Activated carbon pore regulation and control method and monitoring equipment

Abstract

The invention discloses an activated carbon pore regulation and control method and monitoring equipment, and belongs to the technical field of spectrum analys.The method comprises the steps that on the basis of a probability prediction model, spectrum scanning frequency is dynamically adjusted, and real-time global spectrum collection is carried out; the method comprises the following steps: setting a self-adaptive threshold value, extracting time sequence slope and acceleration by a dynamic sliding window, generating a time sequence change matrix, judging an abnormal spectrum, screening and analyzing unexpected signals, realizing accurate identification of known and new unexpected signals, and reversely deducing molecular configuration parameters through a characteristic peak position interval. Calculating a matching degree difference value between the pore size and the molecular configuration, and carrying out matching analysis to ensure the suitability analysis accuracy of the pore characteristics and the pollutant molecules; according to the matching degree difference value, porosity gradient regulation and control are carried out, real-time monitoring is carried out to verify the regulation and control effect, self-evolution is achieved, and the regulation and control efficiency and precision are continuously improved.

Description

Technical Field

[0001] This invention belongs to the field of spectral analysis technology and relates to a method for regulating the pore size of activated carbon and a monitoring device. Background Technology

[0002] In the field of industrial flue gas treatment, activated carbon has become a core material for desulfurization and denitrification technologies due to its high specific surface area, rich pore structure, and surface chemical activity. Traditional activated carbon pore control methods are mostly based on static processes, such as steam activation, chemical impregnation, or high-temperature treatment, which directionally increase the proportion of mesopores through physical or chemical means to improve the diffusion efficiency of reactants. With the increasing refinement of environmental protection requirements, dynamic analysis technology has gradually become a research hotspot. By monitoring the changes in pore structure during the adsorption process of activated carbon in real time and dynamically adjusting the activation conditions in combination with operating parameters, adaptive optimization of the pore structure can be achieved.

[0003] However, existing technologies do not consider the adaptive limitations of dynamic analysis systems in the event of sudden, unexpected signal intrusion. On the one hand, they lack pre-existing models for molecular size-pore matching for pollutants outside the design operating conditions, such as heavy metal vapors and halogenated hydrocarbons. When such special pollutants suddenly appear in the flue gas, they cannot generate pore structure adjustment strategies in real time based on their physicochemical properties, resulting in a severe mismatch between the active sites on the activated carbon surface and the adsorption requirements of pollutants. On the other hand, they do not integrate emergency pore expansion mechanisms for trace pollutants. For example, when facing mercury vapor, they cannot quickly generate sulfur-modified nanoscale active sites through pulsed activator injection, which can easily lead to the risk of pollutant penetration due to the lack of adsorption sites. More importantly, there is a quality control vacuum period during the control transition period after the unexpected signal intrusion. Within the time window from pollutant detection and identification to the completion of pore structure reconstruction, there is a lack of real-time compensation measures for the adsorption performance of activated carbon. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and monitoring device for controlling the pore size of activated carbon. This method identifies unexpected signals and infers molecular configuration, calculates the matching degree by combining pore characteristics, generates a gradient control scheme and adjusts parameters in real time, and implements full-chain monitoring and closed-loop optimization to solve problems such as missing matching models, insufficient emergency mechanisms, and vacuum control.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for controlling the pore size of activated carbon, comprising:

[0007] Based on a probabilistic prediction model, the spectral scanning frequency is dynamically adjusted, and real-time global spectral acquisition is performed.

[0008] An adaptive threshold is set, and a dynamic sliding window is used to extract the temporal slope and acceleration, generate a temporal change matrix, identify abnormal spectra, filter out unexpected signals and analyze them, infer molecular configuration parameters through the characteristic peak spacing, calculate the difference in matching degree between pore size and molecular configuration, and perform matching analysis.

[0009] Based on the matching degree difference, pore gradient is controlled, and the control effect is verified by real-time monitoring.

[0010] Specifically, the steps for adjusting the spectral scanning frequency include:

[0011] When the scanning device is started, a probability prediction model is invoked to predict the probability of unexpected signals appearing in each band in the future period;

[0012] Bands with a probability greater than a probability threshold are identified as potential bands, and the rest are non-potential bands.

[0013] Set the scanning frequency of the potential band to the normal frequency. The non-potential band maintains the regular scanning frequency, and real-time global spectral acquisition is performed simultaneously.

[0014] Specifically, the steps for generating the time-series variation matrix include:

[0015] Acquire historical spectral data for the most recent period and locate the baseline region within the spectral data;

[0016] Calculate the baseline parameters of the baseline region, including the mean and standard deviation, and simultaneously calculate the real-time fluctuation entropy of the current spectral data. Combined with the historical maximum fluctuation entropy, calculate the fluctuation ratio.

[0017] An adaptive threshold is set by combining the baseline parameters with the fluctuation ratio.

[0018] Using the current time as the endpoint, an initial window is set. Based on variable weight linear regression, time decay weights are assigned to the data within the initial window. Weighted least squares method is used for fitting to obtain the basic slope at the current time.

[0019] Calculate the difference in fundamental slope between two consecutive time points, and normalize it using the sampling period to obtain the rate of change of the signal. :

[0020] like The window is expanded to One cycle; if Window minimized to One cycle; among which, Adjust the threshold for the window;

[0021] Based on the adjusted sliding window, the linear equation is fitted using the weighted least squares method with time decay weights to obtain the time-series slope. The second-order difference of the time-series slope at three consecutive time points is calculated and normalized by the square of the sampling period to obtain the acceleration, thus generating the time-series change matrix.

[0022] Specifically, the steps for filtering out unexpected signals include:

[0023] For each wavelength point in the time series variation matrix, if the time series slope is greater than the adaptive threshold or the acceleration is greater than zero, it is marked as a single abnormal point; otherwise, it is judged as a normal fluctuation and the data is stored in the historical spectrum library.

[0024] The band distribution of single anomalies is statistically analyzed. If five or more consecutive wavelength points in a certain band are single anomalies and are not within the characteristic range of the expected pollutant standard spectrum, then it is determined to be an abnormal spectrum and the analysis process is triggered.

