Electroplating wastewater treatment optimization system based on big data analysis

Abstract

The invention discloses an electroplating wastewater treatment optimization system based on big data analysis, and relates to the technical field of wastewater treatment.The system comprises an acquisition module used for acquiring multi-source heterogeneous data; the reliability evaluation module is used for analyzing the multi-source heterogeneous data through a nonlinear kernel space analysis algorithm and screening high-reliability sensing channel data; the interference decoupling module is used for judging the hydraulic impact state according to the change rate of the fluid volume flow and projecting the high-reliability sensing channel data by utilizing orthogonal decomposition in the non-stationary hydraulic impact state to obtain a pure effective signal matrix; the parameter analysis module is used for constructing and solving a target functional based on the pure effective signal matrix to obtain a load coefficient; and the control module is used for controlling the shunting device and the dosing device based on the load coefficient. According to the invention, the problems of difficult non-linear coupling feature recognition, large hydraulic impact interference and control misalignment caused by detection lag are solved.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an optimized system for electroplating wastewater treatment based on big data analysis. Background Technology

[0002] Electroplating, as an indispensable basic process in the modern industrial system, not only endows metal products with anti-corrosion, wear-resistant, and decorative properties, but also generates a large amount of complex, highly toxic, and difficult-to-degrade industrial wastewater. Electroplating wastewater usually contains heavy metal ions and additive pollutants. If it is not treated properly and discharged directly, it will pose a serious threat to the ecological environment and human health. Traditional wastewater treatment modes based on manual experience or simple PID feedback are no longer able to meet the current dual needs of precise pollution control and energy conservation due to their low precision in reagent dosing and poor resistance to load fluctuations. Therefore, building a data-driven electroplating wastewater treatment optimization system to achieve intelligent transformation from pollutant monitoring to process control is of great practical significance for improving treatment efficiency, reducing operating costs, and ensuring stable compliance of effluent standards. Among existing intelligent treatment technologies for electroplating wastewater, Chinese invention patent CN117602688B discloses an intelligent control method and system for zero discharge of electroplating wastewater. This scheme identifies diversion pipes by recognizing the electroplating process, samples and tests the wastewater after primary treatment, and compares simulated ideal treatment data with actual test data to calculate secondary pollution indicators to identify abnormal pipes. Abnormal diversion pipes with high component overlap are then merged into the same secondary treatment tank for centralized treatment. The advantage of this technical solution lies in its refined identification and classification of wastewater components, which optimizes the secondary treatment process to a certain extent, avoids cross-interference between wastewaters of different properties, and helps... While these technologies can improve resource utilization and processing efficiency, they still have limitations in complex industrial applications. The chemical reaction process of electroplating wastewater is extremely complex, and there is often a strong nonlinear coupling relationship between physicochemical parameters and pollutant concentration. Existing technologies mainly rely on linear comparisons or simple component content detection, ignoring the potential nonlinear mapping characteristics between multi-source heterogeneous data. In actual production, operations such as opening drain valves and cleaning plating tanks often cause drastic fluctuations in fluid volume and flow rate. Such non-steady hydraulic shocks can seriously interfere with the authenticity of sensor data, and existing technologies are prone to misinterpreting hydraulic disturbances as water quality fluctuations, thereby affecting the accuracy of control decisions. Summary of the Invention

[0003] The technical problem solved by this invention is that existing technologies mainly rely on linear comparisons or simple component content detection, ignoring the potential nonlinear mapping characteristics between multi-source heterogeneous data. In actual production, operations such as opening drain valves and cleaning plating tanks often cause drastic fluctuations in fluid volume and flow rate. Such non-steady hydraulic shocks can seriously interfere with the authenticity of sensor data. Existing technologies are prone to misjudging hydraulic disturbances as water quality fluctuations, thereby affecting the accuracy of control decisions.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an electroplating wastewater treatment optimization system based on big data analysis, comprising: a data acquisition module, a reliability assessment module, an interference decoupling module, a parameter parsing module, and a control module; The acquisition module is used to acquire multi-source heterogeneous data during the electroplating wastewater treatment process and perform preprocessing. The reliability assessment module is used to analyze the preprocessed multi-source heterogeneous data using a nonlinear kernel space analysis algorithm and to filter high-reliability sensing channel data. The interference decoupling module is used to determine the hydraulic impact state based on the rate of change of fluid volume flow rate in the preprocessed multi-source heterogeneous data, and to project the high-confidence sensing channel data using orthogonal decomposition under the non-stationary hydraulic impact state to obtain a pure and effective signal matrix. The parameter parsing module is used to construct and solve the target functional based on the pure effective signal matrix to obtain the load coefficient; The control module is used to control the diversion device and the dosing device based on the load factor.

[0005] Preferably, the multi-source heterogeneous data includes sensing channel data and pollutant load data; The sensing channel data includes physicochemical data and production condition data; The physicochemical data include the pH value, redox potential, conductivity, turbidity, and fluid volumetric flow rate of the drainage branch pipe. The production status data includes the process parameters of the plating type currently operating on each electroplating production line, the output current value of the plating tank rectifier, and the opening degree of the electric drain valve. The process parameters include plating tank temperature, stirring motor speed, and the electroplating time of the current batch. The pollutant load data includes the mass concentration of heavy metal ions; The preprocessing includes data cleaning, missing value imputation, and Z-score normalization.

[0006] Preferably, the process of screening high-confidence perception channel data includes: A covariate matrix is ​​constructed based on the physicochemical data and production condition data, and a composite response vector is constructed based on the pollutant load data. The process of constructing the covariate matrix includes: Within a preset sliding time window, physicochemical data and production condition data are arranged in chronological order to obtain the corresponding physicochemical data sequence and production condition data sequence. The physicochemical data sequence and production condition data sequence are then concatenated as column vectors to obtain the covariate matrix. Within a preset sliding time window, pollutant load data are arranged in chronological order to obtain the corresponding pollutant load data sequence. The pollutant load data sequence is then concatenated column by column to obtain the composite response vector.

