Multi-point patrolling method, system and device

Abstract

The invention belongs to the technical field of flue gas monitoring, and provides a multipoint patrolling method, system and device. The method comprises the following steps of adaptive sampling and preprocessing, distributed full-dimensional self-diagnosis and fault early warning, fault root cause mining and processing, LIBS spectral analysis and 4D pollution diffusion model construction, environment-data dual-drive dynamic calibration mechanism and data fusion and collaborative analysis. According to the method, LIBS spectral analysis and a CFD model are coupled to construct a 4D pollution diffusion model, and an environment-data dual-drive dynamic calibration mechanism is combined, so that accurate simulation and prediction of a pollution diffusion rule are realized; through space fusion and time trend prediction, a scientific basis is provided for pollution source responsibility division and emission reduction decision, adaptive sampling, multi-modal fault diagnosis and dynamic pollution modeling are integrated, closed-loop and environment-data dual-drive dynamic calibration from data acquisition to decision support is achieved, limitation of a traditional static model is broken through, multi-dimensional collaborative decision support is achieved, and the method is suitable for large-scale popularization and application. And a landing emission reduction scheme is provided.

Description

Technical Field

[0001] This application relates to the field of flue gas monitoring technology, and in particular to a multi-point monitoring method, system and device. Background Technology

[0002] Flue gas monitoring refers to a technology that uses sampling equipment at multiple monitoring points in an industrial flue gas emission system to collect and analyze the concentration of pollutants in the flue gas in real time in order to assess environmental quality and emission compliance. Multi-point monitoring aims to cover a wide monitoring area to ensure the comprehensiveness and continuity of data.

[0003] Currently, existing flue gas monitoring technologies mainly employ single-point fixed monitoring or simple multi-point switching sampling modes, combined with traditional sensors for pollutant detection. For example, using a fixed probe to collect flue gas at a single location, or switching between different monitoring points, relies on simple threshold alarms and experience-based judgment for fault handling.

[0004] However, existing technologies have significant drawbacks: during the sampling process, high humidity can cause flue gas condensation, or background gas interference can lead to inaccurate pollutant concentration measurements; dust deposition can easily clog sampling probes, affecting data continuity; fault diagnosis capabilities are weak, making it difficult to quickly identify abnormal locations and root causes, and single-point failures can lead to overall monitoring failure; pollution diffusion models are disconnected from actual data, making accurate source tracing and dynamic simulation impossible; pollution diffusion models are disconnected from measured data, failing to accurately reflect the spatiotemporal migration patterns of pollutants, and the lack of dynamic integration between environmental parameters and pollutant data results in insufficient reliability of prediction results. Summary of the Invention

[0005] The purpose of this application is to provide a multi-point survey method, system, and apparatus to solve at least one technical problem mentioned in the background art.

[0006] To address the aforementioned technical problems, this application provides a multi-point survey method, employing the following technical solution: A multi-point survey method includes the following steps: S100, Adaptive Sampling and Preprocessing: It continuously collects raw flue gas from multiple monitoring points through time-sequence switching, and simultaneously performs anti-condensation control, dual-cavity isolation anti-interference processing and laser self-cleaning to output clean sample gas; S200, Distributed Full-Dimensional Self-Diagnosis and Fault Early Warning: Based on hardware, data and environmental parameters, a three-level diagnostic network is built. Through residual detection and adaptive threshold mechanism, abnormal points are identified in real time and early warning is triggered. The status labels and abnormal parameter vectors of each monitoring point are output. S300, Fault Root Cause Mining and Handling: Combining acoustic signature, pressure difference, and spectral multimodal data to locate the fault cause, generating a handling plan by matching historical case library through convolutional neural network, and executing a graded response; S400 and LIBS spectral analysis and 4D pollution diffusion model construction: Couple LIBS elemental concentration data with CFD fluid dynamics model, dynamically correct the diffusion coefficient through measured concentration gradient, and construct a 4D pollution diffusion model including spatial three-dimensional and temporal dimensions. S500, Environment-Data Dual-Driven Dynamic Calibration: Based on the Transformer model, environmental parameters and pollutant data are fused together. The weight coefficients of CFD prediction and measured data are dynamically adjusted according to the environmental volatility, and the calibrated pollutant diffusion results are output. S600, Data Fusion and Collaborative Analysis: By spatially fusion, the contribution ratio of each pollution source is calculated, and a time series model is used to predict the short-term pollution trend in the future, generating a comprehensive monitoring report that includes spatial heat maps, responsibility rankings, and emission reduction recommendations.

[0007] To address the aforementioned technical problems, this application also provides a multi-point survey system, comprising: Sampling module: Equipped with an improved sampling probe and gas collection unit, it is used to collect raw flue gas from multiple monitoring points. Through time-series switching, it can achieve continuous sampling at multiple points to obtain raw flue gas. Pre-processing module: Simultaneously performs anti-condensation control, anti-interference treatment, and self-cleaning maintenance during the sampling of raw flue gas, and outputs clean sample gas; Diagnostic early warning module: Based on a three-level network of "probe box - diagnostic edge gateway - expert system", it monitors multi-dimensional parameters of hardware, data and environment, identifies abnormal points through residual calculation and adaptive threshold analysis, and outputs monitoring point status labels and abnormal parameter vectors; Fault handling module: Constructs a multimodal analysis matrix based on abnormal parameter vectors, combines a convolutional neural network model to match historical fault cases, determines the fault type and confidence level, and executes a graded handling scheme according to the confidence level; Spectral analysis and model building module: LIBS spectral technology is used to obtain the elemental characteristics of particulate matter, coupled with a CFD model and the diffusion coefficient is corrected by measured data to build a 4D pollution diffusion model that includes spatial three-dimensional and temporal dimensions; Dynamic calibration module: It integrates environmental parameters and pollutant data through the Transformer model, dynamically adjusts the weights of CFD model prediction results and measured data based on the fluctuation rate of environmental parameters, and outputs the calibrated pollutant diffusion results. Data fusion and analysis module: Spatially fuses the calibrated 4D model data to calculate the contribution ratio of pollution sources, uses time series models to predict short-term pollution trends, and integrates data to output a comprehensive monitoring report.

[0008] To address the aforementioned technical problems, this application also provides a multi-point monitoring device, including a sampling device and a computer device. The sampling device is used to collect flue gas samples from a flue gas emission system, and the computer device is used to analyze the collected flue gas samples. The computer device includes a processor and a memory, and the memory stores computer-readable instructions. When the processor executes the computer-readable instructions, it implements the steps of the method described above.

[0009] The beneficial effects of this invention are as follows: This application provides a multi-point survey method that ensures continuous data coverage from multiple points through time-series switching sampling, dynamically adjusts heating power based on real-time humidity to prevent condensation control, effectively eliminates background interference through dual-cavity isolation design, and avoids dust blockage through laser self-cleaning. This significantly improves the authenticity of the sample gas and the stability of the sampling system, and solves the problems of data distortion and easy equipment failure in traditional sampling.

