Ship exhaust gas multi-component synchronous online monitoring system and method based on laser Raman spectrum

Abstract

The invention belongs to the technical field of gas detection, and relates to a ship exhaust gas multi-component synchronous online monitoring system and method based on a laser Raman spectrum. The system comprises a sampling module, a preprocessing module, an analysis module, a data processing module, a blowback module and a flow measurement module. The method comprises the following steps: starting and self-checking the system; sample collection, pretreatment and flow velocity measurement; laser excitation and signal acquisition; carrying out spectrum analysis and concentration inversion; data output and system maintenance; and performing periodic automatic calibration. According to the invention, a complete monitoring process chain from sampling, pretreatment, flow measurement, analysis to back flushing is creatively integrated. The high-efficiency pretreatment and back flushing design greatly reduces the influence of high-dust and high-humidity environment on the analysis system, and ensures the long-term stability and reliability of the system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of gas detection and relates to a ship exhaust gas multi-component synchronous online monitoring system and method based on laser Raman spectrum. BACKGROUND

[0002] At present, commercial ship CEMS (Continuous Emission Monitoring System) mostly adopts a combination scheme of NDIR (Non-Dispersive Infrared), UV-DOAS (Ultraviolet Differential Optical Absorption Spectroscopy) and electrochemical sensor technologies. The NDIR technology is sensitive to CO2, CO and the like, but one sensor corresponds to only one gas, and multiple sensors are needed for multi-component measurement, which has problems of complex equipment, high cost, cross interference and the like; the UV-DOAS technology is suitable for measurement of SO2, NO and the like, but has no response to symmetrical molecules such as CO2 and CH4; and the electrochemical sensor has a short service life, needs frequent calibration and has insufficient stability. The above-mentioned traditional technologies cannot realize synchronous, rapid and accurate measurement of all key gas components (such as SO2, NO x , CO, CO2, CH4, NH3 and the like) through a single technology platform, and the system reliability faces severe challenges in the harsh environment of high humidity, high particulate matter and strong vibration of a ship.

[0003] The laser Raman spectrum technology has the characteristics of "molecular fingerprint" and can theoretically simultaneously qualitatively and quantitatively measure all molecular gas, which is an ideal solution to the above-mentioned problems. However, direct application of the laboratory technology to online monitoring of ship exhaust gas faces a series of unprecedented technical challenges: 1) the characteristics of high temperature, high humidity and high particulate matter of ship exhaust gas are easy to contaminate and damage precise optical elements; 2) the complex vibration environment of a ship affects the long-term stability of an optical system; and 3) the composition of ship exhaust gas is complex, and its overlapping spectral characteristics need a special analysis algorithm. Therefore, there is an urgent need for a complete system solution that can overcome the above-mentioned challenges and is oriented to ship applications. SUMMARY

[0004] The application aims to overcome the challenges of the harsh environment (high temperature, high humidity, high particulate matter and strong vibration) of a ship to a Raman spectrum system and provide a ship exhaust gas multi-component synchronous online monitoring system and method based on laser Raman spectrum, which realizes synchronous, rapid, accurate and stable online measurement of multiple gas components such as SO2, NO x , CO, CO2, CH4 and NH3.

[0005] To achieve the above object, the application adopts the following technical scheme: a ship exhaust multi-component synchronous online monitoring system based on laser Raman spectrum, comprising a sampling module, a pretreatment module, an analysis module, a data processing module and a flow measurement module; the sampling module adopts a sampling probe with a heating filter to extract original exhaust samples; the pretreatment module performs rapid cooling, dehydration and dust removal on the sample gas collected by the sampling module; the flow measurement module uses an ultrasonic gas flow meter to obtain the gas flow rate of the ship exhaust; the analysis module uses a Raman integrating sphere spectrometer to detect the gas molecular components in the ship exhaust; the data processing module uses a spectrum denoising algorithm to denoise the spectrum data collected by the analysis module and performs baseline correction; a concentration inversion algorithm model is used for fitting calculation to accurately invert the concentration values of each gas component; the spectrum denoising algorithm uses an improved adaptive wavelet threshold denoising method to reconstruct the signal, and the reconstructed signal is input into a Kalman filter to suppress random noise; the improved adaptive wavelet threshold denoising method uses a continuously derivable threshold function for adaptive threshold estimation: ; wherein, is a wavelet coefficient; is a threshold value.