[0007] Preferably, the process of screening high-confidence perception channel data further includes: The correlation between the covariate matrix and the composite response vector is analyzed by nonlinear kernel space analytical algorithm to obtain the weight coefficients of each sensing channel data. The specific process of obtaining the weight coefficients of the data from each sensing channel includes: Construct the kernel matrix of the covariate matrix and the kernel matrix of the composite response vector based on the Gaussian kernel function, and then perform centering processing; Based on the kernel matrix of the covariate matrix after centering and the kernel matrix of the composite response vector, a generalized eigenvalue equation is constructed, and generalized eigenvalue decomposition is performed to select the eigenvector corresponding to the largest eigenvalue. Multiply the kernel matrix of the covariate matrix by the eigenvector to obtain the canonical variable vector; Calculate the Pearson correlation coefficient between the original data column vector corresponding to each sensing channel in the covariate matrix and the canonical variable vector. Use the absolute value of the Pearson correlation coefficient as the canonical loading value of each sensing channel. Normalize the canonical loading value to obtain the weight coefficient of each sensing channel. Perception channel data with weighting coefficients greater than a preset effective threshold are used as high-reliability perception channel data. Within the sliding time window, the physicochemical data sequence and production condition data sequence corresponding to the high-confidence sensing channel data are concatenated as column vectors to obtain the effective signal matrix; At the current moment, the physical and chemical data and production condition data corresponding to the high-confidence perception channel are arranged according to feature dimensions to construct an instantaneous feature vector.

[0008] Preferably, the process of determining the hydraulic impact state specifically includes: Within a preset sliding time window, the first derivative of the fluid volumetric flow rate with respect to time is calculated as the rate of change of the fluid volumetric flow rate. If the fluid volume flow rate change rate does not meet the preset non-stationary hydraulic impact state determination condition, then the sliding time window is determined to be a stationary hydraulic impact state, and the effective signal matrix and composite response vector are directly used as the pure effective signal matrix and pure composite response vector. If the fluid volume flow rate change rate meets the preset non-stationary hydraulic impact state determination condition, then the sliding time window is determined to be a non-stationary hydraulic impact state, and the effective signal matrix and composite response vector are decoupled from interference to obtain a pure effective signal matrix and a pure composite response vector. The condition for determining the non-stationary hydraulic impact state is: the rate of change of fluid volume flow rate at multiple consecutive sampling points exceeds the preset impact threshold.

[0009] Preferably, the interference decoupling process specifically includes: Within a preset sliding time window, the flow regime is divided into an impact flow regime and a steady flow regime based on the rate of change of fluid volumetric flow rate at the sampling point. The first sampling point where the fluid volume flow rate change rate is greater than the impact threshold is taken as the starting node, and the last sampling point where the volume flow rate change rate is greater than the preset impact threshold is taken as the ending node. The starting node and the ending node constitute the impact flow region, and the segment within the sliding time window other than the impact flow region is taken as the steady flow region. In the time axis of the impact flow region, encrypted nodes are constructed at a preset encryption interval, and in the time axis of the steady flow region, sparse nodes are constructed at a preset sparse interval. The encrypted nodes and sparse nodes are arranged in chronological order to construct a non-uniform node vector.

[0010] Preferably, the interference decoupling process further includes: Based on the non-uniform node vector, construct the analytical expression of the B-spline basis function; substitute the sampling time of each sampling point into the analytical expression of the B-spline basis function to calculate and obtain the basis function sampling value vector corresponding to that time; arrange the basis function sampling value vectors corresponding to all sampling points in rows to construct the basis function interference matrix. The basis function interference matrix is ​​orthogonally decomposed to obtain the projection operator, and the residual between the identity matrix and the projection operator is used as the left multiplier of the orthogonal complement matrix. The effective signal matrix and the composite response vector are projected onto the left-multiplied orthogonal complement matrix to obtain the pure effective signal matrix and the pure composite response vector.

[0011] Preferably, the process of obtaining the load factor specifically includes: Based on the linear mapping relationship between the pure composite response vector and the pure effective signal matrix, an objective functional is constructed. The mathematical expression for the objective functional is: ; in, For the target functional, Represents the pure composite response vector. This represents a pure, effective signal matrix. For feature dimension, This represents the predefined non-convex sparse penalty function. The regularization parameter represents the sparse penalty term. express The regularization parameter of the regularization term. Let n be a positive definite weighting matrix, and n be the number of sampling points. The load factor is obtained by iteratively solving the objective functional using a local linear approximation algorithm and a coordinate descent algorithm.

[0012] Preferably, the process of controlling the diversion device includes: The product of the instantaneous concentration response index and the current fluid volumetric flow rate is taken as the instantaneous total pollution flux; The instantaneous concentration response index is the result of the inner product of the load coefficient and the instantaneous feature vector; Based on the instantaneous total pollution flux, trigger control is performed on the diversion device, and the trigger control specifically includes: The shunt device is activated when any of the shunt device activation conditions are met. The shunt device activation conditions include: The fluid volumetric flow rate exceeds the preset hydraulic limit threshold. The instantaneous total pollution flux exceeds the preset maximum treatment capacity threshold.

[0013] Preferably, the process of controlling the dosing device includes: The dosing device is subjected to flux feedforward control, which specifically includes: Calculate the net influent flow rate at the current moment, whereby the net influent flow rate is the fluid volumetric flow rate at the current moment minus the instantaneous diversion flow rate; The product of the net influent flow rate and the instantaneous concentration response index is taken as the effective treatment flux. The theoretical dosing rate is calculated based on the effective treatment flux and the stoichiometric coefficient, and the theoretical dosing rate is converted into a control signal for the dosing device and output.