[0025] Extract the original spectral data corresponding to the abnormal spectra and merge them with the background interference spectral library collected at the same time to generate a matrix to be processed;

[0026] The covariance matrix of the matrix to be processed is calculated by principal component analysis and eigenvalues ​​and eigenvectors are obtained. The top three principal components with a cumulative variance contribution rate of not less than 95% are selected, and the first principal component is removed to generate a denoised spectrum.

[0027] The pre-stored standard spectral library of expected pollutants is called, and a regression model is constructed using partial least squares regression with the standard spectral library as the independent variable and the denoised spectrum as the dependent variable.

[0028] Based on the difference between the denoised spectrum and the model prediction, the residual spectrum is calculated, and the peak intensity of the residual spectrum is normalized to generate the unexpected feature spectrum.

[0029] Specifically, the steps for filtering out unexpected signals also include:

[0030] Baseline correction and moving average smoothing are performed on the unexpected characteristic spectrum to generate a smoothed spectrum. The first derivative is calculated to find candidate peak positions and then screened to generate a preliminary peak position list.

[0031] The smooth spectrum peaks were fitted using a Gaussian-Lorentz mixture function, and the key parameters of each characteristic peak were extracted and sorted by peak position to generate a set of vibrational characteristic parameters.

[0032] Each parameter in the vibration feature parameter set is standardized to obtain a vibration feature vector. The dynamic spectral fingerprint library is called to calculate the cosine similarity between the vibration feature vector and the known unexpected standard vectors in the library. The maximum value is selected as the matching similarity and the corresponding matching spectrum.

[0033] If the matching similarity is not less than the matching threshold, it is determined to be a known unexpected signal; otherwise, it is determined to be a new unexpected signal, and the vibration feature vector is stored in the dynamic spectral fingerprint database.

[0034] Specifically, the steps of matching analysis include:

[0035] For known unexpected signals, the corresponding classification sub-model is activated to convert spectral features into molecular geometric parameters and generate molecular configuration parameters.

[0036] For new unexpected signals, the general sub-model is invoked and the incremental learning mechanism is triggered to generate an initial parameter set. Incremental learning is then carried out based on the initial parameter set to generate a specific sub-model and to generate the molecular configuration parameters of the new unexpected signals.

[0037] Specifically, the steps of the matching analysis also include:

[0038] The small-angle X-ray scattering device and X-ray photoelectron spectroscopy device built into the bed are activated to generate the activated carbon pore feature matrix.

[0039] The size matching degree is calculated based on molecular diffusion theory, the polarity matching degree is calculated using the absolute difference formula, and the volume matching degree is calculated based on molecular packing theory. Weights are set for each matching degree and the overall matching degree is calculated.

[0040] The matching degree difference is calculated based on the preset target matching degree, a secondary matching threshold is set to divide the matching level, and the deduction logic is dynamically adjusted based on the matching level to generate the optimized pore control deduction logic.

[0041] Specifically, the steps for pore gradient regulation include:

[0042] Based on the matching degree difference, the core control target is identified, and the corresponding control methods are retrieved from the preset physical control library and chemical control library. The control priority is set based on the deduction logic, and the corresponding control parameter calculation formula is called with the matching degree difference as input to calculate the parameter gradient and time node, thereby generating a gradient control scheme.

[0043] The control is carried out based on the gradient control scheme. During the control process, the pore distribution is scanned at the feedback interval, the control pore volume ratio is calculated, the actual pore change is calculated based on the pore volume ratio in the matching analysis, and the parameters are automatically adjusted when the actual pore change exceeds the expected deviation.

[0044] After the regulation is completed, a multi-dimensional matching degree evaluation system is invoked to calculate the regulation matching degree difference based on the real-time monitored pore characteristics, so as to determine the regulation effect and make corrections.

[0045] The entire chain of data from this regulation is stored in the pore regulation knowledge base according to category. Cluster analysis is performed on multiple regulation data of the same type of unexpected signals to extract the optimal parameter combination, update the recommended parameters of the gradient regulation scheme, and input the new data into the multi-dimensional matching degree evaluation system and optimize the weight allocation through incremental learning.

[0046] An activated carbon pore regulation and monitoring device includes: a spectral acquisition terminal, a sensor array, a regulation terminal, and a central control terminal;

[0047] The spectral acquisition terminal is used to perform real-time full-domain spectral acquisition of flue gas, generate a three-dimensional raw spectral dataset, and receive instructions from the central control terminal to adjust the scanning frequency.

[0048] The sensor array is used to collect parameters of the activated carbon bed and flue gas.

[0049] The control terminal is used to perform physical and chemical control, receive the gradient control scheme from the central control terminal, implement control according to parameter gradients and time nodes, and provide feedback on the equipment operating status.

[0050] The central control terminal is used to receive spectral data, identify unexpected pollutants and update the dynamic spectral fingerprint database, process sensor array data, generate gradient control schemes, monitor and correct deviations in real time, verify the effect, store the data in the database, and optimize the strategy through cluster analysis and incremental learning.

[0051] Specifically, the central control terminal includes an identification module, an analysis module, and a control and optimization module;

[0052] The identification module is used to adjust the spectral scanning frequency, generate a time-series change matrix based on an adaptive threshold and a dynamic sliding window, determine abnormal points and abnormal spectra, extract a set of vibration feature parameters, and match the set of vibration feature parameters with a dynamic spectral fingerprint database to identify known unexpected signals and new unexpected signals.

[0053] The analysis module is used to infer the molecular configuration parameters of pollutants, analyze and generate the activated carbon pore feature matrix, construct a multi-dimensional matching degree evaluation system, calculate the matching degree difference, and dynamically adjust the pore regulation deduction logic.

[0054] The regulation and optimization module is used to generate a gradient regulation scheme and calculate the actual porosity change. If the deviation exceeds the expected value, the parameters are automatically adjusted. After the regulation is completed, the effect is verified, and cluster analysis and incremental learning are performed on the knowledge base data to optimize the recommended values ​​of the regulation parameters and the weights of the matching degree model.