[0006] Preferably, the baseline correction uses an adaptive iterative reweighted penalized least squares method.

[0007] Preferably, the concentration inversion algorithm model uses a fusion of sparse representation and kernel partial least squares algorithm, specifically including: (1) sparse coding dictionary learning Constructing a dictionary , containing the Raman spectrum of standard gas components, the spectrum to be measured is expressed as: ; wherein, is a sparse concentration vector, is an error; Solve by LASSO model: ; Use the coordinate descent method to iteratively update : ; wherein, is a soft threshold function; is a regularization parameter; (2) take the sparse coding result as the input feature and perform KPLS regression with the real concentration value; the kernel function uses a Gaussian kernel: ; The KPLS model obtains the latent variable by solving an eigenvalue problem, and the final concentration prediction is: ; Wherein, the coefficient is learned by least square; is the number of training samples, is a constant.

[0008] Preferably, the concentration inversion algorithm model is updated online by using an adaptive calibration algorithm; the adaptive calibration algorithm combines recursive least squares and incremental learning, and specifically comprises: For a new data point , the concentration inversion algorithm model weight is updated as follows: ; Wherein, the Kalman gain is: ; The covariance matrix is updated as follows: ; is a forgetting factor, and is 0.95-0.99.

[0009] Preferably, when the new data and the old model have a large deviation, the dictionary and the kernel matrix are updated.

[0010] Preferably, the system further comprises a backflushing module for removing the accumulated soot particles in the sampler.

[0011] The application further provides a ship exhaust gas multi-component synchronous online monitoring method based on laser Raman spectrum. S1: system starting and self-checking: starting the temperature control and active damping system, preheating the laser, the spectrometer and the heat tracing system to the set temperature; S2: sample collection, pretreatment and flow rate measurement: after dust removal, filtration and full-path heat tracing by the pretreatment module, dry and clean sample gas is obtained; at the same time, the ultrasonic flowmeter synchronously measures the exhaust gas flow rate; S3: laser excitation and signal collection: the laser in the analysis module emits laser light, excites the gas molecules in the sample cell to produce Raman scattering, and the spectrum collection unit synchronously collects the spectrum signal; S4: spectrum analysis and concentration inversion: the data processing module intelligently analyzes the spectrum data, inversely calculates the concentration values of each gas component in real time, and calculates the emission rate in combination with the flow rate data.

[0012] Preferably, it also includes data output and system maintenance: outputting concentration and emission data to the ship end and shore-based monitoring center; the back flushing module automatically performs back flushing cleaning operation at a preset period.

[0013] Preferably, it also includes periodic automatic calibration of the system: using standard gas for measurement and calibrating the algorithm model.

[0014] The beneficial effects of the present application are: (1) Technological advancement: based on the "molecular fingerprint" characteristics of laser Raman spectroscopy, a set of systems can realize the simultaneous, rapid and accurate measurement of all key gas pollutants such as NO x , SO2, CO, CO2, NH3, CH4, O2, N2, H2O, etc., breaking through the limitations of traditional multi-technology splicing solutions; (2) Process integrity: innovatively integrating a complete monitoring process chain from sampling, pretreatment, flow measurement, analysis to back flushing. Efficient pretreatment and back flushing design greatly reduces the impact of high dust and high humidity environments on the analysis system, ensuring long-term stability and reliability of the system; (3) System robustness: through comprehensive environmental adaptability design and precise thermal management against water, corrosion, vibration and high temperature, the system can meet the severe challenges of the harsh environment of the ship engine room, ensuring long-term normal and stable operation of the equipment; (4) The present application first combines laser Raman spectroscopy technology with special pretreatment subsystems and special algorithm modules for extreme working conditions of ships to form a complete system that is stable, reliable and commercially available, solving the engineering problems in this specific application field. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a schematic diagram of the overall structure of the system of the present application; Figure 2 is a gas Raman analysis system; Figure 3 is a flowchart of the sampling and pretreatment subsystem in the present application. DETAILED DESCRIPTION

[0016] In order to facilitate the understanding of the present application, the specific embodiments of the present application are further described in detail below in combination with the drawings and specific examples. The following examples or drawings are used to illustrate the present application, but not to limit the scope of the present application.