[0014] The beneficial effects of this invention are as follows: By constructing an orthogonal projection decoupling mechanism based on B-spline basis functions, this invention significantly enhances the anti-interference capability under non-stationary operating conditions. For sudden changes in fluid volumetric flow rate caused by the opening of the drain valve or pump speed changes, the invention uses a sequence of B-spline basis functions generated by non-uniform node vectors to accurately fit the transient waveform of hydraulic impact. Mathematically, this eliminates the influence of fluid physical disturbances on sensor readings, achieving deep decoupling between physical flow field interference and chemical reaction signals. This greatly improves the detection reliability of the system under dynamic hydraulic environments. Furthermore, by constructing a dual objective functional containing SCAD non-convex sparse penalty terms and L2 regularization terms, this invention overcomes the common multicollinearity and non-correlated redundancy problems in sensor arrays, resulting in a calculated load coefficient that possesses both sparsity and numerical stability. Attached Figure Description

[0015] Figure 1 This is a basic flowchart of an electroplating wastewater treatment optimization system based on big data analysis, provided as an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example, refer to Figure 1 This paper presents an optimized system for electroplating wastewater treatment based on big data analysis, which includes: a data acquisition module, a reliability assessment module, an interference decoupling module, a parameter parsing module, and a control module. The acquisition module is used to acquire multi-source heterogeneous data during the electroplating wastewater treatment process and perform preprocessing. The reliability assessment module is used to analyze the preprocessed multi-source heterogeneous data using a nonlinear kernel space analysis algorithm and to filter high-reliability sensing channel data. The interference decoupling module is used to determine the hydraulic impact state based on the rate of change of fluid volume flow rate in the preprocessed multi-source heterogeneous data, and to project the high-confidence sensing channel data using orthogonal decomposition under the non-stationary hydraulic impact state to obtain a pure and effective signal matrix. The parameter parsing module is used to construct and solve the target functional based on the pure effective signal matrix to obtain the load coefficient; The control module is used to control the diversion device and the dosing device based on the load factor.

[0018] This invention solves the problems of inconsistent data quality and frequent hydraulic interference in electroplating wastewater treatment by constructing a closed-loop optimization system that includes reliability assessment, interference decoupling, and parameter analysis. The system first eliminates distorted or low-correlation pseudo-sensing channels through a reliability assessment module, then uses an interference decoupling module to mathematically isolate fluid physical disturbances, and finally constructs a control model based on pure signals. This layered data processing architecture ensures that the load coefficients used by the control module truly reflect the essential requirements of the chemical reaction, thereby significantly improving the robustness and control accuracy of the entire wastewater treatment system and avoiding malfunctions caused by sensor drift or hydraulic fluctuations.

[0019] Multi-source heterogeneous data includes sensing channel data and pollutant load data; The sensing channel data includes physicochemical data and production condition data; Physicochemical data include the pH value, redox potential, conductivity, turbidity, and fluid volumetric flow rate of the drainage branch pipe; Production status data includes the process parameters of the plating type currently in operation for each electroplating production line, the output current value of the plating tank rectifier, and the opening degree of the electric drain valve; Process parameters include plating bath temperature, stirring motor speed, and the plating time of the current batch; Pollutant load data includes the mass concentration of heavy metal ions; Preprocessing includes data cleaning, missing value imputation, and Z-score normalization.

[0020] In one specific embodiment of the present invention, physicochemical data are collected through an online sensor array installed on each branch pipe; Production status data is captured in real time from the PLCs of each production line via industrial Ethernet. Pollutant load data are provided via an online heavy metal analyzer.

[0021] The process of screening high-reliability sensing channel data includes: A covariate matrix is ​​constructed based on physicochemical data and production condition data, and a composite response vector is constructed based on pollutant load data; The process of constructing the covariate matrix includes: Within a preset sliding time window, physicochemical data and production condition data are collected at the sampling frequency. The physicochemical data and production condition data are arranged in chronological order to obtain the corresponding physicochemical data sequence and production condition data sequence. The physicochemical data sequence and production condition data sequence are then concatenated as column vectors to obtain the covariate matrix. Within a preset sliding time window, pollutant load data are collected at the sampling frequency, arranged in chronological order, and the corresponding pollutant load data sequence is obtained. The pollutant load data sequence is then concatenated column by column to obtain a composite response vector, which is a column vector.

[0022] In a specific embodiment of the present invention, the sampling sliding window length is set to... (In this embodiment) (300 sampling points), for each sampling time, the sensing channel data is arranged into a row vector according to the sampling order, and this row vector contains Each feature dimension (in this embodiment) These correspond to pH value, redox potential, conductivity, turbidity, electric drain valve opening, plating tank rectifier output current, stirring motor speed, plating tank temperature, and the plating time of the current batch, respectively. Arrange the row vectors corresponding to the sensing channel data at each sampling time vertically to construct a dimension of The covariate matrix, where each row of the covariate matrix represents all the physicochemical data and production condition data at the same time; Construction dimension The composite response vector, where each element represents the mass concentration of heavy metal ions at the sampling time.

[0023] This invention constructs a covariate matrix and a composite response vector through a sliding time window. This matrix construction method based on time-series splicing preserves the dynamic evolution characteristics of the sensing channel data in the time dimension and effectively solves the alignment problem of data with different sampling frequencies. By transforming single-point data into a matrix form containing the trend of the data lake, it is possible to obtain the lag effect and cumulative effect of pollutant concentration changes, providing a digital carrier containing rich spatiotemporal information for subsequent mining of deep correlations between variables.

