 5011. An improved CNN+DNN combined model is adopted: CNN (Convolutional Neural Network) identifies the initial outline of interference features, and DNN (Deep Neural Network) deeply mines complex patterns.

[0050] A CNN (Neural Network Array) model is a deep learning model specifically designed for processing data with a grid structure, including images and audio spectrograms. In this system, the CNN model is used to identify the initial contours of interfering features, and its construction process is as follows: Data preparation: Collect a large amount of monitoring data containing various interference conditions. This data can be spectral data, sensor signal data, etc., and divide it into training set, validation set and test set.

[0051] Network architecture: It consists of multiple convolutional layers, pooling layers, and fully connected layers. Convolutional layers perform convolution operations by sliding convolution kernels across the data, extracting local features. For example, for spectral data, convolutional kernels can capture feature variation patterns within a specific wavelength range. Pooling layers reduce data dimensionality while preserving key features; for instance, max pooling selects the maximum value in the convolutional data block as the output. Fully connected layers integrate the feature vectors processed by convolution and pooling, outputting the final classification result.

[0052] Training: The CNN model is trained using the training set data. The model's parameters, such as the weights of the convolutional kernels, are continuously adjusted through backpropagation to minimize the error between the model's predictions on the training set and the true labels. During training, the model gradually learns the preliminary contour patterns of interference features. For example, a certain type of interference may exhibit specific intensity variation patterns in certain bands of the spectrum, and the CNN model can capture these patterns.

[0053] CNN models identify local interference features, such as abnormal absorption peaks and noise patterns, through convolutional layers; they output a 256-dimensional feature vector (containing preliminary contour information of the interference) to the DNN model. For example, if a non-standard absorption peak appears in the spectrum at 290nm, the CNN marks it as "suspected interference".

[0054] A DNN model is a neural network containing multiple hidden layers. Fully connected layers are multi-layered neuron structures that take features extracted by a CNN as input and output interference classification results. In a fully connected layer, each neuron is connected to all neurons in the layer above it. In this system, the fully connected layers of the DNN are used to deeply mine complex patterns. After the CNN model identifies the initial outlines of interference features, these features are input into the fully connected layers of the DNN. The fully connected layers perform complex weighted combinations and nonlinear transformations on these features to uncover deeper and more complex interference patterns. Examples include combined features between different types of interference and patterns of interference feature changes over time.

[0055] The fully connected layer of the DNN model integrates nonlinear relationships; the output layer has two neurons, corresponding to the "no interference" and "interference" categories.

[0056] The output is converted into a probability using the Softmax function, and the interference confidence level I is obtained. conf : I conf =P(with interference)=e z干扰 / (e z无干扰 +e z干扰 ) e z干扰 e represents the raw score of the DNN for the "interference" class; z无干扰 This is the raw score of the DNN for the "no interference" class.

[0057] 5012. Basic Model Pre-training: Trained based on a general VOCs dataset, it has the ability to identify general interference. The general dataset contains the spectra of 10,000+ volatile organic compounds.

[0058] A base model refers to a neural network model pre-trained on a large-scale, general-purpose dataset, such as ResNet and VGG models commonly used in image recognition, and BERT models in natural language processing. In this system, a pre-trained model related to spectral data processing or interference recognition is selected as the base. The purpose of training this base model is to enable it to learn general features and patterns. The pre-training process is conducted on large-scale, general-purpose data, such as a large amount of spectral data collected under different environments, which includes various possible interference and normal signal conditions. By training on this data, some basic feature representations can be learned, such as the combination of different spectral features and the general patterns of common interferences.

[0059] 5013. Transfer Learning Trigger: When the interference recognition success rate is <95% (such as when encountering unknown gases), the pre-trained model is invoked for transfer learning to quickly adapt to new interference types.

[0060] The interference identification success rate is usually calculated by the evaluation module in the system. The calculation method is as follows: on the test set, the number of samples in which the model correctly identifies interference is divided by the total number of interference samples in the test set, and then multiplied by 100%. Correct identification means that the interference type predicted by the model is consistent with the actual labeled interference type.

[0061] When a new type of interference emerges, it can be initially identified manually by observing the differences between the newly appearing interference and known interference types. Alternatively, it can be determined using anomaly detection algorithms, such as statistical anomaly detection or machine learning-based isolated forest algorithms. If certain statistical characteristics or patterns in the monitored data differ significantly from known normal and interference patterns, a new type of interference is suspected. For the new interference type, transfer learning is used. The parameters obtained from the pre-training of the base model are used as initial values, and then fine-tuned on a training set containing the new interference type. The training set serves as the dataset for fine-tuning the base model. By performing optimization algorithms such as backpropagation on this training set, the model's parameters are adjusted, enabling the model to quickly adapt to the new interference type.