[0017] Example one, the present embodiment provides a ship exhaust multi-component simultaneous online monitoring system based on laser Raman spectroscopy, such as Figure 1As shown, the system includes a sampling module 101, a pretreatment module 102, an analysis module 103, a data processing module 104, a back flushing module 105, and a flow measurement module 106, and has high environmental adaptability.

[0018] The sampling module 101 considers factors such as temperature, pressure, water content, dust content, and measured components in combination with the flue gas conditions. Considering that the flue gas after combustion is high in temperature and contains a large amount of dust and water, a sampling probe with a heating filter is used to avoid condensate from being precipitated during sampling. The sampling probe is directly installed in the engine exhaust pipe to extract a representative raw exhaust sample.

[0019] Due to the complexity of ship fuel, substances such as oil stains, smoke dust, and oil-water mixtures are generated in the exhaust gas, which can cause blockage and corrosion of the sampler and sampling pipeline, thereby affecting the detection accuracy of the ship exhaust online monitoring equipment. Therefore, a ship exhaust pretreatment module 2 is needed to pretreat the sampling gas to prevent dust from blocking the sampling probe 102. The pretreatment module 102 first uses a cold-dry extraction method to sample the ship exhaust gas after filtering through a heating-type sampling probe, and then transports the sample gas to a condenser through a heat tracing pipe for rapid cooling, dehydration, and dust removal to prevent SO2 and CO2 from dissolving in water and affecting the detection accuracy. The cold-dry extraction method keeps all other components unchanged except for reducing the dust concentration of the gas during sampling. Secondly, the sample gas after cold-dry extraction is further filtered through a fine filter to remove dust particles, and then the flow and pressure of the sample gas are adjusted through a flow meter to make the sample gas become dust-free, water-free, and stable in flow and pressure, and then sent to the analysis module 3 for concentration analysis.

[0020] To meet the detection requirements of complex ship exhaust components, the analysis module 103 carries out ship exhaust component analysis based on Raman sphere integration. The gas detection limit of the Raman integral sphere spectrometer is high, and it can qualitatively and quantitatively detect all molecular gases. The gas molecules in the ship exhaust are detected. The gas Raman analysis system is shown in FIG. 3, which includes a signal collection light path 201, a signal enhancement light path 202, a sample cell 203, a Raman integral sphere spectrometer 204, a CCD 205, a gas pump 206, and a flow meter 207. The signal enhancement light path is composed of a laser, a light shield mirror, a beam splitter, a focusing lens, a collection lens, and other optical devices. The laser is focused on the ship exhaust in the sample cell through a reflecting mirror and a focusing lens to excite the Raman scattering signal of the ship exhaust. The signal is collected by a collection mirror, and the stray light signal is filtered out by a Raman filter. The Raman signal is transmitted to the Raman integral sphere spectrometer 204 for recording, and the analysis results of the gas components and their proportions are given by the Raman spectrum data analysis software.

[0021] The data processing module 104 is an intelligent spectral analysis and control unit. This module is responsible for controlling the coordinated operation of all hardware units and uses a spectral denoising algorithm to denoise the acquired spectral data; it employs baseline correction and concentration inversion algorithm models to accurately invert the concentration values ​​of each gas component, and integrates a self-learning function, allowing for model updates through an adaptive calibration algorithm. The algorithms involved in the data processing module 104 are detailed below: 1. The spectral denoising algorithm employs a fusion of adaptive wavelet threshold denoising and Kalman filtering. Traditional wavelet denoising uses a fixed threshold, which easily leads to signal distortion. This invention combines adaptive wavelet threshold denoising and Kalman filtering to achieve multi-scale noise suppression.