[0055] The beneficial effects of this invention are:

[0056] Alternating scanning with fixed-wavelength ultraviolet spectroscopy and variable-wavelength infrared spectroscopy, combined with a probabilistic prediction model to dynamically adjust the scanning frequency, and accurately capturing unexpected signals through adaptive thresholds and sliding windows, followed by principal component analysis to remove interference and characteristic peak extraction to obtain vibrational parameters, achieves accurate identification of known and new unexpected signals. Simultaneously, molecular configuration parameters are inferred based on vibrational characteristics, and the matching degree difference is calculated using activated carbon pore diffraction patterns, solving the problem of not being able to generate adjustment strategies in real time based on the physicochemical characteristics of unexpected pollutants, and avoiding mismatch between active sites and adsorption requirements. Addressing the lack of emergency pore-expanding mechanisms for trace pollutants, a targeted strategy is formulated based on the matching degree difference, retrieving methods from physical and chemical regulation libraries, generating gradient schemes through regulation parameter calculation formulas, and scanning pore distribution and adjusting parameters in real time during regulation to ensure rapid response to specific adsorption needs and resolve the penetration risk caused by missing adsorption sites. Addressing the quality control vacuum during the regulation transition period, pore changes are monitored through feedback intervals during regulation, and the effect is judged in real time by combining a matching degree evaluation system, initiating fine-tuning or regenerating the scheme, achieving full-chain quality control from detection to reconstruction, and avoiding performance fluctuations during the transition period. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of a method for controlling the pore size of activated carbon.

[0058] Figure 2 This is a flowchart of the process for filtering unexpected signals in this invention;

[0059] Figure 3 This is a flowchart of the matching analysis in this invention;

[0060] Figure 4 This is a structural diagram of an activated carbon pore regulation and monitoring device. Detailed Implementation

[0061] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0062] Example 1

[0063] refer to Figures 1 to 3 As shown in the figure, this embodiment introduces a method for controlling the pore size of activated carbon, including the following steps:

[0064] Step S1: Using an alternating scanning mode of fixed-wavelength ultraviolet spectroscopy and variable-wavelength infrared spectroscopy, the spectral scanning frequency is adjusted by calling a probabilistic prediction model to perform real-time full-domain spectral acquisition of flue gas. An adaptive threshold is set to adapt to background fluctuations and reduce false positives and false negatives. The temporal slope and acceleration are extracted using a dynamic sliding window to generate a temporal change matrix, accurately capturing signal trends and identifying abnormal points and abnormal spectra. Unexpected signals are filtered through the distribution of continuous abnormal points to ensure accurate identification. At the same time, the unexpected signals are analyzed by removing interference, enhancing unexpected features, extracting vibration parameters and matching them with a dynamic fingerprint database to achieve accurate identification of known and new unexpected signals.

[0065] Specifically, the steps for adjusting the spectral scanning frequency include:

[0066] When starting the scanning device, certain short-duration transient unexpected signals, such as mercury vapor from a pipeline leak, are prone to being missed. A pre-trained probability prediction model is invoked to obtain the probability of unexpected signals appearing in each band in the future. If the probability is greater than a preset probability threshold, it is determined to be a potential band; otherwise, it is determined to be a non-potential band. This method predicts potential bands where unexpected signals will appear in the future, and the scanning frequency of these potential bands is increased to the normal frequency. The scanning frequency is increased several times, while non-potential bands maintain the regular scanning frequency to concentrate resources on capturing potential pollutant signals. This improves the capture rate of transient pollutants without increasing the total scanning time, avoiding missed detections due to long scanning intervals. Simultaneously, real-time global spectral acquisition is performed to generate a three-dimensional raw spectral dataset containing wavelength, absorbance, and timestamps, which is then sliced ​​into a two-dimensional spectral matrix according to the timestamps. Each row corresponds to a wavelength, and each column corresponds to the absorbance at a specific time point. The probabilistic prediction model is built upon a long short-term memory network; given historical flue gas parameters, operating load, and raw material composition, it predicts the probability of pollutant occurrence. For timestamps.

[0067] Specifically, the steps for filtering out unexpected signals include:

[0068] Fixed thresholds cannot adapt to fluctuations in flue gas background, such as sudden increases in humidity causing baseline drift, which can easily lead to misjudgments or missed detections. To address this, historical spectral data from the most recent period is acquired, and high-frequency noise is eliminated using the Savitzky-Golay smoothing algorithm. Wavelet filtering is then used to separate the baseline components from the signal. At each wavelength, regions with absorbance fluctuations <5% and no expected pollutant characteristic peaks within 10 consecutive periods are selected to locate baseline regions with stable signals and no target features at each wavelength, such as the water vapor stability region in the infrared region. The arithmetic mean of all signal values ​​within the baseline region is calculated as the baseline mean. Simultaneously, the square root of the average of the squares of the differences between the signal values ​​within the baseline region and the baseline mean is calculated as the baseline standard deviation to reflect the natural fluctuations of the baseline. Real-time fluctuation entropy is calculated by combining the signal variance and kurtosis of the current spectral data to quantify the severity of signal fluctuations. The fluctuation ratio is calculated by combining the historical maximum fluctuation entropy. The sum of the fluctuation ratio and the base coefficient is then used to... Multiply by the baseline standard deviation by a factor of 1 to obtain the volatility value. The sum of the volatility value and the baseline mean is used to obtain the adaptive threshold, which can dynamically amplify or reduce the threshold range by the volatility ratio to adapt to background volatility.

[0069] A sliding window is set based on the signal rate of change, and the size of the sliding window will be adaptively adjusted according to the signal rate of change, including: based on the current time... As the endpoint, extract consecutive segments forward. A spectral data segment from each sampling period contains The spectral signal intensity of each sampling period forms an initial window. Based on variable-weight linear regression, time-decay weights are assigned to the spectral data at different times within the initial window. These weights decay exponentially over time, with the current time having a weight of 1, the previous time having a weight of 0.8, and so on, ensuring that more recent data has a greater impact on the current trend. Weighted least squares regression is then used to fit the equation to the data within the window, thus obtaining... The baseline slope at time step is taken, and the baseline slope at two consecutive time steps is taken. The slope difference is calculated, and the slope difference is normalized in combination with the sampling period to obtain the signal rate of change. The number of periods included in the sliding window is dynamically adjusted based on the absolute value of the signal change rate; if The signal is stable, the slope changes slowly, and the window expands to... Each cycle, smoothing out random noise using more historical data; if The signal fluctuates, the slope changes drastically, and the window shrinks to... Each cycle reduces the dilution of current transient changes by historical data, such as signal jumps during sudden pollutant exposure; among which, Adjust the threshold for the window;

[0070] Based on the adaptively adjusted sliding window, the time decay weight is still used. For the spectral data within the window, the weighted least squares method is used to fit a linear equation and obtain the slope, which is used as the time-series slope at the current time to reflect the true rate of change of the signal within the current window. At the same time, the second difference of the time-series slope at three consecutive time points is calculated and combined with the sampling period squared normalization to obtain the acceleration, thereby generating a time-series change matrix, which includes wavelength, timestamp, time-series slope and acceleration.