[0062] If the interference identification success rate is less than 95%, the transfer learning module is activated, the original data is retained, and an alarm is triggered; the manually calibrated data is added to the training set to trigger online incremental learning.

[0063] 5014. Online Incremental Learning: New gas data (spectrum + concentration) after manual calibration are dynamically added to the training set, and the parameters of the DNN fully connected layer are updated by mini-batch gradient descent (learning rate 0.001, updated once every 100 sets of data) to continuously optimize the model.

[0064] 502. Concentration Correction.

[0065] Concentration C of interfering gas i predicted by neural network output inter,i : C inter,i = f NN (Spectral data, standard spectral library) Among them, f NN is the CNN+DNN combined model function, and the interfering gas i includes SO2, NO x , O2, CO2.

[0066] Based on the extended formula of Beer-Lambert law:

[0067] The original concentration C output by DOAS is corrected to obtain the intermediate concentration C corr , formula:

[0068] Among them, k i is the interference coefficient, with a value range of 0-1, output by the neural network; C is the original concentration measured by DOAS; C inter,i is the concentration of the interfering gas predicted by the neural network.

[0069] The intermediate concentration C corr is dynamically compensated to obtain the corrected concentration C curr , formula:

[0070] Among them: β is the compensation coefficient, obtained by experimental fitting; I conf is the interference confidence level, with a value range of 0-1, output by CNN.

[0071] If I conf > 0.8, trigger the verification of the standard gas and automatically inject the standard gas; And 0.2 ≤ I conf ≤ 0.8, the system executes the interference compensation process; If I<9000032>< 0.2, directly output C curr , to avoid overcorrection; If the verification error > ±3%, recalibrate the DNN model.

[0072] 600. The particulate matter fingerprint and pollution source tracing module analyzes the elemental composition of particulate matter based on LIBS technology and matches the pollution source fingerprint library through the characteristic element ratio decision tree.

[0073] As another embodiment, the particulate matter fingerprint and pollution source tracing module includes: The second LIBS analysis unit bombards the particulate matter with a 50mJ pulsed laser and measures the plasma emission spectrum of 200-800nm; The fingerprint matching unit compares the measured element ratio with the pollution source fingerprint database based on the feature element ratio decision tree, and calculates the matching degree using cosine similarity.

[0074] The processing method for particulate matter fingerprints and pollution source tracing modules is as follows: LIBS technology analysis: 50mJ pulsed laser bombards particulate matter, ionizing it into plasma. The spectrum is measured (2048 channels, 200-800nm) to obtain the intensity of characteristic peaks of elements such as Fe, Ca, and Pb.

[0075] Fingerprint matching: Based on the feature element ratio decision tree, the measured element ratio is compared with the pollution source fingerprint database to determine the pollution source type and proportion.

[0076] Pollution sources refer to the types of industrial facilities that emit particulate matter, including coal-fired power plants, industrial boilers, and vehicle exhaust emissions.

[0077] The feature element ratio decision tree is constructed by analyzing a large number of particulate matter samples emitted from known pollution sources. The construction method is as follows: collect particulate matter samples from different pollution sources, and use multimodal spectral analysis modules and other means to determine the content of various elements in these samples; then, calculate the ratio between different elements, use these element ratios as features, and construct a model using the decision tree algorithm; the decision tree algorithm classifies pollution source samples according to these features, and continuously selects features that can distinguish different pollution source categories to the greatest extent for splitting, and finally constructs a decision tree.

[0078] The pollution source fingerprint database is constructed through long-term accumulation and analysis of particulate matter characteristics emitted from different pollution sources, including the proportions of typical elements from industries such as thermal power and steel. The matching method involves comparing the measured elemental ratios with the fingerprint database and calculating the matching degree using cosine similarity. If the matching degree is <70%, the source is marked as an unknown pollution source, and the data is uploaded to the system to update the fingerprint database.

[0079] By comparing elemental characteristics, the industry type of the emission source can be deduced to achieve pollution source tracing. The output shows the pollution source classification results, such as steel sources accounting for 85%.

[0080] 700. Multi-source collaborative data fusion module: uses the Kriging algorithm for spatial interpolation to generate a pollution distribution heat map, and uses the ARIMA model to predict pollutant concentration and diffusion trend.

[0081] As another embodiment, the multi-source collaborative data fusion module includes: Spatial interpolation units are used to generate a global pollution distribution heat map using the Kriging algorithm; The time series prediction unit uses the ARIMA model to predict the concentration in the next 1 hour and the diffusion trend in the next 72 hours.