[0022] The specific algorithm steps are as follows: (1) Wavelet decomposition Use Daubechies wavelet bases (such as db4) to process the original spectral signal. Multi-scale decomposition is performed to obtain wavelet coefficients. ,in As a scale, For location.

[0023] (2) Adaptive threshold estimation Thresholds for wavelet coefficients at each level are calculated based on Bayesian estimation. : ; in, For the first j The noise standard deviation of the layer wavelet coefficients is estimated using the following formula: ; N This is the signal length.

[0024] (3) Improve the threshold function Using a continuously differentiable threshold function instead of a traditional soft threshold reduces distortion. .

[0025] This function smoothly attenuates small-amplitude noise while preserving the large-amplitude coefficients.

[0026] (4) Kalman filtering post-processing Reconstruct the signal using wavelets The input Kalman filter further suppresses random noise. The state-space model is as follows: ; in, For example, a state vector (such as a spectral baseline). and For the transition matrix and observation matrix, and are process noise and observation noise. The state estimate is updated by the Kalman gain ; ; where, is the error covariance matrix, is the noise covariance.

[0027] 2. Baseline correction uses adaptive iterative reweighted penalized least squares. Traditional baseline correction is poor at adapting to complex baselines. The present invention uses the airPLS algorithm to automatically identify and fit the baseline.

[0028] The objective function is: ; where, is the original spectral intensity, is the baseline estimate, is the weight, is the smoothing parameter (usually set to ), is the number of data points for a single spectrum.

[0029] The iterative steps are: (1) Initialize the weights ; (2) Solve the linear system to obtain the baseline :

[0030] where, is the diagonal weight matrix, is the first-order difference matrix; (3) Update the weights: ; is the standard deviation of the current residual.

[0031] (4) Repeat steps 2-3 until the weights change less than a threshold (e.g. ).

[0032] 3. The concentration inversion algorithm model is a fusion model based on sparse representation and kernel partial least squares (KPLS). For spectral overlap and nonlinear problems, sparse representation and KPLS are combined to improve inversion accuracy. The algorithm steps of the model are as follows: (1) Sparse coding dictionary learning Construct a dictionary ​, Raman spectrum of standard gas components (obtained by experiment or database). The spectrum to be measured is denoted as: ; where, is the sparse concentration vector, is the error.

[0033] Solved by LASSO model: ; Iterative update using coordinate descent method : ; where, is the soft threshold function; is the regularization parameter.

[0034] (2) Take the sparse coding result as input features, and perform KPLS regression with the true concentration value. The kernel function uses Gaussian kernel: ; The KPLS model obtains latent variables by solving eigenvalue problems, and the final concentration prediction is: ; where, is the number of training samples, is a constant; the coefficient is learned by least squares.

[0035] 4. The adaptive calibration algorithm combines recursive least squares (RLS) with incremental learning to achieve online model updating and adapt to long-term drift.

[0036] The algorithm formula is as follows: For new data points , update the concentration inversion model weight : ； where, the Kalman gain is: ; The covariance matrix is updated as: ; where, is the forgetting factor, usually taken as 0.95~0.99.

[0037] At the same time, an incremental learning mechanism is introduced: when the new data has a large deviation from the old model, trigger the dictionary And the update of the nuclear matrix, avoid model degradation.

[0038] The execution flow of the data processing module 104 is as follows: Step 1: input: original Raman spectrum signal ; Step 2: denoising: apply adaptive wavelet-Kalman fusion denoising, output ; Step 3: baseline correction: use airPLS algorithm to fit the baseline , get the net signal ; Step 4: concentration inversion: first get the preliminary concentration by sparse coding, and then nonlinearly correct it by KPLS model, output the final concentration value c ; Step 5: adaptive calibration: according to the real-time data stream, update the model parameters by RLS, and recalibrate the dictionary regularly and kernel function.

[0039] The above improvements make the data processing module realize more efficient and accurate multi-component gas monitoring in complex ship environment.