[0071] For each wavelength point in the time series variation matrix, if the time series slope is greater than the adaptive threshold or the acceleration is greater than zero, it is marked as a single abnormal point; otherwise, it is determined to be a normal fluctuation, the regular scan is maintained, and the data is stored in the historical spectral library to update the baseline parameters.

[0072] The band distribution of single anomalies is statistically analyzed. If five or more consecutive wavelength points in a certain band are single anomalies and are not within the characteristic range of the expected pollutant standard spectrum, then it is determined to be an abnormal spectrum and the analysis process is triggered.

[0073] The original spectral data corresponding to the abnormal spectra were extracted, and a background interference spectral library collected concurrently was imported, containing standard spectra of water vapor and dust under different humidity and dust concentrations. The original spectral data and the background interference spectral library data were merged to generate a matrix to be processed. Principal component analysis was used to calculate the covariance matrix of the matrix to be processed and perform eigenvalue decomposition to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​were sorted in descending order before calculation. The cumulative variance contribution rate is obtained by taking the proportion of the sum of individual eigenvalues ​​to the sum of all eigenvalues. The top three principal components with a cumulative variance contribution rate of not less than 95% are selected, where the first principal component corresponds to background interference and the second and third principal components correspond to pollutant signals.

[0074] The original spectral data of the abnormal spectrum is projected onto the eigenvector space of the three principal components to obtain the principal component score matrix, which is the projection value of the original spectral data onto each principal component. The score of the first principal component is removed, which is to remove the contribution of background interference. Only the scores of the second and third principal components are retained, which are the corresponding pollutant signals. Through the transpose operation of the eigenvector matrix, the retained scores of the second and third principal components are back-projected onto the original wavelength space to obtain a denoised spectrum containing only the pollutant signal. This achieves effective removal of water vapor and dust interference and highlights the characteristic peaks of the unexpected signal.

[0075] The pre-stored standard spectral library of expected pollutants is called. With the standard spectral library as the independent variable and the denoised spectrum as the dependent variable, a regression model is constructed using partial least squares regression. Based on the difference between the denoised spectrum and the model prediction, the residual spectrum is calculated to remove the expected pollutant signal and retain only the unexpected signal features. The peak intensity of the residual spectrum is normalized, and the maximum peak intensity is set to 1 to highlight the weak feature peaks, thus obtaining the enhanced unexpected feature spectrum.

[0076] Baseline correction was performed on unexpected characteristic spectra using polynomial fitting. A baseline trend line was fitted using the flat regions at both ends of the original spectrum as a reference. The baseline trend line was then subtracted from the original spectral data to eliminate baseline tilt caused by instrument drift or background interference. The corrected spectrum was smoothed using a moving average method to remove sawtooth fluctuations caused by high-frequency noise, resulting in a smooth spectrum that ensures the stability of subsequent peak shape identification. The first derivative of the smooth spectrum was calculated to find the extreme points where the derivative is zero. Based on these extreme points, the peaks corresponding to the points where the derivative changes from positive to negative were identified and used as candidate peaks. Extreme points with absorbance values ​​less than three times the baseline standard deviation were excluded to avoid noise misidentifying them as peaks. Candidate peaks with significantly higher peak intensities than baseline fluctuations are retained to generate a preliminary peak list. The smooth spectrum is then fitted with a Gaussian-Lorentz mixture function. For each candidate peak in the preliminary peak list, the peak shape is set to be a mixture of Gaussian and Lorentz functions in a preset ratio. Starting from the preliminary peak, the peak shape parameters are optimized through an iterative algorithm. Based on the sum of squares of the differences between the fitted and measured values, the residual between the fitted curve and the original spectrum is calculated until the residual is less than the residual ratio to ensure fitting accuracy. Key parameters of each characteristic peak are extracted, including peak position, half-width, peak intensity, and peak area, and arranged in ascending order of peak position to generate a set of vibrational characteristic parameters.

[0077] Each parameter in the vibration feature parameter set is standardized and transformed into a dimensionless vibration feature vector. A dynamic spectral fingerprint database is then called to calculate the cosine similarity between the vibration feature vector and known unexpected standard vectors in the database. The maximum cosine similarity is selected as the matching similarity, and the unexpected standard vector corresponding to the matching similarity is used as the matching spectrum. If the matching similarity is not less than the matching threshold, it is directly identified as a known pollutant in the dynamic spectral fingerprint database and determined as a known unexpected signal. Otherwise, it is determined as a new unexpected signal, and the vibration feature vector is stored in the dynamic spectral fingerprint database. The dynamic spectral fingerprint database is constructed by accumulating basic data and is used to store and manage the spectral features and associated information of unexpected signals, providing a basis for identifying pollutants. The system identifies and provides a standard comparison benchmark, and evolves itself through continuous learning. It has a built-in database update mechanism. Once a new, unexpected signal is identified, its corresponding metadata is extracted, including vibration characteristic parameter sets, standardized feature vectors, raw data of the abnormal spectrum, timestamps of occurrence, and operating parameters. A unique identifier is assigned to the new pollutant, stored in a new pollutant sub-database, and associated with its corresponding treatment records, such as subsequent pore control strategies and removal effects, forming a linked data chain. Incremental learning algorithms are used to adjust the historical mean and standard deviation databases, incorporating new pollutant parameters into the statistical scope to ensure the adaptability of subsequent standardization processing. Cluster analysis is performed on the feature vectors of similar pollutants, merging similar features, such as the same pollutant at different concentrations, reducing redundant data and improving comparison efficiency.

[0078] Step S2: Based on the extracted vibrational features, the pre-stored molecular size and vibrational mode correlation algorithm is invoked. The molecular configuration parameters of pollutant molecules are inferred from the feature peak spacing. Then, combined with the online diffraction pattern of the current pore distribution of activated carbon, the matching degree difference between pore size and molecular configuration is calculated for matching analysis. The inference logic is dynamically adjusted according to the matching degree difference to ensure the accuracy of the compatibility analysis between pore features and pollutant molecules.

