Desulfurization optimization control method and system based on multi-modal model

Abstract

The invention discloses a desulfurization optimization control method and system based on a multi-modal model, and relates to the technical field of industrial intelligent control, and the method comprises the steps: constructing a gas-liquid interface boundary point set, and generating a control feature vector; calculating fractal dimensions of control feature vectors by using a box counting method, constructing a disturbance point set, multiplying disturbance intensity by an expansion trend to obtain an unstable diffusion grade, matching key propagation points with nearest disturbance points by using nearest neighbor matching, and generating a control priority sequence; and defining a control unstable diffusion level, defining an optimization objective function, and solving by using constraint optimization to obtain a final control variable. According to the method, a composite strength factor is constructed through an image gradient field and a signal response ratio, a gas-liquid interface boundary point set and a control feature vector are constructed, the robustness and information density of features are improved, a main resonance frequency is recognized by combining a fractal interference field and short-time Fourier transform, key propagation points are screened by using wavelet transform and information entropy analysis, and a high-precision gas-liquid interface is obtained. The desulfurization efficiency and the stability of the reaction tower are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial intelligent control, and in particular to a desulfurization optimization control method and system based on a multimodal model. Background Art

[0002] With the upgrading of the energy industry structure and the continuous improvement of environmental protection requirements, coal-fired power plants, steel mills and other industrial enterprises have an increasing demand for optimized control of flue gas desulfurization systems. Although traditional wet flue gas desulfurization (FGD) technology has advantages such as high reaction rate and stable desulfurization efficiency, it still faces problems such as complex operating conditions, feedback control lag, and single detection dimension in industrial operation. It is difficult to perceive and accurately adjust microscopic parameters such as the state of the gas-liquid reaction interface, foam structure changes, and reaction intensity distribution in real time. The rapid development of technologies such as artificial intelligence, image processing, and industrial Internet of Things has provided basic support for the intelligent control of desulfurization systems driven by multimodal data.

[0003] Existing desulfurization control technologies generally have the following shortcomings: most control methods are based on linear modeling or empirical rules, which fail to reflect the multi-scale coupling characteristics of the reaction system and find it difficult to capture the unsteady diffusion behavior during the disturbance propagation process. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a desulfurization optimization control method and system based on a multimodal model to solve the problem that most control methods are based on linear modeling or empirical rules, fail to reflect the multi-scale coupling characteristics of the reaction system, and have difficulty in capturing the unstable diffusion behavior during the disturbance propagation process.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a desulfurization optimization control method based on a multimodal model, which comprises:

[0008] Collect and preprocess multimodal data, calculate horizontal gradients using the Sobel operator, construct a set of gas-liquid interface boundary points, place alternating electric field induction probes to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal sections, calculate the average composite intensity of each area, and generate a control feature vector.

[0009] The fractal dimension of the control eigenvector is calculated using the box counting method. The fractal dimension and the control eigenvector are weighted to construct a fractal interference field. The maximum STFT amplitude is identified as the main resonant frequency. A set of disturbance points is constructed. The covariance matrix of the disturbance point set is calculated. The maximum eigenvalue is extracted as the main direction expansion trend of the disturbance point. The disturbance intensity and the expansion trend are multiplied to obtain the unstable diffusion level. Based on the wavelet coefficients, the total energy of each scale is calculated. Points with absolute values of wavelet coefficients greater than the average value are screened and marked as key propagation points. The key propagation points are matched with the nearest disturbance point using nearest neighbor matching to generate a control priority sequence.

[0010] Define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables.

[0011] As a preferred solution of the desulfurization optimization control method based on the multimodal model of the present invention, the calculation of the average composite strength of each region and the generation of the control characteristic vector include:

[0012] The slurry foam image is converted into a grayscale image using the weighted average method, the grayscale potential energy is calculated using logarithmic transformation, the horizontal gradient is calculated using the Sobel operator, the gradient threshold is set, and the pixel points that meet the gradient threshold are extracted as the gas-liquid interface boundary using the gradient threshold method to construct the gas-liquid interface boundary point set;

[0013] Extract the boundary points of the gas-liquid interface boundary point set, place an alternating electric field induction probe to emit weak electromagnetic waves, collect the reflected signal amplitude, calculate the signal response ratio, and use low-pass filtering to remove high-frequency noise in the signal response ratio to obtain the filtered signal response ratio;

[0014] Point-by-point product fusion is used to fuse the grayscale potential energy and the corresponding signal response ratio into a composite reaction intensity. The desulfurization tower profile is divided into three layers: upper, middle, and lower regions using the equal division method based on the pixel position. The average composite intensity of each region is calculated, and the average composite intensity is normalized to generate a control feature vector.