[0024] The process of screening high-confidence perception channel data also includes: The correlation between the covariate matrix and the composite response vector is analyzed by nonlinear kernel space analytical algorithm to obtain the weight coefficients of data from each sensing channel. The specific process of obtaining the weight coefficients of each sensing channel data includes: Construct the kernel matrix of the covariate matrix and the kernel matrix of the composite response vector based on the Gaussian kernel function, and then perform centering processing; Based on the kernel matrix of the covariate matrix after centering and the kernel matrix of the composite response vector, a generalized eigenvalue equation is constructed, and generalized eigenvalue decomposition is performed to select the eigenvector corresponding to the largest eigenvalue. Multiply the kernel matrix of the covariate matrix by the eigenvector corresponding to the largest eigenvalue to obtain the canonical variable vector, which represents the dominant nonlinear feature in the high-dimensional reproducing kernel Hilbert space. Calculate the Pearson correlation coefficient between the original data column vector and the canonical variable vector corresponding to each sensing channel in the covariate matrix. Use the absolute value of the Pearson correlation coefficient as the canonical loading value of each sensing channel. Normalize the canonical loading value to obtain the weight coefficient of each sensing channel. Perception channel data with weighting coefficients greater than a preset effective threshold are used as high-reliability perception channel data. Within the sliding time window, the physicochemical data sequence and production condition data sequence corresponding to the high-confidence sensing channel data are concatenated as column vectors to obtain the effective signal matrix; At the current moment, the physical and chemical data and production condition data corresponding to the high-confidence perception channel are arranged according to feature dimensions to construct an instantaneous feature vector.

[0025] In a specific embodiment of the present invention, since the chemical reaction process of electroplating wastewater is complex, for example, the effect of pH value on the reduction reaction of hexavalent chromium is not a simple linear relationship, but a nonlinear relationship. If the correlation is calculated directly in the original data space, key variables will be missed because this nonlinear feature is ignored. Therefore, the Gaussian kernel function is selected to construct the kernel matrix of the covariate matrix and the kernel matrix of the composite response vector, respectively, and the original low-dimensional sensing channel data is mapped to the high-dimensional regeneration kernel Hilbert space. By decomposing the covariance operator, the projection direction with the highest correlation is found in the reproducing kernel Hilbert space. Based on the canonical variables obtained from the decomposition, the canonical loading values ​​(i.e., Pearson correlation coefficients) of the column vectors corresponding to each original sensing channel in the canonical variables are calculated. The absolute values ​​of the calculated canonical loading values ​​of each channel are processed and normalized to obtain the final weight coefficients. The effective threshold in this embodiment is 0.6.

[0026] This invention utilizes a nonlinear kernel space analytical algorithm to replace traditional linear correlation analysis, solving the problem of identifying the strong nonlinear coupling relationship between physicochemical parameters and heavy metal concentration in electroplating wastewater. By mapping low-dimensional data to a high-dimensional regeneration kernel Hilbert space through a Gaussian kernel function and performing generalized eigenvalue decomposition in this space, it can accurately screen out high-confidence sensing channels that truly have chemical indicative significance and effectively eliminate redundant variables that are not essentially related to the target pollutants.

[0027] The process of determining the state of hydraulic impact specifically includes: Within a preset sliding time window, the first derivative of the fluid volumetric flow rate with respect to time is calculated as the rate of change of the fluid volumetric flow rate. If the fluid volume flow rate change rate does not meet the preset non-stationary hydraulic impact state judgment condition, then the sliding time window is judged to be a stationary hydraulic impact state, and the effective signal matrix and composite response vector are directly used as the pure effective signal matrix and pure composite response vector. If the fluid volume flow rate change rate meets the preset non-stationary hydraulic impact state judgment condition, then the sliding time window is determined to be a non-stationary hydraulic impact state. The effective signal matrix and composite response vector are decoupled from interference to obtain a pure effective signal matrix and a pure composite response vector. The condition for determining non-stationary hydraulic impact state is: the rate of change of fluid volume flow rate at multiple consecutive sampling points exceeds the preset impact threshold.

[0028] In one specific embodiment of the present invention, the fluid volumetric flow rate change rate at three consecutive sampling points all exceeds [a certain value]. It is determined that the system has entered a non-steady hydraulic shock state.

[0029] This invention introduces a non-stationary hydraulic impact state determination mechanism based on the fluid volume flow rate change rate. It can identify transient fluid disturbances caused by the opening of the drain valve or the change of pump speed in real time. By setting a clear impact threshold and continuous judgment logic, it can distinguish between simple physical flow field changes and chemical water quality fluctuations. This prevents the oscillation of sensing channel data caused by drastic changes in flow velocity from being misjudged as a sudden change in pollutant concentration. As a result, the control system avoids issuing incorrect dosing or diversion commands during periods of hydraulic instability, thus ensuring the stability of the control logic.

[0030] The interference decoupling process specifically includes: Within a preset sliding time window, the flow regime is divided into an impact flow regime and a steady flow regime based on the rate of change of fluid volumetric flow rate at the sampling point. The first sampling point where the fluid volume flow rate change rate is greater than the impact threshold is taken as the starting node, and the last sampling point where the volume flow rate change rate is greater than the preset impact threshold is taken as the ending node. The starting node and the ending node constitute the impact flow region, and the segment within the sliding time window other than the impact flow region is taken as the steady flow region. In the time axis of the impact flow region, encrypted nodes are constructed at a preset encryption interval, and in the time axis of the steady flow region, sparse nodes are constructed at a preset sparse interval. The encrypted nodes and sparse nodes are arranged in chronological order to construct a non-uniform node vector.

[0031] This invention employs a non-uniform node vector construction strategy based on flow regime division to achieve adaptive capture of hydraulic impact waveforms. In the impact flow regime, dense nodes are used to record high-frequency details of fluid change instants with high fidelity. In the steady flow regime, sparse nodes are used to effectively reduce data redundancy and computational dimensionality.

[0032] The interference decoupling process also includes: Based on non-uniform node vectors, construct the analytical expression of B-spline basis function. Substitute the sampling time of each sampling point into the analytical expression of B-spline basis function to calculate and obtain the basis function sampling value vector corresponding to that time. Arrange the basis function sampling value vectors corresponding to all sampling points in rows to construct a basis function interference matrix. Perform orthogonal decomposition on the basis function disturbance matrix to obtain the projection operator, and use the residual between the identity matrix and the projection operator as the left multiplication of the orthogonal complement matrix. By projecting the effective signal matrix and the composite response vector onto the left multiplier of the orthogonal complement matrix, the pure effective signal matrix and the pure composite response vector are obtained.