[0079] Specifically, the steps of matching analysis include:

[0080] Based on the vibrational feature parameter set, unexpected molecular configuration parameters are inversely deduced, achieving high-precision and high-efficiency conversion from spectral features to molecular configuration. For known unexpected signals, the corresponding classification sub-model is invoked. The sub-model is trained based on quantum chemical data of similar unexpected signals, converting spectral features into molecular geometric parameters. This includes: inversely deducing the molecular spatial configuration, such as linear or planar triangular, through the characteristic peak spacing and peak intensity ratio; using the peak position and bond length correlation table in the classification sub-model, calling the corresponding bond length using the peak position in the vibrational feature set, calculating the minimum circumscribed circle diameter in combination with the molecular spatial configuration, calculating polarity parameters through the characteristic peak position offset, calculating atomic occupancy using VSEPR theory (valence shell electron pair repulsion theory), generating a three-dimensional model in combination with bond angle data, and taking the minimum sphere volume that encloses all atoms as the conservative value of the minimum envelope sphere volume, thereby generating molecular configuration parameters in a unified format.

[0081] For new and unexpected signals, a general sub-model is invoked and an incremental learning mechanism is triggered to generate a dedicated sub-model. A multi-task learning architecture is adopted to process multispectral features simultaneously. Different band weights are dynamically allocated through an attention mechanism to automatically focus on the key bands of unknown feature peaks. Based on the correlation between peak position and bond energy, combined with the influence of operating parameters on molecular vibrational frequency, the preliminary range of molecular dynamic diameter is initially inferred. Through the correlation between peak intensity ratio and molecular symmetry, the fuzzy range of polarity parameters is output. Monte Carlo simulation is used to generate multiple sets of molecular conformations. Combined with the molecular flexibility reflected by the half-width at half-maximum, the volume of the smallest sphere that encloses all atoms is taken as the conservative value of the minimum envelope sphere volume to generate the initial parameter set.

[0082] Incremental learning is carried out based on the initial parameter set to generate a dedicated sub-model. The initial parameter set, original spectral data, and operating condition parameters are encapsulated into an incremental learning data package and transmitted to the model training cluster through the federated learning interface to ensure data privacy. The training cluster adopts a knowledge transfer strategy to freeze the core feature extraction layer in the general sub-model when identifying known unexpected signals, and only fine-tunes the model parameters for the unique features of unknown unexpected signals, such as the unknown peak shape in a specific band.

[0083] Through multiple rounds of iterative optimization, the deviation between the molecular parameters predicted by the dedicated sub-model and the virtual labels generated based on interpolation of molecules with similar structures is compared, and the parameter weights are continuously adjusted. Finally, a dedicated sub-model for unknown and unexpected signals is generated to output an optimized set of molecular parameters. The dedicated sub-model is then validated. By combining the physical correlation between molecular parameters and flue gas diffusion characteristics, it is verified whether the parameters conform to the basic physical properties of gas molecules. The optimized parameters are input into the pore matching degree calculation model. If the output matching degree difference is within a reasonable range, it indicates that the parameters have no significant deviation. At this time, the dedicated sub-model is stored in the system model library for direct use in the subsequent identification of similar pollutants, and molecular configuration parameters for new unexpected signals are generated.

[0084] The built-in small-angle X-ray scattering device of the bed is activated to perform a full-domain scan of the activated carbon bed and generate a pore diffraction pattern. The diffraction pattern is fitted with peaks using Porod's law. By setting a secondary diameter threshold, the pore size is divided into three categories: micropores, mesopores, and macropores. The diffraction pattern is decomposed into scattering signals of micropores, mesopores, and macropores. The pore volume ratio and average ratio of each pore size range are calculated. At the same time, the pore distribution results are corrected by combining the bed pressure drop distribution data. If the pressure drop of a region is higher than the preset ratio of the average, it is determined to be pore blockage, and the pore volume weight of the corresponding pore size is reduced.

[0085] Simultaneously activate the X-ray photoelectron spectroscopy (XPS) device to analyze the distribution density of functional groups on the surface of activated carbon, calculate the polar site density, reflect the adsorption affinity of pores for polar molecules, and finally generate an activated carbon pore feature matrix, including pore size distribution curve, average pore size, and polar site density. Real-time capture of pore dynamic changes provides a current state benchmark for matching degree calculation.

[0086] Based on molecular diffusion theory, the effective pore range is determined so that molecules can enter the pores without being hindered by excessive size. The proportion of pore volume falling within the effective pore range in the pore size distribution of activated carbon is statistically analyzed to calculate the effective pore ratio. The effective pore ratio is then used as the base value and normalized using the Sigmoid function to obtain the size matching degree, which reflects the overall fit between molecular size and pore size.

[0087] The polarity difference between the unexpected signal and the pore surface is calculated using the absolute difference formula. The polarity matching degree is set to be negatively correlated with the polarity difference. An inverse proportional function is used for normalization to map the polarity difference to the polarity matching degree.

[0088] Based on the ratio of total pore volume to the number of pores, the average volume of a single pore is calculated. A volume fit ratio is set by comparing the volume of the minimum envelope sphere with the average volume of a single pore, reflecting the proportional relationship between molecular volume and pore space. Based on molecular packing theory, a volume ratio range is defined to ensure that molecules can fully utilize pore space without affecting adsorption stability due to crowding. The volume fit degree is calculated using a piecewise function. If the volume fit ratio is within the range, the volume fit degree is set to 1. If the volume fit ratio is less than the lower limit of the range, the volume fit degree decreases linearly as the volume fit ratio decreases. If the volume fit ratio is greater than the upper limit of the range, the volume fit degree decreases linearly as the volume fit ratio increases.

[0089] Weights are assigned to each matching degree to construct a multi-dimensional matching degree evaluation system, and a comprehensive matching degree is calculated. Based on the preset target matching degree, the matching degree difference is calculated, and a secondary matching threshold is set to classify the matching levels, including severe mismatch, moderate mismatch, and basic matching. The deduction logic is dynamically adjusted based on the matching level to generate an optimized pore control deduction logic. If there is a severe mismatch, the deduction logic automatically strengthens the "emergency pore enlargement" weight, prioritizing the increase of pore size. If there is a moderate mismatch, the deduction logic focuses on "polarity site control," such as activating potential polarity sites through physical fields. If there is a basic matching, the deduction logic maintains the "fine-tuning mode," only optimizing the pore volume distribution.