[0015] As a preferred solution of the desulfurization optimization control method based on the multimodal model of the present invention, wherein: the said multiplication of the disturbance intensity and the expansion trend to obtain the unstable diffusion level includes:

[0016] The fractal dimension of the control eigenvector is calculated using the box counting method. The fractal dimension and the control eigenvector are weighted to construct the fractal interference field. The STFT amplitude of the fractal interference field is extracted using the short-time Fourier transform. The maximum STFT amplitude is identified as the main resonance frequency. The detection threshold is set and the main resonance frequency is screened. The corresponding time point t ′Mark as disturbance point, extract the control eigenvector in the time window corresponding to the disturbance point, construct the disturbance point set, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point;

[0017] Based on the control eigenvector, the disturbance intensity is calculated, and the unstable diffusion level is obtained by multiplying the disturbance intensity and the expansion trend.

[0018] As a preferred solution of the multimodal model-based desulfurization optimization control method of the present invention, the method of using nearest neighbor matching to match the key propagation point with the nearest disturbance point to generate a control priority sequence includes:

[0019] Calculate the information entropy of the unstable diffusion level, use the Daubechies-4 wavelet basis to calculate the wavelet coefficients of the information entropy, calculate the total energy of each scale based on the wavelet coefficients, calculate the energy proportion of each scale, select the maximum energy scale, and select the points where the absolute value of the wavelet coefficient is greater than the average value and mark them as key propagation points;

[0020] The key propagation point is matched with the nearest disturbance point using nearest neighbor matching to obtain matching pairs. The index of the corresponding disturbance point is searched for each matching pair, the corresponding unstable diffusion level is extracted, and a temporary sequence is generated. The absolute values of the corresponding wavelet coefficients are extracted from the temporary sequence and arranged in descending order to generate a control priority sequence.

[0021] As a preferred solution of the desulfurization optimization control method based on the multimodal model of the present invention, the step of defining the optimization objective function and obtaining the final control variable includes:

[0022] Based on the control priority sequence, the control level of unstable diffusion is defined, the state vector is constructed, and a linear regression model is constructed to predict SO2 concentration and pH value;

[0023] Define the optimization objective function and use constrained optimization to solve the problem and obtain the final control variables.

[0024] As a preferred solution of the desulfurization optimization control method based on the multimodal model of the present invention, the construction of a visual interface to display the final control variables includes:

[0025] Use the front-end framework React.js to build a visual interface to visualize the final control variables;

[0026] Users who have passed real-name verification are allowed to access the information.

[0027] As a preferred solution of the desulfurization optimization control method based on the multimodal model of the present invention, the collecting and preprocessing of multimodal data includes:

[0028] The multimodal data includes slurry foam images, SO2 concentration and pH value data;

[0029] An industrial camera was used to collect slurry foam images at the desulfurization tower observation window, an electrochemical sensor was used to collect SO2 concentration at the desulfurization tower outlet flue, and a pH meter was used to collect pH value from the slurry pool. The collected multimodal data were denoised and normalized.

[0030] In a second aspect, the present invention provides a desulfurization optimization control system based on a multimodal model, comprising:

[0031] The vector collection module is used to collect and preprocess multimodal data, calculate the horizontal gradient using the Sobel operator, construct a set of gas-liquid interface boundary points, place an alternating electric field induction probe to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal parts, calculate the average composite intensity of each area, and generate a control feature vector.

[0032] A hierarchical control module is used to calculate the fractal dimension of the control eigenvector using the box counting method, perform weighted processing on the fractal dimension and the control eigenvector, construct a fractal interference field, identify the maximum STFT amplitude as the main resonant frequency, construct a set of disturbance points, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point, multiply the disturbance intensity and the expansion trend to obtain the unstable diffusion level, calculate the total energy of each scale based on the wavelet coefficients, screen points with absolute values of wavelet coefficients greater than the average value, mark them as key propagation points, use nearest neighbor matching to match the key propagation points to the nearest disturbance point, and generate a control priority sequence;

[0033] The target visualization module is used to define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables.