[0082] The processing method of the multi-source collaborative data fusion module is as follows: 701. Spatial interpolation.

[0083] Using the Kriging algorithm, the semi-mutation function model is as follows:

[0084] Where: γ(h) is the semivariogram value, reflecting the degree of variation between two points with a spatial distance of h; h is the sample point spacing; N is the number of sample point pairs with a spacing of h; z(x i ) and z(x i +h) represent the spatial location x i and x i The pollutant concentration monitoring value at +h, z(x) i )=C curr .

[0085] Using the interpolation formula:

[0086] The estimated value of the interpolation point z(x0) is calculated; where: x0 is the interpolation point; λ i Given point x i Weights for x0; z(x i ) represents the known sample point monitoring value; n represents the number of known points participating in the interpolation.

[0087] Based on the interpolation results, a spatial heat map is generated to visualize the pollution distribution across the entire region.

[0088] The method for generating space heat maps is as follows: Data gridding: The continuously distributed data obtained by interpolation is gridded according to a certain grid resolution, and the interpolated predicted value at the corresponding position of each grid cell is determined.

[0089] Color mapping and visualization rendering: Based on the monitored variable values ​​of the grid cells, i.e., pollutant concentrations, color mapping rules are set. For example, areas with low concentrations are represented by blue, and the color gradually transitions to yellow, orange, red, etc., as the concentration increases. Then, using a visualization library, the gridded data is rendered into a spatial heatmap according to the color mapping rules. Different colored areas intuitively reflect the spatial distribution differences of the monitored variables; for example, red areas represent areas with high pollutant concentrations, and blue areas represent areas with low concentrations.

[0090] 702. Time series forecasting.

[0091] The ARIMA model is used for modeling and predicting time series data. Its model structure is ARIMA(p,d,q), where p is the autoregressive order (determined by the ACF plot), d is the differencing order, and q is the moving average order (determined by the PACF plot). In the ARIMA model, the original pollutant concentration time series data is set as z. t , representing the actual monitored value of pollutant concentration at time t, z t For C in the historical database curr , z t-1 The concentration at a historical moment ti.

[0092] The prediction equation is:

[0093] in, It is a difference sequence of order d; c is a constant term. For autoregressive coefficients, The moving average coefficients are fitted using maximum likelihood estimation. The error term is normally distributed.

[0094] Collect historical monitoring data of pollutant concentrations (z1, z2, ..., z T (T is the total number of historical observation times), and by performing a d-th order difference, we obtain Sequence (t=d+1,d+2…,T); using these differencing data, combined with the maximum likelihood estimation method, determine c, , Including the distribution parameters of the random error term, the ARIMA(p,d,q) model is constructed.

[0095] 703. Prediction of pollutant concentration and diffusion trends.

[0096] 7031. Concentration forecast for the next hour: Given the current time t, to predict the pollutant concentration for the next hour, calculate... The predicted value is denoted as .

[0097]

[0098] Where c is a constant term, The data after differentiation at time t+1-i; This represents the predicted value of the error term at the corresponding time point.

[0099] get Then, by using the inverse difference operation, the predicted value of the pollutant concentration at time t+1 is calculated. This refers to the predicted concentration value for the next hour.

[0100] 7032. Forecast of pollutant diffusion trends in the next 72 hours: Repeat the steps for predicting the next hour as described above, and calculate the predicted pollutant concentration values ​​at times (t+2, t+3, ..., t+72) in sequence; By organizing and analyzing these predicted concentration values ​​for 72 consecutive future moments, such as plotting concentration curves over time to observe whether the concentration is rising, falling, or fluctuating, as well as the rate and magnitude of change, the pollutant diffusion trend over the next 72 hours can be presented. Furthermore, by combining spatial interpolation, meteorological data, and other auxiliary information, the spatial diffusion trend of pollutants can be described more accurately, such as the differences in concentration changes across different regions.

[0101] It should be noted that by obtaining the concentration change trend over time from the perspective of the ARIMA model and then combining it with spatial dimension processing, the diffusion trend of pollutants in space and time can be fully presented.

[0102] 800. Carbon Flow Rate Calculation Module: This module calculates the carbon emission rate using a dual-source fusion model that combines coal consumption inversion with measured CO2, outputting real-time carbon flow rate and cumulative emissions.

[0103] In yet another embodiment, the carbon flow rate calculation module includes: The coal consumption inversion unit calculates coal consumption and inverts carbon flow rate based on boiler operating parameters. The actual measurement calculation unit calculates the actual carbon flow rate based on the flue gas flow rate and CO2 concentration; The fusion unit uses a weighted fusion algorithm to output the final carbon flow rate.

