A resourceful treatment system for heavy metal-containing wastewater from automobile wire harness electroplating

Abstract

The present application relates to the field of intelligent electroplating wastewater treatment, and particularly relates to a resource treatment system for heavy metal-containing electroplating wastewater of automobile wiring harness, comprising: a running parameter acquisition module acquires multi-dimensional real-time parameters of a membrane separation unit and constructs an observation sequence, a hidden state decoding module inputs the sequence into a hidden Markov model to decode and output hidden states of the membrane separation unit, an adaptive backwashing control module controls to perform a pulse backwashing operation when it is determined that the state is reversible pollution and a preset time length is reached, a target function construction module guides concentrated liquid into an electrolysis unit and calculates real-time current efficiency as an extreme optimization target function, a disturbance and gradient calculation module superimposes a sinusoidal disturbance on a control signal to calculate a gradient estimation value of the target function, and an optimization iteration control module updates duty cycle and peak current to adjust pulse power output accordingly. The present application realizes prevention and control of irreversible pollution of the membrane assembly, and improves global adaptive purification energy efficiency of the heavy metal electrolysis recovery process.

Description

Technical Field

[0001] This invention relates to the field of intelligent electroplating wastewater treatment, and more particularly to a resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating. Background Technology

[0002] The electroplating process for automotive wiring harnesses generates toxic wastewater rich in heavy metals such as copper. Proper treatment of this wastewater is crucial for achieving water reuse and secondary recovery of valuable metals. Currently, industrial applications often employ membrane separation technology to deeply concentrate and retain the wastewater to recover pure water. The concentrated heavy metal solution is then introduced into an electrolytic recovery device, where a pulsed power supply reduces metal ions at the cathode into economically valuable solid metals. This creates a comprehensive resource-based treatment system that balances environmental compliance with economic benefits.

[0003] However, existing systems mostly rely on fixed thresholds for passive membrane backwashing and constant parameter electrolysis control. Each processing unit lacks global data linkage and adaptive adjustment capabilities, which not only easily leads to irreversible fouling and damage to the membrane module, but also causes high-energy-consuming hydrogen evolution side reactions due to concentration decrease in the later stage of electrolysis, resulting in low overall system operating efficiency. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating, aiming to improve the problems of membrane fouling and low electrolysis efficiency caused by the lack of adaptive adjustment in existing technologies.

[0005] This invention provides the following technical solution: a resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating, comprising:

[0006] The operation parameter acquisition module is used to obtain the real-time transmembrane pressure difference, membrane flux and water temperature of the membrane separation unit, and construct the observation sequence according to the time dimension;

[0007] The hidden state decoding module is used to input the observation sequence into the hidden Markov model and decode and output the hidden state of the membrane separation unit through the Viterbi algorithm. The hidden state includes a clean state, a reversible contamination state, and an irreversible contamination state.

[0008] An adaptive backwash control module is used to control the membrane separation unit to perform a pulse backwash operation when it is determined that the hidden state at the current moment is the reversible contamination state and the duration reaches a preset time threshold.

[0009] The objective function construction module is used to introduce the heavy metal concentrate discharged from the membrane separation unit into the electrolytic recovery unit, obtain the real-time metal recovery rate and power consumption of the electrolytic recovery unit, calculate the real-time current efficiency and use it as the objective function for extreme value optimization control.

[0010] The perturbation and gradient calculation module is used to superimpose a sinusoidal perturbation signal onto the duty cycle and peak current control signals corresponding to the electrolytic recovery unit, and calculate the gradient estimate of the objective function based on the response fluctuation of the objective function to the sinusoidal perturbation signal.

[0011] The optimization iterative control module is used to update the duty cycle and peak current of the next control cycle based on the gradient estimate and the preset optimization step size, so as to adjust the pulse power output of the electrolytic recovery unit.

[0012] Preferably, in the parameter acquisition module, the step of constructing the observation sequence according to the time dimension includes:

[0013] The Kalman filter operator is used to perform data fusion and denoising on the real-time transmembrane pressure difference, membrane flux and water temperature, and to remove high-frequency random noise caused by fluid pulses;

[0014] Based on the maximum and minimum value normalization mapping function, the denoised physical parameters are projected to the standard numerical range to suppress the weight interference of physical dimensions on probability calculation.

[0015] A sliding time window with a preset time step is constructed, and the normalized data points are concatenated into a time-series feature matrix according to the sampling order to generate the observation sequence.

[0016] Preferably, in the hidden state decoding module, the step of inputting the observation sequence into the hidden Markov model includes:

[0017] Based on the physical decay characteristics of membrane fouling caused by heavy metal wastewater, the state transition probability matrix of the hidden Markov model is initialized and constructed to constrain the evolution logic between different fouling states.

[0018] A Gaussian mixture model is introduced to construct the emission probability density function. By using a weighted linear combination of Gaussian distributions, the statistical probability of generating the current observation data under a specific hidden state is quantified.

[0019] Prior sample data of historical membrane fouling life cycle are extracted, and unsupervised parameter learning is performed on the state transition probability matrix and the emission probability density function using the Baum-Welch algorithm to complete the model calibration.

[0020] Preferably, in the hidden state decoding module, the step of decoding and outputting the hidden state of the membrane separation unit using the Viterbi algorithm includes:

[0021] Based on the initial state probability vector and emission probability density function of the Hidden Markov Model, calculate the local forward log probability of each hidden state in the observation sequence at the initial time step, and initialize the starting point of the dynamic programming.

[0022] Initiate the dynamic programming path recursion process, calculate the path with the maximum cumulative probability to any candidate hidden state at the current time step, and use the backtracking pointer matrix to record the local optimal predecessor node;

[0023] Extract the endpoint state with the highest global cumulative probability at the end of the observation sequence, reconstruct the optimal state transition path by tracing back along the backtracking pointer matrix, and output the hidden state at the current time corresponding to the endpoint.

[0024] Preferably, in the adaptive backwash control module, the step of controlling the membrane separation unit to perform pulse backwashing includes:

[0025] Based on the cumulative duration of the reversible fouling state and the corresponding state probability, combined with the preset mechanical strength tolerance limit of the membrane material, the target pressure peak and pulse frequency of the backwash fluid are dynamically calculated.