[0034] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the desulfurization optimization control method based on a multimodal model as described in the first aspect of the present invention is implemented.

[0035] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the desulfurization optimization control method based on a multimodal model as described in the first aspect of the present invention is implemented.

[0036] The beneficial effects of the present invention are as follows: the present invention constructs a composite intensity factor through the image gradient field and the signal response ratio, constructs a gas-liquid interface boundary point set and a control feature vector, improves the robustness and information density of the features, combines the fractal interference field and the short-time Fourier transform to identify the main resonance frequency, uses the wavelet transform and information entropy analysis to screen the key propagation points, and improves the desulfurization efficiency and the stability of the reaction tower. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a flow chart of the desulfurization optimization control method based on the multimodal model in Example 1.

[0039] Figure 2 Schematic diagram of the desulfurization optimization control system based on the multimodal model in Example 1. DETAILED DESCRIPTION

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0041] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0043] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a desulfurization optimization control method based on a multimodal model, comprising the following steps:

[0044] S1. Collect and preprocess multimodal data, calculate the horizontal gradient using the Sobel operator, construct a set of gas-liquid interface boundary points, place an alternating electric field induction probe to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal sections, calculate the average composite intensity of each area, and generate a control feature vector.

[0045] Specifically, collect multimodal data and perform preprocessing, including:

[0046] The multimodal data includes slurry foam images, SO2 concentration and pH value data;

[0047] An industrial camera was used to collect slurry foam images at the desulfurization tower observation window, an electrochemical sensor was used to collect SO2 concentration at the desulfurization tower outlet flue, and a pH meter was used to collect pH value from the slurry pool. The collected multimodal data were denoised and normalized.

[0048] Slurry foam images can provide visual information of the gas-liquid interface, concentration data reflects the chemical composition of the gas and liquid, and pH data can reveal the acidity and alkalinity of the reaction medium. Multimodal data are denoised and normalized, which not only ensures the accuracy of data quality but also eliminates interference from external noise, thereby ensuring the reliability of subsequent analysis.

[0049] Furthermore, the average composite strength of each region is calculated to generate a control feature vector, including:

[0050] The slurry foam image is converted into a grayscale image using the weighted average method, and the grayscale potential energy is calculated using logarithmic transformation. The formula is:

[0051]

[0052] Where (x, y) is the grayscale potential energy of the pixel (x, y), indicating the density of the gas-liquid distribution at the pixel (x, y), G(x, y) is the grayscale value of the pixel (x, y), and ε is a very small constant to prevent the denominator from being zero in the logarithmic operation;

[0053] The Sobel operator is used to calculate the horizontal gradient, and the gradient threshold is set. The gradient threshold method is used to extract the pixel points that meet the gradient threshold as the gas-liquid interface boundary, and the gas-liquid interface boundary point set is constructed. The formula is:

[0054]

[0055] in is the change of the potential field in the horizontal direction, reflecting the intensity of the change of the gas-liquid interface, ρ is the gradient threshold, and is set using the ROC curve parameter adjustment;

[0056] Extract the boundary points of the gas-liquid interface boundary point set, place an alternating electric field induction probe to emit weak electromagnetic waves, collect the reflected signal amplitude, and calculate the signal response ratio. The formula is:

[0057]

[0058] Where S(t) is the signal response ratio at time t, reflecting the turbidity and particle size concentration of the slurry, V r (t) is the amplitude of the reflected electromagnetic wave at time t, V in The input electromagnetic wave amplitude is collected from the equipment manufacturer's parameters;

[0059] Use low-pass filtering to remove high-frequency noise from the signal response ratio to obtain a filtered signal response ratio;

[0060] Use point-by-point product fusion to fuse the grayscale potential energy and the corresponding signal response ratio into a composite reaction intensity. The formula is:

[0061] M(x,y)=(x,y) · S flit (t x,y ),

[0062] Where M(x,y) is the composite reaction intensity of the pixel (x,y), which represents the comprehensive state of gas-liquid distribution and turbidity, and S flit (t x,y ) is the corresponding time t x,y The filter signal response ratio, time is associated with the boundary point position;

[0063] The desulfurization tower section is divided into three layers: upper, middle, and lower regions using the equal division method based on the pixel position. The average composite intensity of each region is calculated using the formula:

[0064]

[0065] in is the average composite strength of the sth region (upper, middle, and lower layers), A s is the pixel set of the sth region, and s is the number of hierarchical regions;

[0066] Normalize the average composite intensity to generate the control feature vector, the formula is:

[0067]

[0068] Where F(t) is the control characteristic vector at time t, which represents the operating characteristics of the three-layer area in the tower. as well as is the maximum average composite intensity of regions 1, 2, and 3 (upper, middle, and lower layers) in historical data, T is the transposition operation, as well as is the average composite strength of the 1st, 2nd and 3rd regions.

[0069] Extracting boundary points through the gradient threshold method not only accurately describes the morphological changes of the gas-liquid interface, but also reveals the spatial characteristics of the gas-liquid distribution, enables a better understanding of the dynamic behavior of the slurry and gas in the desulfurization tower, and accurately locates the key areas affecting the desulfurization efficiency. By placing an alternating electric field sensing probe to emit weak electromagnetic waves and calculate the signal response ratio, it is possible to detect changes in slurry turbidity and particle size. By combining the grayscale potential energy with the electromagnetic wave signal response ratio to construct a composite reaction intensity, it is possible to more comprehensively describe the comprehensive state of gas-liquid distribution and turbidity, which provides an accurate quantitative standard for the status assessment of various areas of the desulfurization tower. By dividing the operating differences of different areas inside the desulfurization tower, the control system can more specifically optimize the operating conditions of each area, generate control characteristic vectors, and provide accurate reference indicators for subsequent optimization control.

[0070] S2. Use the box counting method to calculate the fractal dimension of the control eigenvector, perform weighted processing on the fractal dimension and the control eigenvector, construct a fractal interference field, identify the maximum STFT amplitude as the main resonance frequency, construct a set of disturbance points, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point, multiply the disturbance intensity and the expansion trend to obtain the unstable diffusion level, calculate the total energy of each scale based on the wavelet coefficients, screen points with absolute values of wavelet coefficients greater than the average value, mark them as key propagation points, use nearest neighbor matching to match the key propagation points to the nearest disturbance point, and generate a control priority sequence;

[0071] Specifically, the disturbance intensity and expansion trend are multiplied to obtain the unstable diffusion level, including:

[0072] The fractal dimension of the control eigenvector is calculated using the box counting method, and the formula is:

[0073]

[0074] Where D is the fractal dimension, reflecting the nonlinear self-similarity of the gas-liquid distribution disturbance, N() is the number of boxes covering the control eigenvector trajectory at the grid scale, and is the grid scale;

[0075] The fractal dimension and the control eigenvector are weighted to construct the fractal interference field. The formula is:

[0076]

[0077] Where F′(t) is the fractal interference field at time t, N s () is the number of boxes of the sth feature under scale, D is the fractal dimension weighting factor, which is set using empirical scale selection;

[0078] The STFT amplitude of the fractal interference field is extracted using short-time Fourier transform, and the formula is:

[0079] Ψ(f,t)=|∑ τ F ′ (τ)·w(τ-t) · e -j2πfτ |,

[0080] Where Ψ(f,t) is the STFT amplitude at time t and frequency f, reflecting the spectral characteristics of the disturbance level, F′(τ) is the fractal interference field at time τ, and w( · ) is the Hamming window function, j is the virtual unit, f is the frequency variable, and τ is the time variable;

[0081] Identify the maximum STFT amplitude as the main resonant frequency, the formula is:

[0082]

[0083] where f * is the main resonant frequency;

[0084] Set the detection threshold and filter the main resonance frequency, and set the corresponding time point t ′ Marked as disturbance point;

[0085] Extract the control feature vector in the time window corresponding to the disturbance point and construct the disturbance point set. The formula is:

[0086]

[0087] in is the set of i-th disturbance points, t ′ i is the i-th disturbance point at time t, Δt is the time window length, which is set using the empirical setting method;