[0040] To prevent oil stains, smoke and other attachments in ship exhaust from affecting the detection accuracy of the sampler, a back flushing module 105 is provided. By reasonable control, the switching action of the valve and other equipment on the pipeline during the sampling analysis and back flushing process is executed, the accumulated soot particles are removed, the sampling is ensured to be smooth and the optical path is clean, the stability of the measurement of the analysis system is ensured, and the problem of easy blockage and large maintenance of the sampler is solved.

[0041] The flow measurement module 106 adopts an ultrasonic gas flowmeter which has good adaptability to wet gas, dirty gas and mixed gas medium. According to the principle that the difference between the propagation speeds of ultrasonic waves in forward flow and reverse flow is related to the flow rate of the medium, the gas flow rate of the ship exhaust is obtained. The emission and reception of ultrasonic waves adopt a double probe mode, which can be used interchangeably or bidirectionally.

[0042] The system's environmental adaptability design is based on the process flow of ship exhaust gas emissions. Research is conducted on the environmental adaptability of ship exhaust gas emission monitoring equipment to ensure long-term stable operation. 1) Temperature adaptability: ① High-temperature sampling probes and tubes are used, with heating and insulation treatment applied to all joints. For example, stainless steel + lightweight refractory materials are used, with a maximum allowable temperature of 400℃; ② Finite element analysis is used to analyze the temperature field of high-temperature components, ensuring the stability and reliability of the overall equipment structure at high temperatures. 2) Humidity adaptability: ① A cold-dry extraction sampling method is adopted, where the collected flue gas samples undergo dust removal through filters, dehydration through condensers, and flow rate adjustment before entering the analysis and measurement module; ② Gas drying tubes composed of Nafion polymers are selected, allowing selective filtration of moisture from the gas samples while keeping most analytes unaffected. 3) Dust and oil resistance: ① To prevent particle sedimentation in the sampler from affecting detection accuracy, backflushing functions are added to the pipeline and sampler itself. The entire sampling gas path is backflushed to prevent contaminants from settling in the sampler and to prevent the impact of sudden temperature changes on the sensor. ② To prevent oil, oil-water mixtures, etc., from clogging the sampling tube, a layer of Teflon is added to the surface of the sampling tube. 4) Corrosion resistance: ① Stainless steel and heat-resistant steel with good corrosion resistance are selected as equipment materials, and the impact of processing methods on the material's corrosion resistance is fully considered. ② The equipment structure is rationally designed to minimize gaps and liquid accumulation.

[0043] Example 2: This example provides a method for online monitoring using the above system, based on the system of Example 1, such as... Figure 3 As shown, the method includes the following steps: S1: System startup and self-test. The system is powered on, the temperature control and active vibration damping system are activated, and the laser, spectrometer and heat tracing system are preheated to the set temperature.

[0044] S2: Sample collection, pretreatment, and flow rate measurement. Waste gas is extracted through the sampling module, and after dust removal, filtration, and full-process heating by the pretreatment module, a dry and clean sample gas is obtained; at the same time, the ultrasonic flow meter measures the waste gas flow rate.

[0045] S3: Laser Excitation and Signal Acquisition. The laser in the analysis module emits a laser beam, which excites gas molecules in the sample cell to produce Raman scattering. The spectral acquisition unit simultaneously acquires the spectral signal.

[0046] S4: Spectral Analysis and Concentration Inversion. The data processing module intelligently analyzes the spectral data, inverts the concentration values ​​of each gas component in real time, and calculates the emission rate by combining it with flow velocity data.

[0047] S5: Data output and system maintenance. Concentration and emission data are output to the ship end and shore-based monitoring center; the backflush module automatically performs backflush cleaning operation at a preset period.

[0048] S6: Periodic automatic calibration. The system automatically switches to standard gas, measures and calibrates the algorithm model to ensure long-term monitoring accuracy.