[0090] Step S3: Based on the matching degree difference, construct a gradient control scheme to control the pore gradient, monitor pore changes in real time and verify the effect, classify and store the obtained data in the database, extract the optimal parameters through cluster analysis, and optimize the model weights through incremental learning to achieve self-evolution, continuously improve the control efficiency and accuracy, and provide reliable support for pollutant adsorption.

[0091] Specifically, the steps for pore gradient regulation include:

[0092] Targeted strategies are formulated based on the matching degree difference, clarifying the core regulatory objectives. Corresponding regulatory methods are retrieved from pre-set physical and chemical regulatory libraries, and regulatory priorities are set based on deductive logic. Using the matching degree difference as input, corresponding regulatory parameter calculation formulas are invoked to calculate specific parameter gradients and time points, generating a gradient-based regulatory scheme. This scheme includes combinations of regulatory methods, parameter gradients, and time points, avoiding blind operation. Regulatory objectives include: for severe mismatch, the core objective is to improve size matching; for moderate mismatch, the core objective is to simultaneously optimize size and polarity matching; for basic matching, the core objective is to maintain the existing matching state through fine-tuning. Simultaneously, for new unexpected signals, a conservative coefficient is added to the original target setting, which is based on the error range derived from molecular parameters, to appropriately amplify the regulatory objective and avoid insufficient regulation due to parameter errors. Regulatory methods include: for severe mismatch, physical regulation is the primary method, prioritizing size adaptation; for moderate mismatch, physical and chemical regulation are combined, taking into account both size and polarity; for basic matching, only physical fine-tuning is used to avoid excessive intervention. The regulatory parameter calculation formulas are constructed based on historical regulatory data and porosity variation patterns.

[0093] Based on the generated gradient control scheme, control is carried out. During the control process, the pore distribution is scanned according to the feedback interval, and the control pore volume ratio is calculated. Based on the pore volume ratio calculated in the matching analysis, the difference is calculated to obtain the actual pore change, so as to determine whether the current control exceeds the expected deviation. Once the actual pore change exceeds the expected deviation, the parameters are automatically adjusted. Real-time feedback correction ensures that pore optimization is carried out according to the target, significantly improving the accuracy of control.

[0094] After the regulation is completed, the multi-dimensional matching degree evaluation system is invoked. Based on the real-time monitored pore characteristics, the geometric matching degree, polarity matching degree, and spatial matching degree are recalculated to obtain the comprehensive matching degree of regulation and the difference in matching degree of regulation. The regulation effect is then determined based on the difference in matching degree.

[0095] If the difference in the control matching degree is less than the first control threshold, it means that the current control has met the standard, and the control is stopped. The key parameters of this control are recorded as the benchmark control scheme, stored in the pore control knowledge base and associated with the corresponding pollutant type, so as to provide a reference for the control of similar pollutants in the future.

[0096] If the difference in the control matching degree is between the first control threshold and the second control threshold, it indicates a slight non-compliance. In this case, a fine-tuning mode is activated, such as low-frequency ultrasonic vibration, which optimizes the channel connectivity through gentle physical action and further reduces pore defects.

[0097] If the difference in the control matching degree is greater than the second control threshold, it indicates that the control is insufficient. The control scheme is regenerated based on the current pore characteristics and executed until the matching degree reaches the standard.

[0098] Database updates and strategy optimizations are performed. The entire chain of data from this regulation is stored in the pore regulation knowledge base by category. Cluster analysis is conducted on multiple regulation data of the same type of unexpected signals to extract the optimal parameter combination. The recommended parameters of the gradient regulation scheme are updated. The new data is then fed into the multi-dimensional matching degree evaluation system. The model weight allocation is optimized through incremental learning to improve the accuracy of subsequent matching degree calculations.

[0099] Example 2

[0100] Please see Figure 4 Another embodiment of the present invention provides: an activated carbon pore regulation and monitoring device, comprising: a spectral acquisition terminal, a sensor array, a regulation terminal and a central control terminal;

[0101] The spectral acquisition terminal is used to perform real-time global spectral acquisition of flue gas using an alternating scanning mode of fixed wavelength ultraviolet spectrum and variable wavelength infrared spectrum, generating a three-dimensional raw spectral dataset containing wavelength, absorbance and timestamp, and slicing it into a two-dimensional spectral matrix according to the timestamp, while receiving instructions from the central control terminal to adjust the scanning frequency.

[0102] The sensor array is used to collect relevant parameters of the activated carbon bed and flue gas, including generating activated carbon pore diffraction patterns through a small-angle X-ray scattering device, analyzing the distribution density of functional groups on the surface of activated carbon through an X-ray photoelectron spectroscopy device, collecting bed pressure drop distribution data through a pressure sensor, and collecting operating parameters such as flue gas temperature, humidity, and dust concentration.

[0103] The control terminal is used to perform pore control operations, including physical control and chemical control. It receives the gradient control scheme from the central control terminal, implements control according to parameter gradients and time nodes, and provides feedback on the equipment operating status during the control process.

[0104] The central control terminal coordinates the operation of the spectral acquisition terminal, sensor array, and control terminal to achieve intelligent management of the entire activated carbon pore control process. It receives spectral data from the spectral acquisition terminal, analyzes and identifies unexpected pollutant signals and updates the dynamic spectral fingerprint database, processes activated carbon pore characteristics and operating condition data collected by the sensor array, infers pollutant molecular configuration parameters, calculates the matching degree difference between pores and pollutants, generates a gradient control scheme based on the matching degree difference, controls the control terminal to perform physical or chemical control operations, monitors the control process in real time and corrects deviations, verifies the control effect, classifies and stores the entire chain of data, and optimizes the system strategy through cluster analysis and incremental learning. Ultimately, it achieves closed-loop management from pollutant identification and pore matching analysis to precise control and effect optimization, ensuring efficient adaptation between activated carbon pores and pollutants, and improving pollutant removal efficiency and system stability.

