Method, device and apparatus for multi-stage purification of ship exhaust gas

Abstract

The application discloses a ship flue gas multistage purification method, device and equipment, through collecting pressure pulsation signals, pollutant concentration signals and ship attitude signals, the pressure pulsation signals are subjected to time-frequency transformation to extract pulsation characteristic spectrum, and the pulsation characteristic spectrum is fused with the pollutant concentration signals to generate a dynamic pollution characteristic library; based on the characteristic library, particle grading clusters are identified, collision probability distribution is generated through flow field analysis, agglomeration characteristics are extracted to realize strengthened separation; according to the pulsation characteristic spectrum, vortex enhancement parameters are obtained to generate an optimized control matrix, so that primary purification is achieved; through droplet concentration characteristics, a seed density threshold value is determined, an adaptive purification sequence is constructed to realize secondary purification; the secondary purification state is coupled with the ship attitude signals to generate an inertial force field distribution, a scour peak point is identified to trigger pulse cleaning; through comparison between a purification stability index and the pollutant concentration signals, a performance deviation value is generated, and based on the deviation value, an optimized adjustment instruction is generated, and ship flue gas multistage purification is completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship pollution emission control, in particular to a ship flue gas multi-stage purification method, device and equipment. BACKGROUND

[0002] With the increasingly stringent standards of the International Maritime Organization on ship emissions, ship flue gas purification has become a major technical challenge faced by the shipping industry. Traditional single purification technology cannot simultaneously meet the efficient removal requirements of various pollutants such as particulate matter, sulfur oxides, and nitrogen oxides, and the purification efficiency fluctuates greatly and the operation stability is poor under complex sea conditions.

[0003] The existing ship exhaust gas treatment method mainly has the following deficiencies: lack of effective use of the inherent dynamic characteristics of the exhaust pipeline, unable to fully utilize the flow field pulsation to promote the capture of pollutants; ignoring the influence of sea navigation conditions on the purification equipment, the treatment effect significantly decreases under dynamic conditions such as ship body rolling; the aggregation and capture mechanism of particulate matter is single, and the removal capacity of submicron fine particles is insufficient; lack of long-term operation guarantee measures, and dirt deposition is easily formed inside the equipment leading to performance degradation. Therefore, it is urgent to develop a ship exhaust gas purification technology that can adapt to complex marine environment, realize the collaborative removal of various pollutants, and has intelligent control capability. SUMMARY

[0004] The present application discloses a ship flue gas multi-stage purification method, device and equipment, aiming to collect and fuse pressure pulsation signals, pollutant concentration signals and ship attitude signals, construct a dynamic pollution feature library, realize efficient agglomeration and fractional purification of particulate matter; identify the scouring peak point by using the inertial force field distribution, trigger pulse cleaning to form a scale control sequence, and ensure long-term stable operation of the equipment; realize adaptive control of the multi-stage purification process through real-time evaluation and optimization adjustment of the performance deviation value, and finally achieve efficient and stable removal of ship flue gas pollutants.

[0005] The present application discloses a ship flue gas multi-stage purification method, device and equipment, aiming to collect and fuse pressure pulsation signals, pollutant concentration signals and ship attitude signals, construct a dynamic pollution feature library, realize efficient agglomeration and fractional purification of particulate matter; identify the scouring peak point by using the inertial force field distribution, trigger pulse cleaning to form a scale control sequence, and ensure long-term stable operation of the equipment; realize adaptive control of the multi-stage purification process through real-time evaluation and optimization adjustment of the performance deviation value, and finally achieve efficient and stable removal of ship flue gas pollutants.

[0006] Collecting pressure pulsation signals, pollutant concentration signals and ship attitude signals of ship emission flue gas, performing time-frequency transformation on the pressure pulsation signals to extract pulsation characteristic spectrum, fusing the pulsation characteristic spectrum with the pollutant concentration signals to generate a dynamic pollution feature library;

[0007] Identifying particle classification clusters based on the dynamic pollution feature library, performing flow field analysis on the particle classification clusters to generate a collision probability distribution, extracting peaks from the collision probability distribution to form an agglomeration feature set, and outputting a strengthened separation control sequence based on the agglomeration feature set;

[0008] The performance evaluation of the enhanced separation control sequence extracts a reference separation index, acquires a vortex enhancement parameter based on the pulsation characteristic spectrum, superimposes the vortex enhancement parameter and the reference separation index to generate an optimization control matrix, and generates a primary purification state through the optimization control matrix;

[0009] The droplet concentration characteristic is extracted from the primary purification state, the seed density threshold is determined based on the droplet concentration characteristic, the adaptive purification sequence is constructed according to the seed density threshold, and the secondary purification state is output by executing the adaptive purification sequence;

[0010] The secondary purification state is coupled with the ship attitude signal to generate an inertial force field distribution, the peak value point of the scour is identified based on the inertial force field distribution, the pulse cleaning is triggered at the peak value point to generate an anti-fouling control sequence, and the purification stability index is acquired by executing the anti-fouling control sequence;

[0011] The performance deviation value is generated by comparing the purification stability index with the pollutant concentration signal, the optimization adjustment instruction is generated based on the performance deviation value, and the ship smoke multi-stage purification intelligent control is completed.

[0012] The second aspect of the present application proposes a ship smoke multi-stage purification device, comprising:

[0013] The signal acquisition module is used for acquiring the pressure pulsation signal, the pollutant concentration signal and the ship attitude signal of the ship exhaust flue gas, extracting the pulsation characteristic spectrum by time-frequency transformation on the pressure pulsation signal, and fusing the pulsation characteristic spectrum and the pollutant concentration signal to generate a dynamic pollution characteristic library;

[0014] The particle polymerization module is used for identifying particle hierarchical clusters based on the dynamic pollution characteristic library, generating a collision probability distribution by flow field analysis on the particle hierarchical clusters, forming an agglomeration characteristic set by peak value extraction on the collision probability distribution, and outputting an enhanced separation control sequence based on the agglomeration characteristic set;

[0015] The primary purification module is used for performance evaluation of the enhanced separation control sequence to extract a reference separation index, acquiring a vortex enhancement parameter based on the pulsation characteristic spectrum, superimposing the vortex enhancement parameter and the reference separation index to generate an optimization control matrix, and generating a primary purification state through the optimization control matrix;

[0016] The secondary purification module is used for extracting a droplet concentration characteristic from the primary purification state, determining a seed density threshold based on the droplet concentration characteristic, constructing an adaptive purification sequence according to the seed density threshold, and outputting a secondary purification state by executing the adaptive purification sequence;

[0017] A fouling control module is configured to couple the secondary purification state with the ship attitude signal to generate an inertial force field distribution, identify a scouring peak point based on the inertial force field distribution, trigger a pulse cleaning at the peak point to generate a fouling control sequence, and execute the fouling control sequence to obtain a purification stability index.

[0018] An intelligent adjustment module is configured to compare the purification stability index with the pollutant concentration signal to generate a performance deviation value, generate an optimized adjustment instruction based on the performance deviation value, and complete the intelligent control of the ship flue gas multi-stage purification.

[0019] A third aspect of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the ship flue gas multi-stage purification method disclosed in the first aspect when executing the program.

[0020] The beneficial effects of the present application are reflected in the following points: first, by fusing the pressure pulsation signal and the pollutant concentration signal to construct a dynamic pollution feature library, the precise identification of the emission law is realized, and the identified pulsation features are converted into vortex enhancement parameters, which enhances the agglomeration effect of particulate matter and improves the targeting of the multi-stage purification. Second, by extracting the seed density threshold from the droplet concentration feature, a direct correlation between droplet distribution and purification efficiency is established, and the seed density threshold can guide the real-time adjustment of the spraying parameters. By constructing an adaptive purification sequence and dynamically adjusting the execution parameters, the secondary purification control based on concentration feedback is realized, and the removal efficiency of fine particulate matter is improved. Finally, the ship attitude signal coupling technology is used to generate an inertial force field distribution, the pulse cleaning is triggered by combining the scouring peak point identification, and the long-term operation stability is ensured by the fouling control sequence, effectively solving the equipment fouling problem under dynamic working conditions at sea, and improving the overall operation reliability and long-term stability of the purification effect.

[0021] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings herein show specific examples of the technical solutions described in the present application, and constitute part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0023] Unless specifically stated or otherwise, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0024] Figure 1 is a flowchart of a ship flue gas multi-stage purification method.

[0025] Figure 2 is a structural diagram of a ship flue multi-stage purification device.

[0026] Figure 3 is a structural diagram of a computer device. DETAILED DESCRIPTION

[0027] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0028] It is to be understood that the terminology "including", "comprising", "consisting" and "consisting essentially of" used in the specification and the appended claims, indicates the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0029] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" or "in other embodiments" or "in some embodiments" in various places throughout this specification are not necessarily referring to the same embodiment, unless otherwise specified. The terms "including", "comprising", "consisting essentially of" and "consisting of" are used interchangeably, unless otherwise specified.

[0030] The technical solutions of the embodiments of the present application are introduced as follows.

[0031] As shown in Figure 1 The present application provides a ship flue multi-stage purification method, which comprises the following steps S110-S160:

[0032] In step S110, the pressure pulsation signal, the pollutant concentration signal and the ship attitude signal of the ship flue are collected, the pulsation characteristic spectrum is extracted by time-frequency transformation on the pressure pulsation signal, and the dynamic pollution characteristic library is generated by fusing the pulsation characteristic spectrum and the pollutant concentration signal.