[0088] Calculate the covariance matrix of the disturbance point set and extract the maximum eigenvalue as the main direction expansion trend of the disturbance point. The formula is:

[0089]

[0090] where θ i is the main direction expansion trend of the i-th disturbance point, is the covariance matrix of the disturbance point set, eig( · ) is the set of eigenvalues of the covariance matrix;

[0091] Based on the control eigenvector, the disturbance intensity is calculated using the formula:

[0092]

[0093] Where V i is the intensity of the i-th disturbance point;

[0094] Multiplying the disturbance intensity and the expansion trend gives the unstable diffusion level, which is:

[0095] ζ i =V i ·θ i ,

[0096] Among them i is the unstable diffusion level of the i-th disturbance point.

[0097] The fractal dimension of the control eigenvector is calculated by the box counting method, which can reveal the nonlinear self-similarity of the gas-liquid distribution disturbance. The fractal dimension D reflects the self-similarity of the disturbance at different scales, which can reveal the dynamic characteristics of the gas-liquid interface changes, and provides a deep understanding of the complex fluid behavior. The fractal interference field is constructed, which enables the system to accurately identify the detailed characteristics of the disturbance in the complex gas-liquid interface disturbance. By introducing the fractal theory, the recognition and response capabilities of the desulfurization system to nonlinear disturbances are improved. By processing the fractal interference field through short-time Fourier transform, the time domain signal of the disturbance can be effectively converted into the frequency domain signal. By identifying the main resonant frequency, the system can quickly locate the disturbance source with the strongest frequency response in the desulfurization tower, thereby achieving efficient dynamic control and significantly improving the system's sensitivity and response to frequency disturbances. The calculation of the covariance matrix helps evaluate the correlation between disturbance points, further revealing the propagation direction and trend of disturbances in the system. By multiplying the disturbance intensity and the expansion trend to obtain the unstable diffusion level, the system can quantify the intensity and propagation direction of the disturbance. The unstable diffusion level provides an accurate disturbance evaluation index for subsequent optimization control, helps the system achieve adaptive adjustment, and improves its ability to respond to complex dynamic environments.

[0098] Furthermore, the nearest neighbor matching is used to match the key propagation points to the nearest disturbance points, generating a control priority sequence, including:

[0099] Calculate the information entropy of the unstable diffusion level. If the disturbance point set is empty, there is no disturbance point, and the information entropy is set to 0. The formula is:

[0100]

[0101] Where H(t) is the information entropy at time t, W t is the set of disturbance points at time t;

[0102] The Daubechies-4 wavelet basis is used to calculate the wavelet coefficient of information entropy. The formula is:

[0103]

[0104] Where C(a,b) is the wavelet coefficient of scale a at time b, reflecting the local time-frequency characteristics of the entropy sequence, a is the decomposition scale, set using the time-frequency resolution matching method, b is the time shift, ψ(t) is the db4 wavelet function at time t, g k′ is a fixed coefficient, db4 standard value, is the db4 scaling function, k′ is the time shift index used to generate the discrete version of the wavelet function, set using the DWT algorithm;

[0105] Based on the wavelet coefficients, the total energy of each scale is calculated as follows:

[0106] E a =∑ b |C(a,b)| 2 ,

[0107] Among them E a is the total energy of scale a;

[0108] Based on the total energy, calculate the energy proportion of each scale and select the maximum energy scale. The formula is:

[0109]

[0110] Among them, P a is the energy proportion of scale a, J is the decomposition scale index set, Scale 2 j energy;

[0111] Based on the maximum energy scale, the points whose absolute values of wavelet coefficients are greater than the average value are screened and marked as key propagation points. The formula is:

[0112]

[0113] where a * is the dominant disturbance scale, indicating the most significant propagation time scale, C(a * ,b k ) is the scale a * , at time b k The wavelet coefficients, Z b is scale a * The total number of time points of the wavelet coefficients;

[0114] Use nearest neighbor matching to match the key propagation point with the nearest disturbance point to obtain a matching pair. For each matching pair, find the index of the corresponding disturbance point, extract the corresponding unstable diffusion level, and generate a temporary sequence. The formula is:

[0115]

[0116] where t kis the disturbance time corresponding to the key propagation point;

[0117] Extract the absolute values of the corresponding wavelet coefficients from the temporary sequence and arrange them in descending order to generate a control priority sequence. The formula is:

[0118] {(t k ,ζ k )},

[0119] where t k To control the disturbance time in the priority sequence, ζ k is the corresponding unstable diffusion level.