[0033] In a specific embodiment of the present invention, the order of the B-spline basis function is set to 3, and the fluid volume flow rate change rate at each sampling point within the current sliding window is calculated. The fluid volume flow rate change rates at all sampling points constitute a fluid volume flow rate change rate sequence. Set the impact threshold (in this embodiment, it is set to ). The process iterates through the sequence of fluid volumetric flow rate changes, taking the first sampling point where the fluid volumetric flow rate change is greater than the impact threshold as the starting node and the last sampling point where the fluid volumetric flow rate change is greater than the impact threshold as the ending node. The starting node and the ending node constitute the impact flow region, and the segment within the sliding window other than the impact flow region is the steady flow region. In the impinging flow region, dense nodes are constructed at a preset density interval (1 s in this embodiment) on the time axis, and sparse nodes are constructed at a preset sparse interval (60 s in this embodiment) on the time axis of the steady flow region. The dense and sparse nodes are arranged in chronological order to construct a non-uniform node vector. ; Based on non-uniform node vectors By using the Cox-de-Boor recursive formula, the analytical expressions of B-spline basis functions are generated stepwise from low to high order, thus obtaining the sequence of analytical expressions formed by the corresponding B-spline basis function expressions. (in The spline order is taken in this embodiment. Each element in the B-spline basis function sequence is an analytical expression of a basis function; The non-uniform node vector defines the distribution density and local support interval of the B-spline basis function expressions on the time axis, and determines the effective range of each basis function expression in the time dimension; Substitute the sampling time of each sampling point into the analytical sequence of B-spline basis functions to obtain the corresponding B-spline basis function sequence. Arrange the B-spline basis function sequences of all sampling points into row vectors to construct a vector with dimension [dimensionality missing]. The basis function interference matrix; The sampling points are those within the sliding time window; The first in the basis function interference matrix Line number The column element is the first Each sampling time Substitute the first The scalar values ​​obtained from the analytical expressions of the basis functions are: each column of the basis function interference matrix represents the numerical evolution of a basis function over the entire time window, and each row represents the value of each basis function at that sampling time.

[0034] This invention utilizes left-multiplication of orthogonal complement matrix to achieve mathematical decoupling between physical interference and chemical signals. It accurately describes the waveform characteristics of hydraulic impact through B-spline basis function sequence and constructs corresponding projection operators, which can force the effective signal matrix to be projected onto the null space of the interference signal. This can filter out pseudo-wave components caused by fluid dynamic factors (such as bubbles and turbulence) in the sensing channel data, providing a clean data source with high signal-to-noise ratio for the calculation of load factor.

[0035] The process of obtaining the load factor specifically includes: Based on the linear mapping relationship between the pure composite response vector and the pure effective signal matrix, an objective functional is constructed; the objective functional is used to determine the optimal value of the load factor. The mathematical expression for the objective functional is: ; in, For the target functional, Represents the pure composite response vector. This represents a pure, effective signal matrix. For feature dimension, This represents the predefined non-convex sparse penalty function. The regularization parameter represents the sparse penalty term. express The regularization parameter of the regularization term. Let n be a positive definite weighting matrix, and n be the number of sampling points. The load factor is obtained by iteratively solving the objective functional using a local linear approximation algorithm and a coordinate descent algorithm.

[0036] In one specific embodiment of the present invention, a linear observation model in the projection subspace is constructed based on the pure effective signal matrix and the pure composite response vector. The linear observation model is used to represent the linear mapping relationship between the pure composite response vector and the pure effective signal matrix. The mathematical expression of the linear observation model is as follows: ; in, For a pure composite response vector, For a pure and effective signal matrix, The load factor to be solved represents the true chemical contribution of each sensing channel data to the pollutant concentration after eliminating hydraulic interference. Let be the regression residual vector, which follows a mean of 0 and a variance of . Gaussian distribution, This represents the random error term caused by sensor noise and hydraulic disturbances outside the ideal linear model. The core objective of this invention in constructing the objective functional is to optimize the load factor to make the sum of squared residuals between the actual observed values ​​and the model predictions (i.e., Minimize energy; In order to obtain the ability (Predicted value) and The smallest difference between (true values) Under ideal conditions, the load factor is usually obtained by constructing the sum of squared residuals. However, considering the multicollinearity (such as the high correlation between pH and ORP values ​​in a specific range) and the potential non-correlation redundancy of some sensors (such as turbidity not contributing to heavy metal concentration under certain operating conditions) in the physicochemical data of electroplating wastewater, there are differences from the ideal conditions for establishing a pure and effective signal matrix; therefore, this invention constructs an objective functional The mathematical expression for the double-penalty objective functional is: ; in, This represents the function value of the double-penalty objective functional to be optimized. Represents the pure composite response vector. This represents a pure, effective signal matrix. The number of feature dimensions. The dimension is The load factor vector to be solved, its elements Corresponding to the Weights of each feature dimension, The transpose operation represents a matrix or vector. The dimension is The positive definite weighting matrix is ​​set as a diagonal matrix in this embodiment, and its diagonal elements are inversely proportional to the rate of change of fluid volumetric flow rate at the corresponding time point, which is used to adjust the confidence weights of different sampling points. This represents the total number of sampling points within the sampling window, and is used as the normalization coefficient. This represents a preset non-convex sparse penalty function (SCAD function in this embodiment), used to apply a threshold value. To achieve variable selection, The regularization parameter (0.01 in this example) represents the sparsity penalty term and is used to control the sparsity of the solution. The regularization parameter (0.005 in this example) represents the L2 regularization term and is used to control the smoothness of the solution. The normalization coefficient refers to the total number of sampling points within the sampling window; The dual-penalty objective functional includes weighted residuals ( ), sparsity penalty term ( ) and L2 regularization term ( ); In the sparsity penalty term The pre-defined non-convex penalty function (SCAD function in this embodiment) is used to reduce the coefficients of insignificant physicochemical data (such as redundant turbidity features) to zero by utilizing the threshold characteristics of the non-convex penalty function, thereby achieving sparsity of the solution and eliminating noise interference. The sparsity penalty term is introduced to automatically compress the weight coefficients of physicochemical parameters that do not significantly contribute to the concentration of heavy metals to zero. Introduction The norm penalty term is for utilizing The smoothing properties of the norm resolve the problem of high correlation between physicochemical data (e.g., the high coupling between pH and ORP values ​​within a specific range). L2 constraints can prevent drastic fluctuations or overfitting of weight coefficients caused by collinearity among dependent variables, thus ensuring that the obtained weight coefficients are numerically smooth and stable. To address the difficulty in solving the non-convex penalty function in the sparse penalty term, a local linear approximation algorithm is employed to approximate the non-convex sparse penalty function. The process of converting it into a weighted L1 penalty term includes: Set the initial load factor vector (Usually initialized to 0), in the... In the next iteration, based on the coefficients obtained in the previous step... The non-convex penalty term is transformed into a weighted L1 penalty term using a local linear approximation algorithm. Specifically, adaptive penalty weights are calculated for each feature dimension. : ; in, Let be the first derivative of the SCAD penalty function. At this point, the original objective functional is transformed into the following weighted Lasso subproblem: ; For the aforementioned weighted Lasso subproblem, a coordinate descent method is used to update the load factors one by one until the norm of the change in the load factor vector between two adjacent iterations is less than a preset convergence threshold (in this embodiment). ), at this time the output This is the globally optimal load factor.