[0033] Specifically, the pressure fluctuation signal of the exhaust gas, the pollutant concentration signal and the ship attitude signal are collected. A high-sensitivity pressure sensor array is arranged in the ship exhaust pipe and chimney system to monitor the pressure fluctuation signal of the exhaust gas in real time. The pressure sensor uses a piezoelectric dynamic pressure sensor with a frequency response range of 0.1 Hz to 10 kHz, which can capture the complete frequency spectrum of diesel engine combustion pulsation and turbulent pulsation. The sensors are distributed in the axial and circumferential directions, with an axial spacing of 0.5 meters and a circumferential spacing of 45 degrees. A three-dimensional monitoring network is formed. The pollutant concentration signal is collected by a multi-component gas analyzer, which monitors the concentrations of nitrogen oxides, sulfur oxides, carbon monoxide, carbon dioxide and particulate matter. The gas analyzer uses non-dispersive infrared absorption and chemiluminescence technology, with a sampling frequency of 1 Hz and a measurement accuracy of ppm level. The ship attitude signal is obtained by an inertial measurement unit, including roll angle, pitch angle, heave displacement, speed and heading information. The inertial measurement unit integrates a three-axis gyroscope and an accelerometer, with an attitude angle measurement accuracy of 0.1 degrees and a displacement measurement accuracy of centimeter level. All signals are recorded synchronously by a high-speed data acquisition system, which is equipped with a GPS clock synchronization module to ensure that the time alignment accuracy of multi-source signals reaches milliseconds. Data preprocessing includes denoising, outlier removal and signal compensation to provide high-quality raw data.

[0034] The pressure fluctuation signal is subjected to time-frequency transformation to extract the pulsation characteristic spectrum. The collected pressure fluctuation signal is passed through a high-pass filter to remove the direct current bias and low-frequency drift, retaining the dynamic components related to diesel engine combustion pulsation. The short-time Fourier transform method is used for time-frequency analysis of the pressure fluctuation signal, converting one-dimensional time domain signal to two-dimensional time-frequency representation. The Hanning window is selected as the analysis window function, with a window length of two firing periods and a window overlap rate of 50%, ensuring the balance of time-frequency resolution. The time-frequency transformation generates a time-frequency spectrum, where the time axis reflects the dynamic process of pressure fluctuation, the frequency axis shows the distribution of each frequency component, and the amplitude is represented by color depth. The pulsation characteristic spectrum is extracted from the time-frequency spectrum, including key features such as fundamental frequency, harmonic components, energy distribution and phase information. The fundamental frequency component of the pulsation characteristic spectrum corresponds to the firing frequency of the diesel engine, which is automatically identified by a peak detection algorithm. The integer multiples of the fundamental frequency are extracted, and the amplitude and phase relationship of these harmonics reflects the regularity of the combustion process. The energy distribution of the pulsation characteristic spectrum is calculated as P(f,t)=|STFT{p(t)}|², where P(f,t) represents the time-frequency energy distribution, STFT represents the short-time Fourier transform, and p(t) represents the pressure fluctuation signal. A pulsation characteristic spectrum matrix is constructed, with each row corresponding to the frequency spectrum characteristics of a time segment and each column corresponding to the time evolution of a specific frequency. The pulsation characteristic spectrum not only contains instantaneous spectrum information, but also retains the trajectory of the spectrum over time, completely describing the time-frequency characteristics of the pressure fluctuation.

[0035] In some embodiments, the fusing the pulsation feature spectrum and the pollutant concentration signal to generate a dynamic pollution feature library comprises: separating a fundamental pulsation component and a high-frequency disturbance component from the pulsation feature spectrum; identifying a pollutant emission period based on the fundamental pulsation component; segmenting the pollutant concentration signal into a plurality of concentration segments with the pollutant emission period as a time window; and arranging the concentration segments in pulsation phase after superimposing the high-frequency disturbance component to form a dynamic pollution feature library.

[0036] The fundamental pulsation component and the high-frequency disturbance component are separated from the pulsation feature spectrum. The pulsation feature spectrum obtained by time-frequency transformation of the pressure pulsation signal is subjected to frequency spectrum analysis to identify the main frequency components in the spectrum. The fundamental pulsation component corresponds to the basic frequency of the diesel engine working cycle, and the calculation formula is: f_base = n x z / (60 x τ), where f_base represents the fundamental frequency (Hz), n represents the engine speed (rpm), z represents the number of cylinders, and τ represents the stroke coefficient (2 for four-stroke). A band-pass filter is used to extract the fundamental frequency and its first three harmonic components, and the filter bandwidth is set to ±10% of the fundamental frequency to ensure complete retention of the fundamental pulsation feature. The high-frequency disturbance component includes components such as turbulent pulsation, combustion noise, and pipe resonance, and the frequency range is usually above 100 Hz. The wavelet packet decomposition method is used to decompose the pulsation feature spectrum into different frequency bands, and the fundamental pulsation component is concentrated in the low-frequency sub-band, and the high-frequency disturbance component is distributed in the high-frequency sub-band. An adaptive threshold method is used in the separation process, and the threshold is dynamically determined according to the energy distribution of each frequency band. The time-domain representation P_base(t) of the fundamental pulsation component and the time-domain representation P_high(t) of the high-frequency disturbance component are established, and the two are superimposed to restore the original pressure pulsation signal. The amplitude, phase and modulation characteristics of the fundamental pulsation, as well as the statistical characteristics of the high-frequency disturbance, including the root mean square value, the peak factor and the probability distribution, are recorded.

[0037] The pollutant emission period is identified based on the fundamental pulsation component. The periodicity of the fundamental pulsation component P_base(t) is analyzed, and the pulsation period T_pulse is calculated using the autocorrelation function. The autocorrelation function is expressed as R(τ)=∫P_base(t)×P_base(t+τ)dt, where R(τ) represents the autocorrelation function value, P_base(t) represents the fundamental pulsation component, τ represents the time delay, and the integration interval covers multiple periods. The τ value corresponding to the first significant peak determines the pulsation period. Considering the operating characteristics of marine diesel engines, pollutant emissions are directly related to the combustion process, and there is a certain phase relationship between the emission period and the fundamental pulsation period. By analyzing the envelope curve of the fundamental pulsation component, the starting point and peak point of each combustion cycle are identified. The relationship between the pulsation phase φ(t) and pollutant generation is established, with the initial combustion corresponding to the peak of nitrogen oxide generation and the later combustion corresponding to the increase in particulate matter emissions. The instantaneous phase of the fundamental pulsation is extracted using the Hilbert transform, and the phase range is normalized to [0, 2π]. The firing order of the multi-cylinder engine is identified, the phase difference of each cylinder is determined, and the complete emission period characteristics are established. The relationship between the pollutant emission period T_emission and the fundamental pulsation period is T_emission=k×T_pulse, where T_emission represents the pollutant emission period, k represents the integer multiple coefficient, and T_pulse represents the pulsation period. The identified emission period parameters, including period length, period stability, and phase drift characteristics, are recorded.

[0038] The pollutant concentration signal is segmented into multiple concentration segments using the pollutant emission period as the time window. According to the identified pollutant emission period T_emission, a time window division strategy is established. The window starting point is aligned with the phase zero point of the fundamental pulsation to ensure that each window covers a complete emission period. The continuous pollutant concentration signal C(t) is segmented, and the i-th concentration segment is defined as C_i(t)=C(t), t∈[(i-1)×T_emission,i×T_emission], where C_i(t) represents the i-th concentration segment, C(t) represents the original concentration signal, and t represents the time variable. Considering the influence of ship movement and environmental disturbances, a dynamic window adjustment mechanism is adopted, and the window length is adjusted in real time according to the identified period. Each concentration segment contains the concentration time series of multiple pollutants, maintaining the synchronization of each pollutant component. The segmentation boundaries are smoothed to reduce the truncation effect using a window function, and the Hanning window or Blackman window is selected. An index system for concentration segments is established, recording the timestamp, corresponding operating parameters, and environmental conditions of each segment. The statistical characteristics of each concentration segment are calculated, including average concentration, peak concentration, concentration change rate, and integrated emission. The concentration segment sequence {C_1,C_2,...,C_n} is formed, each segment retaining the original time resolution and multi-component information.

[0039] The concentration segments are superimposed with the high-frequency disturbance components and arranged according to the fluctuation phase to form a dynamic pollution feature library. The high-frequency disturbance component P_high_i(t) corresponding to each concentration segment is extracted to ensure the time alignment accuracy. The superimposition relationship F_i(t)=C_i(t)+α×P_high_i(t) is established, where F_i(t) represents the superimposed feature, C_i(t) represents the concentration segment, P_high_i(t) represents the corresponding high-frequency disturbance component, and a represents the coupling coefficient. The coupling coefficient is determined by cross-correlation analysis, considering the sensitivity differences of different pollutants to pressure fluctuations. The fluctuation phase attribute of each superimposed feature is calculated, and the phase value is obtained from the instantaneous phase of the fundamental frequency fluctuation component. All feature segments are rearranged according to the fluctuation phase, and the phase range [0, 2π] is equally divided into M phase intervals, usually M=36 corresponding to a 10-degree phase resolution. A three-dimensional feature matrix is established as the data structure of the dynamic pollution feature library, with dimensions corresponding to the phase, time and pollutant species. Each element in the feature library contains the concentration value, pressure fluctuation intensity and corresponding ship attitude parameters. The dynamic updating mechanism of the feature library is realized, and the newly collected data is added to the corresponding phase interval after being processed according to the same process. The dynamic pollution feature library completely records the coupling relationship between pressure fluctuations and pollutant emissions, providing comprehensive emission feature data support.

[0040] In step S120, the particle classification clusters are identified based on the dynamic pollution feature library, the flow field analysis is performed on the particle classification clusters to generate a collision probability distribution, the peak value of the collision probability distribution is extracted to form an agglomeration feature set, and a strengthened separation control sequence is output based on the agglomeration feature set.