[0026] The target pressure peak and pulse frequency are converted into a pneumatic command sequence for a programmable logic controller, and a closed-loop control algorithm is used to drive the backwash pump and the multi-way reversing valve to switch in coordination.

[0027] The slope of the transmembrane pressure difference recovery during the backwash cycle is monitored in real time. By calculating the exponential smoothing of the recovery slope, the preset time threshold for the next operating cycle is adaptively corrected to complete the backwash closed-loop control.

[0028] Preferably, in the objective function construction module, the step of calculating the real-time current efficiency and using it as the objective function for extreme value optimization control includes:

[0029] The theoretical metal deposition rate within a time slice is calculated using an ion concentration difference sensing model and Faraday's law, and the actual effective metal deposition rate is estimated by combining cathode dynamic weighing data.

[0030] The transient output voltage and current sequence of the pulse power supply module is collected, the active power consumption in the time slice is calculated using the discrete integral operator, and the actual metal deposition per unit energy consumption is calculated to obtain the real-time current efficiency.

[0031] A multi-objective optimization mapping function based on a penalty mechanism is constructed. On the basis of maximizing the real-time current efficiency, a nonlinear penalty constraint term characterizing the cell temperature rise and hydrogen evolution intensity is introduced to reconstruct the objective function.

[0032] Preferably, in the disturbance and gradient calculation module, the step of superimposing a sinusoidal disturbance signal onto the duty cycle and peak current control signals corresponding to the electrolytic recovery unit includes:

[0033] Initialize two local oscillators with orthogonal phase and coprime frequencies to generate dual-channel high-frequency sinusoidal perturbation carriers for decoupling multidimensional control variables;

[0034] The local variance and signal-to-noise ratio of the objective function are extracted, and the perturbation amplitude is dynamically calculated through an adaptive scaling operator, so that the perturbation energy has a nonlinear inverse mapping relationship with the current degree of optimization convergence.

[0035] The scaled dual-channel high-frequency sinusoidal perturbation carrier is synchronously fed forward and injected into the DC control reference baseline of the duty cycle and peak current using a digital-to-analog conversion matrix to generate a synthetic drive signal.

[0036] Preferably, in the perturbation and gradient calculation module, the step of calculating the gradient estimate of the objective function includes:

[0037] The transient response sequence of the objective function output by the electrolytic recovery unit after being excited by a control signal superimposed with the sinusoidal disturbance signal is extracted, and the DC bias baseline and low-frequency drift component are filtered out using a high-pass digital filter.

[0038] The transient response sequence after filtering out the DC component is multiplied in the time domain synchronously with the sinusoidal disturbance signal, and the sensitivity information of the objective function is shifted to the baseband spectrum through the synchronous demodulation mechanism.

[0039] The baseband spectrum data is input into a first-order inertial low-pass integral filter to smoothly extract the DC effective component in the demodulated signal, and the approximate first-order partial derivative vector of the objective function with respect to each control variable is constructed as the gradient estimate.

[0040] Preferably, in the optimization iterative control module, the step of updating and calculating the duty cycle and peak current for the next control cycle includes:

[0041] An adaptive gradient optimization algorithm with a first-order momentum acceleration mechanism is introduced to perform exponential moving average filtering on the gradient estimate, thereby suppressing local gradient oscillations caused by the mass transfer delay of the electroplating solution.

[0042] Multiply the filtered smooth gradient vector by the feedforward dynamic learning rate step size, and use the gradient ascent optimization rule to perform incremental iterative updates on the duty cycle and peak current basic parameters at the current moment.

[0043] The updated parameters are forcibly truncated using a hardware safety boundary projection operator. After verifying that the parameters do not exceed the physical limits of the electrolytic power supply, the parameters are overwritten to the pulse control register to complete the pulse power supply output regulation of the electrolytic recovery unit.

[0044] The present invention has the following beneficial effects:

[0045] 1. In this invention, the Hidden Markov Model and the Extreme Value Optimization Algorithm are deeply integrated, breaking the bottleneck of traditional water treatment systems that rely on fixed thresholds for passive operation. Through the global linkage of membrane fouling state decoding and electrolysis pulse power supply adaptive optimization, pure water closed-loop circulation and automated and efficient extraction of heavy metals are achieved while ensuring compliant wastewater treatment, thereby improving the overall operating efficiency and economic added value of the equipment.

[0046] 2. In this invention, a non-invasive online monitoring mechanism for membrane health status is constructed. The Viterbi algorithm is used to decode the microscopic reversible fouling stage from the macroscopic observation sequence, thereby triggering an adaptive pulse backwashing operation before the fouling deteriorates irreversibly. This avoids the waste of pure water and high-pressure physical damage caused by traditional timed blind cleaning, reduces operation and maintenance costs, and extends the actual service life of the separation membrane module.

[0047] 3. In this invention, in response to the pain point that serious hydrogen evolution side reactions are easily triggered in the later stage of electrolysis, an extreme value optimization control algorithm is introduced to inject sinusoidal disturbances into the control signal and dynamically extract gradients. Under the constraints of multiple safety penalty mechanisms, the duty cycle and peak current of the pulse power supply are continuously iteratively adjusted, so that the electrolytic cell is locked at the optimal current efficiency operating point in real time, which reduces the unit purification energy consumption and ensures the dense purity of the cathode deposited metal. Attached Figure Description

[0048] Figure 1 This is an architectural diagram of a resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating, as proposed in this invention. Detailed Implementation

[0049] The technical solutions in 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.

[0050] In an embodiment of the present invention, the present invention provides a resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating, such as... Figure 1 As shown, it includes:

[0051] The operation parameter acquisition module is used to obtain the real-time transmembrane pressure difference, membrane flux and water temperature of the membrane separation unit, and construct the observation sequence according to the time dimension;

[0052] Furthermore, in the parameter acquisition module, the steps for constructing the observation sequence according to the time dimension include:

[0053] The Kalman filter operator is used to perform data fusion and denoising on the real-time transmembrane pressure difference, membrane flux and water temperature, and to remove high-frequency random noise caused by fluid pulses;

[0054] Based on the maximum and minimum value normalization mapping function, the denoised physical parameters are projected to the standard numerical range to suppress the weight interference of physical dimensions on probability calculation.