[0120] By using the Daubechies-4 wavelet basis to calculate the wavelet coefficients of information entropy, the system can extract the energy characteristics of the disturbance signal at different scales. The key propagation points play a vital role in the disturbance propagation process. By matching the nearest disturbance point, the system can provide an accurate control priority sequence. By generating the control priority sequence in real time, the system can quickly locate and process the most important disturbance points, avoiding excessive response to unimportant disturbances.

[0121] S3. Define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables;

[0122] Specifically, define the optimization objective function and obtain the final control variables, including:

[0123] Based on the control priority sequence, the control unstable diffusion level is defined and the state vector is constructed. The formula is:

[0124]

[0125] S(t)=[F(t),ζ′(t)] T ,

[0126] where ζ′(t) is the control unstable diffusion level at time t, and S(t) is the state vector at time t;

[0127] Collect historical SO2 concentration and pH value data, generate training sets for SO2 concentration and pH value respectively, construct linear regression models, use the training sets for SO2 concentration and pH value respectively, train the linear regression models, use the loss function and Adam optimizer to iteratively optimize the model parameters, and use the trained linear regression models to predict SO2 concentration and pH value;

[0128] Define the optimization objective function, the formula is:

[0129]

[0130] Where J is the optimization objective function value, Δt ′ is the prediction window length, and pH(t) are the predicted values of SO2 concentration and pH value at time t, respectively. and pH ref The target values of SO2 concentration and pH value are respectively, based on the national emission standard GB 28662-2012, the emission standard of air pollutants from thermal power plants, and the average value setting based on the optimal reaction conditions of wet flue gas desulfurization, u m (t) is the mth controlled variable, including limestone slurry flow rate and air flow rate, which is set using the empirical rule;

[0131] Use constrained optimization to solve and obtain the final control variables. The formula is:

[0132] {u ′ m (t)}=argminJ,stg(S(t),u m (t))≤0,

[0133]

[0134] where u ′ m (t) is the final control variable, g(S(t),u m (t)) is the constraint function, which is set using empirical rules, u 1,max and u 2,max are the upper limit of slurry flow rate and the upper limit of air flow rate, respectively, which are set using historical data. non , O slur and O air are the non-negativity constraint, slurry flow constraint and air flow rate constraint respectively, u1(t) and u2(t) are the limestone slurry flow rate and air flow rate at time t respectively.

[0135] The unstable diffusion level represents the comprehensive characteristics of disturbance intensity and propagation trend. By constructing a state vector based on it, the system disturbance can be closely combined with the subsequent optimization control decision, and the sensitivity of the control model to unstable behavior can be enhanced. Compared with the traditional PID control method that relies on fixed set points, the introduction of the state vector provides a unified mathematical expression framework for multi-variable coupling and dynamic change processes. The linear regression model is used to predict the desulfurization efficiency and the acidity and alkalinity of the absorption tower liquid phase in the future time window, which helps to intervene in the control strategy in advance. The objective function integrates the core performance indicators of the desulfurization system (SO2 concentration, slurry pH) and the national emission standards (GB 28662-2012), and combined with the optimal process setting values under actual working conditions, the optimization model strikes a balance between compliance and process feasibility. By minimizing the difference between the system emission prediction value and the target value through the objective function, the mathematical modeling of multi-objective optimization is realized, and a systematic solution path is provided for multivariable control problems under constraints. The introduction of non-negativity constraints avoids the occurrence of physically unreasonable negative flow control strategies. By introducing empirically set constraint functions, the adaptability of the model to the actual operating environment is improved, ensuring that the solution results can be called and executed by the actual PLC or DCS system.

[0136] Furthermore, a visualization interface is constructed to display the final control variables, including:

[0137] Use the front-end framework React.js to build a visual interface to visualize the final control variables;

[0138] Users who have passed real-name verification are allowed to access the information.