[0041] Specifically, the particle classification clusters are identified based on the dynamic pollution feature library. Multi-dimensional feature data of particulate matter concentration are extracted from the dynamic pollution feature library, including particle number concentration, mass concentration and particle size distribution function in different particle size ranges. The particle size range covers 10 nanometers to 10 microns, which is divided into multiple particle size channels according to logarithmic interval to ensure capturing the complete particle size spectrum. Cluster analysis algorithm is used to classify the particulate matter features to identify particle groups with similar motion characteristics and physical properties. The cluster feature vector includes particle average particle size, geometric standard deviation, number concentration and mass concentration, etc. K-means clustering algorithm is used for particle classification, and the number of clusters K is determined by the elbow rule and the silhouette coefficient. The first classification cluster corresponds to the ultrafine particle mode, mainly derived from gas phase nucleation in the fuel combustion process. The second classification cluster corresponds to the accumulation mode particles, formed by the collision of ultrafine particles through Brownian motion. The third classification cluster corresponds to the coarse particle mode, including unburned carbon particles, lubricating oil splashes and particles generated by mechanical wear. The characteristic parameters of each classification cluster are calculated, including median particle size, geometric mean particle size, arithmetic mean particle size, particle size span, particle number concentration, mass concentration and volume concentration. The spatiotemporal distribution map of the classification cluster is established to show the distribution characteristics, concentration gradient and evolution law of different classification clusters in the exhaust pipe.

[0042] The flow field analysis of the particle classification cluster is performed to generate the collision probability distribution. A three-dimensional flow field calculation domain of the exhaust pipe is constructed to accurately describe the structural characteristics of the pipe geometry, bends, expansion sections and contraction sections. The effects of exhaust temperature gradient, pressure pulsation and wall heat transfer on the flow field are considered to establish complete boundary conditions. The computational fluid dynamics method is used to solve the flow field, and the control equations include continuity equation, momentum equation and energy equation. The k-ε model is used to close the turbulent flow, and the enhanced wall function is used to process the near-wall region. The identified particle classification clusters are introduced into the flow field as discrete phases, and the Euler-Lagrange two-way coupling method is used. The particle motion equation is: mp(dvp / dt) = FD + FG + FB + FT + FM, where mp represents the particle mass, vp represents the particle velocity vector, FD represents the drag force, FG represents the gravity, FB represents the Brownian force, FT represents the thermophoretic force, and FM represents the Magnus force. The drag force coefficient considers the comprehensive effects of particle Reynolds number, Mach number and shape factor, and the Brownian force uses the random walk model to simulate the diffusion motion of nanoparticles. The relative motion and collision parameters between particles of different classification clusters are calculated to establish a complete collision kernel function. Based on the particle trajectory and concentration distribution, the Monte Carlo method is used to statistically analyze the collision events at each point in space. A high-resolution three-dimensional collision probability distribution field is generated, and the adaptive encryption technology is used to improve the resolution in high gradient areas.

[0043] In some embodiments, the peak extraction on the collision probability distribution forms the agglomeration feature set, including: performing two-dimensional scanning on the collision probability distribution to identify probability contour lines; determining a high-probability aggregation region based on the probability contour lines; generating an agglomeration path by guiding and planning the particle motion path using the high-probability aggregation region; and converting the agglomeration path to form the agglomeration feature set.

[0044] The probability contour lines are identified by two-dimensional scanning on the collision probability distribution. A complete set of two-dimensional projections is formed by systematically generating axial, radial and circumferential slice sequences in the three-dimensional collision probability distribution field. The axial slices are uniformly distributed along the pipeline centerline, the radial slices are arranged at equal angular intervals, and the circumferential slices are selected at characteristic cross-section positions. On each two-dimensional probability distribution slice, the improved marching square algorithm is used to extract the probability contour lines, which takes into account the probability values of the grid vertices and edge midpoints to improve the smoothness of the contour lines. A multi-level probability threshold is set to extract the contour line family, covering the complete range from low probability to high probability, ensuring that all important probability features are captured. The extracted contour lines are smoothed using spline interpolation techniques to eliminate the jagged effect caused by grid discretization. The topological features of the contour lines are identified, including closed contour lines, open contour lines, bifurcation points and convergence points, and a hierarchical structure of the contour lines is established. The geometric parameters of each contour line are calculated, including the perimeter, curvature distribution, minimum curvature radius, number of inflection points and shape complexity index. The enclosed area, centroid position, principal axis direction and aspect ratio of the closed contour lines are calculated, which reflect the morphological features of the high-probability region. A time series of contour lines is established to track the evolution of the contour line morphology with changes in operating conditions, and stable and dynamic regions are identified.

[0045] For example, the high-probability aggregation region is determined based on the probability contour lines, including: performing density analysis on the probability contour lines to identify contour line clusters; extracting convergence centers based on the spacing distribution of the contour line clusters; expanding the convergence centers outward to form an influence domain; and determining the high-probability aggregation region by boundary fusion on the influence domain.

[0046] Density analysis of contour lines identifies contour clusters. A two-dimensional grid is established to cover the entire analysis domain, and the grid size is adaptively adjusted according to the minimum feature scale of the contour lines. The number of contour lines passing through each grid cell is calculated to form a contour line density field. The contour line density is defined as ρ_line = ∑(L_i / A_cell), where ρ_line represents the contour line density, L_i represents the length of the i-th contour line within the grid cell, A_cell represents the area of the grid cell, and the summation is taken over all contour lines passing through the cell. The discrete density values are smoothed using the kernel density estimation method, with a Gaussian kernel and a bandwidth determined by cross-validation. The watershed algorithm is applied to the density field to identify local maximum regions, each corresponding to the center of a contour cluster. A density threshold is set to filter out significant contour clusters, and the threshold is adaptively determined based on the statistical properties of the density distribution, typically choosing the mean plus two times the standard deviation. The region growing algorithm is used to expand from the cluster center, and the connected regions with density greater than the threshold are grouped into the same contour cluster. The characteristic parameters of the contour cluster are calculated, including the number of contour lines, the average spacing, the density peak, the spatial range, and the anisotropy degree. The adjacency relationship graph of the contour clusters is established to analyze the interaction and influence range between clusters.

[0047] Convergence centers are extracted based on the spacing distribution of contour clusters. The contour lines within each contour cluster are sorted according to their probability values from low to high. The normal distance between adjacent contour lines is measured to construct a spacing sequence d = {d(1,2), d(2,3),..., d(n-1,n)}, where d(i,i+1) represents the average normal distance between the i-th and i+1-th contour lines. The spacing change rate is calculated as r(i) = [d(i-1,i) - d(i,i+1)] / d(i,i+1), with positive values indicating a decreasing spacing, i.e., converging contour lines, and negative values indicating an increasing spacing, i.e., diverging contour lines. Monotonically decreasing segments in the spacing sequence are identified, which correspond to the convergence regions of the contour lines. The convergence center is located at the position with the smallest spacing, and the precise coordinates are determined by interpolation. The convergence intensity index is calculated as S_conv = ∑[max(r(i),0)], where the summation is taken over the entire spacing sequence, and this index quantifies the severity of convergence. The spatial range of convergence is evaluated, defined as the distance from the convergence center to the position where the spacing starts to increase. The position coordinates, convergence intensity, influence range, and corresponding probability level of each convergence center are recorded. A hierarchical structure of convergence centers is established, with the main convergence centers having the highest convergence intensity and the largest influence range.

[0048] The convergence center is expanded outward to form an influence domain. From each convergence center, ray tracing is performed in multiple directions to determine the boundaries of the influence domain. The ray directions are uniformly distributed in three-dimensional space, and the Fibonacci spiral method is used to generate direction vectors to ensure uniformity of spatial sampling. Along each ray, the change in probability value is tracked, and when the probability value falls below a specified proportion of the convergence center probability value, the influence radius in that direction is determined. The determination criterion for the influence radius is R(θ,φ)=min{r:P(r,θ,φ)≤α×P_center}, where R represents the influence radius, θ and φ represent the direction angle, P(r,θ,φ) represents the probability value at a distance r along that direction, P_center represents the probability value of the convergence center, and α represents the attenuation coefficient. The continuous influence domain boundary surface is reconstructed from discrete directional sampling through radial basis function interpolation methods. The geometric characteristics of the influence domain, including volume, surface area, equivalent radius, and shape factor, are calculated. The anisotropy of the influence domain is analyzed, and the principal axis direction and axis length ratio are determined through principal component analysis. The probability gradient field inside the influence domain is established, and the fastest and slowest directions of probability decline are identified. The stability of the influence domain is evaluated, and the sensitivity of the boundary to parameter changes is determined through perturbation analysis.

[0049] The influence domain is subjected to boundary fusion to determine high-probability aggregation zones. Adjacent influence domains in space are identified, and an adjacency relationship matrix between the influence domains is established. The minimum distance between the boundaries of adjacent influence domains is calculated, and the fast nearest point algorithm is used to improve computational efficiency. When the minimum distance between the boundaries is less than a specified threshold, the adjacent influence domains are marked as pairs to be fused. Morphological dilation operations are performed to expand the boundaries of the influence domains, and the spherical structure element is selected with a radius that is adaptively adjusted according to the boundary distance. The overlapping influence domains after dilation are merged using Boolean AND operations to form preliminary fusion regions. The boundaries of the fusion regions are smoothed using the Laplace smoothing algorithm to eliminate irregularities in the boundaries. The probability distribution of the fusion regions is recalculated, and the probability values in the overlapping regions are processed using a weighted average method. The connectivity and integrity of the fusion regions are verified to ensure that there are no probability voids inside. The final high-probability aggregation zones have smooth three-dimensional boundaries and continuous internal probability distributions. The identifier, spatial range, center position, volume, average probability value, and number of original influence domains contained by each aggregation zone are recorded.