[0010] This method constructs a distributed, multi-dimensional diagnostic network that combines quantum residual detection and multimodal analysis matrix to achieve accurate identification of anomalies and rapid location of root causes of faults, accurately tracing the source of pollution. The convolutional neural network model matches fault types based on massive historical cases, and the hierarchical processing scheme significantly improves the system's fault tolerance and maintenance efficiency, overcoming the shortcomings of traditional diagnostics such as lag and low accuracy.

[0011] This method couples LIBS spectral analysis with CFD models to construct a 4D pollution diffusion model. Combined with an environment-data dual-driven dynamic calibration mechanism, it achieves accurate simulation and prediction of pollution diffusion patterns. Through spatial fusion and temporal trend prediction, the output comprehensive monitoring report provides a scientific basis for the division of pollution source responsibilities and emission reduction decisions, and promotes the upgrade of flue gas monitoring from single-point static monitoring to full-area dynamic and accurate monitoring.

[0012] This method integrates adaptive sampling, multimodal fault diagnosis, and dynamic pollution modeling to achieve a closed loop from data acquisition to decision support. It features dual-driven dynamic calibration based on environment and data, breaks through the limitations of traditional static models, provides multi-dimensional collaborative decision support, and offers feasible emission reduction solutions. Attached Figure Description

[0013] 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.

[0014] Figure 1 This is a flowchart of a multi-point flue gas monitoring method provided in an embodiment of this application; Figure 2 This is an exemplary system architecture diagram to which this application can be applied; Figure 3 This is a schematic diagram of the structure of the multi-point flue gas monitoring device provided in the embodiments of this application; Figure 4 This is a schematic diagram of the sampling device provided in the embodiments of this application. Detailed Implementation

[0015] 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.

[0016] 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.

[0017] 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.

[0018] A multi-point survey method includes steps S100-S600, wherein: S100, Adaptive Sampling and Preprocessing: It continuously collects raw flue gas from multiple monitoring points through time-sequence switching, and simultaneously performs anti-condensation control, dual-cavity isolation anti-interference processing and laser self-cleaning to output clean sample gas.

[0019] In another embodiment, step S100, the time-sequence switching sampling includes: setting multiple monitoring points in the flue gas emission system, installing a sampling probe at each monitoring point, and the gas collection unit automatically switching the sampling point according to a preset time sequence. Anti-condensation control dynamically adjusts the heating power based on real-time humidity and switches to a backup probe when there is a risk of condensation. Anti-interference processing employs synchronous sampling in both main and auxiliary chambers, eliminating spectral interference through background gas difference.

[0020] The raw flue gas (containing SO2 and NO) at each monitoring point in the flue gas emission system xThe following methods are used for the pretreatment of (NH3, high humidity water vapor, dust) and other pollutants: 101. Timing switching sampling.

[0021] Multiple monitoring points are set up in the flue gas emission system, and a sampling probe is installed at each monitoring point to collect raw flue gas at the monitoring point location. The gas collection unit automatically switches the sampling point according to a preset time sequence of 30 seconds / point to cover all monitoring points and ensure the continuity of data from multiple points.

[0022] A gas collection unit is a pipeline fitting system for collecting and distributing gas, used to achieve the orderly collection of flue gas from multiple monitoring points.

[0023] In one implementation method, each monitoring point is independently equipped with a gas collection unit. Each gas collection unit consists of one or more main gas pipes connected in parallel with several branch pipes, and is equipped with valves, pressure regulators, and other accessories. During operation, each gas collection unit is activated sequentially according to a preset time sequence of 30 seconds per point. When a gas collection unit is activated, its branch pipe connects to the sampling port of the corresponding monitoring point, thereby collecting the raw flue gas from that monitoring point into the main gas pipe. The collected raw flue gas is then transmitted to the subsequent processing stage through pipelines, ensuring the continuity and stability of multi-point data acquisition.

[0024] As another implementation method, a gas collection unit is set in the flue gas emission pipeline. The gas collection unit adopts a stainless steel multi-way valve. The multi-way valve has multiple interfaces, each of which is connected to the sampling pipeline of a monitoring point. The stepper motor is driven by a timing controller to rotate the valve core in a preset order and connect it to the sampling port of the corresponding monitoring point, thereby collecting the original flue gas of each monitoring point in sequence.

[0025] 102. Anti-condensation control.

[0026] The sampling probe integrates a miniature heating system and a humidity sensor. The heating system uses a heating ring with an adjustable power of 50-200W. The humidity sensor has an accuracy of ±1%RH. According to the formula:

[0027] The output power P of the heat tracing system is adjusted in real time to prevent flue gas condensation. Here, 50 is the base heating power (W); 3 is the temperature compensation coefficient (W / %); H is the real-time humidity (%); and 30 is the humidity threshold (%).

[0028] When H ≤ 30%, the system maintains a base heating power of 50W to ensure that the sampling tube temperature is above the dew point. The heating system is activated only when H > 30%; if the humidity is still > 30% after heating, the power is first increased to 200W and maintained for 30 seconds; if it still does not meet the standard, the system automatically switches to the backup probe to avoid condensation affecting the authenticity of the sample gas.

[0029] 103. Anti-interference measures.

[0030] This program aims to monitor a variety of pollutants harmful to the environment and human health. Target pollutants in flue gas include SO2 and NO. x NH3 and particulate matter.

[0031] A dual-chamber isolation sampling design is adopted, in which the raw flue gas is simultaneously sent into the main chamber and the auxiliary chamber. The main chamber collects the flue gas containing the target pollutants, while the auxiliary chamber filters the raw flue gas through activated carbon or molecular sieves, and simultaneously collects the background gas that has been filtered to remove the target pollutants. The difference between the two is used to calculate and eliminate background interference, such as the influence of inherent components in the air on SO2 and NO. x To eliminate spectral interference, the true concentration C of the target pollutant in the original flue gas was obtained. 真实 The formula is as follows:

[0032] Among them, C 主腔 Original flue gas concentration, in ppm, C 副腔 The background gas concentration is expressed in ppm, and k is a correction factor determined through standard gas experiments.

[0033] 104. Self-cleaning maintenance: Dust particles in the flue gas, especially CaSO4 and carbon black, can deposit at the sampling port of the sampling probe, forming a light-blocking layer. A laser self-cleaning unit cleans the sampling probe to prevent clogging by dust and other impurities from affecting sampling accuracy.

[0034] The laser self-cleaning unit uses a 650nm semiconductor laser with a power of 3W. It scans the sampling port once every hour at an angle of ±30°. If the dust concentration in the monitored environment is >50mg / m³, it will trigger on-demand startup to avoid dust clogging the probe.

