Electroplating wastewater treatment agent reduction prediction method and system based on multi-modal data

Abstract

The application relates to the technical field of electroplating wastewater treatment, and discloses an electroplating wastewater treatment agent reduction prediction method and system based on multi-modal data. The method comprises the following steps: collecting wastewater pH and an original image, extracting color distribution and turbid particle density, and completing data cleaning; the original image is analyzed layer by layer to obtain a color gradient matrix and a particle space characteristic matrix, and a state description vector is formed by fusing the pH; if the density value of the central region exceeds a threshold value, the corresponding weight is increased, a strengthened description vector is obtained, and the strengthened description vector is decomposed into a feature subset; an initial dosage is generated by matching the feature importance ranking with historical samples; the optimized dosage is obtained by combining data fluctuation and color gradient to complete correction; a treatment process is simulated, data is iteratively updated according to simulation deviation, and a refined state description is generated; preset prediction parameters are dynamically adjusted, the dosage interval is further optimized, and a stable dosing scheme is determined. The method can improve the treatment effect of the agent.

Description

Technical Field

[0001] This invention relates to the field of electroplating wastewater treatment technology, and in particular to a method and system for predicting the reduction of electroplating wastewater treatment agents based on multimodal data. Background Technology

[0002] Currently, as a key supporting field for industrial manufacturing, the electroplating industry generates wastewater containing a large amount of heavy metal ions, cyanides, acidic and alkaline substances, and organic additives. If not effectively treated, this wastewater will pose a serious threat to the water environment and ecological security. Therefore, the efficient treatment of electroplating wastewater has become an important research direction for industrial environmental protection and green production.

[0003] In existing technologies, electroplating wastewater treatment primarily relies on chemical precipitation as the core process. This involves adding agents such as acid-base regulators, coagulants, and heavy metal scavengers to remove pollutants. The precision of agent dosing directly determines the treatment effect, operating costs, and sludge production. In the context of industrial intelligence and the increasing prevalence of big data processing technology, current electroplating wastewater agent dosing control methods still largely depend on a combination of online pH monitoring and manual experience. Some automated systems use only single physicochemical indicators as the basis for dosing, lacking effective collection, quantitative analysis, and big data processing integration of visual characteristics such as wastewater color, turbidity, and suspended solids distribution. This makes it difficult to comprehensively characterize the complex dynamic state of the wastewater. Such single-dimensional monitoring and control models cannot fully utilize multi-source data and big data processing capabilities to achieve deep information integration. This results in the inability to dynamically adapt agent dosage to changes in water quality, commonly leading to overdosing and resource waste, or underdosing and substandard effluent, severely restricting the stability of the electroplating wastewater treatment process.

[0004] Therefore, existing technologies make it difficult to control the amount of pesticide used, resulting in poor treatment effects. Summary of the Invention

[0005] This invention provides a method and system for predicting the reduction of chemicals used in electroplating wastewater treatment based on multimodal data, so as to accurately control the amount of chemicals used and improve the treatment effect.

[0006] In a first aspect, to address the aforementioned technical problems, this invention provides a method for predicting the reduction of electroplating wastewater treatment chemicals based on multimodal data, comprising:

[0007] The pH values ​​and raw image data of electroplating wastewater samples were collected and preprocessed to obtain initial monitoring data;

[0008] A single-channel grayscale image is extracted from the initial monitoring data, and a layered analysis is performed on the single-channel grayscale image to obtain a color gradient matrix. Turbid particles are located based on the color gradient matrix to obtain a particle spatial characteristic matrix. The color gradient matrix, the particle spatial characteristic matrix, and the pH value are fused to obtain a state description vector.

[0009] The density value of the central region is extracted based on the state description vector. If the density value of the central region exceeds a preset density threshold, the density value of the central region is weighted to obtain a strengthening weight coefficient. The state description vector is adjusted based on the strengthening weight coefficient to obtain a strengthening description vector. The strengthening description vector is decomposed to obtain a feature subset.

[0010] Calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial dosage.

[0011] The data fluctuation range is extracted from the initial delivery volume, and the initial delivery volume is corrected by combining the color gradient matrix with the data fluctuation range to obtain the optimized delivery volume.

[0012] Based on the optimized dosage, the wastewater treatment process is simulated and calculated to obtain a description of the refining state.

[0013] The optimized dosage is corrected based on the description of the refining state to obtain a stable dosage of the treatment agent.

[0014] Secondly, the present invention provides a prediction system for reducing the amount of chemicals used in electroplating wastewater treatment based on multimodal data, comprising:

[0015] The initial monitoring data acquisition and cleaning module is used to collect the pH value and raw image data of electroplating wastewater samples and perform preprocessing to obtain initial monitoring data;

[0016] The state description vector generation module is used to extract a single-channel grayscale image from the initial monitoring data, perform layered analysis on the single-channel grayscale image to obtain a color gradient matrix, locate turbid particles based on the color gradient matrix to obtain a particle spatial characteristic matrix, and fuse the color gradient matrix, the particle spatial characteristic matrix and the pH value to obtain a state description vector.

[0017] The enhanced description vector generation and decomposition module is used to extract the central region density value based on the state description vector. If the central region density value exceeds a preset density threshold, the central region density value is weighted to obtain an enhanced weight coefficient. The state description vector is adjusted according to the enhanced weight coefficient to obtain an enhanced description vector. The enhanced description vector is decomposed to obtain a feature subset.

[0018] The initial deployment quantity generation module is used to calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial deployment quantity.

[0019] The optimized delivery volume correction module is used to extract the data fluctuation range from the initial delivery volume, and combine the color gradient matrix with the data fluctuation range to correct the initial delivery volume to obtain the optimized delivery volume;

[0020] The simulation and data iteration update module is used to simulate and calculate the wastewater treatment process based on the optimized dosage to obtain a description of the refining state.

[0021] The stable dosage scheme determination module is used to correct the optimized dosage based on the refining state description to obtain a stable dosage of the treatment agent.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention collects acidity and alkalinity data and original image data, extracts multi-dimensional features such as color gradient and particle spatial distribution, and combines the central region density value threshold judgment and weight enhancement to achieve accurate characterization of wastewater pollution status, solving the problem that existing technologies rely on only a single indicator and cannot adapt to local pollution anomalies.

[0024] (2) The present invention generates the initial dosage by ranking the importance of features and accurately matching historical samples, combined with data fluctuations and dynamic correction of color gradients, and then continuously optimizes through simulation and iterative updates, so as to avoid the waste of reagents and secondary pollution caused by excessive addition of existing technologies, or the substandard treatment caused by insufficient addition.