[0050] The high-probability aggregation zones are used to guide the planning of particle motion paths to generate agglomeration paths. An array of particle release points is set at the inlet cross-section of the flow field, covering different radial positions and circumferential angles to ensure representative spatial sampling. For each released virtual particle, the distance to all high-probability aggregation zones is calculated, and a distance field D(x,y,z) is established. A potential energy field U=-k×exp(-D / λ) is constructed, where U represents the potential energy, k represents the potential energy intensity coefficient, D represents the distance to the nearest aggregation zone, and λ represents the characteristic length scale. The potential energy gradient is calculated to generate a guide force field: F_guide= The guiding force points to the direction of the fastest potential energy decrease, i.e. the direction of the aggregation zone. The guiding force is superimposed with other forces such as fluid drag, gravity, etc. to solve the modified particle motion equation. The fourth-order Runge-Kutta method is used to numerically integrate the motion equation, and the time step is adaptively adjusted according to the local flow field characteristics. The particle position is monitored in real time, and the time and position are recorded when the particle enters the high-probability aggregation zone. The complete three-dimensional trajectory of the particle from release to entry into the aggregation zone is tracked, and the geometric and dynamic parameters of the path are calculated. The particle trajectories that successfully reach the aggregation zone are screened, and divergent trajectories and excessively long trajectories are removed to form an effective agglomeration path set. Cluster analysis is performed on the agglomeration paths to identify the main path patterns and branch structures.

[0051] The agglomeration feature set is formed according to the conversion of the agglomeration paths. Each agglomeration path is represented using a parametric curve, and a B-spline or NURBS curve is used to fit the discrete trajectory points. An appropriate number of control points are selected to balance the smoothness of the curve and the fidelity of the trajectory, and the control point positions are determined by the least squares method. The differential geometric features of the path are calculated, including the spatial distribution of arc length parameterization, tangent vector, principal normal vector, binormal vector, curvature and torsion. The integral features of the path are extracted, including total length, average curvature, maximum curvature, curvature variation rate, total torsion and helicity. The kinematic features of the particle along the path are analyzed, and the velocity distribution, acceleration distribution, residence time distribution and kinetic energy change are calculated. A comprehensive feature vector is established to describe each path: F_path = [L, κ_avg, κ_max, τ_total, v_avg, v_max, E_kin, θ_total], where L represents the path length, κ_avg and κ_max represent the average and maximum curvature, τ_total represents the total motion time, v_avg and v_max represent the average and maximum velocity, E_kin represents the kinetic energy change, and θ_total represents the total deflection angle. Statistical analysis is performed on the feature vectors of all paths to calculate the mean, standard deviation, correlation coefficient matrix and principal components. The hierarchical clustering method is used to group the feature vectors, and each group represents a typical agglomeration behavior pattern. A structured agglomeration feature set is formed, containing feature vectors, corresponding path parameters and applicable conditions.

[0052] The enhanced separation control sequence is output based on the agglomeration feature set. Through the high-efficiency agglomeration mode in the agglomeration feature set, a first group of control sequences is output: adjust the mainstream velocity to the optimal range specified by the agglomeration feature set, and make the particles move along the predetermined agglomeration path. According to the vortex intensity requirement shown by the agglomeration feature set, the control sequence of the vortex generator is output, including the blade angle adjustment instruction and the rotation speed setting value. The collision hotspot position in the agglomeration feature set is converted into a nozzle activation sequence, and the high-collision-probability area is timed to spray to promote particle agglomeration. Based on the time characteristics of the agglomeration feature set, the pulsation control sequence is output, and the pulsation frequency and phase are synchronized with the dynamic characteristics of the agglomeration path. The spatial distribution information extracted from the agglomeration feature set generates a guide vane adjustment sequence to guide the airflow to form a flow field structure conducive to enhanced separation. According to the agglomeration requirements of different particle size sections in the agglomeration feature set, a grading control sequence is output to specifically enhance the separation effect of a specific particle size range. The stability index of the agglomeration feature set is converted into a feedback control sequence to adjust the control parameters in real time to maintain the optimal agglomeration state. All control instructions are integrated to form a complete enhanced separation control sequence, and the sequence format includes time stamp, control object, action instruction and target value. The output enhanced separation control sequence is directly used for the actuator to realize precise separation enhancement control based on the agglomeration feature set.

[0053] In step S130, the performance of the enhanced separation control sequence is evaluated, the baseline separation index is extracted, the vortex enhancement parameters are obtained according to the pulsation feature spectrum, the vortex enhancement parameters are superimposed with the baseline separation index to generate an optimization control matrix, and the first purification state is generated through the optimization control matrix.

[0054] Specifically, performance evaluation of the enhanced separation control sequence extracts baseline separation indicators. The enhanced separation control sequence is executed, and data throughout the separation process is collected. Monitoring parameters include inlet particle concentration, outlet particle concentration, particle size distribution changes, pressure loss, and energy consumption. Calculate the separation efficiency: η_i = (C_in,i - C_out,i) / C_in,i, where η_i represents the separation efficiency of the i-th particle size segment, C_in,i represents the inlet concentration of the particle size segment, and C_out,i represents the outlet concentration. Establish a separation efficiency spectrum for multiple particle size segments, ranging from nanometers to microns, divided into logarithmically equal-interval particle size channels. Evaluate the overall separation efficiency, obtained by weighted average of the efficiency of each particle size segment, and the weight factor reflects the mass proportion of the particle size segment in the total particulate matter. Analyze the time stability of the separation efficiency, calculate the statistical parameters of the efficiency fluctuation, and evaluate the robustness of the control sequence. Measure the system pressure loss, including static pressure loss and dynamic pressure loss, with pressure sensors placed at multiple locations on the inlet and outlet of the separator. Calculate the energy consumption index, which reflects the energy consumption required for unit volume of processing capacity, taking into account the fan power and flow rate. Establish a performance evaluation matrix containing multi-dimensional indicators such as separation efficiency, pressure loss, energy consumption, and processing capacity. Extract the baseline separation indicator set, including the separation efficiency of each particle size segment, the total efficiency, the pressure loss coefficient, and the specific energy consumption, which form the basis for optimization.

[0055] According to the pulsation characteristic spectrum, obtain the vortex enhancement parameters. From the dynamic contamination characteristic library, retrieve the pulsation characteristic spectrum data, including the frequency components, amplitude distribution, and phase relationship of the pressure pulsation. Identify the dominant frequency in the pulsation characteristic spectrum, which corresponds to the spectral peak position with the maximum energy density. Calculate the pulsation intensity index by integrating the frequency energy, which reflects the overall intensity level of the pulsation. Analyze the vortex generation mechanism induced by pulsation, which produces a resonance enhancement effect when the pulsation frequency matches the characteristic frequency of the flow field. Determine the Strouhal number of vortex enhancement: St = f_dom x D / V, where St represents the Strouhal number (dimensionless), f_dom represents the dominant frequency (Hz), D represents the characteristic length such as the pipe diameter (m), and V represents the average flow velocity (m / s). Establish the relationship between vortex intensity and pulsation parameters, which presents nonlinear characteristics with the changes of pulsation intensity and frequency. Extract the vortex enhancement parameter vector, including key parameters such as frequency, intensity, dimensionless number, circulation intensity, and phase information. Analyze the variation law of vortex enhancement parameters under different operating conditions, and establish the mapping relationship between parameters and operating conditions. Calculate the influence of vortex enhancement on particle motion, including increased radial migration speed, prolonged residence time, and increased collision frequency. Form a vortex enhancement parameter library to record the parameter combinations and enhancement effects under typical operating conditions.

[0056] The vortex enhancement parameters are superimposed with the benchmark separation index to generate an optimized control matrix. A superimposition mechanism is established to quantify the vortex enhancement effect as an increment of separation efficiency, and the enhancement function form is determined by fitting. The structure of the optimized control matrix is constructed, with the row dimension corresponding to different combinations of control parameters, and the column dimension corresponding to performance indicators and constraint conditions. The matrix elements are obtained through comprehensive calculation, including benchmark efficiency, vortex enhancement increment, and implementation cost. The control parameters include main flow velocity, fluctuation frequency, fluctuation amplitude, guide structure angle, and injection parameters, among others. The performance indicators include separation efficiency of each particle size segment, total efficiency, pressure loss, energy consumption, and processing capacity. The constraint conditions include maximum allowable pressure loss, energy consumption limit, equipment pressure capacity, and noise standard. Normalization is used to unify the dimensions of different indicators, and all indicators are mapped to a unified interval for comparison. A comprehensive performance index is calculated by weighted summation to integrate multiple performance indicators, with the weights reflecting the relative importance of each indicator. The optimized control matrix contains complete performance evaluations of all feasible control schemes, supporting multi-objective optimization decisions.

[0057] In some embodiments, generating a primary purification state through the optimized control matrix includes: constructing a separation efficiency surface based on the optimized control matrix; determining control accuracy by performing precision analysis on the separation efficiency surface; forming an operation sequence by performing optimal path search on the separation efficiency surface according to the control accuracy; and setting states with efficiency values exceeding a threshold in the operation sequence as primary purification states.

[0058] A separation efficiency surface is constructed based on the optimized control matrix. Control parameters and corresponding separation efficiency data are extracted from the optimized control matrix to form a set of multi-dimensional data points. The main control parameters are selected as the independent variables of the surface, and the flow rate and fluctuation frequency are usually selected as the basic coordinates of the two-dimensional surface, while other parameters are processed through slicing or projection. A continuous efficiency surface is constructed using a radial basis function interpolation method, and the interpolation function adopts a Gaussian function or a multiple quadratic function form. The interpolation coefficients are determined by solving a system of linear equations to ensure that the surface passes through all the sampling points and maintains smoothness. When extended to high-dimensional cases, a hypersurface is constructed, and the tensor product method is used to handle high-dimensional interpolation to reduce computational complexity. Contour maps and three-dimensional visualizations of the surface are generated to visually display the trends and gradient distributions of efficiency changes with control parameters. The gradient field of the surface is calculated to identify the direction and rate of change with the fastest efficiency growth. Local extreme points of the surface, including maximum, minimum, and saddle points, are analyzed, and these feature points correspond to special working states. A mathematical expression or lookup table structure of the surface is established to support real-time efficiency prediction and optimization calculations. The interpolation accuracy of the surface is verified through cross-validation to ensure that the prediction error is within an acceptable range.