[0035] After the above processing, a clean sample gas with no condensation and low interference is output to meet the requirements of subsequent analysis.

[0036] S200, Distributed Full-Dimensional Self-Diagnosis and Fault Early Warning: Based on hardware, data and environmental parameters, a three-level diagnostic network is constructed. Through residual detection and adaptive threshold mechanism, abnormal points are identified in real time and early warning is triggered. The status labels and abnormal parameter vectors of each monitoring point are output.

[0037] In another embodiment, in step S200, the diagnostic network adopts a three-tier architecture of "probe box - diagnostic edge gateway - expert system" and blockchain notarization. The deviation between the current parameter value and the historical mean is calculated using a quantum residual algorithm. This deviation is then combined with an adaptive threshold to trigger a hybrid expert system. The adaptive threshold is dynamically updated using historical thresholds and the absolute value of the current residual. Status labels include normal, suspected fault, and confirmed fault. A repeated detection mechanism is initiated for suspected fault points to reduce the false positive rate.

[0038] The self-diagnosis and fault early warning methods are as follows: 1. Distributed diagnostic network architecture.

[0039] It consists of a three-tier architecture of "probe box - diagnostic edge gateway - expert system", which transmits data via CAN bus (high transmission rate and strong anti-interference) and combines blockchain evidence storage to ensure that parameters cannot be tampered with (ensuring data traceability).

[0040] 2. Core algorithm: Quantum residual detection.

[0041] 201. Calculate the residual r between each parameter and the normal operating condition. i ;

[0042] residual r i This reflects the degree of deviation of the current parameter from the historical mean, where X i This refers to the current parameter value, i.e., the current value of a parameter (such as temperature, pressure, concentration, etc.) monitored in real time; μ hist σ is the historical mean. hist This represents the historical standard deviation.

[0043] 202. Construct an adaptive threshold θ based on historical data threshold It is used to determine whether the residual is abnormal.

[0044] The decision-making logic is as follows: If r i >3θ threshold The status is determined to be a confirmed fault; the MoE (Hybrid Expert) system is triggered to integrate flow control, optics, and temperature control expert modules for collaborative analysis; If the condition 10 < r is satisfied for 5 consecutive times i <3θ threshold The status is determined to be a suspected fault; preventive maintenance is initiated, such as replacing aging parts in advance. Otherwise: the state is determined to be normal; only the adaptive threshold θ is updated.

[0045] 203. Adaptive threshold dynamic update:

[0046] The updated threshold θ is calculated. threshold，new , where θ threshold，old The threshold before the update, |r i | represents the absolute value of the current residual.

[0047] The detection threshold is dynamically adjusted to enable the system to adapt to changes in operating conditions, such as equipment aging and environmental fluctuations.

[0048] 204. Fault confirmation mechanism.

[0049] For suspected outliers, such as r i >3θ threshold The system initiates three repeated checks with a 10-second interval. If all three checks fail, the system is considered to have confirmed the fault, thus avoiding false positives.

[0050] After the above processing, the status labels for each monitoring point are output, including normal, suspected fault, and confirmed fault. The multi-dimensional vector composed of the abnormal parameters of the confirmed fault points is the abnormal parameter vector, which reflects the hardware or data characteristics of the fault.

[0051] S300, Fault Root Cause Mining and Handling: Combining acoustic signature, pressure difference, and spectral multimodal data to locate the cause of the fault, generating a handling plan by matching historical case library through convolutional neural network, and executing a graded response.

[0052] In another embodiment, step S300 involves constructing a multimodal analysis matrix comparing acoustic signature MFCC offset, differential pressure oscillation frequency, and spectral confidence levels. Anomaly vectors are matched against a historical fault feature library using a CNN to output the fault type and confidence level. When the confidence level exceeds a threshold, a backup module is automatically switched; otherwise, a maintenance order is pushed. The default confidence threshold is 90%.

[0053] The troubleshooting methods are as follows: 301. Multimodal analysis matrix.

[0054] By combining multi-dimensional data such as acoustic signature, spectrum, pressure difference, and signal characteristics, a matrix of "fault phenomenon - analysis mode - diagnostic logic" is constructed, as shown in the example below: When the sampling flow rate drops suddenly, the acoustic fingerprint analysis + differential pressure spectrum analysis mode is used; compared with the normal state Mel-frequency cepstral coefficient (MFCC), if the MFCC offset is >15% and the differential pressure oscillation frequency is ∈[50,100Hz], it is determined that the diaphragm is ruptured; the solution is to switch to the backup flow path and push the spare parts replacement order. NO x When data jumps, a confidence comparison analysis mode of quantum dot spectroscopy and DOAS (differential absorption spectroscopy) is used; if the DOAS confidence score is <0.6 and the quantum dot confidence score is >0.9, the DOAS optical window is determined to be contaminated; the solution is to activate laser self-cleaning and extend the purging time to 15 seconds. When communication is intermittent, the signal constellation diagram analysis mode is used; if the detected QPSK phase offset is >π / 8, electromagnetic interference is determined, such as that caused by the high-frequency electric furnace in the secondary combustion chamber; the solution is to switch the anti-interference protocol and add a shielding sleeve.

[0055] 3011. Voiceprint analysis involves collecting sound signals from a device during operation and extracting acoustic features to identify the device's status. The method for voiceprint analysis of the sampling system's voiceprint signals is as follows: 30111. Acquire audio signals and extract segments containing human voices or device-specific sounds through voice activity detection; 30112. Calculate Mel-frequency cepstral coefficients (MFCC):

[0056] Where X(t,k) is the spectral energy of the k-th filter in the t-th frame; H(k) is the triangular filter bank; DCT is the discrete cosine transform; and K is the total number of filters.

[0057] 30113. Compare the Euclidean distance between the real-time MFCC and the historical benchmark database, and calculate the Mel-frequency cepstral coefficient shift ΔMFCC:

[0058] in, This is the real-time acquired MFCC coefficient vector; is the MFCC coefficient vector in the historical benchmark library; N is the total dimension of the MFCC coefficients.

[0059] 3012. Differential pressure spectrum analysis involves performing spectral analysis on the fluctuation signal of the pressure difference in the gas path to identify the pressure change pattern. The method for differential pressure spectrum analysis of the sampled gas path differential pressure signal is as follows: 30121. Collect differential pressure sensor data and construct a time series graph; 30122. Calculation of differential pressure oscillation frequency: The frequency domain dominant frequency is extracted using Fast Fourier Transform (FFT), yielding the spectrum P(f):

[0060] P(f) is the amplitude of the differential pressure signal at frequency f, reflecting the energy of that frequency component. Here, t is time; p(t) is the differential pressure timing signal; and f is the frequency.