[0104] The processing method of the carbon flow rate calculation module is as follows: 801. Coal consumption inversion carbon flow rate F c1 According to the formula:

[0105] The carbon flow rate based on coal consumption is calculated in kgCO2 / h, where: M coal Coal consumption, unit: kg / h; C coal The carbon content of coal is expressed as 0.55 kg C / kg coal, obtained from industrial analysis; 44 / 12 is the molar mass ratio of CO2 to C.

[0106] 802. Measured CO2 carbon flow rate F c2 According to the formula:

[0107] The carbon flow rate based on measured CO2 is calculated, in units of kgCO2 / h, where: Q std Flue gas flow rate under standard conditions, unit C CO2CO2 volume concentration, expressed as a percentage; 1.96 × 10⁻⁶ -3 For % to The conversion factor.

[0108] 803. Fusion carbon flow rate F c : The dynamic fusion coefficient k is:

[0109] Where tanh is the hyperbolic tangent function, ensuring that k is between 0.15 and 0.25: when the load D > 500 t / h, k ≈ 0.15, and when the load D < 300 t / h, k ≈ 0.25.

[0110] According to the fusion formula:

[0111] The real-time carbon flow rate F was calculated. c , unit kgCO2 / h.

[0112] Cumulative carbon emissions are the result of integrating the real-time carbon flow rate over time.

[0113] in, Let be the real-time carbon flow rate at time t. Let n be the time interval, and n be the number of intervals corresponding to the cumulative duration.

[0114] If DCS communication is interrupted, the LSTM model is activated to estimate M based on the coal consumption data of the past 24 hours. coal After communication is restored, correct the data using actual figures. When the boiler is shut down, the cumulative emissions are frozen; they will resume accumulating after restarting.

[0115] 900. Digital Twin Maintenance Module: This module predicts the remaining lifespan of equipment and outputs maintenance priorities by combining a physical wear and tear model with an LSTM neural network.

[0116] As another embodiment, the digital twin maintenance module includes: The life prediction unit combines a physical loss model and an LSTM neural network to predict the remaining life of a component. The health assessment unit outputs equipment maintenance priorities based on health indicator formulas.

[0117] Taking the light source as an example, the processing method of the digital twin maintenance module is as follows; 901. Based on the physical loss model:

[0118] The remaining lifespan (RUL) of the light source was calculated. light , in hours, where: t max=10000 hours, theoretical lifespan; I0=1000μW, initial intensity; T0=25℃, standard temperature; k1=0.8, loss coefficient, fitted with 100+ light source lifespan data.

[0119] 902. Calculate the final remaining lifetime (RUL) using an LSTM neural network for correction. final The remaining lifespan (RUL) of the light source was calculated. light , in hours, where: t max =10000 hours, theoretical lifespan; I0=1000μW, initial intensity; T0=25℃, standard temperature; k1=0.8, loss coefficient, fitted with 100+ light source lifespan data.

[0120] 902. Calculate the final remaining lifetime (RUL) using an LSTM neural network for correction. final :

[0121] Among them, RUL LSTM The output value is the LSTM neural network value, which is output by learning the relationship between historical lifetimes and real-time parameters.

[0122] 903. Health Indicators: Based on the formula:

[0123] The health index HI is calculated, with a value ranging from 0 to 100. A lower value indicates a more severe degree of equipment degradation. Where: I is the real-time light source intensity, I0 is the initial light source intensity, and R... decay denoted as the quantum dot fluorescence intensity decay rate, T as the real-time temperature of the gas cell, T0 as the standard operating temperature, and w1, w2, and w3 as weighting coefficients.

[0124] Based on the remaining life (RUL) of each component final Based on the health indicator HI, determine the maintenance priority: High priority: RUL final <100 hours or HI<40; triggers an emergency alarm, prompting shutdown for maintenance or switching to backup components; Medium priority: 100≤RUL final ≤1000 hours or 40≤HI≤70; trigger an alert and generate a maintenance plan; Low priority: RUL final >1000 hours or HI>70; Normal operation, daily status report pushed.

[0125] 1000. Blockchain evidence storage module, used to ensure the immutability of data, achieves encrypted evidence storage of monitoring data by improving the PBFT consensus algorithm and data signature technology.

[0126] As another embodiment, the blockchain evidence storage module includes: a signature unit that generates a data digest using the SHA3-256 hash algorithm and generates a data signature using ECDSA signature technology; The consensus unit employs an improved PBFT algorithm to achieve rapid consensus within a 5-node regional consensus group.