[0105] Specifically, the central control terminal includes an identification module, an analysis module, and a control and optimization module;

[0106] The identification module processes the spectral data output by the spectral acquisition terminal, calls the probability prediction model to adjust the spectral scanning frequency to improve the capture rate of transient pollutants, sets an adaptive threshold to adapt to flue gas background fluctuations, extracts the temporal slope and acceleration through a dynamic sliding window to generate a temporal change matrix, identifies abnormal points and abnormal spectra, analyzes the background interference of abnormal spectra, strengthens unexpected features, extracts the vibration feature parameter set, matches the vibration feature parameter set with the dynamic spectral fingerprint database, identifies known unexpected signals and new unexpected signals, and updates the dynamic spectral fingerprint database.

[0107] The analysis module, based on the vibration characteristic parameter set output by the identification module, calls the classification sub-model or general sub-model to infer the molecular configuration parameters of pollutants, receives the activated carbon pore diffraction pattern and pressure drop data output by the sensor array, analyzes and generates the activated carbon pore characteristic matrix, constructs a multi-dimensional matching degree evaluation system, calculates the size matching degree, polarity matching degree, spatial matching degree and comprehensive matching degree difference, and dynamically adjusts the pore control deduction logic based on the matching degree difference.

[0108] The regulation and optimization module, based on the matching degree difference and deductive logic output by the analysis module, formulates targeted regulation strategies. It retrieves regulation methods from the physical and chemical regulation libraries and sets priorities, generates gradient regulation schemes by calling the regulation parameter calculation formulas, controls the regulation terminal to execute the regulation schemes, receives pore change data from the sensor array in real time, calculates the actual pore change, and automatically adjusts parameters if the deviation exceeds the expected value. After regulation, it calls a multi-dimensional matching degree evaluation system to verify the effect. If the target is met, the regulation parameters are recorded as the benchmark scheme and stored in the knowledge base; if the target is not met, the scheme is regenerated until the target is met. Cluster analysis and incremental learning are performed on the knowledge base data to optimize the recommended values ​​of regulation parameters and the weights of the matching degree model, realizing the system's self-evolution.

[0109] Working principle and effects:

[0110] By alternately scanning flue gas with fixed-wavelength ultraviolet spectroscopy and variable-wavelength infrared spectroscopy, and using a probabilistic prediction model to adjust the scanning frequency, the potential unexpected signal bands are focused. By combining adaptive thresholds and adaptive sliding windows, the temporal slope and acceleration are extracted to form a temporal change matrix, anomalies are marked and abnormal spectra are screened, and then principal component analysis is used to remove background interference and partial least squares regression is used to remove expected signals. Vibration feature parameters are extracted and matched with a dynamic spectral fingerprint database to achieve the identification of known and new unexpected signals.

[0111] Subsequently, molecular configuration parameters were inferred based on vibrational characteristics. Simultaneously, the pore characteristics of activated carbon were obtained through small-angle X-ray scattering and X-ray photoelectron spectroscopy. The size, polarity, volume matching degree and comprehensive difference were calculated as the basis for regulation. Based on the matching degree difference, a targeting strategy was formulated. Methods were retrieved from a preset library and combined with the parameter calculation formula to generate a gradient scheme containing method combinations, parameter gradients and time nodes.

[0112] During regulation, porosity changes are monitored at feedback intervals, and parameters are automatically adjusted when deviations exceed expectations. After the regulation ends, the effect is judged through a multi-dimensional evaluation system. If the standard is met, it is saved as the benchmark scheme; if it is slightly substandard, fine-tuning is initiated; if it is insufficient, the scheme is regenerated. All data in the chain is stored in the knowledge base, and parameters and model weights are optimized through cluster analysis and incremental learning to form a closed-loop optimization mechanism.

[0113] Real-time capture of unexpected pollutants is achieved through adaptive signal analysis. A precise control benchmark is established based on dynamic calculation of molecular-pore matching degree. Combined with gradient control and closed-loop update mechanism, dynamic adaptation between pore characteristics and pollutant demand is ensured, thus solving the control adaptability problem when unexpected signals invade.

[0114] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for controlling the pore size of activated carbon, characterized in that, include: Based on a probabilistic prediction model, the spectral scanning frequency is dynamically adjusted, and real-time global spectral acquisition is performed. An adaptive threshold is set, and a dynamic sliding window is used to extract the temporal slope and acceleration, generate a temporal change matrix, identify abnormal spectra, filter out unexpected signals and analyze them, infer molecular configuration parameters through the characteristic peak spacing, calculate the difference in matching degree between pore size and molecular configuration, and perform matching analysis. Based on the matching degree difference, pore gradient is controlled, and the control effect is verified by real-time monitoring.

2. The method for controlling the pore size of activated carbon according to claim 1, characterized in that, The steps for adjusting the spectral scanning frequency include: When the scanning device is started, a probability prediction model is invoked to predict the probability of unexpected signals appearing in each band in the future period; Bands with a probability greater than a probability threshold are identified as potential bands, and the rest are non-potential bands. Set the scanning frequency of the potential band to the normal frequency. The non-potential band maintains the regular scanning frequency, and real-time global spectral acquisition is performed simultaneously.

3. The method for controlling the pore size of activated carbon according to claim 2, characterized in that, The steps for generating the time series variation matrix include: Acquire historical spectral data for the most recent period and locate the baseline region within the spectral data; Calculate the baseline parameters of the baseline region, including the mean and standard deviation, and simultaneously calculate the real-time fluctuation entropy of the current spectral data. Combined with the historical maximum fluctuation entropy, calculate the fluctuation ratio. An adaptive threshold is set by combining the baseline parameters with the fluctuation ratio. Using the current time as the endpoint, an initial window is set. Based on variable weight linear regression, time decay weights are assigned to the data within the initial window. Weighted least squares method is used for fitting to obtain the basic slope at the current time. Calculate the difference in fundamental slope between two consecutive time points, and normalize it using the sampling period to obtain the rate of change of the signal. : like The window is expanded to One cycle; if Window minimized to One cycle; among which, Adjust the threshold for the window; Based on the adjusted sliding window, the linear equation is fitted using the weighted least squares method with time decay weights to obtain the time-series slope. The second-order difference of the time-series slope at three consecutive time points is calculated and normalized by the square of the sampling period to obtain the acceleration, thus generating the time-series change matrix.