[0139] By building a visualization interface using React.js, users can view the dynamic changes of the final control variables (such as slurry flow and air flow rate) in real time. The front-end interface display not only improves data transparency, but also helps users understand the optimization decision-making process.

[0140] This embodiment also provides a desulfurization optimization control system based on a multimodal model, including:

[0141] The vector collection module is used to collect and preprocess multimodal data, calculate the horizontal gradient using the Sobel operator, construct a set of gas-liquid interface boundary points, place an alternating electric field induction probe to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal parts, calculate the average composite intensity of each area, and generate a control feature vector.

[0142] A hierarchical control module is used to calculate the fractal dimension of the control eigenvector using the box counting method, perform weighted processing on the fractal dimension and the control eigenvector, construct a fractal interference field, identify the maximum STFT amplitude as the main resonant frequency, construct a set of disturbance points, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point, multiply the disturbance intensity and the expansion trend to obtain the unstable diffusion level, calculate the total energy of each scale based on the wavelet coefficients, screen points with absolute values of wavelet coefficients greater than the average value, mark them as key propagation points, use nearest neighbor matching to match the key propagation points to the nearest disturbance point, and generate a control priority sequence;

[0143] The target visualization module is used to define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables.

[0144] This embodiment also provides a computer device, which is suitable for the desulfurization optimization control method based on a multimodal model, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the desulfurization optimization control method based on a multimodal model proposed in the above embodiment.

[0145] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0146] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the desulfurization optimization control method based on a multimodal model as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0147] In summary, the present invention constructs a composite intensity factor through the image gradient field and signal response ratio, constructs a gas-liquid interface boundary point set and a control feature vector, improves the robustness and information density of the features, combines the fractal interference field and short-time Fourier transform to identify the main resonance frequency, and uses wavelet transform and information entropy analysis to screen key propagation points, thereby improving the desulfurization efficiency and the stability of the reaction tower.

[0148] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A desulfurization optimization control method based on a multimodal model, characterized by: include, Collect and preprocess multimodal data, calculate horizontal gradients using the Sobel operator, construct a set of gas-liquid interface boundary points, place alternating electric field induction probes to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal sections, calculate the average composite intensity of each area, and generate a control feature vector. The fractal dimension of the control eigenvector is calculated using the box counting method. The fractal dimension and the control eigenvector are weighted to construct a fractal interference field. The maximum STFT amplitude is identified as the main resonant frequency. A set of disturbance points is constructed. The covariance matrix of the disturbance point set is calculated. The maximum eigenvalue is extracted as the main direction expansion trend of the disturbance point. The disturbance intensity and the expansion trend are multiplied to obtain the unstable diffusion level. Based on the wavelet coefficients, the total energy of each scale is calculated. Points with absolute values of wavelet coefficients greater than the average value are screened and marked as key propagation points. The key propagation points are matched with the nearest disturbance point using nearest neighbor matching to generate a control priority sequence. Define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables.

2. The desulfurization optimization control method based on a multimodal model according to claim 1, characterized in that: The calculation of the average composite intensity of each region and the generation of the control feature vector include: The slurry foam image is converted into a grayscale image using the weighted average method, the grayscale potential energy is calculated using logarithmic transformation, the horizontal gradient is calculated using the Sobel operator, the gradient threshold is set, and the pixel points that meet the gradient threshold are extracted as the gas-liquid interface boundary using the gradient threshold method to construct the gas-liquid interface boundary point set; Extract the boundary points of the gas-liquid interface boundary point set, place an alternating electric field induction probe to emit weak electromagnetic waves, collect the reflected signal amplitude, calculate the signal response ratio, and use low-pass filtering to remove high-frequency noise in the signal response ratio to obtain the filtered signal response ratio; Point-by-point product fusion is used to fuse the grayscale potential energy and the corresponding signal response ratio into a composite reaction intensity. The desulfurization tower profile is divided into three layers: upper, middle, and lower regions using the equal division method based on the pixel position. The average composite intensity of each region is calculated, and the average composite intensity is normalized to generate a control feature vector.