[0025] (3) This invention forms a fully automated closed loop from data acquisition, feature extraction, interval generation to simulation verification, parameter iteration and scheme determination, without much human intervention. At the same time, it fully records the parameters and adjustment basis of each link, solving the problems of excessive human intervention, large error and untraceable process in the existing technology. Attached Figure Description

[0026] Figure 1This is a schematic diagram of the electroplating wastewater treatment agent reduction prediction method based on multimodal data provided in the first embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the electroplating wastewater treatment reagent reduction prediction system based on multimodal data provided in the second embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Reference Figure 1 The first embodiment of the present invention provides a method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data, comprising the following steps:

[0030] S11, Collect the pH value and raw image data of electroplating wastewater samples and perform preprocessing to obtain initial monitoring data;

[0031] S12, extract a single-channel grayscale image from the initial monitoring data, perform layered analysis on the single-channel grayscale image to obtain a color gradient matrix, locate turbid particles according to the color gradient matrix to obtain a particle spatial characteristic matrix, and fuse the color gradient matrix, the particle spatial characteristic matrix and the pH value to obtain a state description vector.

[0032] S13, extract the density value of the central region according to the state description vector. If the density value of the central region exceeds the preset density threshold, increase the weight of the density value of the central region to obtain the enhancement weight coefficient. Adjust the state description vector according to the enhancement weight coefficient to obtain the enhancement description vector. Decompose the enhancement description vector to obtain the feature subset.

[0033] S14, calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial dosage;

[0034] S15, extract the data fluctuation range from the initial delivery volume, and combine the color gradient matrix with the data fluctuation range to correct the initial delivery volume to obtain the optimized delivery volume;

[0035] S16, Based on the optimized dosage, the wastewater treatment process is simulated and calculated to obtain a description of the refining state;

[0036] S17, The optimized dosage is corrected according to the description of the refining state to obtain a stable dosage of the treatment agent.

[0037] In step S11, the acidity / alkalinity values ​​and raw image data of the electroplating wastewater samples are collected and preprocessed to obtain initial monitoring data, including:

[0038] The pH value and original image data of the electroplating wastewater sample are obtained. The original image data is converted to color space to obtain color image data. Histogram statistics are performed on the color image data to obtain color distribution characteristics.

[0039] The color image data is converted to grayscale to obtain a single-channel grayscale image. The particle outline is extracted from the single-channel grayscale image based on the color distribution features to obtain the particle density value. If the particle density value is greater than the preset turbidity particle density judgment threshold, the particle density value is determined to be abnormal and an abnormality is marked to obtain wastewater treatment historical samples.

[0040] Data cleaning was performed on the pH value and the particle density value to obtain initial monitoring data.

[0041] A pH sensor was used to collect the acidity and alkalinity values ​​of electroplating wastewater samples in real time. Simultaneously, a high-resolution image acquisition device was used to acquire raw image data of the wastewater samples, with the acquisition frequency set to once every 5 minutes. The acquired raw image data was uniformly scaled to a standard size of 1000×1000 pixels for easy subsequent image region segmentation. The precise acquisition time was automatically recorded as a timestamp for each acquisition, ensuring that each set of acidity / alkalinity data and image data has a unique corresponding time identifier.

[0042] The HSV color space conversion algorithm is used to convert the original RGB format image to HSV format as the target format, obtaining color image data. Histogram statistics are performed on the hue (H) component of the color image data, dividing the hue range from 0 to 179 into 10 consecutive intervals. All pixels in the image are traversed, and the number of pixels falling within each hue sub-interval is counted. The percentage of pixels in each sub-interval is then calculated by dividing the number of pixels in each sub-interval by the total number of pixels in the image. The color distribution features are represented by the pixel percentage of each hue interval, stored in a one-dimensional array of length 10, where each element corresponds to the pixel percentage of a specific hue interval. This one-dimensional array is bound to the current acquisition timestamp, allowing for precise indexing and retrieval of pixel changes based on the color distribution features through the timestamp.

[0043] The image after HSV color space conversion is converted to grayscale, transforming the multi-channel image into a single-channel grayscale image. The Canny edge detection algorithm is then used to extract particle contours. A low threshold of 50 and a high threshold of 150 are initially set, and continuous closed contours of the particles are detected and preserved to accurately locate turbid particles. Effective particles are then selected based on color distribution characteristics. The dominant color range with the highest proportion in the color distribution is used as the water background color. False particle areas where pixels within the contour region fall into the dominant color range of the water are removed, retaining only the contours of real particles that are significantly different from the water background color.

[0044] The image is divided into fixed regions. A single wastewater image is divided into rectangular regions of the same size using a uniform grid. The standard unit region is set to 100×100 pixels, corresponding to an actual physical area of ​​1 cm². The total number of turbid particles detected in each standard unit region is counted. The arithmetic mean of the particle density values ​​of all divided regions is calculated. This average value is used as the turbid particle density index of the current image, with units of particles / cm². This turbid particle density index is then linked to the same timestamp.

[0045] The threshold for determining turbidity particle density is set based on stable operating conditions in the electroplating industry and normal data from the past 30 days, specifically 80 particles / cm². The calculated turbidity particle density value is compared with this threshold. When the density value exceeds the threshold, the particle density value at that moment is considered abnormal. Based on the timestamp, the abnormal particle density value is associated with the wastewater treatment history records for the same time, and an anomaly marker is added to the relevant records, forming a wastewater treatment history sample containing the anomaly marker, which is stored in a preset database. The preset database stores the pH value, particle density value, and reagent dosage of each timestamp for the wastewater treatment history sample.

[0046] The system retrieves pH and particle density values ​​from the past 30 days from a pre-defined database and uses a box plot (IQR) method to identify outliers. Taking pH anomaly identification as an example, the median and interquartile range (IQR) of the historical data are first calculated. When the median pH value is 5.0 and the IQR is 1.2, the anomaly range is calculated according to the box plot anomaly judgment rules. Data values ​​less than the median minus 1.5 times the IQR or greater than the median plus 1.5 times the IQR are considered outliers, corresponding to values ​​less than 2.2 or greater than 7.8. The real-time collected pH values ​​are then substituted into this range for judgment. For example, if the collected pH value is 1.8, this value is less than 2.2 and falls within the anomaly range, thus being identified as outlier data and removed from the dataset. Particle density values ​​are identified and removed using the same method and correlated with the corresponding single-channel grayscale image to obtain initial monitoring data.

[0047] In step S12, a single-channel grayscale image is extracted from the initial monitoring data, and the single-channel grayscale image is subjected to layered analysis to obtain a color gradient matrix. Turbid particles are located based on the color gradient matrix to obtain a particle spatial characteristic matrix. The color gradient matrix, the particle spatial characteristic matrix, and the pH value are then fused to obtain a state description vector, including:

[0048] A single-channel grayscale image is extracted from the initial monitoring data, and the single-channel grayscale image is subjected to layered analysis to obtain the depth gradient value;

[0049] The depth gradient values ​​are filled into matrix elements to obtain the color gradient matrix. The average value of all elements in the color gradient matrix is ​​obtained.