4. The method for controlling the pore size of activated carbon according to claim 3, characterized in that, The steps for filtering out unexpected signals include: For each wavelength point in the time series variation matrix, if the time series slope is greater than the adaptive threshold or the acceleration is greater than zero, it is marked as a single abnormal point; otherwise, it is judged as a normal fluctuation and the data is stored in the historical spectrum library. The band distribution of single anomalies is statistically analyzed. If five or more consecutive wavelength points in a certain band are single anomalies and are not within the characteristic range of the expected pollutant standard spectrum, then it is determined to be an abnormal spectrum and the analysis process is triggered. Extract the original spectral data corresponding to the abnormal spectra and merge them with the background interference spectral library collected at the same time to generate a matrix to be processed; The covariance matrix of the matrix to be processed is calculated by principal component analysis and eigenvalues ​​and eigenvectors are obtained. The top three principal components with a cumulative variance contribution rate of not less than 95% are selected, and the first principal component is removed to generate a denoised spectrum. The pre-stored standard spectral library of expected pollutants is called, and a regression model is constructed using partial least squares regression with the standard spectral library as the independent variable and the denoised spectrum as the dependent variable. Based on the difference between the denoised spectrum and the model prediction, the residual spectrum is calculated, and the peak intensity of the residual spectrum is normalized to generate the unexpected feature spectrum.

5. The method for controlling the pore size of activated carbon according to claim 4, characterized in that, The specific steps for filtering unexpected signals also include: Baseline correction and moving average smoothing are performed on the unexpected characteristic spectrum to generate a smoothed spectrum. The first derivative is calculated to find candidate peak positions and then screened to generate a preliminary peak position list. The smooth spectrum peaks were fitted using a Gaussian-Lorentz mixture function, and the key parameters of each characteristic peak were extracted and sorted by peak position to generate a set of vibrational characteristic parameters. Each parameter in the vibration feature parameter set is standardized to obtain a vibration feature vector. The dynamic spectral fingerprint library is called to calculate the cosine similarity between the vibration feature vector and the known unexpected standard vectors in the library. The maximum value is selected as the matching similarity and the corresponding matching spectrum. If the matching similarity is not less than the matching threshold, it is determined to be a known unexpected signal; otherwise, it is determined to be a new unexpected signal, and the vibration feature vector is stored in the dynamic spectral fingerprint database.

6. The method for controlling the pore size of activated carbon according to claim 5, characterized in that, The steps of matching analysis include: For known unexpected signals, the corresponding classification sub-model is activated to convert spectral features into molecular geometric parameters and generate molecular configuration parameters. For new unexpected signals, the general sub-model is invoked and the incremental learning mechanism is triggered to generate an initial parameter set. Incremental learning is then carried out based on the initial parameter set to generate a specific sub-model and to generate the molecular configuration parameters of the new unexpected signals.

7. The method for controlling the pore size of activated carbon according to claim 6, characterized in that, The steps of matching analysis also include: The small-angle X-ray scattering device and X-ray photoelectron spectroscopy device built into the bed are activated to generate the activated carbon pore feature matrix. The size matching degree is calculated based on molecular diffusion theory, the polarity matching degree is calculated using the absolute difference formula, and the volume matching degree is calculated based on molecular packing theory. Weights are set for each matching degree and the overall matching degree is calculated. The matching degree difference is calculated based on the preset target matching degree, a secondary matching threshold is set to divide the matching level, and the deduction logic is dynamically adjusted based on the matching level to generate the optimized pore control deduction logic.

8. The method for controlling the pore size of activated carbon according to claim 7, characterized in that, The steps for pore gradient control include: Based on the matching degree difference, the core control target is identified, and the corresponding control methods are retrieved from the preset physical control library and chemical control library. The control priority is set based on the deduction logic, and the corresponding control parameter calculation formula is called with the matching degree difference as input to calculate the parameter gradient and time node, thereby generating a gradient control scheme. The control is carried out based on the gradient control scheme. During the control process, the pore distribution is scanned at the feedback interval, the control pore volume ratio is calculated, the actual pore change is calculated based on the pore volume ratio in the matching analysis, and the parameters are automatically adjusted when the actual pore change exceeds the expected deviation. After the regulation is completed, a multi-dimensional matching degree evaluation system is invoked to calculate the regulation matching degree difference based on the real-time monitored pore characteristics, so as to determine the regulation effect and make corrections. The entire chain of data from this regulation is stored in the pore regulation knowledge base according to category. Cluster analysis is performed on multiple regulation data of the same type of unexpected signals to extract the optimal parameter combination, update the recommended parameters of the gradient regulation scheme, and input the new data into the multi-dimensional matching degree evaluation system and optimize the weight allocation through incremental learning.

9. An activated carbon pore control monitoring device, used to implement the activated carbon pore control method as described in any one of claims 1-8, characterized in that, include: Spectrum acquisition terminal, sensor array, control terminal, and central control terminal; The spectral acquisition terminal is used to perform real-time full-domain spectral acquisition of flue gas, generate a three-dimensional raw spectral dataset, and receive instructions from the central control terminal to adjust the scanning frequency. The sensor array is used to collect parameters of the activated carbon bed and flue gas. The control terminal is used to perform physical and chemical control, receive the gradient control scheme from the central control terminal, implement control according to parameter gradients and time nodes, and provide feedback on the equipment operating status. The central control terminal is used to receive spectral data, identify unexpected pollutants and update the dynamic spectral fingerprint database, process sensor array data, generate gradient control schemes, monitor and correct deviations in real time, verify the effect, store the data in the database, and optimize the strategy through cluster analysis and incremental learning.

10. The activated carbon pore control and monitoring device according to claim 9, characterized in that: The central control terminal includes an identification module, an analysis module, and a regulation and optimization module; The identification module is used to adjust the spectral scanning frequency, generate a time-series change matrix based on an adaptive threshold and a dynamic sliding window, determine abnormal points and abnormal spectra, extract a set of vibration feature parameters, and match the set of vibration feature parameters with a dynamic spectral fingerprint database to identify known unexpected signals and new unexpected signals. The analysis module is used to infer the molecular configuration parameters of pollutants, analyze and generate the activated carbon pore feature matrix, construct a multi-dimensional matching degree evaluation system, calculate the matching degree difference, and dynamically adjust the pore regulation deduction logic. The regulation and optimization module is used to generate a gradient regulation scheme and calculate the actual porosity change. If the deviation exceeds the expected value, the parameters are automatically adjusted. After the regulation is completed, the effect is verified, and cluster analysis and incremental learning are performed on the knowledge base data to optimize the recommended values ​​of the regulation parameters and the weights of the matching degree model.

Images