[0037] This invention constructs a dual objective functional comprising a SCAD non-convex sparsity penalty term and an L2 regularization term, and solves it using local linear approximation and coordinate descent. This solves the balance problem between multicollinearity and feature selection. The SCAD penalty function achieves sparse selection of variables while avoiding excessive compression of important feature coefficients by traditional L1 regularization, ensuring the unbiasedness of the estimation. The L2 term effectively smooths the weight distribution among highly correlated variables (such as pH and ORP), preventing model overfitting. This dual constraint mechanism ensures that the calculated load coefficients have both clear physical sparsity and numerical smoothness and stability, and can truly reflect the contribution rate of each sensing channel data to pollutant concentration.

[0038] The process of controlling the diversion device includes: The product of the instantaneous concentration response index and the current fluid volumetric flow rate is taken as the instantaneous total pollution flux; The instantaneous concentration response index is the result of the inner product of the load coefficient and the instantaneous eigenvector; Based on the instantaneous total pollution flux, trigger control is implemented on the diversion device. The trigger control specifically includes: The shunt device is activated when any of the shunt device activation conditions are met. The shunt device activation conditions include: The fluid volumetric flow rate exceeds the preset hydraulic limit threshold. The instantaneous total pollution flux exceeds the preset maximum treatment capacity threshold.

[0039] In a specific embodiment of the present invention, the load factor essentially characterizes the contribution weight of each physicochemical data to the heavy metal concentration. Through inner product operation, the high-frequency physicochemical signal is mapped in real time to the estimated value of the heavy metal concentration, i.e., the instantaneous concentration response index. Multiply the instantaneous concentration response index calculated above with the current fluid volumetric flow rate to obtain the instantaneous total pollution flux. The instantaneous total pollution flux intuitively quantifies the actual chemical load at the current moment, providing a physical benchmark for subsequent diversion protection and precise dosing.

[0040] This invention solves the problem of detection lag in online analyzers by obtaining the instantaneous concentration response index through inner product operation. Combined with the total flux obtained by flow calculation, it intuitively quantifies the current chemical load. Based on this, the dual-threshold diversion judgment can ensure that the protection mechanism is triggered in time when the influent load exceeds the physical or chemical treatment limit, thus eliminating the risk of effluent exceeding the standard due to instantaneous load.

[0041] The process of controlling the dosing device includes: The dosing device is subjected to flux feedforward control, which specifically includes: Calculate the net inflow rate at the current moment. The net inflow rate is the fluid volumetric flow rate at the current moment minus the instantaneous diversion flow rate. The product of the net influent flow rate and the instantaneous concentration response index is taken as the effective treatment flux. The theoretical dosing rate is calculated based on the effective treatment flux and the stoichiometric coefficient, and then the theoretical dosing rate is converted into a control signal for the dosing device for output.

[0042] In a specific embodiment of the present invention, when the fluid volume flow rate is greater than a preset hydraulic limit threshold, the amount of water to be diverted is the difference between the fluid volume flow rate and the preset hydraulic limit threshold. When the instantaneous total pollution flux exceeds the preset maximum treatment capacity threshold, the amount of water to be diverted is the ratio of the load to be intercepted to the instantaneous concentration response index, and the load to be intercepted is the difference between the instantaneous total pollution flux and the maximum treatment capacity threshold. When the fluid volumetric flow rate is greater than the preset hydraulic limit threshold and when the instantaneous total pollution flux is greater than the preset maximum treatment capacity threshold, the maximum diversion volume is selected as the diversion volume. The instantaneous diversion flow rate of the diversion device is collected in real time, and the difference between the current fluid volume flow rate and the instantaneous diversion flow rate is used as the net influent flow rate. The product of the net influent flow rate and the instantaneous concentration response index is taken as the effective treatment flux, and the product of the effective treatment flux and the stoichiometric coefficient is taken as the theoretical dosing rate. The PLC for electroplating wastewater treatment takes the theoretical dosing rate as the target rate and converts it into the frequency of the frequency converter of the dosing pump. It controls the dosing for the net influent flow rate that is not diverted and drives the dosing pump to dosing. In this embodiment, the hydraulic limit threshold is The data source is the ratio of the design volume of the wastewater treatment reaction sedimentation tank to the minimum hydraulic retention time, and the maximum treatment capacity threshold is... The data source is the ratio of the maximum dosing rate of the dosing device to the stoichiometric coefficient, where the stoichiometric coefficient is... (For every 1g of contaminant removed, the following dosage is required) (dose of diluted reagent in water), the basic dosage is... The data source is the flow rate corresponding to the lowest stable operating frequency of the dosing pump inverter.