[0050] The color gradient matrix is ​​used to identify turbid particles and obtain local particle density. The local particle density is then filled into matrix elements to obtain a particle spatial characteristic matrix. The density values ​​of the central region and the edge region are extracted from the particle spatial characteristic matrix.

[0051] The pH value, the average color gradient value, the density value of the central region, and the density value of the edge region are combined to generate a state description vector.

[0052] Extract the single-channel grayscale image corresponding to the current timestamp from the initial monitoring data. Divide the single-channel grayscale image into uniform blocks, using a 10×10 fixed grid both horizontally and vertically. Each block is 100×100 pixels, corresponding to an actual physical area of ​​1 cm², consistent with the area division in step S11, ensuring uniform area size. Simultaneously, perform depth layering based on pixel brightness, with brightness values ​​ranging from 0 to 100. 0–35 is defined as deep, 36–65 as mid-level, and 66–100 as shallow. Subtract the average brightness of the deep layers from the average brightness of the shallow layers within the same block, and take the absolute value to obtain the depth gradient value for that block. Arrange the depth gradient values ​​according to the horizontal and vertical indices of the blocks to construct a 10×10 two-dimensional color gradient matrix. Rows correspond to horizontal block indices, and columns correspond to vertical block indices. Each element in the matrix represents only the depth gradient of its corresponding block.

[0053] Based on the color gradient matrix, turbid particles are located. For each block in the color gradient matrix, the average gradient of that block and its eight neighboring blocks is calculated as the local average gradient. If the gradient value of a block is higher than the local average gradient, the block is determined to be the particle center, and the coordinates of the gradient maximum point are used as the particle center coordinates. If a particle spans multiple blocks, the block with the largest gradient value is taken as its region. Each particle belongs to only one region to avoid duplicate counting. This step uses the 10×10 uniform grid already divided in S11 without re-dividing the regions. The number of particles in each block is counted to obtain the local particle density of each block. The unit of local particle density is consistent with S11, which is particles / cm². A 10×10 particle spatial characteristic matrix is ​​constructed according to the row and column positions of the blocks, and its dimensions directly correspond to the number of regions.

[0054] The pH value corresponding to the current timestamp in the initial monitoring data is read. The average color gradient value is obtained by averaging all elements of the color gradient matrix. The density values ​​of the central and edge regions are extracted from the particle spatial characteristic matrix. The density value of the central region is obtained by averaging the values ​​of the four central blocks of the matrix, and the density value of the edge region is obtained by averaging the values ​​of all outer blocks of the matrix. The pH value, average color gradient value, central region density value, and edge region density value are then concatenated in sequence into a four-dimensional vector to obtain the state description vector.

[0055] In step S13, the density value of the central region is extracted based on the state description vector. If the density value of the central region exceeds a preset density threshold, the density value of the central region is weighted to obtain a strengthening weight coefficient. The state description vector is adjusted according to the strengthening weight coefficient to obtain a strengthening description vector. The strengthening description vector is then decomposed to obtain a feature subset, including:

[0056] The density value of the central region is extracted based on the state description vector. If the density value of the central region is greater than the preset density threshold, the preset weight coefficient is adjusted to obtain the enhancement weight coefficient.

[0057] The state description vector is adjusted according to the enhancement weight coefficient to obtain the enhancement description vector;

[0058] The enhanced description vector is subjected to feature decomposition to obtain a feature subset.

[0059] The density value of the central region is extracted from the state description vector and compared with a preset density threshold. If the density value of the central region is greater than the preset density threshold, it is determined that local pollution exceeds the standard; if the density value of the central region is less than or equal to the preset density threshold, it is determined that the distribution is normal. The preset density threshold is set to 120 particles / cm², based on the fact that during normal operation of electroplating wastewater, the density value of the central region does not exceed 120 particles / cm². Exceeding this threshold indicates that the wastewater is severely polluted locally, and the influence weight of this feature needs to be strengthened to match the subsequent process adjustment requirements.

[0060] When the density value in the central region exceeds a preset threshold, the spatial distribution characteristics of the particles need to be weighted more heavily. This is to strengthen the influence of this abnormal feature in subsequent reagent prediction and ensure that process adjustments can accurately target local contamination problems. The density values ​​in the central and peripheral regions directly reflect the spatial distribution characteristics of the particles, and the excessive density value in the central region is the core cause of this anomaly. Therefore, the weights of both need to be increased simultaneously to fully reflect the abnormal state of particle aggregation. The deviation rate between the density value in the central region and the preset density threshold is calculated. The deviation rate is equal to the central region density value minus the preset density threshold, divided by the preset density threshold. The weight coefficient of the particle spatial distribution characteristics is set to the original weight coefficient divided by the deviation rate. The weight coefficients of the physicochemical characteristics and color gradient characteristics are proportionally reduced to their original proportions, so that the sum of the three equals 1. Simultaneously, the original weight coefficient of the particle spatial distribution characteristics is 0.5, corresponding to the sum of the weights of the central and peripheral region densities.

[0061] After applying the enhancement weight coefficients to the original feature values, the enhancement description vector is obtained using the S12 state description vector construction method. The enhancement description vector is then decomposed using SVD, where each element corresponds to the pH value, average color gradient value, central region density value, and edge region density value, respectively. The decomposition process follows the standard SVD algorithm procedure: first, the column matrix of the four-dimensional vector is constructed, converting the enhancement description vector into a 4×1 column matrix. Multiplying this column matrix by its transpose yields a 4×4 covariance matrix. Standard SVD decomposition is then performed on this covariance matrix to solve for the singular values, left singular vector, and right singular vector. Based on the contribution rate of the singular values, three core singular vectors are selected, with a contribution rate threshold of 85%. Each singular vector is treated as an independent feature subset, achieving effective decomposition of the original four-dimensional vector. The three feature subsets obtained after decomposition are the physicochemical feature subset, the color gradient feature subset, and the particle spatial distribution feature subset.

[0062] The singular value contribution rate threshold is set to 85%, a common standard for industrial feature engineering and data dimensionality reduction. After performing SVD decomposition on the 4×4 covariance matrix, the singular values ​​are sorted from largest to smallest. Regardless of whether the first one or two singular values ​​have reached the 85% contribution rate, the first three singular vectors are forcibly selected as core feature vectors to ensure output structure stability. According to the preset mapping rules, the first singular vector corresponds to the physicochemical feature subset, the second singular vector corresponds to the color gradient feature subset, and the third singular vector corresponds to the particle spatial distribution feature subset. These three together constitute a complete feature subset.