[0055] A sliding time window with a preset time step is constructed, and the normalized data points are concatenated into a time-series feature matrix according to the sampling order to generate an observation sequence.

[0056] Specifically, during the operation of the membrane separation unit in the automotive wiring harness electroplating wastewater treatment system, pressure sensors, flow meters, and temperature sensors installed on the pipes and membrane modules acquire raw measurement data of transmembrane pressure difference, membrane flux, and water temperature in real time at a fixed sampling frequency. Because the operation of the high-pressure pump and the pulse scouring of the fluid in the heavy metal wastewater treatment system cause random high-frequency fluctuations in the sensor data, the measurement data from these three dimensions are constructed into an observation vector and input into a Kalman filter operator to perform data fusion and denoising.

[0057] Set at The actual measurement vector for each discrete time step is ,in This represents the measured transmembrane pressure difference. This indicates the measured membrane flux. The measured water temperature is represented. The prediction and update equations for the Kalman filter are constructed. Since the changes in the membrane's working fouling state are relatively slow compared to the sampling frequency, both the state transition matrix and the observation matrix are set to identity matrices. First, prior state estimation and error covariance prediction are performed:

[0058] ;

[0059] ;

[0060] Then, the Kalman gain is calculated and the posterior state and error covariance are updated by combining the actual measurement vector:

[0061] ;

[0062] ;

[0063] ;

[0064] In the above Kalman filter formula, For the first The prior state estimation vector of the step. This is the optimal posterior state estimation vector from the previous step. Let be the prior error covariance matrix. This is the error covariance matrix from the previous step. To characterize the slowly varying properties of a membrane system, the process noise covariance matrix is ​​used. To calculate the Kalman gain, This is the actual measurement vector at the current moment, containing the transmembrane pressure difference, membrane flux, and water temperature. The measurement noise covariance matrix caused by the fluid pulses from the high-pressure pump is given. It is the identity matrix. The output is used to calculate the optimal state estimation vector, which includes the denoised transmembrane pressure difference, membrane flux, and water temperature.

[0065] The unit for transmembrane pressure difference is typically megapascals (MPA), the unit for membrane flux is typically liters per square meter per hour (L / m² / h), and the unit for water temperature is degrees Celsius (°C). These significant differences in the dimensions of these physical quantities can severely interfere with the weight allocation in subsequent probabilistic models. Therefore, a maximum-minimum normalized mapping function is used to estimate the optimal state vector after denoising. Each physical parameter in the equation undergoes a linear transformation independently:

[0066] ;

[0067] In this normalization formula, These are the normalized physical parameter values. This refers to the current values ​​of transmembrane pressure, membrane flux, or water temperature after noise reduction. and These are the minimum and maximum values ​​of the corresponding physical parameters recorded during the historical operation of the electroplating wastewater membrane separation unit. After processing with this formula, all physical parameters are proportionally projected to a standard value range of zero to one.

[0068] Set a number of time steps to A sliding time window. As the system sampling time progresses, the system extracts data from the current moment according to the physical sampling sequence. The length is calculated backwards. All normalized data points. These time-series multidimensional vectors are concatenated column-wise to generate a three-row vector. Column time series feature matrix :

[0069] ;

[0070] In this matrix sequence, For the current moment The normalized transmembrane pressure difference, membrane flux, and water temperature form a three-dimensional column vector. This generated time-series feature matrix is ​​the observation sequence constructed along the time dimension, serving as the input basis for subsequent model decoding.

[0071] This step filters out high-frequency data noise caused by fluid mechanical fluctuations and unifies the dimensions of heterogeneous sensor data, providing a smooth and standardized input sequence for subsequent hidden Markov models, which can reduce the system's misjudgment rate of membrane fouling status.

[0072] The hidden state decoding module is used to input the observation sequence into the Hidden Markov Model and decode the hidden state of the membrane separation unit through the Viterbi algorithm. The hidden states include clean state, reversible contamination state and irreversible contamination state.

[0073] Furthermore, in the hidden state decoding module, the step of inputting the observation sequence into the hidden Markov model includes:

[0074] Based on the physical decay characteristics of membrane fouling caused by heavy metal wastewater, the state transition probability matrix of the hidden Markov model is initialized and constructed to constrain the evolution logic between different fouling states.

[0075] A Gaussian mixture model is introduced to construct the emission probability density function. By using a weighted linear combination of Gaussian distributions, the statistical probability of generating the current observation data under a specific hidden state is quantified.

[0076] Prior sample data of historical membrane fouling life cycle are extracted, and unsupervised parameter learning is performed on the state transition probability matrix and emission probability density function using the Baum-Welch algorithm to complete the model calibration.

[0077] Furthermore, in the hidden state decoding module, the step of decoding the hidden state of the output membrane separation unit using the Viterbi algorithm includes:

[0078] Based on the initial state probability vector and emission probability density function of the Hidden Markov Model, calculate the local forward log probability of each hidden state in the observation sequence at the initial time step, and initialize the starting point of the dynamic programming.

[0079] Initiate the dynamic programming path recursion process, calculate the path with the maximum cumulative probability to any candidate hidden state at the current time step, and use the backtracking pointer matrix to record the local optimal predecessor node;

[0080] Extract the endpoint state with the highest global cumulative probability at the end of the observation sequence, reconstruct the optimal state transition path by tracing back along the backtracking pointer matrix, and output the hidden state at the current time corresponding to the endpoint.