[0127] The processing method of the blockchain evidence storage module is as follows: 1001. Data Signature: The data digest is generated using the SHA3-256 hash algorithm, and the digest is signed with a private key using ECDSA signature technology to generate a data signature, ensuring data integrity.

[0128] 1002. Consensus Mechanism: Improved PBFT consensus algorithm, adopting a three-phase protocol: Request → Pre-Prepare → Prepare → Commit. Dynamic node management, achieving consensus time T. consensus :

[0129] Where: t base The base time; k is the coefficient; N node This specifies the number of nodes; the default is 5.

[0130] 1003. Evidence Storage: Anchored to the environmental monitoring chain every 10 minutes.

[0131] The evidence storage and processing ensures data immutability and meets environmental regulatory traceability requirements. It outputs local database files, cloud-synchronized data, and blockchain-encrypted data packets.

[0132] 1100. System control and communication module, used for intelligent control and alarm, adjusts the opening of the ammonia injection valve based on PID algorithm, and triggers multi-level alarms according to the level of abnormality.

[0133] In yet another embodiment, the system control and communication module includes: The intelligent control unit adjusts the opening of the ammonia injection valve based on the PID (proportional-integral) algorithm, with a response time of less than 20 seconds. The alarm unit triggers audible and visual alarms, SMS notifications, or platform push notifications based on the level of abnormality.

[0134] The ammonia injection valve is a key component in a flue gas denitrification system. Its core function is to remove nitrogen oxides from the flue gas by controlling the amount of ammonia injected, i.e., by adjusting the opening degree to control the flow rate of ammonia into the flue. In denitrification processes such as selective catalytic reduction, ammonia acts as a reducing agent to react with NO. x The reaction, under the action of a catalyst, produces harmless nitrogen and water, thereby reducing NO levels. x Emission concentration.

[0135] The opening degree of the ammonia injection valve directly determines the amount of ammonia injected: the larger the opening degree, the more ammonia is injected. x Higher removal efficiency may lead to excessive ammonia escape, causing secondary pollution; a smaller opening results in insufficient ammonia injection, leading to higher NO levels. x Emissions may exceed standards. Therefore, precise control of the ammonia injection valve opening is crucial for balancing denitrification efficiency and preventing secondary pollution; its goal is to reduce NO emissions. x The concentration was kept stable at the target value.

[0136] The processing method of the system control and communication module is as follows: 1101. Intelligent ammonia injection control: based on NH3 and NO x Real-time concentration, the opening of the ammonia injection valve is adjusted through a PID optimization algorithm:

[0137] Where: V NH3 Ammonia injection valve opening degree (%); K p K i The proportional / integral coefficient is determined through experimental debugging; NO x target =50mg / m³, target concentration; NO x For real-time monitoring of NO x Concentration, from the output C of the adaptive interference compensation module. curr .

[0138] 1102. Multi-level alarm: Minor abnormalities, such as RUL final =500 hours, alarm method: audible and visual alarm + SMS notification to maintenance personnel; if the abnormality is not resolved within 30 minutes, the alarm level will be automatically upgraded.

[0139] In case of serious anomalies, such as pollutant concentration exceeding the standard by 30%, quantum dot fluorescence intensity decreasing by more than 30% requiring coating replacement, or light source damage, the alarm will be triggered by linking the enterprise's DCS system, prompting a reduction in production load, and uploading the information to the environmental protection platform.

[0140] The overall system process steps are as follows: Step 1. Sampling and pretreatment: Collect raw flue gas from the flue and remove interfering substances such as dust and moisture.

[0141] In-situ extraction sampling, the probe integrates 180℃ heating and quantum dot coating, and the flue gas is purified through five-stage gradient filtration to provide clean samples for subsequent analysis and reduce interference.

[0142] Step 2. Multimodal spectral analysis: Simultaneously measure the gas concentration and particulate element composition in the flue gas.

[0143] The study combines DOAS and quantum dot enhancement technologies to measure gas concentrations; analyzes particulate matter elements using LIBS technology; and obtains basic pollutant data.

[0144] Step 3. Adaptive Interference Compensation: Identify and eliminate the influence of interfering gases, and correct the target gas concentration.

[0145] The CNN+DNN model is used to extract interference features; transfer learning is introduced to identify unknown interference, and the model is dynamically updated through online incremental learning; concentration is corrected based on the extended Beer-Lambert law; and the measurement error is controlled within ±2%, thus solving the accuracy problem caused by complex / unknown interference.

[0146] Step 4. Particulate matter fingerprinting and pollution source tracing: Analyze the elemental composition of particulate matter and match it with the characteristics of pollution sources.