Claims

1. A ship exhaust multi-component synchronous online monitoring system based on laser Raman spectrum, characterized in that: The system includes a sampling module, a preprocessing module, an analysis module, a data processing module, and a flow measurement module. The sampling module uses a sampling probe with a heated filter to extract raw exhaust gas samples. The preprocessing module rapidly cools, dehydrates, and removes dust from the sample gas collected by the sampling module. The flow measurement module uses an ultrasonic gas flow meter to obtain the gas velocity of the ship's exhaust gas. The analysis module uses a Raman integrating sphere spectrometer to detect the gas molecular components in the ship's exhaust gas. The data processing module uses a spectral denoising algorithm to denoise the spectral data collected by the analysis module and performs baseline correction; it also uses a concentration inversion algorithm model for fitting calculations to accurately invert the concentration values ​​of each gas component. The spectral denoising algorithm uses an improved adaptive wavelet threshold denoising method to reconstruct the signal, and the reconstructed signal is input into a Kalman filter to suppress random noise. The improved adaptive wavelet threshold denoising method uses a continuously differentiable threshold function for adaptive threshold estimation. ； wherein is a wavelet coefficient; is a threshold value.

2. The synchronous online monitoring system for multiple components of ship exhaust gas based on laser Raman spectroscopy according to claim 1, characterized in that: Baseline correction employs an adaptive iterative reweighted penalized least squares method.

3. The synchronous online monitoring system for multiple components of ship exhaust gas based on laser Raman spectroscopy according to claim 1, characterized in that: The concentration inversion algorithm model described above combines sparse representation with kernel partial least squares algorithm, specifically including: (1) Sparse coding dictionary learning Building a dictionary Raman spectra of standard gas components, spectra to be measured Represented as: ; in, For sparse concentration vectors, For error; Solving using the LASSO model: ； Iterative updates using coordinate descent method : ； in, This is a soft thresholding function; For regularization parameters; (2) The sparse coding result As input features, KPLS regression was performed with the true concentration values; a Gaussian kernel was used as the kernel function. ； The KPLS model obtains latent variables by solving the eigenvalue problem, and the final concentration prediction is: ； Among them, coefficient Learning through least squares; It is the number of training samples. It is a constant.

4. The synchronous online monitoring system for multiple components of ship exhaust gas based on laser Raman spectroscopy according to claim 1, characterized in that: An adaptive calibration algorithm is used to update the concentration inversion algorithm model online; the adaptive calibration algorithm combines recursive least squares with incremental learning, specifically including: For new data points Update the weights of the concentration inversion algorithm model. : ； Among them, Kalman gain for: ； covariance matrix Updated to: ； The forgetting factor is set to 0.95~0.

99.

5. The synchronous online monitoring system for multiple components of ship exhaust gas based on laser Raman spectroscopy according to claim 4, characterized in that: When the new data deviates significantly from the old model, the dictionary is triggered. And the update of the kernel matrix.

6. The synchronous online monitoring system for multi-component marine exhaust gas based on laser Raman spectroscopy according to any one of claims 1-5, characterized in that: It also includes a backflushing module to remove dust particles accumulated in the sampler.

7. A method for simultaneous online monitoring of multiple components in ship exhaust gas based on laser Raman spectroscopy, characterized in that, The system according to claim 6 includes the following steps: S1: System Start-up and Self-Test: Start the temperature control and active vibration damping system, and preheat the laser, spectrometer and heat tracing system to the set temperature; S2: Sample collection, pretreatment and flow rate measurement: After dust removal, filtration and full-process heating by the pretreatment module, dry and clean sample gas is obtained; at the same time, the ultrasonic flow meter measures the waste gas flow rate. S3: Laser Excitation and Signal Acquisition: The laser in the analysis module emits a laser beam, which excites the gas molecules in the sample cell to produce Raman scattering, and the spectral acquisition unit simultaneously acquires the spectral signal. S4: Spectral Analysis and Concentration Inversion: The data processing module intelligently analyzes the spectral data, inverts the concentration values ​​of each gas component in real time, and calculates the emission rate by combining the flow rate data.

8. The method according to claim 7, characterized in that, It also includes data output and system maintenance: outputting concentration and emission data to the ship's end and shore-based monitoring center; the backflushing module automatically performs backflushing cleaning operations according to a preset cycle.

9. The method according to claim 7, characterized in that, It also includes periodic automatic calibration of the system: using standard gases for measurement and calibration of the algorithm model.

Images