3. The desulfurization optimization control method based on a multimodal model according to claim 2, characterized in that: The above mentioned level of unstable diffusion is obtained by multiplying the disturbance intensity and the expansion trend, including: The fractal dimension of the control eigenvector is calculated using the box counting method. The fractal dimension and the control eigenvector are weighted to construct the fractal interference field. The STFT amplitude of the fractal interference field is extracted using the short-time Fourier transform. The maximum STFT amplitude is identified as the main resonance frequency. The detection threshold is set and the main resonance frequency is screened. The corresponding time point t ′ Mark as disturbance point, extract the control eigenvector in the time window corresponding to the disturbance point, construct the disturbance point set, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point; Based on the control eigenvector, the disturbance intensity is calculated, and the unstable diffusion level is obtained by multiplying the disturbance intensity and the expansion trend.

4. The desulfurization optimization control method based on a multimodal model according to claim 3, characterized in that: The method uses nearest neighbor matching to match the key propagation point with the nearest disturbance point to generate a control priority sequence, including: Calculate the information entropy of the unstable diffusion level, use the Daubechies-4 wavelet basis to calculate the wavelet coefficients of the information entropy, calculate the total energy of each scale based on the wavelet coefficients, calculate the energy proportion of each scale, select the maximum energy scale, and select the points where the absolute value of the wavelet coefficient is greater than the average value and mark them as key propagation points; The key propagation point is matched with the nearest disturbance point using nearest neighbor matching to obtain matching pairs. The index of the corresponding disturbance point is searched for each matching pair, the corresponding unstable diffusion level is extracted, and a temporary sequence is generated. The absolute values of the corresponding wavelet coefficients are extracted from the temporary sequence and arranged in descending order to generate a control priority sequence.

5. The desulfurization optimization control method based on a multimodal model according to claim 4, characterized in that: The definition of the optimization objective function to obtain the final control variables includes: Based on the control priority sequence, the control level of unstable diffusion is defined, the state vector is constructed, and a linear regression model is constructed to predict SO2 concentration and pH value; Define the optimization objective function and use constrained optimization to solve the problem and obtain the final control variables.

6. The desulfurization optimization control method based on a multimodal model according to claim 5, characterized in that: The constructed visual interface displays the final control variables, including: Use the front-end framework React.js to build a visual interface to visualize the final control variables; Users who have passed real-name verification are allowed to access the information.

7. The desulfurization optimization control method based on a multimodal model according to claim 1, characterized in that: The collecting of multimodal data and preprocessing thereof include: The multimodal data includes slurry foam images, SO2 concentration and pH value data; An industrial camera was used to collect slurry foam images at the desulfurization tower observation window, an electrochemical sensor was used to collect SO2 concentration at the desulfurization tower outlet flue, and a pH meter was used to collect pH value from the slurry pool. The collected multimodal data were denoised and normalized.

8. A desulfurization optimization control system based on a multimodal model, based on the desulfurization optimization control method based on a multimodal model according to any one of claims 1 to 7, characterized in that: include, The vector collection module is used to collect and preprocess multimodal data, calculate the horizontal gradient using the Sobel operator, construct a set of gas-liquid interface boundary points, place an alternating electric field induction probe to emit weak electromagnetic waves, calculate the signal response ratio, divide the desulfurization tower section into equal parts, calculate the average composite intensity of each area, and generate a control feature vector. A hierarchical control module is used to calculate the fractal dimension of the control eigenvector using the box counting method, perform weighted processing on the fractal dimension and the control eigenvector, construct a fractal interference field, identify the maximum STFT amplitude as the main resonant frequency, construct a set of disturbance points, calculate the covariance matrix of the disturbance point set, extract the maximum eigenvalue as the main direction expansion trend of the disturbance point, multiply the disturbance intensity and the expansion trend to obtain the unstable diffusion level, calculate the total energy of each scale based on the wavelet coefficients, screen points with absolute values of wavelet coefficients greater than the average value, mark them as key propagation points, use nearest neighbor matching to match the key propagation points to the nearest disturbance point, and generate a control priority sequence; The target visualization module is used to define the control level of unstable diffusion, construct the state vector, define the optimization objective function, use constrained optimization to solve, obtain the final control variables, and build a visualization interface to display the final control variables.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the desulfurization optimization control method based on the multimodal model are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the desulfurization optimization control method based on a multimodal model according to any one of claims 1 to 7 are implemented.

Images