[0059] The control accuracy is determined by analyzing the precision of the separation efficiency surface. A dense sampling is performed on the efficiency surface to generate a validation dataset. The sampling points are selected using Latin hypercube design to ensure uniform coverage of the space. The interpolation error is calculated and the error distribution characteristics are analyzed, including the mean absolute error, root mean square error, and maximum error. A spatial distribution map of the error is established to identify high-error areas, which usually correspond to locations with sharp changes in curvature. The sensitivity of the control parameters is analyzed: where S_i represents the sensitivity coefficient of the i-th parameter (dimensionless), denotes the partial derivative of efficiency with respect to the parameter, x_i represents the parameter value, and η represents the efficiency value. High sensitivity parameters require more precise control. The control accuracy requirements for each parameter are determined based on the sensitivity analysis. The physical limitations of the actuators are considered, such as valve positioning accuracy, frequency converter resolution, etc., to determine the achievable control accuracy. The relationship between control accuracy and efficiency uncertainty is established, and the error propagation theory is used to evaluate the impact of control error on the final efficiency. A hierarchical control strategy is developed, with precise control for high sensitivity parameters and coarse control for low sensitivity parameters. Control accuracy specifications are established, clearly defining the control range, accuracy requirements, and adjustment steps for each parameter. The precision analysis results, including parameter sensitivity ranking, control accuracy requirements, and error tolerance, are recorded.

[0060] An optimal path search is performed on the separation efficiency surface based on the control accuracy to form the operation sequence. The starting point of the search is set as the current working state, and the target point is the optimal point on the efficiency surface or the sub-optimal point that meets the constraints. Considering the control accuracy constraints, the continuous parameter space is discretized into a grid of reachable states, with the grid spacing determined by the control accuracy. A path search algorithm is used to find the optimal trajectory from the starting point to the target point, taking into account path length, efficiency improvement, and transition cost. A state transition cost function is defined, considering factors such as parameter change amplitude, transition time, and switching energy consumption. A heuristic function is designed to accelerate the search process, with the heuristic value based on the estimated cost to the target point. Dynamic constraints such as parameter change rate limits and prohibited regions are handled during the search process to ensure the feasibility of the path. When the search reaches the target point or meets the convergence conditions, the complete path is generated by backtracking. The continuous path is discretized into an operation sequence of time-parameter pairs, with each node containing the execution time and control parameter setting value. The time arrangement of the sequence is optimized, considering system response time and stability requirements to avoid parameter mutations.

[0061] The efficiency value of the state in the operation sequence exceeding the threshold value is set as a primary purification state. All state points in the operation sequence are traversed, and the separation efficiency value corresponding to each state is calculated. The efficiency threshold of the primary purification is set, which is determined according to the emission standard and engineering requirements, and is usually higher than the benchmark efficiency by a certain percentage. Identify the state points that meet the efficiency requirements, which constitute the candidate set of the primary purification state. The stability of the candidate state is analyzed, and the sensitivity of the efficiency to the parameter perturbation is calculated, and the state with good stability is preferred. The maintainability of the candidate state is evaluated, including factors such as energy consumption level, equipment load and operation difficulty. The state transition relationship is established, and the reachability and transition cost from the current state to each candidate state are analyzed. The state with the best comprehensive performance is selected as the standard primary purification state, balancing between efficiency, stability and economy. The working envelope of the primary purification state is defined, allowing the parameters to fluctuate within a certain range while maintaining the purification effect. The monitoring index system of the primary purification state is formulated, including the lower limit of the efficiency, the upper limit of the pressure loss and the energy consumption threshold. The complete definition of the primary purification state is generated, including all control parameter settings, expected performance indicators and maintenance conditions.

[0062] In step S140, droplet concentration features are extracted for the primary purification state, a seed density threshold is determined based on the droplet concentration features, an adaptive purification sequence is constructed according to the seed density threshold, and the adaptive purification sequence is executed to output a secondary purification state.

[0063] Specifically, droplet concentration features are extracted for the primary purification state. Under the working conditions of the primary purification state, a laser scattering measurement system is used to monitor the spatial distribution and concentration change of the droplets. The measurement system includes a multi-wavelength laser emitter and a high-speed CCD camera array, covering visible and near-infrared wavelengths, which can distinguish droplets of different particle sizes. The Mie scattering theory is used to analyze the scattering light intensity distribution, and the particle size spectrum and number concentration of the droplets are inversely calculated. The droplet particle size ranges from sub-micron to hundred-micron, and is divided into particle size intervals according to geometric progression. The spatial distribution function C_d(x, y, z, D) of the droplet concentration is calculated, where C_d represents the droplet concentration (pieces / m³), x, y, z represent the spatial coordinates, and D represents the droplet diameter. Statistical features of the droplet concentration are extracted, including average concentration, concentration gradient, spatial non-uniformity and temporal fluctuation. The interaction region of droplets and particulate matter is analyzed, and the action range of mechanisms such as inertial collision, Brownian diffusion and electrophoretic capture is identified. A droplet concentration feature vector is established, including total droplet concentration, average particle size, particle size dispersion, spatial coverage and effective action volume. The evaporation rate and survival time of the droplets are evaluated, considering the influence of temperature, humidity and air flow velocity on the droplet life. A droplet concentration feature map is formed, showing the droplet distribution density and capture potential in different regions.

[0064] Determine seed density threshold based on droplet concentration characteristics. Analyze the nucleation ability of droplets as condensation cores, droplet surface provides condensation sites for water vapor and condensable components. Calculate the critical supersaturation S_c = exp(4σM / (ρRTD)), where S_c represents critical supersaturation (dimensionless), σ represents surface tension (N / m), M represents molecular weight (kg / mol), ρ represents liquid density (kg / m³), R represents gas constant (J / (mol·K)), T represents temperature (K), D represents droplet diameter (m). When the environmental supersaturation exceeds the critical value, the droplets begin to grow rapidly and capture surrounding particles. Evaluate the seed efficiency of droplets of different particle sizes, small droplets have high specific surface area but evaporate quickly, large droplets are stable but few in number. Establish the relationship curve between seed density and purification efficiency, determine the optimal working interval by experimental data fitting. The determination of the seed density threshold considers multiple constraints, including droplet consumption rate, water mist load limit and demisting ability of downstream equipment. Analyze the spatial distribution requirements of seed density, higher seed density is required in high pollution areas, and lower density can be used in clean areas to save resources. Determine the hierarchical seed density threshold, set the corresponding droplet seed density for particles of different particle sizes. Record the seed density threshold parameter set, including minimum density, standard density, maximum density and corresponding application conditions.

[0065] In some embodiments, the adaptive purification sequence is constructed according to the seed density threshold, including: evaluating the purification load based on the seed density threshold, the purification load including particle concentration, droplet distribution density and air flow velocity; generating a spray control boundary according to the purification load; time division of the spray control boundary to form a purification sequence.

[0066] The purification load is evaluated based on the seed density threshold, including particulate matter concentration, droplet distribution density, and airflow velocity. From the seed density threshold, the droplet generation rate and replenishment rate required to maintain the density are calculated. The particulate matter concentration directly affects the droplet consumption rate, and under high concentration conditions, droplets quickly increase and settle by capturing particles. A droplet consumption model is established: dN_d / dt = -K_cap x N_d x N_p, where N_d represents the droplet number concentration (pieces / m³), N_p represents the particulate matter number concentration (pieces / m³), K_cap represents the capture coefficient (m³ / s), and t represents time (s). The droplet distribution density determines the spatial coverage effect of purification, and it is necessary to ensure that the droplet density in the key area is higher than the seed density threshold. Airflow velocity affects the transport and distribution of droplets, and high-speed airflow causes droplets to quickly cross the purification area, reducing the capture efficiency. A comprehensive purification load index is calculated, considering the coupling and mutual influence of the three factors. A load grading system is established, dividing the purification load into three grades: light load, standard load, and heavy load. The resource demand under different load conditions is evaluated, including water supply, compressed air consumption, and energy demand. The spatiotemporal variation characteristics of the load are analyzed, and the time and location of the load peak are identified. A load prediction method is developed based on historical data and current trends to predict future load changes. A purification load evaluation report is formed, including real-time load values, load grades, and resource demand forecasts.

[0067] A spray control boundary is generated based on the purification load. Based on the evaluated purification load, the working range and limit capacity of the spray system are determined. The spray control boundary defines the controllable parameter space of the system under different load conditions. The minimum spray boundary corresponds to the light load condition, at which only the basic droplet seed density needs to be maintained, and the nozzle opening rate and spray pressure are at a low level. The maximum spray boundary corresponds to the heavy load condition, and the system operates at maximum capacity, with all available nozzles fully open and the spray pressure reaching the upper limit of the design. Between the minimum and maximum boundaries, a continuous control surface is established to describe the relationship between the spray parameters and the load. The control boundary considers multiple constraint conditions, including the atomization performance curve of the nozzle, the pressure loss characteristics of the pipeline, and the spatial distribution requirements of the droplets. Computational fluid dynamics is used to simulate the droplet motion trajectory and concentration distribution under different spray configurations. The nozzle arrangement scheme is optimized to ensure uniform droplet coverage within the control boundary. A boundary safety margin is established to avoid control instability caused by operating near the limit boundary. A spray control boundary graph is generated to display the feasible control parameter combinations in the form of a three-dimensional surface.