[0147] By using LIBS technology to obtain the characteristic peaks of particulate matter elements and comparing them with a pollution source fingerprint database, the source of pollutants can be identified, providing a basis for precise governance.

[0148] Step 5. Multi-source collaborative data fusion: Integrate data from multiple monitoring points to predict pollutant diffusion trends.

[0149] The Kriging algorithm is used for spatial interpolation; the ARIMA model is used to predict the diffusion trend over the next 72 hours and the concentration over 1 hour; thus achieving comprehensive pollution status monitoring and trend early warning. Step 6. Dynamic carbon flow rate calculation: Real-time calculation of carbon emission rate and cumulative amount.

[0150] Based on a dual-source fusion model of coal consumption inversion and measured CO2, dynamic coefficients are used to adapt to boiler load, improving calculation accuracy; providing accurate data for carbon emission management and meeting environmental protection and emission reduction requirements.

[0151] Step 7. Digital Twin Maintenance: Predict the remaining lifespan of equipment components and generate a maintenance plan.

[0152] By combining physical loss models and LSTM neural networks, the final remaining lifespan is calculated to determine maintenance priorities; predictive maintenance is achieved to reduce equipment downtime and ensure stable system operation.

[0153] Step 8. Blockchain Evidence Storage: Store all monitoring data to ensure it cannot be tampered with.

[0154] Employing an improved PBFT consensus algorithm and SHA3-256 signature technology, edge nodes generate signatures, regional nodes verify them, and the environmental protection chain is anchored every 10 minutes; ensuring data authenticity and traceability, and meeting environmental regulatory requirements.

[0155] Step 9. Intelligent Control and Alarm: Monitor system status in real time and trigger control strategies and alarms.

[0156] Based on NO x The concentration is adjusted by a PID algorithm to control the amount of ammonia injected; multiple alarms are triggered according to the level of abnormality; pollutant emissions are controlled in a timely manner, and abnormal situations are responded to quickly to reduce environmental risks.

[0157] 01. This system achieves continuous and accurate monitoring of flue gas emissions by updating the following key parameters in real time: 011. Basic flue gas parameters: The raw flue gas flow rate, temperature, and pressure are collected in real time by the data acquisition module. SO2, NO x The concentrations of pollutants such as CO and particulate matter are output after correction by the ultraviolet spectral acquisition module and the adaptive interference compensation module.

[0158] 012. Dynamic control parameters: The ammonia injection valve opening degree is determined by the system control and communication module based on NO. x Dynamic adjustment of concentration and carbon flow rate calculation results; Ammonia injection volume by zone, and a flue gas zone pollution distribution map based on the multi-source fusion module.

[0159] 013. Carbon Flow and Source Tracing Parameters: The carbon emission rate and carbon flow rate calculation module combines gas concentration, flow rate, and fuel data for real-time calculation. Particulate matter fingerprint features: The particulate matter fingerprint module updates the pollution source identifier database and matches pollution sources.

[0160] 014. Prediction and Maintenance Parameters: Equipment health is assessed by the digital twin module, which updates the equipment wear model using sensor data. The system's abnormal threshold is dynamically adjusted by the adaptive interference compensation module to adjust the interference factor threshold.

[0161] 015. Evidence Preservation and Communication Parameters: Data hash value; the blockchain notarization module generates timestamps for notarizing all key parameters. The system control module updates network latency, command response time, and other parameters related to the communication protocol status.

[0162] 02. Each module forms a closed-loop collaborative system through triple coupling of data flow, control flow, and trust flow, as detailed below: 021. Data Flow: Dynamic Looping Throughout the Entire Link.

[0163] 0211. Data Acquisition, Preprocessing, and Analysis: The data acquisition module acquires the raw flue gas, the flue gas preprocessing module purifies and stabilizes the flow, the ultraviolet spectroscopy module extracts spectral features, the multimodal analysis module analyzes pollutant concentrations, and the adaptive interference compensation module corrects environmental interference, such as temperature and humidity fluctuations, and outputs high-precision data.

[0164] 0212. Integration, Calculation, and Source Tracing: The multi-source fusion module integrates and corrects the concentration data, particulate matter fingerprints, and flow information. The carbon flow rate calculation module calculates the carbon emission rate based on this information. The particulate matter fingerprint module compares the pollution source feature database to pinpoint the emission source, such as steel plants or coal-fired power plants.

[0165] 022. Control flow: the neural center from perception to execution.

[0166] 0221. Decision-making, execution, and feedback: The system control and communication module receives carbon flow rate and NO... x Based on the concentration and digital twin prediction results, the opening command of the ammonia injection valve is dynamically generated, such as precise ammonia adjustment by zone. After execution, the adaptive interference compensation module monitors the ammonia escape rate in real time and feeds it back to the control module to form a closed-loop regulation.