[0063] In step S14, calculating the similarity between the feature subset and preset wastewater treatment historical samples, filtering out the preset wastewater treatment historical samples whose similarity exceeds a preset similarity threshold, and generating an initial deployment amount includes:

[0064] The feature subset is standardized to obtain a standard feature subset, and the similarity between the standard feature subset and the preset wastewater treatment historical samples is calculated. The wastewater treatment historical samples with similarity greater than the preset similarity threshold are selected to obtain the optimal matching sample.

[0065] Extract the dosage of the drug from the optimal matching sample from the preset database, and calculate the initial dosage based on the dosage of the drug.

[0066] The current wastewater feature subset is standardized according to the standardization rules of historical wastewater treatment sample data in the preset database. Min-max standardization is used to map the feature values ​​to the [0,1] interval. Then, the KNN algorithm is called with K set to 5. After multiple verifications, K=5 can balance matching efficiency and accuracy and avoid single-sample bias. The Euclidean distance between the current feature subset and the feature subsets of all historical samples in the preset database is calculated, and 5 historical samples with similarity greater than or equal to the preset similarity threshold are selected.

[0067] The preset similarity threshold is set to 90%. Verified with extensive historical data, samples with a similarity of 90% or higher show a deviation of less than 10% in their physicochemical characteristics, particle spatial distribution characteristics, and color gradient characteristics from the current wastewater characteristics. This ensures that the corresponding reagent dosage and treatment effect data have direct reference value, effectively avoiding reagent dosage prediction bias caused by excessive sample differences. The 90% threshold conforms to the conventional standard for matching similar operating conditions in industrial wastewater treatment. It can screen samples that highly match the current wastewater conditions while avoiding insufficient matching samples and subsequent interval calculations due to an excessively high threshold (e.g., 95%), while also reserving reasonable space for subsequent threshold reduction. If there are fewer than 5 samples with a similarity of 90% or higher, the similarity threshold is increased to 85% to ensure the quantity and reliability of matching samples. The 85% reduction threshold has also been verified, ensuring sufficient sample size while maintaining sample correlation, avoiding statistical bias caused by insufficient samples.

[0068] Extract the dosage data of the five best-matched samples, calculate the mean and standard deviation of the dosage, and combine them with a 95% confidence level to generate the upper and lower limit ranges of the dosage requirement. The upper limit of the dosage is equal to the sum of the mean of the dosage data and 1.96 times the standard deviation of the dosage data, and the lower limit of the dosage is equal to the difference between the mean of the dosage data and 1.96 times the standard deviation of the dosage data. The mean of the dosage data is the initial dosage.

[0069] In step S15, extracting the data fluctuation amplitude from the initial delivery volume, and combining the color gradient matrix with the data fluctuation amplitude to correct the initial delivery volume to obtain an optimized delivery volume includes:

[0070] Calculate the dispersion of the initial deployment amount to obtain the fluctuation range;

[0071] Calculate the average value of all elements in the color gradient matrix to obtain the average gradient value. If the average gradient value is not within a preset gradient range, the initial delivery amount is adjusted based on the data fluctuation amplitude to obtain the optimized delivery amount. If the average gradient value is within the preset gradient range, there is no need to adjust the initial delivery amount.

[0072] Based on the initial delivery data of 5 optimal matching samples, the dispersion of the initial delivery volume, i.e., the standard deviation, is calculated, then divided by the average of the initial delivery volume, and converted into a percentage to obtain the fluctuation range.

[0073] The color gradient matrix obtained in the previous steps is used to calculate the average color gradient of all regions and determine whether it exceeds the preset gradient range [20, 60]. Simultaneously, the differences in color depth gradient between the edge and central regions are analyzed to clarify the spatial distribution characteristics of gradient changes. For example, if the color gradient value in the central region is higher than that in the edge region, and the average gradient value exceeds the preset gradient range, it indicates that the concentration of suspended particulate pollutants in the central region is higher, and the initial dosage for that region needs to be increased accordingly. For instance, if the current real-time monitored color gradient value of suspended particulate matter is 68, exceeding the preset range [20, 60], and the color gradient value of the central region (75) is significantly higher than that of the edge region (52), it is determined that the color gradient change of suspended particulate matter is abnormal, and the initial dosage needs to be corrected. If the real-time gradient value is 35, within the preset range, it is determined that the gradient change is normal, and no correction of the reagent range is needed; the range output by S14 can be directly used.

[0074] The preset gradient range [20, 60] is set based on the statistical regularity of similar historical operating conditions and process operation experience. This range is based on the average color gradient of similar electroplating wastewater under stable and compliant treatment conditions in the historical database. By statistically analyzing the gradient data distribution range of a large number of normally operating samples, the normal fluctuation range at a 95% confidence level is taken as the upper and lower limits. At the same time, combined with the correspondence between suspended solids concentration and color gradient in the actual treatment process, the gradient value of 20 is set as the lower limit, corresponding to the normal operating condition with low suspended solids concentration, and the gradient value of 60 is set as the upper limit, corresponding to the critical operating condition where the suspended solids concentration is close to the treatment limit. Exceeding this range indicates that the suspended solids particles are abnormally aggregated and the pollution load deviates from the normal level.

[0075] Based on the data fluctuation analysis and color gradient analysis results, corrections are made in two scenarios. In the first scenario, if the gradient value is higher than the preset upper limit of 60, it indicates that the concentration of suspended particulate pollutants is too high. Considering the adjustment range for the exceedance, if the exceedance is ≤10%, it corresponds to a slight exceedance. In this case, the pollutant concentration is high but still within the adjustable range, so a smaller correction range is adopted, increasing the upper limit by 3%-5% and the lower limit by 2%-3%. The increase refers to multiplying by 1 and adding a percentage. If the exceedance is >10%, it corresponds to a severe exceedance, requiring a larger adjustment to the initial dosage. The upper limit is increased by 6%-8% and the lower limit by 4%-6%. Simultaneously, considering the data fluctuation of the initial dosage, if the fluctuation is >5%, it means that the initial dosage dispersion between historical matching samples is large, indicating poor stability between the current water quality state and historical samples. Therefore, the upper limit is further increased by 1%-2% to ensure adaptation to abnormal pollutant concentrations. In the second scenario, if the color gradient change falls within a preset gradient range and the data fluctuation is ≤5% with good consistency among historical samples, the original interval can be directly used, and the drug demand interval output by S14 will be adopted directly. If the data fluctuation is >5%, there is some uncertainty, so the upper and lower limits are fine-tuned by ±1%-2% to ensure that the interval adapts to data fluctuations and avoids dosage deviation. Based on the data fluctuation range and the color gradient matrix, the drug demand interval is corrected to obtain the corrected interval, and the average value of this interval is taken as the optimized dosage.