[0068] The spray control boundary is time-sequenced to form a purification sequence. The continuous control boundary is discretized into a finite number of control states, each corresponding to a specific spray configuration. The time-sequencing takes into account the dynamic response characteristics of the system, ensuring smoothness and controllability of state switching. Trigger conditions for state transitions are defined, and based on real-time monitoring of purification load and efficiency feedback, it is determined when to switch. The state transition path is designed, with preference given to paths with small parameter changes and low switching costs. A time window mechanism is established, with each control state maintaining a minimum time to avoid frequent switching. A gradual adjustment strategy is developed, with large parameter changes implemented gradually through multiple intermediate states. The mechanical response time of the spray system, such as valve actuation time and pressure build-up time, is taken into account. Standard purification sequence templates are generated, including start-up sequences, steady-state operation sequences, and shutdown sequences. Special sequences are designed for special working conditions, such as rapid response sequences and energy-saving operation sequences. A monitoring mechanism for sequence execution is established, with real-time tracking of sequence execution progress and effect deviation. A complete purification sequence library is formed, with each sequence containing detailed execution steps, time scheduling, and expected effects.

[0069] The adaptive purification sequence is executed to output a secondary purification state. The constructed adaptive purification sequence is loaded into the control system and automatically executed through PLC or DCS. Key parameters are continuously monitored during execution, including droplet density, particulate matter concentration reduction rate, and system pressure. The actual purification effect is compared with the expected target in real time, the efficiency deviation is calculated, and it is determined whether adjustment is needed. When the purification efficiency stabilizes within the target range, the system enters the secondary purification state. The characteristic of the secondary purification state is that the particulate matter concentration is significantly reduced, and the droplet capture process reaches dynamic equilibrium. In the secondary purification state, fine particulate matter is effectively removed through the condensation growth mechanism of droplets. Stability indicators of the secondary purification state are monitored, including efficiency fluctuation amplitude, droplet consumption rate, and energy consumption level. State maintenance strategies are established to maintain the secondary purification effect through fine-tuning of control parameters. Running data of the secondary purification state are recorded to form a state characteristic file for subsequent analysis and optimization. The overall effect of the secondary purification is evaluated, including total dust removal efficiency, grading efficiency curve, and pollutant emission reduction. A secondary purification state report is output, including detailed performance indicators, operating parameters, and improvement suggestions.

[0070] In step S150, the secondary purification state is coupled with the ship attitude signal to generate an inertial force field distribution, the peak point of scour is identified based on the inertial force field distribution, the pulse cleaning is triggered at the peak point to generate an anti-fouling control sequence, and the anti-fouling control sequence is executed to obtain the purification stability indicators.

[0071] Specifically, the secondary purification state is coupled with the ship attitude signal to generate an inertial force field distribution. Current flow field operating parameters are extracted from the secondary purification state, including air flow velocity distribution, droplet concentration field, and particulate matter deposition state. These characteristic parameters of the secondary purification state constitute the initial conditions for the coupling calculation. Real-time data streams of the ship attitude signal are obtained synchronously. The ship attitude signal includes roll angle, pitch angle, heave displacement, and three-axis angular velocity components. The signal sampling rate matches the purification system control frequency. The flow field parameters of the secondary purification state are spatiotemporally coupled with the ship attitude signal to establish a dynamic description in a moving coordinate system. The coupling process is realized through coordinate transformation, which converts the ship attitude signal into inertial acceleration acting on the flow field of the secondary purification state. The inertial acceleration field generated by coupling is calculated: a_inertia=-a_boat-ω×(ω×r)-dω / dt×r, where a_inertia represents the inertial acceleration vector (m / s²), a_boat represents the ship center of mass acceleration vector (m / s²), ω represents the angular velocity vector (rad / s), r represents the position vector relative to the center of mass (m), and × represents the vector cross product operation. The inertial acceleration field is multiplied by the density distribution of the secondary purification state to generate a three-dimensional inertial force field distribution: F_inertia=ρ×a_inertia, where F_inertia represents the unit volume inertial force (N / m³), and ρ represents the fluid density in the secondary purification state (kg / m³). The inertial force field distribution exhibits dynamic characteristics with changes in the ship attitude signal, and periodic disturbances are superimposed on the stable flow field of the secondary purification state.

[0072] In some embodiments, the identification of the scour peak point based on the inertial force field distribution includes: performing vector decomposition on the inertial force field distribution to form a tangential force and a normal force distribution; extracting wall surface action strength from the normal force distribution to establish a strength gradient map; correlating and analyzing the strength gradient map with the tangential force distribution to identify a stress concentration area; and determining the scour peak point based on the strength ranking of the stress concentration area.

[0073] The inertial force field distribution is decomposed into tangential and normal force distributions. A local coordinate system is established at each wall surface position of the inertial force field distribution, with the coordinate axes including the wall surface normal and two orthogonal tangents, ensuring accurate decomposition of the inertial force field distribution. The vector projection of the inertial force field distribution onto the normal is F_n = F_inertia · n, where F_n represents the normal force component decomposed from the inertial force field distribution (N / m³), F_inertia represents the inertial force field distribution vector, n represents the unit normal vector, and · represents the vector dot product. The tangential force is separated from the inertial force field distribution: F_t = F_inertia - F_n × n, ensuring that the tangential force is completely located in the wall tangent plane, and F_t represents the tangential force distribution vector. The normal force distribution directly acts on the wall to produce a squeezing effect, changing the adhesion conditions of the dirt layer in the secondary purification state. The tangential force distribution produces a shearing effect along the wall, producing a scouring effect on the deposits in the inertial force field distribution area. The spatial variation of the tangential force distribution is calculated, and the tangential force is significantly enhanced in the high gradient area of the inertial force field distribution. The relative size of the normal and tangential force distributions is analyzed to determine the dominant type of action in the inertial force field distribution. The force decomposition results are extracted at typical geometric positions, and the inertial force field distribution at positions such as bends and expansion sections exhibits special decomposition characteristics.

[0074] The wall surface action intensity is extracted from the normal force distribution to establish an intensity gradient map. The wall surface action intensity per unit area is calculated based on the normal force distribution: σ_n = F_n × δ, where σ_n represents the wall surface action intensity derived from the normal force distribution (Pa), F_n represents the normal force distribution value (N / m³), and δ represents the boundary layer characteristic thickness (m). The action intensity generated by the normal force distribution on the wall surface includes both steady and fluctuating parts, with the steady part determined by the average normal force distribution and the fluctuating part reflecting the normal force distribution fluctuations caused by ship motion. The rate of change of the wall surface action intensity generated by the normal force distribution in space is calculated, and the intensity gradient is calculated using the finite difference method. The intensity gradient map is constructed from the gradient distribution of the wall surface action intensity, with the gradient vector pointing in the direction of the fastest increase in action intensity. The intensity gradient map shows the spatial non-uniformity of the normal force distribution converted into wall surface action, with high gradient areas corresponding to rapidly changing stress fields. Characteristic structures are identified in the intensity gradient map, including gradient peak lines, gradient valley lines, and gradient bifurcation points, which are directly related to the spatial pattern of the normal force distribution. The time evolution of the intensity gradient map is evaluated, and the intensity gradient map driven by the normal force distribution exhibits corresponding dynamic characteristics as the ship attitude changes. The distribution characteristics of the gradient values in the intensity gradient map are statistically analyzed to determine threshold parameters for subsequent processing.

[0075] The intensity gradient map and the tangential force distribution are analyzed to identify the stress concentration zones. The spatial correspondence between the intensity gradient map and the tangential force distribution is established, and the two mechanical quantities are extracted at the same position for correlation analysis. A comprehensive stress index is defined to combine the gradient value in the intensity gradient map and the tangential force value in the tangential force distribution through weighted combination to form a unified stress measure. In the correlation analysis, the high stress regions are searched, which satisfy both the high gradient in the intensity gradient map and the large shear in the tangential force distribution. The intensity gradient map provides stress variation information, and the tangential force distribution provides erosion capacity information. The actual stress concentration zones are determined by the correlation of the two. The identified stress concentration zones reflect the key positions under the action of inertial force, and these regions have both high stress level and strong erosion effect. The characteristic parameters of each stress concentration zone are calculated, including the maximum gradient value obtained from the intensity gradient map and the maximum shear force obtained from the tangential force distribution. The spatial range of the stress concentration zone is evaluated, and the region boundary is determined by the contour lines of the intensity gradient map and the tangential force distribution. The formation mechanism of the stress concentration zone is analyzed, and the composite effect is generated by the superposition of the local peak value in the intensity gradient map and the concentrated region in the tangential force distribution.

[0076] The intensity ranking of the stress concentration zones is used to determine the peak points of erosion. The comprehensive intensity values of all identified stress concentration zones are extracted, which comprehensively reflect the stress level and erosion potential of the stress concentration zones. The stress concentration zones are ranked in descending order according to the comprehensive intensity values, and the stress concentration zones with the highest intensity are given priority. The local maximum points are precisely located within each high-intensity stress concentration zone, and these points are the potential peak points of erosion. The peak positions are extracted from the intensity distribution of the stress concentration zones, and the gradient search method is used to determine the precise coordinates. The effectiveness of the peak points of erosion is verified, and the significant peaks are selected based on the intensity ranking results of the stress concentration zones. The erosion capacity index of each peak point of erosion is calculated, which is determined by the local intensity and range of the stress concentration zone. According to the intensity ranking, the peak points of erosion that need to be focused on are determined, and usually several points with high intensity ranking are selected. The distribution law of the peak points of erosion in the stress concentration zone is analyzed, and some stress concentration zones may contain multiple peak points of erosion. The complete information of each peak point of erosion is recorded, including the spatial position, peak intensity, belonging stress concentration zone, and intensity ranking position.