[0061] Search for the peak amplitude point in the spectrum graph; its horizontal axis represents the main frequency f. dom main frequency value f dom It can be associated with the severity of the fault; for example, the higher the frequency, the more severe the blockage.

[0062] The spectrum is concentrated in the low frequency range, f dom <1Hz, status is normal; High-frequency oscillation, f dom A frequency greater than 5Hz indicates valve blockage or leakage, and the status is faulty.

[0063] 3013. Quantum dot spectroscopy analysis utilizes the fluorescence response characteristics of quantum dot materials to specific gases to detect concentration, while DOAS is based on the characteristic absorption of gas molecules to ultraviolet / visible light to detect concentration. Quantum dot confidence is calculated based on fluorescence intensity stability, while DOAS confidence is calculated through spectral fitting error; the smaller the error, the higher the confidence.

[0064] Simultaneous acquisition of NO output from quantum dot spectrometer x Concentration confidence level, NO output from DOAS system x Concentration confidence levels were compared and analyzed. If the DOAS confidence level was <0.6 and the quantum dot confidence level was >0.9, it was determined to be DOAS optical window contamination.

[0065] 3014. A signal constellation diagram is a distribution map of a digital modulated signal in the complex plane, where each point represents the phase and amplitude of a symbol. Constellation diagram analysis determines signal quality by identifying the degree of offset of symbol points. The method for performing signal constellation diagram analysis on the complex plane distribution of the QPSK modulated signal used for communication between the monitoring system and the backend is as follows: Collect constellation diagrams of communication signals and measure the phase of actual symbol points. QPSK ideal phase Calculate the phase deviation between the actual sign point and the ideal position for 0, π / 2, π, and 3π / 2. If the phase deviation is greater than π / 8, it is determined to be electromagnetic interference.

[0066] 302. Matching of Convolutional Neural Network (CNN) Models.

[0067] Input an anomaly parameter vector, match it with a multimodal feature library using CNN, train the data to include 100,000+ historical fault cases, and output the fault type and fault type confidence.

[0068] CNN is a deep learning model that includes convolutional layers, pooling layers, and fully connected layers. It extracts MFCC coefficients from voiceprint signals, feature peak intensities from spectral data, oscillation frequencies from differential pressure signals, and phase shifts from communication signals. The convolutional layers of the CNN perform convolution operations on these features, and the pooling layers reduce the dimensionality to finally obtain a fused multimodal feature vector.

[0069] The fault type is a historical fault case type based on multimodal feature matching, such as diaphragm rupture, optical window contamination, and electromagnetic interference. The CNN compares the input abnormal parameter vector with the case features in the feature library and outputs the fault type with the highest matching degree, thus identifying the root cause of the fault.

[0070] The CNN output layer calculates the probability distribution of various faults using the softmax function, and the probability corresponding to the fault type with the highest matching degree is the fault type confidence.

[0071] 303. Graded processing scheme.

[0072] When the confidence level of the fault type is greater than the threshold, a preset operation is automatically executed, such as switching to a backup light source. If the fault persists after processing, cascading backup is triggered, switching to the backup monitoring point and pushing a manual repair order, which includes the cause of the fault and the processing steps.

[0073] S400 and LIBS spectral analysis and 4D pollution diffusion model construction: Coupled with LIBS elemental concentration data and CFD fluid dynamics model, the diffusion coefficient is dynamically corrected by measured concentration gradient to construct a 4D pollution diffusion model including spatial three-dimensional and temporal dimensions.

[0074] In another embodiment, in step S400, LIBS spectral analysis emits laser pulses to obtain particulate matter elemental fingerprints, calculates elemental concentrations through characteristic peak intensities, and analyzes the pollutant source composition. The coupling of CFD and LIBS includes: obtaining the initial pollutant concentration based on the initial CFD simulation, correcting the CFD diffusion coefficient using the measured concentration gradient from LIBS, recalculating the corrected simulated concentration, and using the adjoint method to invert the pollution source location. The 4D pollution diffusion model stores the concentration fields at different time steps in chronological order to form a four-dimensional matrix, and renders it using a WebGL engine to obtain a dynamic heatmap of pollutants, used to trace historical diffusion paths at any location.

[0075] The analysis and model building methods are as follows: 401. LIBS (Laser-Induced Breakdown Spectroscopy) Online Analysis.

[0076] LIBS is a technique that uses laser pulses to excite sample gas to generate plasma, and analyzes the plasma emission spectrum to obtain elemental composition and concentration, which can quickly identify the source characteristics of pollutants.

[0077] A 50 mJ / pulse laser is emitted every 5 minutes to obtain the elemental fingerprint of particulate matter, such as Fe and Ca, characteristic elements of steel sources, and Cl and Zn, characteristic elements of waste incineration sources; the elemental concentration is calculated by the intensity of characteristic peaks, such as Fe corresponding to the 238.2 nm peak and Cl corresponding to the 134.7 nm peak.

[0078] 402. Dynamic coupling.

[0079] 4021. Initial Simulation.

[0080] The fundamental CFD (Computational Fluid Dynamics) model is a virtual computational model of "flue gas flow-pollutant diffusion" built based on the three-dimensional layout of a factory (including pipes and fans). It solves the fluid motion equations through numerical simulation. The goal is to replace real physical experiments with mathematical equations and numerical algorithms, simulating flue gas flow patterns and pollutant diffusion paths in a low-cost and efficient manner. Flue gas flow patterns include velocity, pressure, and temperature distribution. The Reynolds stress model (RSM) is a sub-model in CFD used to simulate turbulent flow. Because flue gas flow is turbulent, RSM can more accurately describe the impact of turbulent fluctuations on the flow.

[0081] Navier-Stokes equations:

[0082] Where t is time, in seconds; θ / θt is the time partial derivative; u is the velocity vector of the flue gas, in meters per second; u i It is the component of the velocity vector in the i-direction, with units of m / s; i = x, y, z, corresponding to the three-dimensional directions in space; Here, ρ is the gradient operator; ρ is the flue gas density, in kg / m³. 3 p is the flue gas pressure, in Pa; τ ij The stress tensor is expressed in Pa; g i The component of gravitational acceleration in the i-direction, in m / s². 2 ; RSM obtains the three-dimensional velocity field u(x,y,z) and pressure distribution p(x,y,z) of flue gas by numerically iterating through the Navier-Stokes equations.

[0083] Pollutant diffusion equation:

[0084] Where C is the pollutant concentration; u is the velocity vector solved by RSM; and D is the diffusion coefficient. This is a diffusion term.

[0085] When the CFD model does not correct for the diffusion coefficient, the default parameter D is used. base Calculation: The basic diffusion coefficient D base Substituting u obtained from the RSM solution into the pollutant diffusion equation, we obtain the initial simulated concentration C of the pollutant. CFD .

[0086] 4022.LIBS actual measurement correction.