[0167] 0222. Predictive Optimization: The digital twin maintenance module continuously simulates the equipment's operating status, such as the wear trend of the ammonia injection valve, and provides early warnings of faults. Based on this, the system control module adjusts the maintenance plan to avoid downtime risks.

[0168] 023. Trust Flow: Full-lifecycle trust assurance.

[0169] Evidence storage, verification, and traceability: The blockchain evidence storage module generates tamper-proof evidence for all key parameters, including concentration, carbon flow rate, ammonia injection instructions, and equipment status, allowing regulatory agencies to trace data at any time. In the event of an exceedance, the pollution source tracing results from the particulate matter fingerprint module are combined with the blockchain evidence storage to provide a judicial-grade chain of evidence.

[0170] Using the above method, the carbon flow rate calculation module unifies the goals of pollution reduction and carbon reduction, and the ammonia injection valve control no longer relies solely on NO. x Instead of focusing on concentration, it focuses on optimizing carbon emissions collaboratively; digital twin modules upgrade equipment from passive maintenance to predictive self-healing, reducing the risk of monitoring interruptions and breaking down traditional fragmentation. Data collection-analysis-control-evidence storage-prediction-recontrol forms a self-evolving ecosystem, such as blockchain-stored data feeding back into the training of digital twin models, thereby constructing an intelligent closed loop. Shifting from post-event penalties to process assurance: blockchain evidence storage ensures data authenticity, enabling precise source tracing and accountability, forcing enterprises to proactively optimize emission control, and reshaping regulatory logic.

[0171] Based on the same inventive concept as the continuous emission monitoring system for flue gas provided in the embodiments of this application, the embodiments of this application also provide a continuous emission monitoring device for flue gas. If there is anything unclear about the content of the device embodiment, please refer to the corresponding content in the method embodiment.

[0172] A continuous emission monitoring device for flue gas includes a processor 1 and a memory 2. The processor 1 processes instructions stored in the memory 2 to implement the functions of the aforementioned continuous emission monitoring system for flue gas. The memory 2 stores program instructions for performing data acquisition, flue gas pretreatment, ultraviolet spectral acquisition, multimodal spectral analysis, adaptive interference compensation, particulate matter fingerprinting and pollution source tracing, multi-source collaborative data fusion, carbon flow rate calculation, digital twin maintenance, blockchain evidence storage, and system control and communication. The device communicates with a computer via a network interface.

[0173] The device includes a processor 1, a memory 2, a communication interface, a display screen, and an input device connected via a system bus. The processor 1 provides computing and control capabilities. The memory 2 includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad located on the computer device casing, or an external keyboard, touchpad, or mouse.

[0174] It should be noted that processor 1 and memory 2 can be integrated into the computer device, or they can be used as a computing device in conjunction with other computer devices.

[0175] In another embodiment, the continuous emission monitoring device further includes a sampling component 3, a preprocessing component 4, a spectral detection component 5, a data transmission component 6, and an execution component 7 connected to the processor 1. The sampling component 3 is used to collect raw flue gas in the flue. The preprocessing component 4 is used to perform dust removal and dehumidification treatment on the raw flue gas. The spectral detection component 5 is used to acquire the absorption spectrum and fluorescence intensity data of the flue gas. The data transmission component 6 is used to realize data interaction between the device and external equipment. The execution component 7 is used to perform ammonia injection valve adjustment and alarm operations according to the instructions of the processor 1.

[0176] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A continuous emission monitoring system for flue gas, characterized in that, include: The data acquisition module collects raw flue gas and simultaneously monitors the sampling flow rate. It uses an in-situ extraction sampling probe with a CdSe / ZnS quantum dot coating to specifically adsorb NH3. The flue gas pretreatment module removes dust and dehumidifies the raw flue gas, and outputs clean and stable test gas and fluorescence signals excited by the quantum dot coating. The ultraviolet spectral acquisition module emits ultraviolet light to the gas to be tested and receives the transmitted light signal to obtain the raw spectral data; simultaneously, it irradiates the quantum dot coating with ultraviolet excitation light to collect the fluorescence signal. The multimodal spectral analysis module combines DOAS technology, Stern-Volmer fluorescence quenching principle and LIBS technology to calculate gas concentration and particulate matter elemental composition; The adaptive interference compensation module identifies interference features by improving the convolutional neural network and corrects the gas concentration based on a dynamic compensation algorithm. The particulate matter fingerprinting and pollution source tracing module analyzes the elemental composition of particulate matter based on LIBS technology and matches the pollution source fingerprint database through a feature element ratio decision tree. The multi-source collaborative data fusion module uses Kriging spatial interpolation and ARIMA model to predict pollution diffusion trends. The system control and communication module adjusts the opening of the ammonia injection valve based on the PID algorithm and triggers multi-level alarms according to the level of abnormality.