[0076] In step S16, the wastewater treatment process is simulated and calculated based on the optimized dosage to obtain a refining state description, including:

[0077] Based on the optimized delivery amount, a virtual simulation is performed to generate a sequence of simulation results, and the pH value, the depth gradient value, and the local particle density are collected simultaneously to obtain real-time monitoring data;

[0078] The deviation between the simulation result sequence and the real-time monitoring data is calculated. If the deviation is greater than a preset deviation threshold, the real-time monitoring data is iteratively updated to generate a refined state description.

[0079] The master equation is established based on the physicochemical mechanisms of the wastewater treatment process. Uncertain parameters in the model are identified and corrected using historical wastewater treatment samples. Based on optimized dosage, simulation parameters are set, using the interval average as the baseline dosage rate. The total dosage rate is split into 60% for the central region and 40% for the peripheral region. Wastewater treatment process parameters, particle spatial characteristic matrix, and color gradient data are input to construct a 1:1 virtual simulation model of the actual wastewater treatment system. The wastewater treatment process parameters mainly include the effective volume of the reaction tank, hydraulic retention time, stirring intensity, water flow velocity, temperature control range, sedimentation time in the sedimentation zone, and reagent diffusion coefficient. These parameters are all taken from actual on-site operating conditions to construct a 1:1 virtual simulation environment that matches the real system, ensuring that the simulated dynamics are consistent with the actual process. During the simulation, the diffusion process of the agent after its release, the reaction process with suspended particles, and the sedimentation process of pollutants are simulated in real time. Simulation data is recorded every 10 minutes to obtain a sequence of simulation results, including pH value, depth gradient value of different areas in the reaction tank, and local particle density. These results correspond one-to-one with the monitoring indicators of the real-time monitoring data to ensure the comparability of the simulation results.

[0080] The diffusion of the reagent in the reaction tank is described by a convection-diffusion equation. The initial reagent concentration is 0, meaning there is no reagent in the tank at the initial moment. The dosing point is determined by the optimized dosage and location. The reagent concentration gradient along the outflow direction is zero, convection is dominant, and the reagent does not penetrate the tank wall. The reaction between the reagent and suspended particles is simplified to a second-order reaction kinetic, applicable to coagulation or sedimentation reactions. It is solved independently on each spatial grid cell, and the reagent concentration consumed in the reaction is fed back into the diffusion equation. The pollutant sedimentation process adopts a modified Stokes sedimentation model, considering the flocculation effect between particles. The sedimentation process causes particles to migrate from the upper water layer to the bottom. Correspondingly, the amount of sediment in each spatial grid cell decreases over time, and this decrease enters the bottom sediment layer.

[0081] Different regions in the reaction tank correspond to image regions defined in step S11, with each region corresponding to a depth gradient value and a local particle density. Real-time monitoring data also includes pH values, depth gradient values ​​for different regions, and local particle densities. pH values ​​are acquired using a pH sensor, while the depth gradient values ​​and local particle densities for different regions are obtained using the same method as in step S12.

[0082] The reaction tank is divided into grid cells according to the image division method in step S11. Each cell corresponds to an actual physical region and is one-to-one with the image block. Based on the particle spatial characteristic matrix and color gradient matrix in step S12, an initial depth gradient value is assigned to each grid cell. An initial pH value is assigned based on real-time pH sensor data. Based on the dosing point location, a reagent concentration boundary condition is applied to the corresponding grid cell. The tank inlet and outlet and the wall are set according to the aforementioned rules. The diffusion equation, reaction equation and sedimentation equation are solved sequentially in each time step using the finite difference method or the finite volume method. The depth gradient value and local particle density of each grid cell are updated. Every set time step, the above variables of each grid cell are saved to form a simulation result sequence.

[0083] Three core indicators—pH value, depth gradient value, and local particle density—corresponding to the simulation result sequence and real-time monitoring data were selected, and independent preset deviation thresholds were set for each indicator: the pH value preset deviation threshold was 5%, which was set based on the stable control requirements of pH value of electroplating wastewater and the sensor measurement error range to ensure the accuracy of pH value control; the depth gradient value preset deviation threshold was 8%, which was set based on the correspondence between wastewater suspended solids concentration and color gradient and the allowable error range of image feature extraction; and the local particle density preset deviation threshold was 10%, which was set based on the particle detection statistical error and the allowable fluctuation range of flocculation reaction area distribution.

[0084] The simulation results sequence selects indicators corresponding to the real-time monitoring data, including pH value, depth gradient value, and local particle density. The deviation value is calculated by subtracting the measured value from the simulated value, dividing by the measured value, and taking the absolute value. The average deviation value of the indicator is then calculated by averaging the deviation values ​​in different regions, and used as the deviation value for this simulation. The calculated average deviation value of each indicator is compared with the corresponding preset deviation threshold. Eight cases are identified and iteratively processed based on the combination of indicators exceeding the limit. According to the type of indicator combination exceeding the limit, corresponding data supplementation and iterative update operations are performed. All operations follow a unified time window stitching rule, supplementing the real-time monitoring data within the previous 30 minutes at a sampling frequency of 5 minutes, for a total of 6 groups, which are then longitudinally stitched with the original monitoring data. The original monitoring data includes the pH value collected in step S11, the depth gradient value obtained in step S12, and the local particle density.

[0085] If all three indicators are within limits, the simulation is considered successful, the optimized deployment amount is reasonable, no iteration is needed, and it can be directly used to determine the final deployment amount plan. If only the pH value deviation exceeds the limit, the latest 30 minutes of real-time pH value data is added, for a total of 6 time points. The added pH sequence is concatenated with the original pH value, the average pH value is recalculated, and the pH component in the state description vector is updated. The dimension of the state description vector does not change; only the corresponding component is replaced. Then, SVD decomposition is re-executed, increasing the weight of the singular vector corresponding to the pH feature by 1.5 times, while correspondingly reducing the weights of other features to keep the normalized sum at 1, generating a refined state description: "The pH value deviation exceeds the limit, caused by the low weight of the pH value feature; the proportion of the pH value in feature fusion needs to be increased."

[0086] If only the depth gradient value deviation exceeds the limit, supplement with the latest 30 minutes of color gradient data, with each time point containing a 10×10 grid of depth gradient matrix. Concatenate the supplemented gradient data with the original color gradient matrix along the time axis, recalculate the average depth gradient value for each grid cell to obtain an updated color gradient matrix, and update the average color gradient value in the state description vector. Then, re-execute SVD decomposition, increasing the weight of the singular vector corresponding to the color gradient feature by 1.5 times, and correspondingly decreasing the weights of other features, generating a refined state description: "Color gradient index deviation exceeds the limit, caused by insufficient gradient extraction accuracy; color gradient feature weights and extraction accuracy need to be increased."