[0087] The LIBS elemental concentration distribution is used as a boundary condition every 30 seconds to correct the diffusion coefficient D. formula:

[0088] The corrected diffusion coefficient D is obtained after calculation. new , where D base Based on the basic diffusion coefficient, This represents the actual concentration gradient measured by LIBS. The ideal concentration gradient is used for CFD simulation.

[0089] The CFD model uses the corrected diffusion coefficient D new Recalculate: D new Replace D base Substituting the pollutant diffusion equation, we obtain the corrected simulated concentration C. CFD-new .

[0090] By correcting D using LIBS measured data, CFD is upgraded from ideal simulation to measured and calibrated simulation, ultimately achieving accurate pollution source inversion and 4D visualization, solving the core pain point of the disconnect between industrial flue gas monitoring theory and practice.

[0091] 4023. Pollution Source Inversion: Introducing the adjoint method, through the objective function:

[0092] The location s of the pollution source is retrieved using the adjoint method. Where, min s To minimize the pollution source location s, the goal is to find the pollution source location that minimizes the error; C obs To measure concentrations accurately, concentration data from all measurement methods, including LIBS, traditional sensors, and manual sampling, are integrated; C CFD The initial simulated concentration; C CFD-new To correct the simulated concentration.

[0093] Output of 403.4D model.

[0094] A 4D model is a four-dimensional spacetime model created by adding a time dimension (t) to a three-dimensional space (x, y, z). The 4D model construction process is as follows: 4031. CFD simulation + diffusion coefficient correction: every 30 seconds, the corrected D... new The pollutant transport equation is recalculated to obtain the concentration field C(x,y,z,t) at the current time step. now ); 4032. Time series overlay: The concentration fields at different time steps are stored in chronological order to form a four-dimensional matrix of "three spatial dimensions + one temporal dimension".

[0095] The 4D model outputs a spatiotemporal matrix C(x,y,z,t), which is a four-dimensional data set of pollutant concentration C varying with space and time.

[0096] By rendering the spatiotemporal matrix using the WebGL engine, a dynamic heat map of pollutants is obtained, which can trace the historical diffusion path of any location over 12 hours, thus realizing the visualization of the pollutant diffusion process.

[0097] By deeply integrating fluid mechanics principles, CFD numerical algorithms, LIBS measured data, and WebGL visualization, a leap from blind monitoring to precise source tracing, dynamic simulation, and intuitive decision-making is achieved.

[0098] S500, Environment-Data Dual-Driven Dynamic Calibration: Based on the Transformer model, environmental parameters and pollutant data are fused together. The weight coefficients of CFD prediction and measured data are dynamically adjusted according to the environmental volatility, and the calibrated pollutant diffusion results are output.

[0099] In another embodiment, in step S500, the environmental data includes wind speed, ambient temperature, and relative humidity, while the pollutant data is the elemental concentration distribution obtained from LIBS spectral analysis. The environmental temporal series and the pollutant spatial matrix are linked using a Transformer multi-head attention mechanism. Weighting coefficients are calculated based on wind speed and humidity fluctuation rates, and CFD predictions and LIBS measured data are dynamically fused. When the environmental fluctuation rate is >10%, measured data is prioritized; when the environmental fluctuation rate is <5%, the model prediction weights are strengthened.

[0100] The dynamic calibration method is as follows: The Transformer model is a deep learning model based on a self-attention mechanism. It can simultaneously process the temporal features of environmental data and the spatial features of pollutant data. It achieves dynamic fusion of multi-source data by calculating the correlation weights between different features. The training data comes from historical environmental-pollutant linkage cases, and the attention parameters are optimized through backpropagation. The Transformer model includes an input layer, a processing layer, and an output layer.

[0101] 501. Input layer: Input environmental data and pollutant data.

[0102] Environmental data: collected by meteorological sensors in the monitoring area, including wind speed (accuracy ±0.1m / s), ambient temperature (±0.5℃), and relative humidity (±1%RH); Pollutant data: Elemental concentration distribution obtained from LIBS spectral analysis in step S400 is synchronously accessed.

[0103] 502. Processing layer: Transformer model fusion.

[0104] 5021. Model Structure: Self-attention mechanism: Calculate the association weights between environmental time series and pollutant spatial matrix; Multi-headed attention: capturing the dynamic correlation between environmental temporal sequences and pollutant spatial matrices, such as the accelerating effect of high humidity on Cl diffusion.

[0105] Environmental time series e env (t) represents real-time data on environmental parameters changing over time, derived from environmental sensors deployed at monitoring points, and used to describe the dynamic evolution of the environmental state. Environmental time series e env (t) is a sequence indexed by timestamps, with data in the form of: e env (t)={T(t),H(t),u(t),P(t)} Where T(t) is the temperature at time t, H(t) is the relative humidity at time t, u(t) is the wind speed at time t, and P(t) is the atmospheric pressure at time t.

[0106] Pollutant spatial matrix C pollutant (x,y,z,t) represents the spatial distribution data of pollutant concentrations, derived from spectral analysis techniques (LIBS, DOAS) and computational fluid dynamics (CFD) models, used to describe the spatial diffusion characteristics of pollutants. pollutant (x,y,z,t) is a matrix indexed by a spatial grid, with the data format as follows:

[0107] Where c is the spatial grid coordinate. ij (t) represents the pollutant concentration at the grid point in the i-th row and j-th column at time t; m and n are the dimensions of the spatial grid.

[0108] It should be noted that environmental time-series data needs to be normalized and denoised to eliminate the influence of sensor noise and environmental fluctuations. Concentration data of the pollutant spatial matrix needs to be interpolated to fill in missing values, and PCA dimensionality reduction should be used to reduce spatial dimensions and decrease computational complexity.

[0109] 5022. Training data: Historical environmental-pollutant linkage data (100,000+ samples) from the past 6 months were used to optimize model parameters with the goal of minimizing prediction error.

[0110] 503. Output layer: Adaptive weight correction.

[0111] The core of adaptive weight adjustment is the dynamic fusion of CFD model predictions. 4D Compared with LIBS measured data C LIBS The weighting coefficient w(t) of the two parameters is adjusted in real time based on the fluctuation range of environmental parameters, such as wind speed and temperature change rate.

[0112] 5031. Environmental time series e env The volatility of w(t) directly determines the magnitude of w(t). The formula for volatility ΔE(t) is:

[0113] Where Δt is the time interval; e env (t) represents the environmental parameter vector at the current moment.

[0114] 5032. Formula for weighting coefficient w(t):

[0115] Where ΔE(t) is the environmental parameter volatility; k=0.5 is the sensitivity coefficient; w(t) takes a value of 0-1, representing the weight of the CFD model prediction.

[0116] 5033. The adaptive adjustment logic is as follows: When environmental fluctuations are large, such as ΔE(t) > 10%, w(t) automatically decreases, enhancing the LIBS measured data C.Libs The weights are adjusted to reduce model error.