2. The system according to claim 1, characterized in that, Also includes: The carbon flow rate calculation module calculates real-time carbon flow rate and cumulative emissions based on a dual-source fusion model of coal consumption inversion and measured CO2. The digital twin maintenance module combines physical wear and tear models with LSTM neural networks to predict the remaining lifespan of equipment and output maintenance priorities. The blockchain evidence storage module achieves encrypted evidence storage of monitoring data by improving the PBFT consensus algorithm and data signature technology.

3. The system according to claim 1, characterized in that, The multimodal spectral analysis module includes: The gas concentration analysis unit calculates SO2 and NO based on DOAS technology and Beer-Lambert's law. x Original concentrations of O2 and CO2; The NH3 calibration unit calculates the corrected NH3 concentration using the humidity-corrected Stern-Volmer formula. The first LIBS analysis unit uses a laser to bombard particulate matter and collects plasma emission spectra to calculate the elemental composition of the particulate matter.

4. The system according to claim 1, characterized in that, The adaptive interference compensation module includes: The interference identification unit uses a CNN+DNN combined model to extract interference features and outputs the interference confidence score. The concentration correction unit dynamically corrects gas concentration based on the extended Beer-Lambert law. The transfer learning unit invokes the pre-trained model for transfer learning and triggers online incremental learning when the interference recognition success rate is less than a preset success rate threshold.

5. The system according to claim 1, characterized in that, The particulate matter fingerprinting and pollution source tracing module includes: The second LIBS analysis unit measures the plasma emission spectrum from 200 to 800 nm by bombarding particulate matter with a 50 mJ pulsed laser. The fingerprint matching unit compares the measured element ratio with the pollution source fingerprint database based on the feature element ratio decision tree, and calculates the matching degree using cosine similarity.

6. The system according to claim 1, characterized in that, The multi-source collaborative data fusion module includes: Spatial interpolation units are used to generate a global pollution distribution heat map using the Kriging algorithm; The time series prediction unit uses the ARIMA model to predict future concentration and diffusion trends.

7. The system according to claim 2, characterized in that, The carbon flow rate calculation module includes: The coal consumption inversion unit calculates the coal consumption and inverts the carbon flow rate based on boiler operating parameters. The actual measurement calculation unit calculates the actual carbon flow rate based on the flue gas flow rate and CO2 concentration; The fusion unit uses a dynamic weighted fusion algorithm to output the final carbon flow rate.

8. The system according to claim 2, characterized in that, The digital twin maintenance module includes: The life prediction unit combines a physical loss model and an LSTM neural network to predict the remaining life of a component. The health assessment unit outputs equipment maintenance priorities based on health indicator formulas.

9. The system according to claim 2, characterized in that, The blockchain evidence storage module includes: The signature unit generates a data digest using the SHA3-256 hash algorithm and a data signature using ECDSA signature technology. The consensus unit employs an improved PBFT algorithm to achieve rapid consensus within a 5-node regional consensus group. The system control and communication module includes: The intelligent control unit adjusts the opening degree of the ammonia injection valve based on the PID algorithm; The alarm unit triggers audible and visual alarms, SMS notifications, or platform push notifications based on the level of abnormality.

10. A continuous emission monitoring device for flue gas, characterized in that, The system includes a processor and a memory. The processor is used to process instructions stored in the memory to implement the functions of a continuous emission monitoring system for flue gas according to any one of claims 1-8. The memory stores program instructions for performing data acquisition, flue gas pretreatment, ultraviolet spectral acquisition, multimodal spectral analysis, adaptive interference compensation, particulate matter fingerprinting and pollution source tracing, multi-source collaborative data fusion, carbon flow rate calculation, digital twin maintenance, blockchain evidence storage, and system control and communication.

11. The apparatus according to claim 10, characterized in that, It also includes a sampling component, a preprocessing component, a spectral detection component, a data transmission component, and an execution component connected to the processor. The sampling component is used to collect raw flue gas in the flue. The preprocessing component is used to perform dust removal and dehumidification treatment on the raw flue gas. The spectral detection component is used to acquire the absorption spectrum and fluorescence intensity data of the flue gas. The data transmission component is used to realize data interaction between the device and external equipment. The execution component is used to perform ammonia injection valve adjustment and alarm operations according to the processor's instructions.

Images