[0087] If only the particle spatial distribution density value deviates beyond the limit, supplement with the most recent 30 minutes of real-time data on particle density in the central and edge regions, totaling 6 time points, each containing density values ​​of a 10×10 grid. Concatenate the new data with the original particle spatial characteristic matrix along the time axis, recalculate the time-averaged density value of each grid cell to obtain an updated particle spatial characteristic matrix, recalculate the density values ​​of the central and edge regions, and update the corresponding components in the state description vector. Then, re-execute SVD decomposition, increasing the weight of the singular vector corresponding to the density value of the central region by 1.5 times, and correspondingly reducing the weights of other features, generating a refined state description: "Particle spatial density index deviation exceeds the limit, caused by low particle spatial distribution feature weights; particle density weights and region identification accuracy need to be increased."

[0088] If both pH and color gradient values ​​exceed limits, then simultaneously supplement the latest 30-minute pH values ​​(6 time points) and color gradient data (a 10×10 gradient matrix for each time point). Concatenate the pH value sequence with the original data and recalculate the average pH value; concatenate the color gradient matrix along the time dimension and recalculate the average depth gradient value for each grid cell to obtain an updated color gradient matrix. Update the pH and average color gradient values ​​in the state description vector. Re-execute SVD decomposition, simultaneously increasing the weights of both pH and color gradient features by 1.3 times (the increase for a single feature is slightly lower than that for a single feature when both exceed limits, to avoid over-enhancement), then normalize to generate a refined state description: "Both pH and color gradient values ​​exceed limits due to insufficient weight allocation and low extraction accuracy of the two types of features; both types of features need to be strengthened simultaneously."

[0089] If both pH and particle spatial density exceed limits, real-time data for both pH and particle density are simultaneously supplemented, with 6 sets of data for each time point. Outliers are removed from the supplemented data using the box plot method described in step S11, and then concatenated with the original data. The pH and density values ​​for the center and edge regions in the state description vector are updated. Weight enhancement is re-executed, with the pH feature weight increased by 1.3 times and the particle spatial distribution feature weight increased by 1.5 times, because exceeding particle density limits more directly reflects local pollution. After normalization, SVD decomposition is performed again to generate a refined state description: "Both pH and particle spatial density exceed limits due to insufficient weight allocation of the two types of features; simultaneous enhancement of physicochemical and spatial distribution features is necessary."

[0090] If both the color gradient value and the particle spatial density value exceed the limits, then supplement the color gradient and particle density data simultaneously, with 6 sets of time points for each. After concatenating the data over time, recalculate the average depth gradient and the particle spatial characteristic matrix, and update the state description vector. Re-execute SVD decomposition, and simultaneously increase the weights of both the color gradient feature and the particle spatial distribution feature by 1.5 times, because both are visual features and need to be strengthened equally, generating a refined state description: "Both color gradient and particle spatial density values ​​exceed the limits, caused by insufficient accuracy in visual feature extraction; the weights of both types of image features need to be increased simultaneously."

[0091] If all three indicators exceed the limits, then all three types of data are simultaneously supplemented: pH value, color gradient data, and particle density data, with 6 sets of time points for each. After concatenation, all four components of the state description vector (pH value, average color gradient value, central region density value, and edge region density value) are recalculated. SVD decomposition is re-executed, and all feature weights are uniformly increased by 1.2 times to improve the overall performance and avoid excessive dominance by a single feature. After normalization, the weights are adaptively adjusted, that is, the central region density threshold judgment and enhancement in step S13 are repeated. Finally, a refined state description is generated: "The pH value, color gradient, and particle spatial distribution density all exceed the limits, which is caused by overall data deviation and unreasonable feature weights. Iterative optimization of parameters throughout the entire process is required."

[0092] It should be noted that the number of iterations should not exceed 3. If the indicators still exceed the limits after 3 iterations, the iteration should be suspended and corrected using process experience values. At the same time, the iteration parameters and adjustment process should be fully recorded to ensure that the process is reproducible and traceable.

[0093] In step S17, the optimized dosage is corrected according to the refining state description to obtain a stable dosage of the treatment agent, including:

[0094] By combining the preset prediction parameters and the refined state description, the optimized delivery amount is corrected to obtain the candidate delivery amount;

[0095] If the candidate delivery volume meets the preset stability condition, then the interval mean of the candidate delivery volume is calculated to obtain the stable delivery volume.

[0096] Based on the refined state description, the causes of the deviations are identified. In step S16, it is determined which indicators (pH value, color gradient value, particle spatial density value) have simulation deviations exceeding preset thresholds, and these deviations are directly mapped to preset cause categories based on the combination type of the excess (single indicator, double indicator, or triple indicator). For example, a single indicator exceeding the limit corresponds to low feature weights or insufficient extraction accuracy; a double indicator exceeding the limit corresponds to insufficient allocation of weights for the two types of features; and a triple indicator exceeding the limit corresponds to unreasonable overall data deviation and feature weights. Simultaneously, the system records the specific deviation value for each indicator, such as a particle spatial density deviation of 9.2% and a color gradient deviation of 7.5%. Therefore, by simply parsing the keywords in the refined state description and reading the recorded deviation values ​​of each indicator, the causes of the deviations can be identified. After identifying the causes of the deviations, such as feature extraction deviations, unreasonable preset prediction parameters, or deviations between the simulation model and the actual scene, the preset prediction parameters that need adjustment are determined, including historical data matching similarity thresholds and color gradient correction magnitudes. Subsequently, based on the historical data matching results from S14, the preset prediction parameters are dynamically adjusted, including historical data matching similarity thresholds and color gradient correction magnitudes. A nonlinear adjustment function is defined, which uses the deviation value δ (relative deviation percentage) as the independent variable to map the parameter adjustment amount. The parameter adjustment amount is equal to the baseline adjustment range multiplied by the ratio of the current deviation value to the maximum allowable deviation raised to the power of γ, where γ is the nonlinear exponent. Specifically, the similarity threshold baseline adjustment range is 2%, and the color gradient correction baseline is 3%. The current deviation value is taken as the maximum relative deviation among the out-of-limit indicators, for example, 9.2%. The maximum allowable deviation is set to 20%, and values ​​exceeding this value are directly adopted based on process experience. The nonlinear exponent is set to 0.7, making adjustments more sensitive to small deviations and tending towards saturation for large deviations.