[0081] Specifically, during the operation of the membrane separation unit in heavy metal wastewater treatment, the physical state of the membrane module undergoes unidirectional irreversible degradation as pollutants accumulate. The set of hidden states of the membrane separation unit is defined as... ,in Indicates a clean state. This represents a reversible pollution state. This represents the irreversible fouling state. Based on the physical decay characteristics of membrane fouling, the fouling process typically follows a unidirectional evolution from clean to reversible and then to irreversible. The state transition probability matrix of the Hidden Markov Model is initialized accordingly. Elements in the matrix Indicates the membrane state from Transfer to The probability of state transition. The specific formula is as follows:

[0082] ;

[0083] In this matrix formula, This indicates the probability that the membrane remains clean. This represents the probability that the membrane transitions from a clean state to a reversibly fouled state. This represents the probability that the membrane transitions from a clean state to an irreversibly fouled state. This represents the probability that the membrane will recover from a reversibly fouled state to a clean state. This indicates the probability that the membrane remains in a reversibly fouled state. This represents the probability that the membrane transitions from a reversibly fouled state to an irreversibly fouled state. This represents the probability that the membrane will recover from an irreversibly fouled state to a clean state. This represents the probability that the membrane recovers from an irreversibly fouled state to a reversibly fouled state. This represents the probability that the membrane remains in an irreversibly fouled state. During initialization, higher initial probability values ​​are assigned to the diagonal elements and the adjacent upper right element, while the probability of the lower triangular elements is set to a minimum value close to zero. This constrains the evolution logic between different fouling states to conform to the actual physical process of heavy metal wastewater clogging the membrane pores. To quantify the statistical probability of generating current multidimensional observation data including transmembrane pressure difference, membrane flux, and water temperature under a specific hidden state, a Gaussian mixture model is introduced to construct the emission probability density function. For the hidden state... Its emission probability density function pass The weighted linear combination of Gaussian distributions is represented as:

[0084] ;

[0085] In this formula, For the current time step The input is an observed feature vector containing transmembrane pressure difference, membrane flux, and water temperature. The number of components in the Gaussian mixture model. For state Next The mixture weights of the Gaussian components are equal to one. Let Gaussian be the multivariate probability density distribution function. The mean vector of this Gaussian component represents the statistical center of the sensor data under this state. The covariance matrix characterizes the range of fluctuations in the relationship between variables such as transmembrane pressure and flux. A sequence of prior observational data covering the historical membrane fouling lifecycle, from new membrane operation to deep fouling and decommissioning, is obtained. The Baum-Welch algorithm is used to perform unsupervised parameter learning on the aforementioned state transition probability matrix and emission probability density function. In the expectation-maximization iteration update step, the re-estimation formula for the state transition probability is:

[0086] ;

[0087] In this revaluation formula, The updated state transition probability. The total length of the historical observation sequence. Given the observation sequence and current model parameters, the time step is... In state And time step In state The joint probability, For time step In state The marginal probabilities are calculated. The parameters of the hidden Markov model and Gaussian mixture model of the system are calibrated by iteratively updating the log-likelihood function until it converges.

[0088] During online system operation, the hidden states of the output membrane separation unit are decoded using the Viterbi algorithm. This is based on the initial state probability vector of the calibrated Hidden Markov Model. Using the emission probability density function, calculate the observation sequence at the initial time step. Local forward log probability corresponding to each hidden state Start by initializing the dynamic programming and initialize the backtracking pointer. :

[0089] ;

[0090] ;

[0091] Then, the dynamic programming path recursion process is initiated to calculate the transition to the current time step. Any candidate hidden state Maximum cumulative log probability path And using the backtracking pointer matrix Record the locally optimal predecessor node:

[0092] ;

[0093] ;

[0094] in the formula For the membrane in its initial state The prior probability, The state reached at the previous time step The maximum cumulative probability, These are the elements of the state transition matrix.

[0095] At the last time step of the current input observation sequence Extract the endpoint state with the highest global cumulative probability. :

[0096] ;

[0097] Then, a reverse tracing operation is performed along the backtracking pointer matrix to reconstruct the optimal state transition path:

[0098] ;

[0099] in the formula To reverse trace and reconstruct the time steps The optimal hidden state. Through reverse derivation, the system outputs the final time step. Corresponding endpoint state This refers to the actual hidden state of the membrane separation unit at the current moment, that is, the specific determination result of whether the membrane is currently in a clean state, a reversibly fouled state, or an irreversibly fouled state.

[0100] This step enables the accurate inference of the evolution stage of micro-fouling inside the membrane, which cannot be directly measured by the naked eye, through readily available apparent continuous monitoring data such as pressure and flow rate. This effectively improves the accuracy and timeliness of the system's determination of the fouling status of reverse osmosis or ultrafiltration membranes.

[0101] The adaptive backwash control module is used to control the membrane separation unit to perform pulse backwashing when it is determined that the hidden state at the current moment is a reversible contamination state and the duration reaches a preset time threshold.

[0102] Furthermore, in the adaptive backwash control module, the steps for controlling the membrane separation unit to perform pulse backwashing operations include:

[0103] Based on the cumulative duration of the reversible fouling state and the corresponding state probability, combined with the preset mechanical strength tolerance limit of the membrane material, the target pressure peak and pulse frequency of the backwash fluid are dynamically calculated.

[0104] The target pressure peak and pulse frequency are converted into a pneumatic command sequence for a programmable logic controller, and a closed-loop control algorithm is used to drive the backwash pump and the multi-way reversing valve to switch in coordination.

[0105] The transmembrane pressure differential recovery slope during the backwash cycle is monitored in real time. By calculating the exponential smoothing of the recovery slope, the preset time threshold for the next operating cycle is adaptively corrected to complete the backwash closed-loop control.

[0106] Specifically, when the system's Hidden Markov Model determines that the current hidden state is a reversible fouling state, and the duration of this state continuously reaches a preset time threshold, the system actively controls the membrane separation unit to perform a pulse backwash operation. To avoid physical tearing damage to the membrane module caused by excessive water pressure, the system dynamically calculates the target peak pressure and pulse frequency of the backwash fluid based on the cumulative duration of the reversible fouling state and its corresponding state probability, combined with the preset mechanical strength tolerance limit of the membrane material. The formula for calculating the target peak pressure is:

[0107] ;

[0108] In this pressure calculation formula, The target peak pressure of the backwash fluid is dynamically calculated and its unit is megapascals (MPa). This is the standard reference backwash pressure for the membrane separation unit. This is the preset pressure scaling factor. This represents the state probability corresponding to the current reversible contamination state output by the Hidden Markov Model. This represents the cumulative duration of the reversible contamination state. The preset mechanical strength tolerance limit of the membrane material is used as the maximum permissible water pressure hard boundary for forced truncation. The system synchronously calculates the pulse frequency under the corresponding state using a linear mapping formula:

[0109] ;

[0110] In this frequency calculation formula, The calculated backwash target pulse frequency is expressed in Hertz. This is the initial reference pulse frequency. This is the frequency growth factor.