[0117] When the environment is stable, such as ΔE(t) < 5%, w(t) automatically increases, giving full play to the advantages of the CFD model. The model prediction is more reliable in a stable environment.

[0118] 5034.C pollutant It is the spatial distribution matrix of pollutant concentrations, such as C 4D and C LIBS The role of w(t) is to fuse these two types of data to generate the calibrated pollutant diffusion result C. calib Fusion formula: C calib (x,y,z,t)=w(t)·C 4D (x,y,z,t)+(1-w(t))·C LIBS (x,y,z,t) Among them, C 4D The initial simulation results for CFD; C LIBS These are the measured interpolation results from LIBS. C calib The calibration results are based on a dual-driven environment and pollutant model, with the weights of the predicted and measured data adjusted for dynamic equilibrium.

[0119] 504. The weighted coordination workflow is as follows: 5041. Environmental Perception: Sensors collect data e env (t), such as wind speed and temperature; 5042. Fluctuation Analysis: Calculate ΔE(t) to determine whether the environment is stable; 5043. Weight Adjustment: Update w(t) based on ΔE(t); 5044. Data Fusion: Using w(t) to fuse predicted C 4D and the measured C LIBS , to obtain C calib This allows for precise distribution of pollutants.

[0120] S600, Data Fusion and Collaborative Analysis: By spatially fusion, the contribution ratio of each pollution source is calculated, and a time series model is used to predict the short-term pollution trend in the future, generating a comprehensive monitoring report that includes spatial heat maps, responsibility rankings, and emission reduction recommendations.

[0121] In another embodiment, in step S600, spatial fusion quantifies the responsibility ratio of each pollution source using a contribution rate formula. Temporal trend prediction employs the ARIMA model, predicting future concentration trends based on historical concentration sequences, for example, predicting the trend over the next 15 minutes.

[0122] The data fusion and collaborative analysis methods are as follows: 601. Spatial integration.

[0123] The maximum regional value of multi-source superimposed pollution was calculated based on the calibrated 4D model, and the contribution ratio of each pollution source was clarified.

[0124] Formula for pollution contribution percentage:

[0125] The contribution rate of the i-th pollution source is calculated, where N is the number of pollution sources; C i Let be the concentration of the i-th pollution source.

[0126] By using the controlled variable method or source tracing and inversion, the contribution of individual pollution sources can be separated and compared with the total concentration to quantify the responsibility of each pollution source for pollution, which helps to quickly locate the main pollution sources. For example, sources with a contribution rate of >50% are the key targets for treatment.

[0127] 602. Time trend prediction.

[0128] Using the ARIMA model (autoregressive integral moving average model), the time dimension data of the calibrated 4D model is input, and the training set is the concentration sequence of the past 30 days. The model predicts the pollution change trend in the next 15 minutes, such as the warning of concentration increase / decrease.

[0129] The ARIMA model is a time series forecasting model that uses differencing to process non-stationary data and combines autoregression and moving average to capture data trends in order to predict short-term changes in pollutant concentrations.

[0130] 6021.ARIMA(p,d,q) consists of three core parts: Autoregression (p): Predicts the current value using the concentration at p past time points (capturing trends); Difference (d): Difference the non-stationary data d times (to make the data stationary); Moving average (q): Corrects the current forecast using the past q forecast errors (capturing random fluctuations).

[0131] 6022. Prediction Steps: 60221. Data Preprocessing: Perform an ADF test on the pollutant concentration time series (such as SO2 concentration collected every minute) to determine whether it is stationary; If it is not stationary, perform d differences.

[0132] 60222. Parameter selection: Determine the optimal p, d, q using the AIC / BIC criteria.

[0133] 60223. Model Fitting and Prediction: Fit the ARIMA model with historical data (such as the concentration in the past hour); output the concentration prediction value for the next 15 minutes.

[0134] Through spatial and temporal fusion, the output is: Multi-source superimposed pollution distribution: By integrating the CFD simulated concentrations and LIBS measured concentrations of various pollution sources, a heat map of pollution distribution after multi-source superposition is obtained, and the area with the most severe pollution is identified. Dynamically calibrated concentration field: 4D concentration field C after dual-driven calibration by environment and data calib It has higher accuracy than a single model, providing a high-precision data foundation for subsequent analysis.

[0135] Collaborative analysis, based on the fused data, outputs survey results, which include: Pollution source contribution ratio: Calculate the responsibility ratio of each pollution source using the contribution rate formula to clarify emission reduction priorities; 15-minute trend prediction: Concentration prediction output by the ARIMA model to guide real-time scheduling.

[0136] Comprehensive monitoring report: Includes: spatial pollution distribution heat map, 15-minute trend curve over time, pollution source contribution ranking based on responsibility, and emission reduction recommendations. The emission reduction recommendations are optimized solutions targeting major pollution sources, such as prioritizing the treatment of chimney A, which is expected to reduce emissions by 30% and thus lower the total concentration by 18%.

[0137] Based on the same inventive concept as the multi-point survey method provided in the embodiments of this application, the embodiments of this application also provide a multi-point survey system. If there is anything unclear about the content in the system embodiments, please refer to the corresponding content in the method embodiments.

[0138] A multi-point survey system, comprising: Sampling module: Equipped with an improved sampling probe and gas collection unit, it is used to collect raw flue gas from multiple monitoring points. Through time-series switching, it can achieve continuous sampling at multiple points to obtain raw flue gas. Pre-processing module: Simultaneously performs anti-condensation control, anti-interference treatment, and self-cleaning maintenance during the sampling of raw flue gas, and outputs clean sample gas; Diagnostic early warning module: Based on a three-level network of "probe box - diagnostic edge gateway - expert system", it monitors multi-dimensional parameters of hardware, data and environment, identifies abnormal points through residual calculation and adaptive threshold analysis, and outputs monitoring point status labels and abnormal parameter vectors; Fault handling module: Constructs a multimodal analysis matrix based on abnormal parameter vectors, combines a convolutional neural network model to match historical fault cases, determines the fault type and confidence level, and executes a graded handling scheme according to the confidence level; Spectral analysis and model building module: LIBS spectral technology is used to obtain the elemental characteristics of particulate matter, coupled with a CFD model and the diffusion coefficient is corrected by measured data to build a 4D pollution diffusion model that includes spatial three-dimensional and temporal dimensions; Dynamic calibration module: It integrates environmental parameters and pollutant data through the Transformer model, dynamically adjusts the weights of CFD model prediction results and measured data based on the fluctuation rate of environmental parameters, and outputs the calibrated pollutant diffusion results. Data fusion and analysis module: Spatially fuses the calibrated 4D model data to calculate the contribution ratio of pollution sources, uses time series models to predict short-term pollution trends, and integrates data to output a comprehensive monitoring report.