[0097] If the deviation is due to insufficient color gradient correction, the correction range will be increased by 2%-3%. If the deviation is due to insufficient historical data matching accuracy, the similarity threshold will be adjusted from 90% to 88%, and the number of matching samples K will be increased. The calculation involves first adding the old K value to the deviation value divided by 5% and rounding down, then taking the smaller of the calculated result and 10. For example, if the deviation value of S16 is 9.2% due to insufficient color gradient correction, the color gradient correction range will be adjusted from 3%-8% to 5%-10% based on the refined state description, ensuring that the influence of color gradient features is fully considered. Simultaneously, the historical data matching similarity threshold will be adjusted to 88% to improve matching accuracy.

[0098] Based on the adjusted preset prediction parameters, the optimized dosage is further optimized in two cases. In the first case, if S16 does not initiate an iterative update, the upper and lower limits of the interval are fine-tuned based on the optimized dosage and the adjusted preset prediction parameters. The new lower limit is equal to the original lower limit multiplied by the number within parentheses minus the difference between 0.5 times the parameter adjustment amount and the baseline adjustment range; the new upper limit is equal to the original upper limit multiplied by the number within parentheses plus the sum of 0.5 times the parameter adjustment amount and the baseline adjustment range. This ensures that the fluctuation range is ≤3%. Simultaneously, the dosage of the best-matched sample in the historical data matching results is referenced. If this dosage falls outside the adjusted interval, the interval is narrowed by 10% towards that dosage, and the final interval average is taken as the candidate dosage.

[0099] In the second scenario, S16 initiates an iterative update. Based on the reasons for deviations obtained from the refined state description, the drug demand interval is recalculated, following the interval calculation logic of S14. Specifically, an updated feature subset, derived from the state description vector output in the last iteration of step S16 through SVD decomposition, is used. This subset is then compared with the historical database to recalculate the similarity, employing an adjusted similarity threshold and K value. Next, based on the drug dosage of the selected optimal matching sample, the drug demand interval is recalculated according to the formula in step S14. This interval is then further corrected using the adjusted color gradient correction magnitude (if color gradient-related deviations exist), ultimately yielding a candidate dosage interval. The average of this interval is then taken as the candidate dosage.

[0100] The stability of candidate dosages is assessed. If the fluctuation range is ≤3%, and the simulation deviation values ​​corresponding to all indicators are less than or equal to the preset deviation threshold, and the similarity with historical wastewater treatment sample data is ≥85%, then it is determined to be a stable dosage range. The average value of the range is taken as the final stable dosage. At the same time, the execution parameters of the plan are clearly defined, including the dosage rate, dosage location, and dosage time, to ensure that the plan can be directly implemented. If the fluctuation range after further optimization is >3%, the parameter adjustment and dosage optimization steps are repeated until the stability judgment criteria are met. This process is repeated a maximum of 2 times to avoid infinite optimization. If the criteria are still not met after 2 repetitions, the median dosage of the five most similar samples in the historical database is used as a temporary stable plan, based on process experience. The optimization process is recorded, including each parameter adjustment value, candidate dosage, and judgment result, and incorporated into the historical database to improve the data chain.

[0101] The fluctuation range refers to the percentage obtained by dividing the width of the dosage range (i.e., the difference between the upper and lower limits) by the mean of the range. It reflects the stability and repeatability of the candidate dosage scheme. When the fluctuation range does not exceed 3%, it means that the dosage range is relatively narrow, the dosage results obtained by different matching samples or different correction methods are highly consistent, and the reagent dosage scheme has good convergence and operability. If the fluctuation range is too large, for example, exceeding 5%, it indicates that the suggested dosage given by different historical samples or different correction strategies are significantly different, and directly using the mean may lead to unstable actual treatment results. The 3% value comes from the statistical analysis of the acceptable error range of reagent dosage in actual electroplating wastewater treatment projects. Under the premise of ensuring that the effluent meets the standards, the allowable fluctuation of reagent dosage is usually within ±3%.

[0102] In summary, this invention discloses a method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data. By integrating water quality and image multimodal features, dynamic weight optimization, and simulation iteration correction, this invention achieves accurate prediction of the dosage of electroplating wastewater treatment agents, thereby improving the treatment effect.

[0103] Reference Figure 2 The second embodiment of the present invention provides a prediction system for reducing the amount of chemicals used in electroplating wastewater treatment based on multimodal data, comprising:

[0104] The initial monitoring data acquisition and cleaning module is used to collect the pH value and raw image data of electroplating wastewater samples and perform preprocessing to obtain initial monitoring data;

[0105] The state description vector generation module is used to extract a single-channel grayscale image from the initial monitoring data, perform layered analysis on the single-channel grayscale image to obtain a color gradient matrix, locate turbid particles based on the color gradient matrix to obtain a particle spatial characteristic matrix, and fuse the color gradient matrix, the particle spatial characteristic matrix and the pH value to obtain a state description vector.

[0106] The enhanced description vector generation and decomposition module is used to extract the central region density value based on the state description vector. If the central region density value exceeds a preset density threshold, the central region density value is weighted to obtain an enhanced weight coefficient. The state description vector is adjusted according to the enhanced weight coefficient to obtain an enhanced description vector. The enhanced description vector is decomposed to obtain a feature subset.

[0107] The initial deployment quantity generation module is used to calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial deployment quantity.

[0108] The optimized delivery volume correction module is used to extract the data fluctuation range from the initial delivery volume, and combine the color gradient matrix with the data fluctuation range to correct the initial delivery volume to obtain the optimized delivery volume;

[0109] The simulation and data iteration update module is used to simulate and calculate the wastewater treatment process based on the optimized dosage to obtain a description of the refining state.

[0110] The stable dosage scheme determination module is used to correct the optimized dosage based on the refining state description to obtain a stable dosage of the treatment agent.

[0111] It should be noted that the electroplating wastewater treatment agent reduction prediction system based on multimodal data provided in this embodiment of the invention is used to execute all the process steps of the electroplating wastewater treatment agent reduction prediction method based on multimodal data in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0112] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0113] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting the reduction of electroplating wastewater treatment chemicals based on multimodal data, characterized in that, include: The pH values ​​and raw image data of electroplating wastewater samples were collected and preprocessed to obtain initial monitoring data; A single-channel grayscale image is extracted from the initial monitoring data, and a layered analysis is performed on the single-channel grayscale image to obtain a color gradient matrix. Turbid particles are located based on the color gradient matrix to obtain a particle spatial characteristic matrix. The color gradient matrix, the particle spatial characteristic matrix, and the pH value are fused to obtain a state description vector. The density value of the central region is extracted based on the state description vector. If the density value of the central region exceeds a preset density threshold, the density value of the central region is weighted to obtain a strengthening weight coefficient. The state description vector is adjusted based on the strengthening weight coefficient to obtain a strengthening description vector. The strengthening description vector is decomposed to obtain a feature subset. Calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial dosage. The data fluctuation range is extracted from the initial delivery volume, and the initial delivery volume is corrected by combining the color gradient matrix with the data fluctuation range to obtain the optimized delivery volume. Based on the optimized dosage, the wastewater treatment process is simulated and calculated to obtain a description of the refining state. The optimized dosage is corrected based on the description of the refining state to obtain a stable dosage of the treatment agent.

2. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 1, characterized in that, The initial monitoring data, including the collected pH values ​​and raw image data of electroplating wastewater samples, is obtained through preprocessing. The pH value and original image data of the electroplating wastewater sample are obtained. The original image data is converted to color space to obtain color image data. Histogram statistics are performed on the color image data to obtain color distribution characteristics. The color image data is converted to grayscale to obtain a single-channel grayscale image. The particle outline is extracted from the single-channel grayscale image based on the color distribution features to obtain the particle density value. If the particle density value is greater than the preset turbidity particle density judgment threshold, the particle density value is determined to be abnormal and an abnormality is marked to obtain wastewater treatment historical samples. Data cleaning was performed on the pH value and the particle density value to obtain initial monitoring data.

3. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 1, characterized in that, The process involves extracting a single-channel grayscale image from the initial monitoring data, performing layered analysis on the single-channel grayscale image to obtain a color gradient matrix, locating turbid particles based on the color gradient matrix to obtain a particle spatial characteristic matrix, and fusing the color gradient matrix, the particle spatial characteristic matrix, and the pH value to obtain a state description vector, including: A single-channel grayscale image is extracted from the initial monitoring data, and the single-channel grayscale image is subjected to layered analysis to obtain the depth gradient value; The depth gradient values ​​are filled into matrix elements to obtain the color gradient matrix. The average value of all elements in the color gradient matrix is ​​obtained. The color gradient matrix is ​​used to identify turbid particles and obtain local particle density. The local particle density is then filled into matrix elements to obtain a particle spatial characteristic matrix. The density values ​​of the central region and the edge region are extracted from the particle spatial characteristic matrix. The pH value, the average color gradient value, the density value of the central region, and the density value of the edge region are combined to generate a state description vector.

4. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 1, characterized in that, The process involves extracting a central region density value from the state description vector. If the central region density value exceeds a preset density threshold, the central region density value is weighted to obtain a strengthening weight coefficient. The state description vector is then adjusted based on the strengthening weight coefficient to obtain a strengthened description vector. Finally, the strengthened description vector is decomposed to obtain a feature subset, including: The density value of the central region is extracted based on the state description vector. If the density value of the central region is greater than the preset density threshold, the preset weight coefficient is adjusted to obtain the enhancement weight coefficient. The state description vector is adjusted according to the enhancement weight coefficient to obtain the enhancement description vector; The enhanced description vector is subjected to feature decomposition to obtain a feature subset.

5. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 2, characterized in that, The process of calculating the similarity between the feature subset and preset wastewater treatment historical samples, filtering out preset wastewater treatment historical samples whose similarity exceeds a preset similarity threshold, and generating an initial deployment amount includes: The feature subset is standardized to obtain a standard feature subset, and the similarity between the standard feature subset and the preset wastewater treatment historical samples is calculated. The wastewater treatment historical samples with similarity greater than the preset similarity threshold are selected to obtain the optimal matching sample. Extract the dosage of the drug from the optimal matching sample from the preset database, and calculate the initial dosage based on the dosage of the drug.

6. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 1, characterized in that, The step of extracting data fluctuation amplitude from the initial delivery volume, and combining the color gradient matrix with the data fluctuation amplitude to correct the initial delivery volume to obtain an optimized delivery volume includes: Calculate the dispersion of the initial deployment amount to obtain the fluctuation range; Calculate the average value of all elements in the color gradient matrix to obtain the average gradient value. If the average gradient value is not within a preset gradient range, the initial delivery amount is adjusted based on the data fluctuation amplitude to obtain the optimized delivery amount. If the average gradient value is within the preset gradient range, there is no need to adjust the initial delivery amount.

7. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 3, characterized in that, Based on the optimized dosage, the wastewater treatment process is simulated and calculated to obtain a description of the refining state, including: Based on the optimized delivery amount, a virtual simulation is performed to generate a sequence of simulation results, and the pH value, the depth gradient value, and the local particle density are collected simultaneously to obtain real-time monitoring data; The deviation between the simulation result sequence and the real-time monitoring data is calculated. If the deviation is greater than a preset deviation threshold, the real-time monitoring data is iteratively updated to generate a refined state description.

8. The method for predicting the reduction of electroplating wastewater treatment agents based on multimodal data according to claim 1, characterized in that, The step of correcting the optimized dosage based on the refining state description to obtain a stable dosage of the treatment agent includes: By combining the preset prediction parameters and the refined state description, the optimized delivery amount is corrected to obtain the candidate delivery amount; If the candidate delivery volume meets the preset stability condition, then the interval mean of the candidate delivery volume is calculated to obtain the stable delivery volume.

9. A predictive system for reducing the amount of chemicals used in electroplating wastewater treatment based on multimodal data, characterized in that, include: The initial monitoring data acquisition and cleaning module is used to collect the pH value and raw image data of electroplating wastewater samples and perform preprocessing to obtain initial monitoring data; The state description vector generation module is used to extract a single-channel grayscale image from the initial monitoring data, perform layered analysis on the single-channel grayscale image to obtain a color gradient matrix, locate turbid particles based on the color gradient matrix to obtain a particle spatial characteristic matrix, and fuse the color gradient matrix, the particle spatial characteristic matrix and the pH value to obtain a state description vector. The enhanced description vector generation and decomposition module is used to extract the central region density value based on the state description vector. If the central region density value exceeds a preset density threshold, the central region density value is weighted to obtain an enhanced weight coefficient. The state description vector is adjusted according to the enhanced weight coefficient to obtain an enhanced description vector. The enhanced description vector is decomposed to obtain a feature subset. The initial deployment quantity generation module is used to calculate the similarity between the feature subset and the preset wastewater treatment historical samples, filter out the preset wastewater treatment historical samples whose similarity exceeds the preset similarity threshold, and generate the initial deployment quantity. The optimized delivery volume correction module is used to extract the data fluctuation range from the initial delivery volume, and combine the color gradient matrix with the data fluctuation range to correct the initial delivery volume to obtain the optimized delivery volume; The simulation and data iteration update module is used to simulate and calculate the wastewater treatment process based on the optimized dosage to obtain a description of the refining state. The stable dosage scheme determination module is used to correct the optimized dosage based on the refining state description to obtain a stable dosage of the treatment agent.

Images