[0111] After obtaining the target parameters, the system converts the target pressure peak and pulse frequency into a pneumatic command sequence that the programmable logic controller (PLC) can recognize. It then uses a proportional-integral-derivative (PID) closed-loop control algorithm to drive the inverter of the backwash pump and the multi-way directional valve to perform coordinated switching. The PLC reads the actual backwash pressure of the pipeline network from the pressure sensor in real time, compares it with the target pressure peak to generate a pressure error, and calculates the frequency converter adjustment command through the closed-loop control algorithm.

[0112] ;

[0113] In this closed-loop control formula, For a moment The frequency control command signal output to the backwash pump inverter is used to adjust the speed of the water pump motor in real time. For a moment The pressure error, i.e., the target pressure peak value. The difference between the measured pressure and the actual feedback pressure. This is the proportional gain coefficient, used to accelerate the system's response to pressure drops. This is the integral gain coefficient, used to eliminate steady-state pipeline pressure deviations. The differential gain coefficient is used to suppress the fluid water hammer overshoot effect generated during high-frequency switching of the multi-way directional valve.

[0114] After completing the backwash cycle and restoring filtered permeate, the system monitors and collects data on the transmembrane pressure difference in real time, calculating the transmembrane pressure difference recovery slope for that cycle to evaluate the actual effectiveness of physical cleaning. To eliminate random fluid fluctuations during a single cleaning operation, the system uses an exponential smoothing algorithm to calculate the historical trend of the current transmembrane pressure difference recovery slope.

[0115] ;

[0116] In this exponential smoothing calculation formula, For the current number The slope of the transmembrane pressure difference recovery after smoothing during each backwash cycle. This is the original recovery slope calculated based on actual measurements during the current cycle. This is the smooth recovery slope of the previous cycle recorded by the system. These are the smoothing weight coefficients defined in the interval between zero and one. After obtaining the smoothing slope, the system compares it with the calibration value and adaptively adjusts the preset time threshold for the next running cycle.

[0117] ;

[0118] In this time threshold correction formula, This is to update the preset time threshold used for the next running cycle after calculation. The preset time threshold is the actual time used in the current period. The time step correction factor is set. This is the recovery slope of the reference pressure under the pre-calibrated ideal new membrane state. After this calculation is completed, the new threshold is overwritten into the control system to form a complete backwash closed-loop control.

[0119] This implementation step enables precise backwashing on demand based on the degree of heavy metal fouling. While effectively removing heavy metal complex sludge from the membrane surface, it avoids the risk of membrane fiber breakage caused by fixed high-pressure cleaning and significantly reduces the consumption of backwash pure water.

[0120] The objective function construction module is used to introduce the heavy metal concentrate discharged from the membrane separation unit into the electrolytic recovery unit, obtain the real-time metal recovery rate and power consumption of the electrolytic recovery unit, calculate the real-time current efficiency and use it as the objective function for extreme value optimization control.

[0121] Furthermore, in the objective function construction module, the step of calculating the real-time current efficiency and using it as the objective function for extreme value optimization control includes:

[0122] The theoretical metal deposition rate within a time slice is calculated using an ion concentration difference sensing model and Faraday's law, and the actual effective metal deposition rate is estimated by combining cathode dynamic weighing data.

[0123] The transient output voltage and current sequence of the pulse power supply module is collected, the active power consumption within the time slice is calculated using the discrete integral operator, and the actual metal deposition under unit energy consumption is calculated accordingly to obtain the real-time current efficiency.

[0124] A multi-objective optimization mapping function based on a penalty mechanism is constructed. On the basis of maximizing real-time current efficiency, a nonlinear penalty constraint term characterizing the cell temperature rise and hydrogen evolution intensity is introduced to reconstruct the objective function.

[0125] Specifically, after the concentrated heavy metal solution discharged from the membrane separation unit is introduced into the electrolytic recovery unit, the system first obtains the real-time metal recovery rate of the electrolytic recovery unit. An ion concentration difference sensing model is constructed using online ion concentration meters installed at both the inlet and outlet of the electrolytic cell, and the theoretical metal deposition amount within a time slice is calculated using Faraday's law.

[0126] ;

[0127] In this theoretical precipitation formula, This represents the theoretical amount of metal precipitation within the time slice, expressed in grams. The molar mass of the heavy metals to be extracted is determined. The valence of this heavy metal ion is given. is Faraday's constant. This represents the transient current function of the electrolytic cell. The system synchronously reads the mass data from the dynamic weighing sensor on the cathode suspension device to estimate the actual effective metal deposition rate. The formula for calculating the actual effective metal deposition rate is:

[0128] ;

[0129] In this actual rate formula, The actual effective metal precipitation rate is expressed in grams per second. This represents the total physical mass of the cathode collected at the end of the current time slice. The total physical mass of the cathode at the start of the time slice. The set time slice length.

[0130] The system acquires the transient output voltage and current sequences of the pulse power supply module using a high-frequency data acquisition card, and calculates the active energy consumption within that time slice using a discrete integral operator. The formula for calculating the active energy consumption is:

[0131] ;

[0132] In this integral formula, This represents the active electrical energy consumption within a time slice, measured in joules. This represents the total number of sampling points within the time slice. For the first The transient output voltage at each sampling point is expressed in volts. For the first The transient output current at each sampling point is expressed in amperes. This is the sampling period of the high-frequency data acquisition card. Based on this, the actual metal deposition per unit of energy consumption is calculated to obtain the real-time current efficiency.

[0133] ;

[0134] In this efficiency formula, The real-time current efficiency is defined as the actual amount of heavy metal extracted per unit of energy consumption, representing the number of grams of heavy metal extracted per joule of electrical energy consumed.