[0043] This invention implements feedforward dosing control based on effective treatment flux. By calculating the product of the net influent flow rate after deducting the diversion and the concentration index, the actual total mass of pollutants entering the reaction tank is accurately obtained, and the theoretical dosing rate is directly calculated based on the stoichiometric ratio. This control method avoids cost waste and secondary pollution caused by excessive dosing, and also prevents substandard treatment caused by excessive dosing, significantly improving the dynamic response speed and control accuracy of the dosing system.

[0044] Existing technologies often rely on linear correlation analysis to screen variables, which easily overlooks the nonlinear sensitivity of physicochemical parameters such as pH and ORP to heavy metal ion concentration. This invention introduces a nonlinear kernel space analytical algorithm, using Gaussian kernel functions to map low-dimensional raw data to a high-dimensional regenerated kernel Hilbert space, and constructs and decomposes generalized eigenvalue equations in this space. This allows for in-depth exploration of the potential nonlinear isomorphic relationship between sensing channels and pollutant load. This approach significantly improves the accuracy of screening high-confidence sensing channels, ensures that the input variables for subsequent model construction have real chemical indicative significance, and effectively avoids the omission or misselection of key features. This invention significantly enhances the anti-interference capability under non-stationary operating conditions by constructing an orthogonal projection decoupling mechanism based on B-spline basis functions. For sudden changes in fluid volume flow rate caused by the opening of the drain valve or the change of pump speed, the transient waveform of hydraulic impact is accurately fitted by the B-spline basis function sequence generated by non-uniform node vectors. The effective signal matrix is ​​then projected onto the null space of hydraulic interference by left multiplying by the orthogonal complement matrix. This process mathematically eliminates the influence of fluid physical disturbance on sensor readings, realizes deep decoupling between physical flow field interference and chemical reaction signals, and ensures that the calculation of the load coefficient only reflects the actual contribution rate of chemical reaction, thereby greatly improving the detection reliability of the system in dynamic hydraulic environment. This invention overcomes the common problems of multicollinearity and uncorrelated redundancy in sensor arrays by constructing a dual objective functional that includes a SCAD non-convex sparse penalty term and an L2 regularization term. The introduction of the SCAD penalty function ensures unbiased estimation of large coefficients while achieving variable selection, while the L2 norm smooths the weight distribution among highly correlated variables, making the calculated load coefficient both sparsity and numerical stability. The instantaneous total pollution flux feedforward control mechanism built based on this load coefficient realizes the leap from concentration control to mass flux control. It can accurately calculate the theoretical dosing rate based on the instantaneous flux and stoichiometry with a second-level response, realizing lag-free feedforward precision dosing and diversion protection. While significantly reducing reagent waste, it eliminates the risk of instantaneous excessive emissions caused by load fluctuations.

[0045] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. 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 storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes; it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the protection scope of the present invention.

Claims

1. A big data analysis-based optimization system for electroplating wastewater treatment, characterized in that, include: The module includes a data acquisition module, a reliability assessment module, an interference decoupling module, a parameter parsing module, and a control module. The acquisition module is used to acquire multi-source heterogeneous data during the electroplating wastewater treatment process and perform preprocessing. The reliability assessment module is used to analyze the preprocessed multi-source heterogeneous data using a nonlinear kernel space analysis algorithm and to filter high-reliability sensing channel data. The interference decoupling module is used to determine the hydraulic impact state based on the rate of change of fluid volume flow rate in the preprocessed multi-source heterogeneous data, and to project the high-confidence sensing channel data using orthogonal decomposition under the non-stationary hydraulic impact state to obtain a pure and effective signal matrix. The parameter parsing module is used to construct and solve the target functional based on the pure effective signal matrix to obtain the load coefficient; The control module is used to control the diversion device and the dosing device based on the load factor.

2. The electroplating wastewater treatment optimization system based on big data analysis of claim 1, wherein, The multi-source heterogeneous data includes sensing channel data and pollutant load data; The sensing channel data includes physicochemical data and production condition data; The physicochemical data include the pH value, redox potential, conductivity, turbidity, and fluid volumetric flow rate of the drainage branch pipe. The production status data includes the process parameters of the plating type currently operating on each electroplating production line, the output current value of the plating tank rectifier, and the opening degree of the electric drain valve. The process parameters include plating tank temperature, stirring motor speed, and the electroplating time of the current batch. The pollutant load data includes the mass concentration of heavy metal ions; The preprocessing includes data cleaning, missing value imputation, and Z-score normalization.

3. The electroplating wastewater treatment optimization system based on big data analysis of claim 2, wherein, The process of screening high-reliability sensing channel data includes: A covariate matrix is ​​constructed based on the physicochemical data and production condition data, and a composite response vector is constructed based on the pollutant load data. The process of constructing the covariate matrix includes: Within a preset sliding time window, physicochemical data and production condition data are arranged in chronological order to obtain the corresponding physicochemical data sequence and production condition data sequence. The physicochemical data sequence and production condition data sequence are then concatenated as column vectors to obtain the covariate matrix. Within a preset sliding time window, pollutant load data are arranged in chronological order to obtain the corresponding pollutant load data sequence. The pollutant load data sequence is then concatenated column by column to obtain the composite response vector.