[0139] In another embodiment, the sampling module includes a sampling unit and a gas collection unit; the sampling unit contains sampling probes for multiple monitoring points, and the gas collection unit uses a timing controller to drive a multi-way valve or an independent pipeline system to achieve point switching every 30 seconds; The pretreatment module includes: an anti-condensation unit, an anti-interference unit, and a self-cleaning unit; the anti-condensation unit integrates a heating ring and a humidity sensor, and adjusts the heating power based on real-time humidity; the anti-interference unit adopts a dual-cavity isolation structure, with the main cavity collecting target pollutant flue gas and the secondary cavity collecting background gas filtered by activated carbon or molecular sieves; the self-cleaning unit is equipped with a 650nm wavelength laser scanning component, which initiates cleaning at a fixed cycle or as needed; The diagnostic early warning module includes a parameter acquisition unit, a residual calculation unit, a threshold update unit, and a status determination unit. The parameter acquisition unit acquires hardware, data, and environmental parameters via a CAN bus. The residual calculation unit analyzes the deviation of the parameters from the historical average. The threshold update unit dynamically adjusts the adaptive threshold. The status determination unit outputs normal, suspected fault, or confirmed fault labels and triggers collaborative analysis by the expert system. The spectral analysis and model building module includes a LIBS analysis unit, a CFD simulation unit, and a 4D model integration unit. The LIBS analysis unit periodically emits lasers to obtain the intensity of elemental characteristic peaks. The CFD simulation unit builds a basic model based on the three-dimensional layout of the factory and corrects the diffusion coefficient by combining LIBS data. The 4D model integration unit stores the concentration fields at different time steps in time sequence, forms a four-dimensional matrix, and renders a dynamic heat map through the WebGL engine.

[0140] The pretreatment module outputs clean sample gas to the diagnostic early warning module for real-time anomaly detection.

[0141] The diagnostic early warning module outputs status labels (including normal, suspected fault, and confirmed fault) and abnormal parameter vectors (for confirmed fault points) for each monitoring point. Among them, the monitoring data under normal conditions will be used as valid monitoring data and participate in subsequent spectral analysis and model construction; while the abnormal parameter vectors will be transmitted to the fault handling module for fault root cause mining.

[0142] After determining the fault type and handling plan through multimodal analysis and convolutional neural network model, the fault handling module will perform hierarchical processing, generate processing instructions and feed them back to the sampling module and preprocessing module. The processing instructions include automatic switching to backup equipment and pushing maintenance orders. For example, the instruction to start the backup probe will be sent to the sampling module.

[0143] Monitoring data under normal conditions are entered into the spectral analysis and model building module, and a 4D pollution diffusion model is constructed by combining LIBS spectral data.

[0144] The 4D model output by the spectral analysis and model building module is fused with environmental parameters in the dynamic calibration module, and the model weights are dynamically adjusted through the Transformer algorithm.

[0145] The pollutant diffusion results output by the dynamic calibration module flow to the data fusion analysis module, which combines spatial fusion and temporal trend prediction to finally generate a comprehensive monitoring report.

[0146] Based on the same inventive concept as the multi-point survey method provided in the embodiments of this application, the embodiments of this application also provide a multi-point survey device. If there is anything unclear about the content in the device embodiment, please refer to the corresponding content in the method embodiment.

[0147] A multi-point monitoring device includes a sampling device 700 and a computer device 800. The sampling device 700 is used to collect flue gas samples from various monitoring points in a flue gas emission system. The computer device 800 is used to analyze the collected flue gas samples. The computer device includes a processor 801 and a memory 802. The memory 802 stores computer-readable instructions. When the processor 801 executes the computer-readable instructions, it implements the steps of the method described above.

[0148] Specifically, the sampling device 700 includes a probe box 701 and a control box 702 located at each flue gas monitoring point. A gas collection unit 703 is connected to the air inlet of the probe box 701. The control box 702 provides pneumatic and electrical control for the probe box 701 and the gas collection unit 703. The gas collection unit 703 integrates a gas path module, pneumatic shut-off valves 1 / 2 / 3, a heat tracing pipeline interface, a heating temperature control unit, and an insulation shell. In the process preceding the probe's flue gas collection, a pre-treatment device is used for flue gas anti-condensation control, dual-chamber isolation anti-interference treatment, and probe laser self-cleaning treatment. A jet pump inside the probe box 701 provides power to extract the flue gas sample. The sample enters the probe box 701 via the gas collection unit 703. The pneumatic shut-off valves 1 / 2 / 3 in the gas collection unit 703 can be controlled via relay settings in the control box 702 to achieve individual or mixed sampling of one, two, or three channels.

[0149] The computer device 800 includes a memory 802, a processor 801, and a network interface 803 that are interconnected via a system bus. It should be noted that only a computer device with components 801-803 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device 800 described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, embedded devices, etc.

[0150] Computer device 800 can be a computer or a cloud server, etc. Computer device 800 can interact with users through a keyboard, mouse, remote control, touchpad, or voice control device.

[0151] The memory 802 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory, random access memory, static random access memory, read-only memory, electrically erasable programmable read-only memory, programmable read-only memory, magnetic memory, disk, optical disk, etc.

[0152] The processor 801 may be a central processing unit, a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. This processor is typically used to control the overall operation of the device. In the embodiments of this application, the processor is used to execute computer-readable instructions stored in the memory or to process data, such as executing computer-readable instructions of the methods provided in the embodiments of this application.

[0153] The network interface 803 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device and other electronic devices.

[0154] 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 multi-point survey method, characterized in that, The method includes the following steps: S100 continuously collects raw flue gas from multiple monitoring points through time-series switching, and simultaneously performs anti-condensation control, dual-cavity isolation anti-interference treatment and laser self-cleaning, and outputs clean sample gas. S200 constructs a three-level diagnostic network based on hardware, data, and environmental parameters. It identifies abnormal points in real time and triggers early warnings through residual detection and adaptive threshold mechanism, and outputs the status label and abnormal parameter vector of each monitoring point. The S300 system combines acoustic signature, differential pressure, and spectral multimodal data to locate the cause of the fault, generates a processing plan by matching historical case databases through a convolutional neural network, and executes a graded response. By combining S400, LIBS elemental concentration data, and CFD fluid dynamics model, the diffusion coefficient is dynamically corrected by measured concentration gradient, and a 4D pollution diffusion model including spatial three-dimensional and temporal dimensions is constructed. S500, based on the Transformer model, integrates environmental parameters and pollutant data, dynamically adjusts the weighting coefficients of CFD prediction and measured data according to environmental volatility, and outputs calibrated pollutant diffusion results. S600 calculates the contribution ratio of each pollution source through spatial fusion, uses a time series model to predict future short-term pollution trends, and generates a comprehensive monitoring report that includes spatial heat maps, responsibility rankings, and emission reduction recommendations.