[0135] Because high-frequency pulsed electrolysis is prone to triggering severe hydrogen evolution side reactions and causing a rapid rise in the temperature of the electroplating bath when the concentration of heavy metal ions decreases in the later stages, a multi-objective optimization mapping function based on a penalty mechanism is constructed for the system. Based on the fundamental term maximizing the real-time current efficiency, a nonlinear penalty constraint term characterizing the bath temperature rise and hydrogen evolution intensity is introduced, reconstructing the final objective function for subsequent extreme value optimization control:

[0136] ;

[0137] In the objective function formula, This is the objective function measure value for extreme value optimization control. The temperature of the electrolyte in real time is obtained by a temperature sensor installed inside the electrolytic cell, and the unit is degrees Celsius. This is the preset upper limit threshold for safe process temperature. This is the weighting coefficient for temperature exceeding the limit penalty. The hydrogen evolution intensity equivalent current is calculated by the gas flow meter at the top of the tank. This is the nonlinear amplification factor for hydrogen evolution. This represents the penalty weighting coefficient for the hydrogen evolution side reaction. The system will maximize this objective function. As the global iterative direction of the extreme value optimization algorithm.

[0138] This step introduces a multidimensional nonlinear penalty mechanism involving temperature and side reactions, transforming a single energy efficiency index into a global optimization target that integrates safety and purification quality. This ensures high-purity precipitation of heavy metals at extremely low concentrations and suppresses ineffective heat consumption.

[0139] The perturbation and gradient calculation module is used to superimpose a sinusoidal perturbation signal onto the duty cycle and peak current control signals corresponding to the electrolytic recovery unit, and calculate the gradient estimate of the objective function based on the response fluctuation of the objective function to the sinusoidal perturbation signal.

[0140] Furthermore, in the disturbance and gradient calculation module, the step of superimposing a sinusoidal disturbance signal onto the duty cycle and peak current control signals corresponding to the electrolytic recovery unit includes:

[0141] Initialize two local oscillators with orthogonal phase and coprime frequencies to generate dual-channel high-frequency sinusoidal perturbation carriers for decoupling multidimensional control variables;

[0142] The local variance and signal-to-noise ratio of the objective function are extracted, and the perturbation amplitude is dynamically calculated through an adaptive scaling operator, so that the perturbation energy has a nonlinear inverse mapping relationship with the current degree of convergence of the optimization.

[0143] The scaled dual-channel high-frequency sinusoidal perturbation carrier is synchronously fed forward and injected into the DC control reference baseline of duty cycle and peak current using a digital-to-analog conversion matrix to generate a synthetic drive signal.

[0144] Furthermore, in the perturbation and gradient calculation module, the step of calculating the gradient estimate of the objective function includes:

[0145] The transient response sequence of the objective function output by the electrolytic recovery unit after being excited by a control signal superimposed with a sinusoidal disturbance signal is extracted, and the DC bias baseline and low-frequency drift components are filtered out using a high-pass digital filter.

[0146] The transient response sequence after filtering out the DC component is multiplied in the time domain synchronously with the sinusoidal disturbance signal, and the sensitivity information of the objective function is shifted to the baseband spectrum through the synchronous demodulation mechanism.

[0147] The baseband spectrum data is input into a first-order inertial low-pass integral filter to smoothly extract the DC effective component in the demodulated signal, and the approximate first-order partial derivative vector of the objective function with respect to each control variable is constructed as the gradient estimate.

[0148] Specifically, in the perturbation and gradient calculation module for extreme value optimization in the electrolytic recovery unit, the system first initializes two local oscillators with orthogonal phases and coprime frequencies. The first perturbation carrier frequency used for duty cycle adjustment is set to... The second disturbance carrier frequency used for peak current regulation is To ensure complete decoupling of multidimensional control variables in subsequent demodulation, the two frequencies must strictly satisfy the coprime condition. The system dynamically calculates the disturbance amplitude using an adaptive scaling operator, and extracts the local variance and signal-to-noise ratio of the extreme value optimization control objective function in real time, resulting in a nonlinear inverse mapping relationship between the disturbance energy and the current optimization convergence degree. The dynamic amplitude calculation formula is as follows:

[0149] ;

[0150] In this amplitude scaling formula, For the calculated first The dynamic disturbance amplitude of each control variable, when Equal to the disturbance amplitude corresponding to the duty cycle, when It equals the disturbance amplitude of the corresponding peak current. This is the preset initial reference disturbance amplitude. This is the amplitude attenuation adjustment coefficient. The local variance of the objective function, which includes real-time current efficiency characteristics, within the historical sliding window is used. A scaled dual-channel high-frequency sinusoidal perturbation carrier is synchronously fed forward and injected onto the DC control reference baseline, which contains the duty cycle and peak current, using a digital-to-analog conversion matrix to generate a synthetic drive signal.

[0151] ;

[0152] ;

[0153] In the formula for synthesizing the driving signal This is the duty cycle control signal that is ultimately output to the pulse power supply, and its dimension is percentage. This is the peak current control signal that is ultimately output to the pulse power supply, and its dimension is ampere. and These are the duty cycle reference value and peak current reference value of the extreme value optimization algorithm in the current iteration cycle, respectively. This is a time variable for system operation.

[0154] After applying the aforementioned synthetic driving signal to the electrolytic recovery unit, the system extracts the transient response sequence of the output objective function. A high-pass digital filter is used to remove the DC bias baseline and low-frequency process drift components from the sequence, yielding the change in the target response after high-pass filtering. The transient response sequence after filtering out the DC component is then multiplied in the time domain synchronously with the corresponding sinusoidal perturbation signal. The sensitivity information of the objective function is then shifted to the baseband spectrum via a synchronous demodulation mechanism.

[0155] ;

[0156] ;

[0157] In this demodulation formula, The time-domain multiplication demodulated signal is the duty cycle dimension. The demodulated signal is a time-domain multiplication of the peak current dimension. The obtained baseband spectrum data is input into a first-order inertial low-pass integral filter to smoothly extract the DC effective component from the demodulated signal. The approximate first-order partial derivative vector of the objective function with respect to each control variable is constructed as the final gradient estimate.