4. The electroplating wastewater treatment optimization system based on big data analysis of claim 3, wherein, The process of screening high-confidence perception channel data also includes: The correlation between the covariate matrix and the composite response vector is analyzed by nonlinear kernel space analytical algorithm to obtain the weight coefficients of each sensing channel data. The specific process of obtaining the weight coefficients of the data from each sensing channel includes: Construct the kernel matrix of the covariate matrix and the kernel matrix of the composite response vector based on the Gaussian kernel function, and then perform centering processing; Based on the kernel matrix of the covariate matrix after centering and the kernel matrix of the composite response vector, a generalized eigenvalue equation is constructed, and generalized eigenvalue decomposition is performed to select the eigenvector corresponding to the largest eigenvalue. Multiply the kernel matrix of the covariate matrix by the eigenvector to obtain the canonical variable vector; Calculate the Pearson correlation coefficient between the original data column vector corresponding to each sensing channel in the covariate matrix and the canonical variable vector. Use the absolute value of the Pearson correlation coefficient as the canonical loading value of each sensing channel. Normalize the canonical loading value to obtain the weight coefficient of each sensing channel. Perception channel data with weighting coefficients greater than a preset effective threshold are used as high-reliability perception channel data. Within the sliding time window, the physicochemical data sequence and production condition data sequence corresponding to the high-confidence sensing channel data are concatenated as column vectors to obtain the effective signal matrix; At the current moment, the physical and chemical data and production condition data corresponding to the high-confidence perception channel are arranged according to feature dimensions to construct an instantaneous feature vector.

5. The electroplating wastewater treatment optimization system based on big data analysis of claim 4, wherein, The process of determining the state of hydraulic impact specifically includes: Within a preset sliding time window, the first derivative of the fluid volumetric flow rate with respect to time is calculated as the rate of change of the fluid volumetric flow rate. If the fluid volume flow rate change rate does not meet the preset non-stationary hydraulic impact state determination condition, then the sliding time window is determined to be a stationary hydraulic impact state, and the effective signal matrix and composite response vector are directly used as the pure effective signal matrix and pure composite response vector. If the fluid volume flow rate change rate meets the preset non-stationary hydraulic impact state determination condition, then the sliding time window is determined to be a non-stationary hydraulic impact state, and the effective signal matrix and composite response vector are decoupled from interference to obtain a pure effective signal matrix and a pure composite response vector. The condition for determining the non-stationary hydraulic impact state is: the rate of change of fluid volume flow rate at multiple consecutive sampling points exceeds the preset impact threshold.

6. The big data analysis based electroplating wastewater treatment optimization system of claim 5, wherein, The interference decoupling process specifically includes: Within a preset sliding time window, the flow regime is divided into an impact flow regime and a steady flow regime based on the rate of change of fluid volumetric flow rate at the sampling point. The first sampling point where the fluid volume flow rate change rate is greater than the impact threshold is taken as the starting node, and the last sampling point where the volume flow rate change rate is greater than the preset impact threshold is taken as the ending node. The starting node and the ending node constitute the impact flow region, and the segment within the sliding time window other than the impact flow region is taken as the steady flow region. In the time axis of the impact flow region, encrypted nodes are constructed at a preset encryption interval, and in the time axis of the steady flow region, sparse nodes are constructed at a preset sparse interval. The encrypted nodes and sparse nodes are arranged in chronological order to construct a non-uniform node vector.

7. The big data analysis based electroplating wastewater treatment optimization system of claim 5, wherein, The interference decoupling process also includes: Based on the non-uniform node vector, construct the analytical expression of the B-spline basis function; substitute the sampling time of each sampling point into the analytical expression of the B-spline basis function to calculate and obtain the basis function sampling value vector corresponding to that time; arrange the basis function sampling value vectors corresponding to all sampling points in rows to construct the basis function interference matrix. The basis function interference matrix is ​​orthogonally decomposed to obtain the projection operator, and the residual between the identity matrix and the projection operator is used as the left multiplier of the orthogonal complement matrix. The effective signal matrix and the composite response vector are projected onto the left-multiplied orthogonal complement matrix to obtain the pure effective signal matrix and the pure composite response vector.

8. The big data analysis based electroplating wastewater treatment optimization system of claim 7, wherein, The process of obtaining the load factor specifically includes: Based on the linear mapping relationship between the pure composite response vector and the pure effective signal matrix, an objective functional is constructed. The mathematical expression for the objective functional is: ； in, For the target functional, Represents the pure composite response vector. This represents a pure, effective signal matrix. For feature dimension, This represents the predefined non-convex sparse penalty function. The regularization parameter represents the sparse penalty term. express The regularization parameter of the regularization term. Let n be a positive definite weighting matrix, and n be the number of sampling points. The load factor is obtained by iteratively solving the objective functional using a local linear approximation algorithm and a coordinate descent algorithm.

9. The electroplating wastewater treatment optimization system based on big data analysis as described in claim 8, characterized in that, The process of controlling the diversion device includes: The product of the instantaneous concentration response index and the current fluid volumetric flow rate is taken as the instantaneous total pollution flux; The instantaneous concentration response index is the result of the inner product of the load coefficient and the instantaneous feature vector; Based on the instantaneous total pollution flux, trigger control is performed on the diversion device, and the trigger control specifically includes: The shunt device is activated when any of the shunt device activation conditions are met. The shunt device activation conditions include: The fluid volumetric flow rate exceeds the preset hydraulic limit threshold. The instantaneous total pollution flux exceeds the preset maximum treatment capacity threshold.

10. The electroplating wastewater treatment optimization system based on big data analysis as described in claim 9, characterized in that, The process of controlling the dosing device includes: The dosing device is subjected to flux feedforward control, which specifically includes: Calculate the net influent flow rate at the current moment, whereby the net influent flow rate is the fluid volumetric flow rate at the current moment minus the instantaneous diversion flow rate; The product of the net influent flow rate and the instantaneous concentration response index is taken as the effective treatment flux. The theoretical dosing rate is calculated based on the effective treatment flux and the stoichiometric coefficient, and the theoretical dosing rate is converted into a control signal for the dosing device and output.

Images