[0077] The pulse cleaning triggered at the peak point generates the anti-fouling control sequence. The pulse cleaning device is installed at the identified peak point position, and the spatial coordinates of the peak point determine the precise installation position of the device. The triggering criterion of the peak point is designed, and the pulse cleaning is started when the real-time monitoring shows that the stress state of the peak point reaches the preset condition. The parameters of the pulse cleaning are determined according to the characteristics of the peak point, and high-intensity peak points are configured with high-pressure short pulses, and medium peak points are configured with medium-pressure long pulses. The pulse cleaning triggered at the peak point forms a time-sequenced action, and the triggering time of each peak point is independently determined according to its stress accumulation. A coordination mechanism for multiple peak points is established, and the pulse cleaning of adjacent peak points is staggered in time to avoid mutual interference. The dynamic characteristics of the peak point are considered, and the intensity of the peak point changes periodically with the movement of the ship, and the pulse cleaning is triggered at the intensity peak, which has the best effect. A complete anti-fouling control sequence is generated, and the triggering condition, pulse parameters and execution time sequence of each peak point are clearly defined in the sequence. The anti-fouling control sequence includes active cleaning and passive waiting states, and the state is switched according to the real-time state of the peak point. The resource allocation of the anti-fouling control sequence is optimized, and the frequency and intensity of the pulse cleaning are reasonably arranged under the premise of ensuring the cleaning effect.

[0078] In some embodiments, the executing the anti-fouling control sequence obtains a purification stability index, including: extracting pulse frequency and pulse intensity parameters from the anti-fouling control sequence; applying the pulse intensity parameters in time sequence to form a vibration cleaning wave; real-time monitoring the effect of the vibration cleaning wave to obtain a pressure drop recovery curve; comparing the pressure drop recovery curve with a reference curve to generate a purification stability index.

[0079] The pulse frequency and pulse intensity parameters are extracted from the anti-fouling control sequence. The data structure of the anti-fouling control sequence is analyzed, and each control instruction in the anti-fouling control sequence contains complete pulse description. The pulse frequency information in the anti-fouling control sequence is extracted: f_pulse=N / T, where f_pulse represents the pulse frequency (Hz) extracted from the anti-fouling control sequence, N represents the number of pulses in the sequence, and T represents the sequence time length (s). The pulse intensity parameters are read from the anti-fouling control sequence, including peak pressure, duration and rise time, which define the physical characteristics of each pulse. The distribution rule of the pulse parameters in the anti-fouling control sequence is analyzed, and the pulses at different positions may have different frequency and intensity configurations. The pulse density of the anti-fouling control sequence is counted, and dense pulses correspond to intensified cleaning sections, and sparse pulses correspond to maintained cleaning sections. The consistency of the pulse parameters in the anti-fouling control sequence is evaluated, and a sequence with stable parameters produces a predictable cleaning effect. The correspondence between the pulse parameters and the positions of the anti-fouling control sequence is established to provide accurate parameter index for subsequent execution.

[0080] The pulse intensity parameters are applied in sequence to form the vibration cleaning wave. Control signals are generated according to the extracted pulse parameters, with the pulse frequency determining the time interval of the signals and the pulse intensity determining the amplitude of the signals. The control signals are applied in sequence according to the timing requirements of the pulse parameters, with each pulse being triggered accurately at the specified time. The pulse parameters are converted into physical actions through actuators, with high-voltage pulses generating pressure waves and vibration pulses generating mechanical fluctuations. Multiple pulse parameters are applied in sequence and superimposed in the medium to form a composite vibration cleaning wave. The waveform of the vibration cleaning wave is determined by the pulse parameters, with the pulse intensity affecting the amplitude and the pulse frequency affecting the periodic characteristics. The propagation characteristics of the vibration cleaning wave in the pipeline are analyzed, including wave speed, attenuation, and reflection phenomena. The timing arrangement of the pulse parameters is optimized to maximize the intensity of the vibration cleaning wave in the target area. The actual waveform of the vibration cleaning wave is monitored to verify the accuracy of the pulse parameter application and waveform fidelity.

[0081] The effect of the vibration cleaning wave is monitored in real time to obtain the pressure drop recovery curve. Pressure sensors are installed in the system to monitor the pressure change under the action of the vibration cleaning wave, with the sensors located at the inlet and outlet positions to measure the pressure difference. The vibration cleaning wave acts on the dirt layer to produce a peeling effect, effectively cleaning and reducing the flow resistance, which is manifested as a decrease in pressure drop. The pressure drop data is recorded in real time: ΔP(t) = P_in(t) - P_out(t), where ΔP(t) represents the pressure drop (Pa) at time t under the action of the vibration cleaning wave. The continuous pressure drop data is plotted into a pressure drop recovery curve, with the horizontal axis representing the action time of the vibration cleaning wave and the vertical axis representing the pressure drop value. The pressure drop recovery curve reflects the cumulative effect of the vibration cleaning wave, with a decreasing curve indicating effective cleaning and a flat curve indicating saturated cleaning. The characteristic parameters of the pressure drop recovery curve are analyzed, including the initial pressure drop, the final pressure drop, the recovery rate, and the recovery time constant. The shape of the pressure drop recovery curve is related to the characteristics of the vibration cleaning wave, with strong vibration producing rapid recovery and weak vibration leading to slow recovery. The actual effect of the vibration cleaning wave is evaluated through the pressure drop recovery curve, providing feedback information for optimizing the pulse parameters.

[0082] The pressure drop recovery curve is compared with the baseline curve to generate a purification stability index. The baseline curve is established under ideal conditions, representing the pressure drop recovery process under the optimal cleaning effect. The measured pressure drop recovery curve is compared with the baseline curve point by point, and the difference between the two curves is calculated. The curve deviation is defined as: ε(t) = |ΔP_actual(t) - ΔP_baseline(t)| / ΔP_baseline(t), where ε(t) represents the relative deviation of the pressure drop recovery curve at time t from the baseline curve. The purification stability index is calculated by integration: SI = 1 - ∫ε(t)dt / T, where SI represents the purification stability index (between 0 and 1), and the integral covers the entire cleaning period T. The purification stability index reflects the closeness of the pressure drop recovery curve to the baseline curve, and the higher the index, the more stable the purification process. Analyze the reasons for the deviation of the pressure drop recovery curve from the baseline curve, including changes in dirt characteristics, decay of cleaning effect, and drift of system parameters. According to the value of the purification stability index, the stability level is divided, and high index corresponds to stable operation, and low index prompts intervention. By comparing the pressure drop recovery curve with the baseline curve, the cleaning strategy is continuously optimized to improve the purification stability index.

[0083] Step S160, compare the purification stability index with the pollutant concentration signal to generate a performance deviation value, generate an optimized adjustment instruction based on the performance deviation value, and complete the intelligent control of the multi-stage purification of ship flue gas.

[0084] Specifically, a performance deviation value is generated by comparing the purification stability index with the pollutant concentration signal. The purification stability index reflects the system's ability to maintain purification effectiveness, and the index varies between 0 and 1, with a value close to 1 indicating system stability. The current pollutant concentration signal is collected simultaneously, including the real-time concentration values of major pollutants such as particulate matter, sulfur dioxide, and nitrogen oxides. The correspondence between the purification stability index and the pollutant concentration signal is established, and the two types of data are extracted at the same time point for comparative analysis. When the purification stability index is high, the pollutant concentration signal should theoretically show low concentration, and the two are negatively correlated. The normalized pollutant concentration level is calculated: C_norm = C_actual / C_limit, where C_norm represents the normalized concentration (dimensionless), C_actual represents the measured value of the pollutant concentration signal (mg / m³), and C_limit represents the emission limit value (mg / m³). The purification stability index is compared with the normalized pollutant concentration, and in the ideal state, the sum of the two should be close to 1. The performance deviation value is defined as: Δ_perf = |SI - (1 - C_norm)|, where Δ_perf represents the performance deviation value (dimensionless), SI represents the purification stability index, and the formula quantifies the gap between the actual performance and the ideal performance of the system. The performance deviation value reflects the degree of mismatch between the purification stability index and the pollutant concentration signal, and the greater the deviation, the more the system performance deviates from the expected performance. Analyze the causes of performance deviation, and if the purification stability index is normal but the pollutant concentration signal is abnormal, it may be caused by external disturbance, otherwise it indicates internal failure. Establish a comprehensive performance deviation value for multiple pollutants, and process the deviation contributions of different pollutants through weighted averaging.

[0085] The performance deviation value is used to generate an optimized adjustment instruction. The system running state is determined according to the size and change trend of the performance deviation value. If the performance deviation value continues to increase, it indicates that immediate intervention is needed. The performance deviation value is set with a hierarchical threshold. A slight deviation only needs fine tuning, a moderate deviation needs significant adjustment, and a serious deviation triggers an emergency response. The composition of the performance deviation value is analyzed to determine whether the deviation is caused by insufficient purification stability or excessive pollutant concentration. Different adjustment strategies are developed for different reasons. When the stability is insufficient, the anti-fouling cleaning is enhanced, and when the concentration is excessive, the purification intensity is increased. The adjustment amount is calculated based on the performance deviation value: ΔU = K_p × Δ_perf + K_i × ∫Δ_perfdt, where ΔU represents the control adjustment amount, K_p represents the proportional coefficient, K_i represents the integral coefficient, and ∫Δ_perfdt represents the time integral of the deviation value. The proportional term provides a fast response, and the integral term eliminates the steady-state error. The combination of the two achieves accurate adjustment. The adjustment amount is decomposed into specific execution parameters, including spray pressure adjustment, pulse frequency correction, flow rate optimization, and other aspects. Structured optimization adjustment instructions are generated, each containing target parameters, adjustment direction, adjustment amplitude, and execution timing. Considering the response characteristics of the system, large adjustments are implemented through multiple steps to avoid system oscillation. The priority of the adjustment instruction is established, and the execution order is determined based on the urgency and impact of the performance deviation value. The feasibility of the adjustment instruction is verified to ensure that the adjusted parameters are within the device capacity and do not violate safety constraints. After executing the optimized adjustment instruction, the system enters a comprehensive running state of multi-stage purification, and the functions of each stage of purification work together. The first stage of purification removes large particle pollutants through basic separation, the second stage of purification captures fine particles using liquid droplets, the anti-fouling mechanism ensures long-term stable operation, and the optimization adjustment ensures optimal overall performance. The multi-stage purification process forms a complete processing chain. The ship exhaust gas passes through pressure pulsation regulation, vortex enhanced separation, liquid droplet coagulation and capture, pulse anti-fouling cleaning, and adaptive optimization adjustment in sequence, and finally completes the intelligent control of multi-stage purification of ship exhaust gas.