[0158] ;

[0159] ;

[0160] In the gradient estimation formula, and These represent the approximate first-order partial derivatives of the objective function with respect to the duty cycle control signal and the peak current control signal, respectively, which are the real-time gradient information required for system optimization. The inertial time constant of the low-pass filter, This represents the DC gain of the filter.

[0161] This step, through the injection and synchronous demodulation extraction of high-frequency carrier signals, enables the online dynamic calculation of the partial derivatives of the objective function, solving the technical problem that the gradient cannot be directly calculated due to the unknown nonlinear mechanism during the electrolytic purification of complex electroplating solutions.

[0162] The optimization iterative control module is used to update and calculate the duty cycle and peak current of the next control cycle based on the gradient estimate and the preset optimization step size, so as to adjust the pulse power output of the electrolytic recovery unit.

[0163] Furthermore, in the optimization iterative control module, the steps for updating and calculating the duty cycle and peak current for the next control cycle include:

[0164] An adaptive gradient optimization algorithm with a first-order momentum acceleration mechanism is introduced to perform exponential moving average filtering on the gradient estimate, thereby suppressing local gradient oscillations caused by the mass transfer delay of the electroplating solution.

[0165] Multiply the filtered smooth gradient vector by the feedforward dynamic learning rate step size, and use the gradient ascent optimization rule to perform incremental iterative updates on the duty cycle and peak current basic parameters at the current moment.

[0166] The updated parameters are forcibly truncated using a hardware safety boundary projection operator. After verifying that the parameters do not exceed the physical limits of the electrolytic power supply, the parameters are overwritten to the pulse control register to complete the pulse power supply output regulation of the electrolytic recovery unit.

[0167] Specifically, in the iterative update phase of the extremum optimization in the electrolytic recovery unit, the system receives the first gradient estimate of the pulse power supply duty cycle and the second gradient estimate of the peak current calculated and output from the previous perturbation cycle. Due to the long liquid-phase mass transfer delay characteristic inside the heavy metal wastewater electrolyzer, directly using the original gradient estimate would cause local high-frequency oscillations in the optimization path. Therefore, an adaptive gradient optimization algorithm with a first-order momentum acceleration mechanism is introduced to perform exponential moving average filtering on the above gradient estimates.

[0168] ;

[0169] ;

[0170] In this exponential moving average filter formula, For the first The duty cycle smoothing gradient is calculated for each control cycle. The peak current smoothing gradient is calculated for the same period. and These are the corresponding smoothing gradient values ​​for the previous control cycle. This is the original gradient estimate of the duty cycle input for this control cycle. This is the original gradient estimate of the peak current. A first-order momentum decay coefficient, set between zero and one, is used to adjust the weighting of historical gradient information retention. After obtaining the smoothed gradient vector, the system multiplies it by the feedforward dynamic learning rate step size and uses the gradient ascent optimization algorithm to perform incremental iterative updates on the duty cycle and peak current baseline parameters at the current moment. The feedforward dynamic learning rate exhibits an adaptive decay law as the system optimization process progresses, balancing initial search speed with later convergence accuracy. The specific calculation formula for incremental iterative updates is:

[0171] ;

[0172] ;

[0173] In the incremental iterative update formula The pulse duty cycle base parameter, which is calculated and prepared for use in the next control cycle, is updated and is expressed as a percentage. This is the basic parameter for the duty cycle of the actual output in the current cycle. The basic parameters of the peak current for the next control cycle are updated to be in the form of amperes. These are the basic parameters for the peak current of the current cycle. For the first Duty cycle of each control cycle, feedforward dynamic learning rate step size This corresponds to the peak current feedforward dynamic learning rate step size. To prevent the updated electrical parameters from exceeding the hardware's capacity and causing short circuits or equipment damage, a hardware safety boundary projection operator is used to forcibly truncate the updated parameters. The system pre-sets the physical lower and upper limits of the pulse power supply duty cycle, and simultaneously sets the safe lower and upper limits of the peak current. The projection calculation formula for the forced truncation protection is as follows:

[0174] ;

[0175] ;

[0176] In the formula for safe projection, This is the final valid duty cycle output command after boundary verification. and These represent the minimum and maximum physical limits of the system's allowed duty cycle, respectively. This is the final, valid peak current output command. and These represent the minimum safe peak current and the maximum rated peak current allowed to be output by the electrolytic power supply, respectively. After verifying that the system parameters do not exceed the physical limits of the electrolytic power supply, the final legal duty cycle output command and the final legal peak current output command are overwritten into the dedicated control register of the pulse power controller via the industrial communication bus, completing the low-level hardware adjustment of the pulse power supply output of the electrolytic recovery unit.

[0177] This step, through momentum gradient smoothing and hard boundary protection mechanisms, ensures the stability and convergence of the electrical parameter optimization trajectory, and avoids equipment downtime due to command exceeding limits.

[0178] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating, characterized in that, include: The operation parameter acquisition module is used to obtain the real-time transmembrane pressure difference, membrane flux and water temperature of the membrane separation unit, and construct the observation sequence according to the time dimension; The hidden state decoding module is used to input the observation sequence into the hidden Markov model and decode and output the hidden state of the membrane separation unit through the Viterbi algorithm. The hidden state includes a clean state, a reversible contamination state, and an irreversible contamination state. An adaptive backwash control module is used to control the membrane separation unit to perform a pulse backwash operation when it is determined that the hidden state at the current moment is the reversible contamination state and the duration reaches a preset time threshold. The objective function construction module is used to introduce the heavy metal concentrate discharged from the membrane separation unit into the electrolytic recovery unit, obtain the real-time metal recovery rate and power consumption of the electrolytic recovery unit, calculate the real-time current efficiency and use it as the objective function for extreme value optimization control. The perturbation and gradient calculation module is used to superimpose a sinusoidal perturbation signal onto the duty cycle and peak current control signals corresponding to the electrolytic recovery unit, and calculate the gradient estimate of the objective function based on the response fluctuation of the objective function to the sinusoidal perturbation signal. The optimization iterative control module is used to update the duty cycle and peak current of the next control cycle according to the gradient estimate and the preset optimization step size, so as to adjust the pulse power output of the electrolytic recovery unit. The step of superimposing a sinusoidal disturbance signal onto the duty cycle and peak current control signal corresponding to the electrolytic recovery unit includes: Initialize two local oscillators with orthogonal phase and coprime frequencies to generate dual-channel high-frequency sinusoidal perturbation carriers for decoupling multidimensional control variables; The local variance and signal-to-noise ratio of the objective function are extracted, and the perturbation amplitude is dynamically calculated through an adaptive scaling operator, so that the perturbation energy has a nonlinear inverse mapping relationship with the current degree of optimization convergence. The scaled dual-channel high-frequency sinusoidal perturbation carrier is synchronously fed forward and injected into the DC control reference baseline of the duty cycle and peak current using a digital-to-analog conversion matrix to generate a synthetic drive signal.

2. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the parameter acquisition module, the step of constructing the observation sequence according to the time dimension includes: The Kalman filter operator is used to perform data fusion and denoising on the real-time transmembrane pressure difference, membrane flux and water temperature, and to remove high-frequency random noise caused by fluid pulses; Based on the maximum and minimum value normalization mapping function, the denoised physical parameters are projected to the standard numerical range to suppress the weight interference of physical dimensions on probability calculation. A sliding time window with a preset time step is constructed, and the normalized data points are concatenated into a time-series feature matrix according to the sampling order to generate the observation sequence.

3. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the hidden state decoding module, the step of inputting the observation sequence into the hidden Markov model includes: Based on the physical decay characteristics of membrane fouling caused by heavy metal wastewater, the state transition probability matrix of the hidden Markov model is initialized and constructed to constrain the evolution logic between different fouling states. A Gaussian mixture model is introduced to construct the emission probability density function. By using a weighted linear combination of Gaussian distributions, the statistical probability of generating the current observation data under a specific hidden state is quantified. Prior sample data of historical membrane fouling life cycle are extracted, and unsupervised parameter learning is performed on the state transition probability matrix and the emission probability density function using the Baum-Welch algorithm to complete the model calibration.

4. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the hidden state decoding module, the step of decoding and outputting the hidden state of the membrane separation unit using the Viterbi algorithm includes: Based on the initial state probability vector and emission probability density function of the Hidden Markov Model, calculate the local forward log probability of each hidden state in the observation sequence at the initial time step, and initialize the starting point of the dynamic programming. Initiate the dynamic programming path recursion process, calculate the path with the maximum cumulative probability to any candidate hidden state at the current time step, and use the backtracking pointer matrix to record the local optimal predecessor node; Extract the endpoint state with the highest global cumulative probability at the end of the observation sequence, reconstruct the optimal state transition path by tracing back along the backtracking pointer matrix, and output the hidden state at the current time corresponding to the endpoint.

5. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the adaptive backwash control module, the step of controlling the membrane separation unit to perform pulse backwashing includes: Based on the cumulative duration of the reversible fouling state and the corresponding state probability, combined with the preset mechanical strength tolerance limit of the membrane material, the target pressure peak and pulse frequency of the backwash fluid are dynamically calculated. The target pressure peak and pulse frequency are converted into a pneumatic command sequence for a programmable logic controller, and a closed-loop control algorithm is used to drive the backwash pump and the multi-way reversing valve to switch in coordination. The slope of the transmembrane pressure difference recovery during the backwash cycle is monitored in real time. By calculating the exponential smoothing of the recovery slope, the preset time threshold for the next operating cycle is adaptively corrected to complete the backwash closed-loop control.

6. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the objective function construction module, the step of calculating the real-time current efficiency and using it as the objective function for extreme value optimization control includes: The theoretical metal deposition rate within a time slice is calculated using an ion concentration difference sensing model and Faraday's law, and the actual effective metal deposition rate is estimated by combining cathode dynamic weighing data. The transient output voltage and current sequence of the pulse power supply module is collected, the active power consumption in the time slice is calculated using the discrete integral operator, and the actual metal deposition per unit energy consumption is calculated to obtain the real-time current efficiency. A multi-objective optimization mapping function based on a penalty mechanism is constructed. On the basis of maximizing the real-time current efficiency, a nonlinear penalty constraint term characterizing the cell temperature rise and hydrogen evolution intensity is introduced to reconstruct the objective function.

7. The resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the perturbation and gradient calculation module, the step of calculating the gradient estimate of the objective function includes: The transient response sequence of the objective function output by the electrolytic recovery unit after being excited by a control signal superimposed with the sinusoidal disturbance signal is extracted, and the DC bias baseline and low-frequency drift component are filtered out using a high-pass digital filter. The transient response sequence after filtering out the DC component is multiplied in the time domain synchronously with the sinusoidal disturbance signal, and the sensitivity information of the objective function is shifted to the baseband spectrum through the synchronous demodulation mechanism. The baseband spectrum data is input into a first-order inertial low-pass integral filter to smoothly extract the DC effective component in the demodulated signal, and the approximate first-order partial derivative vector of the objective function with respect to each control variable is constructed as the gradient estimate.

8. A resource-based treatment system for heavy metal-containing wastewater from automotive wiring harness electroplating according to claim 1, characterized in that, In the optimization iterative control module, the step of updating and calculating the duty cycle and peak current for the next control cycle includes: An adaptive gradient optimization algorithm with a first-order momentum acceleration mechanism is introduced to perform exponential moving average filtering on the gradient estimate, thereby suppressing local gradient oscillations caused by the mass transfer delay of the electroplating solution. Multiply the filtered smooth gradient vector by the feedforward dynamic learning rate step size, and use the gradient ascent optimization rule to perform incremental iterative updates on the duty cycle and peak current basic parameters at the current moment. The updated parameters are forcibly truncated using a hardware safety boundary projection operator. After verifying that the parameters do not exceed the physical limits of the electrolytic power supply, the parameters are overwritten to the pulse control register to complete the pulse power supply output regulation of the electrolytic recovery unit.

Images