2. The method according to claim 1, characterized in that, In S100, the time-series switching sampling includes: setting up multiple monitoring points in the flue gas emission system, installing a sampling probe at each monitoring point, and automatically switching the sampling point position according to a preset time sequence by the gas collection unit. The anti-condensation control dynamically adjusts the heating power based on real-time humidity and switches to a backup probe when there is a risk of condensation. The anti-interference processing adopts synchronous sampling of the main and auxiliary cavities, and eliminates spectral interference by using the background gas difference value.

3. The method according to claim 1, characterized in that, In the S200, the diagnostic network adopts a three-level architecture of "probe box - diagnostic edge gateway - expert system" and blockchain evidence storage; The deviation between the current parameter value and the historical mean is calculated by the quantum residual algorithm. Combined with the adaptive threshold to trigger the hybrid expert system, the adaptive threshold is dynamically updated by the historical threshold and the absolute value of the current residual. The status labels include normal, suspected fault, and confirmed fault. Repeated testing is initiated for suspected fault points.

4. The method according to claim 1, characterized in that, In S300, a multimodal analysis matrix is ​​constructed to compare acoustic signature MFCC offset, differential pressure oscillation frequency, and spectral confidence. By matching anomaly vectors with a historical fault feature library using CNN, the fault type and confidence level are output. When the confidence level is greater than the threshold, the backup module is automatically switched; otherwise, a maintenance order is pushed.

5. The method according to claim 1, characterized in that, In S400, LIBS spectral analysis emits laser pulses to obtain particulate matter elemental fingerprints, calculates elemental concentrations through characteristic peak intensities, and analyzes pollutant source components. The coupling of CFD and LIBS includes: obtaining the initial concentration of pollutants based on the initial CFD simulation, correcting the CFD diffusion coefficient by combining the measured concentration gradient of LIBS, recalculating the corrected simulated concentration, and using the adjoint method to invert the location of the pollution source; The 4D pollution diffusion model stores the concentration fields at different time steps in chronological order to form a four-dimensional matrix, and then renders it with the WebGL engine to obtain a dynamic heat map of pollutants, which can be used to trace the historical diffusion path of any location.

6. The method according to claim 1, characterized in that, In the S500, the environmental data includes wind speed, ambient temperature, and relative humidity, and the pollutant data is the elemental concentration distribution obtained from LIBS spectral analysis. The environmental temporal sequence and pollutant spatial matrix are correlated through the Transformer multi-head attention mechanism; Weighting coefficients are calculated based on wind speed and humidity fluctuation rates, and CFD predictions and LIBS measured data are dynamically integrated. When the environmental volatility is greater than 10%, measured data should be used first, and when the environmental volatility is less than 5%, the prediction weight of the model should be strengthened.

7. The method according to claim 1, characterized in that, In S600, spatial integration quantifies the responsibility ratio of each pollution source through a contribution rate formula; The time trend prediction uses the ARIMA model to predict future concentration trends based on historical concentration sequences.

8. A multi-point survey system, applied to the method described in any one of claims 1-7, characterized in that, include: Sampling module: Equipped with an improved sampling probe and gas collection unit, it is used to collect raw flue gas from multiple monitoring points. Through time-series switching, it can achieve continuous sampling at multiple points to obtain raw flue gas. Pre-treatment module: Simultaneously performs anti-condensation control, anti-interference treatment, and self-cleaning maintenance during the sampling of raw flue gas, and outputs clean sample gas; Diagnostic early warning module: Based on a three-level network of "probe box - diagnostic edge gateway - expert system", it monitors multi-dimensional parameters of hardware, data and environment, identifies abnormal points through residual calculation and adaptive threshold analysis, and outputs monitoring point status labels and abnormal parameter vectors; Fault handling module: Constructs a multimodal analysis matrix based on abnormal parameter vectors, combines a convolutional neural network model to match historical fault cases, determines the fault type and confidence level, and executes a graded handling scheme according to the confidence level; Spectral analysis and model building module: LIBS spectral technology is used to obtain the elemental characteristics of particulate matter, coupled with a CFD model and the diffusion coefficient is corrected by measured data to build a 4D pollution diffusion model that includes spatial three-dimensional and temporal dimensions; Dynamic calibration module: It integrates environmental parameters and pollutant data through the Transformer model, dynamically adjusts the weights of CFD model prediction results and measured data based on the fluctuation rate of environmental parameters, and outputs the calibrated pollutant diffusion results. Data fusion and analysis module: Spatially fuses the calibrated 4D model data to calculate the contribution ratio of pollution sources, uses time series models to predict short-term pollution trends, and integrates data to output a comprehensive monitoring report.

9. The system according to claim 8, characterized in that, The sampling module includes a sampling unit and a gas collection unit; the sampling unit contains sampling probes for multiple monitoring points, and the gas collection unit uses a timing controller to drive a multi-way valve or an independent pipeline system to achieve point switching every 30 seconds. The pretreatment module includes: an anti-condensation unit, an anti-interference unit, and a self-cleaning unit; the anti-condensation unit integrates a heating ring and a humidity sensor, and adjusts the heating power based on real-time humidity; the anti-interference unit adopts a dual-cavity isolation structure, with the main cavity collecting target pollutant flue gas and the secondary cavity collecting background gas filtered by activated carbon or molecular sieves; the self-cleaning unit is equipped with a 650nm wavelength laser scanning component, which initiates cleaning at a fixed cycle or as needed; The diagnostic early warning module includes a parameter acquisition unit, a residual calculation unit, a threshold update unit, and a status determination unit. The parameter acquisition unit acquires hardware, data, and environmental parameters via a CAN bus. The residual calculation unit analyzes the deviation of the parameters from the historical average. The threshold update unit dynamically adjusts the adaptive threshold. The status determination unit outputs normal, suspected fault, or confirmed fault labels and triggers collaborative analysis by the expert system. The spectral analysis and model building module includes a LIBS analysis unit, a CFD simulation unit, and a 4D model integration unit. The LIBS analysis unit periodically emits lasers to obtain the intensity of elemental characteristic peaks. The CFD simulation unit builds a basic model based on the three-dimensional layout of the factory and corrects the diffusion coefficient by combining LIBS data. The 4D model integration unit stores the concentration fields at different time steps in time sequence, forms a four-dimensional matrix, and renders a dynamic heat map through the WebGL engine.

10. A multi-point survey device, comprising sampling equipment and computer equipment, characterized in that, The sampling device is used to collect flue gas samples from various monitoring points in the flue gas emission system. The computer device is used to analyze the collected flue gas samples. The computer device includes a processor and a memory. The memory stores computer-readable instructions. When the processor executes the computer-readable instructions, it implements the steps of the method as described in any one of claims 1 to 7.

Images