[0086] To perform the above-mentioned method embodiment corresponding to a ship exhaust gas multi-stage purification method, to realize the corresponding functions and technical effects. Referring to Figure 2 , Figure 2 The structure block diagram of a ship exhaust gas multi-stage purification device 200 provided by an embodiment of the present application is shown. For ease of illustration, only the parts related to the present embodiment are shown. The ship exhaust gas multi-stage purification device 200 provided by an embodiment of the present application includes:

[0087] The signal acquisition module 201 is configured to acquire the pressure pulsation signal, the pollutant concentration signal and the ship attitude signal of the ship exhaust gas, perform time-frequency transformation on the pressure pulsation signal to extract the pulsation characteristic spectrum, and fuse the pulsation characteristic spectrum with the pollutant concentration signal to generate a dynamic pollution characteristic library.

[0088] A particle polymerization module 202 is configured to identify a particle hierarchical cluster based on the dynamic contamination feature library, perform flow field analysis on the particle hierarchical cluster to generate a collision probability distribution, perform peak extraction on the collision probability distribution to form an agglomeration feature set, and output a strengthened separation control sequence based on the agglomeration feature set.

[0089] A primary purification module 203 is configured to perform performance evaluation on the strengthened separation control sequence to extract a baseline separation index, obtain a vortex enhancement parameter based on the pulsation feature spectrum, superimpose the vortex enhancement parameter and the baseline separation index to generate an optimized control matrix, and generate a primary purification state through the optimized control matrix.

[0090] A secondary purification module 204 is configured to extract a droplet concentration feature from the primary purification state, determine a seed density threshold based on the droplet concentration feature, construct an adaptive purification sequence according to the seed density threshold, and execute the adaptive purification sequence to output a secondary purification state.

[0091] A scale prevention control module 205 is configured to couple the secondary purification state and the ship attitude signal to generate an inertial force field distribution, identify a scour peak point based on the inertial force field distribution, trigger a pulse cleaning at the peak point to generate a scale prevention control sequence, and execute the scale prevention control sequence to obtain a purification stability index.

[0092] An intelligent adjustment module 206 is configured to compare the purification stability index and the pollutant concentration signal to generate a performance deviation value, generate an optimized adjustment instruction based on the performance deviation value, and complete the ship flue gas multi-stage purification intelligent control.

[0093] The ship flue gas multi-stage purification device 200 described above can implement a ship flue gas multi-stage purification method described above. The options in the method embodiment described above are also applicable to this embodiment, which will not be described in detail here. The remaining content of the present embodiment can refer to the content of the method embodiment described above, and will not be described in detail in this embodiment.

[0094] As shown in Figure 3 The third embodiment of the present application also provides a computer device, which includes a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, and the processor 302 implements the steps of the ship flue gas multi-stage purification method of the first embodiment of the present application when executing the program.

[0095] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, purposes and effects of the present application, so that the public can understand the disclosure of the present application more thoroughly and comprehensively, and the protection scope of the present application is not limited thereby.

[0096] The above embodiments are also not exhaustive enumeration based on the present application, in addition to which there can be a plurality of other embodiments not listed. Any substitution and improvement made without violating the concept of the present application is within the scope of the present application.

Claims

1. A method for multi-stage purification of a ship exhaust gas, characterized in that, The method comprises the following steps: Collecting pressure fluctuation signals, pollutant concentration signals and ship attitude signals of ship exhaust flue gas, performing time-frequency transformation on the pressure fluctuation signals to extract fluctuation characteristic spectrum, fusing the fluctuation characteristic spectrum and the pollutant concentration signals to generate a dynamic pollution characteristic library; Identifying particle classification clusters based on the dynamic pollution characteristic library, performing flow field analysis on the particle classification clusters to generate collision probability distribution, extracting peak values from the collision probability distribution to form agglomeration characteristic set, and outputting enhanced separation control sequence based on the agglomeration characteristic set; Performing efficiency evaluation on the enhanced separation control sequence to extract baseline separation index, obtaining vortex enhancement parameters from the fluctuation characteristic spectrum, superimposing the vortex enhancement parameters and the baseline separation index to generate an optimized control matrix, and generating a primary purification state through the optimized control matrix; Extracting droplet concentration characteristics from the primary purification state, determining seed density threshold based on the droplet concentration characteristics, constructing an adaptive purification sequence according to the seed density threshold, and executing the adaptive purification sequence to output a secondary purification state; Coupling the secondary purification state and the ship attitude signal to generate an inertial force field distribution, identifying a scouring peak point based on the inertial force field distribution, triggering pulse cleaning at the peak point to generate an anti-fouling control sequence, and executing the anti-fouling control sequence to obtain a purification stability index; Comparing the purification stability index with the pollutant concentration signal to generate a performance deviation value, generating an optimized adjustment instruction based on the performance deviation value, and completing intelligent control of ship flue gas multi-stage purification.

2. The method of claim 1, wherein, The method for fusing the fluctuation characteristic spectrum and the pollutant concentration signal to generate a dynamic pollution characteristic library comprises the following steps: Separating the fluctuation characteristic spectrum into a fundamental frequency fluctuation component and a high-frequency disturbance component; Identifying pollutant emission period based on the fundamental frequency fluctuation component; Segmenting the pollutant concentration signal into multiple concentration segments with the pollutant emission period as a time window; Superimposing the concentration segments and the high-frequency disturbance component, and arranging them according to fluctuation phase to form a dynamic pollution characteristic library.

3. The method of claim 1, wherein, The method for extracting peak values from the collision probability distribution to form an agglomeration characteristic set comprises the following steps: Performing two-dimensional scanning on the collision probability distribution to identify probability contour lines; Determining a high-probability aggregation area based on the probability contour lines; Using the high-probability aggregation area to guide and plan particle motion paths to generate agglomeration paths; Converting the agglomeration paths to form an agglomeration characteristic set.

4. The method of claim 1, wherein, The method for generating a primary purification state through the optimized control matrix comprises the following steps: Constructing a separation efficiency surface based on the optimized control matrix; Performing precision analysis on the separation efficiency surface to determine control precision; According to the control precision, performing optimal path search on the separation efficiency surface to form an operation sequence; Setting a state with efficiency value exceeding a threshold value in the operation sequence as a primary purification state.

5. The method of claim 1, wherein, The method for constructing an adaptive purification sequence according to the seed density threshold comprises the following steps: Evaluating purification load based on the seed density threshold, the purification load including particulate matter concentration, droplet distribution density and air flow velocity; Generating a spraying control boundary according to the purification load; Performing time sequence segmentation on the spraying control boundary to form a purification sequence.

6. The method of claim 1, wherein, The peak point of the erosion is identified based on the inertial force field distribution, and the peak point of the erosion includes: The inertial force field distribution is vector-decomposed to form a tangential force distribution and a normal force distribution; A wall surface action intensity is extracted from the normal force distribution to establish an intensity gradient graph; The intensity gradient graph is associated with the tangential force distribution to identify a stress concentration area; The peak point of the erosion is determined based on the intensity ranking of the stress concentration area.

7. The method of claim 1, wherein, The purification stability index is obtained by executing the anti-fouling control sequence, and the purification stability index includes: Pulse frequency and pulse intensity parameters are extracted from the anti-fouling control sequence; The pulse intensity parameters are applied in time sequence to form a vibration cleaning wave; The effect of the vibration cleaning wave is monitored in real time to obtain a pressure drop recovery curve; The purification stability index is generated by comparing the pressure drop recovery curve with a reference curve.

8. The method of claim 3, wherein, The high-probability aggregation area is determined based on the probability contour, and the high-probability aggregation area includes: The probability contour is analyzed to identify an isocline cluster; A convergence center is extracted based on the spacing distribution of the isocline cluster; The convergence center is expanded outward to form an influence domain; The high-probability aggregation area is determined by boundary fusion of the influence domain.

9. A multi-stage purification device for ship exhaust gas, characterized in that It includes: The signal acquisition module is used to collect the pressure pulsation signal, the pollutant concentration signal and the ship attitude signal of the ship exhaust flue gas, to extract the pulsation characteristic spectrum by time-frequency transformation of the pressure pulsation signal, to fuse the pulsation characteristic spectrum with the pollutant concentration signal to generate a dynamic pollution feature library; The particle aggregation module is used to identify particle hierarchical clusters based on the dynamic pollution feature library, to generate a collision probability distribution by flow field analysis of the particle hierarchical clusters, to form an agglomeration feature set by peak extraction of the collision probability distribution, and to output a strengthened separation control sequence based on the agglomeration feature set; The first purification module is used to extract a reference separation index by efficiency evaluation of the strengthened separation control sequence, to obtain a vortex enhancement parameter according to the pulsation characteristic spectrum, to generate an optimized control matrix by superimposing the vortex enhancement parameter and the reference separation index, and to generate a first purification state through the optimized control matrix; The secondary purification module is used to extract a droplet concentration feature from the first purification state, to determine a seed density threshold based on the droplet concentration feature, to construct an adaptive purification sequence according to the seed density threshold, and to output a secondary purification state by executing the adaptive purification sequence; The anti-fouling control module is used to couple the secondary purification state with the ship attitude signal to generate an inertial force field distribution, to identify a peak point of the erosion based on the inertial force field distribution, to trigger pulse cleaning at the peak point to generate an anti-fouling control sequence, and to obtain a purification stability index by executing the anti-fouling control sequence; The intelligent adjustment module is used to generate a performance deviation value by comparing the purification stability index with the pollutant concentration signal, to generate an optimized adjustment instruction based on the performance deviation value, and to complete the intelligent control of ship flue gas multi-stage purification.

10. A computer device, comprising: It includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and realize the method of any one of claims 1 to 8 when executing the computer program.

Images