Heavy metal waste liquid electro-deposition recycling treatment method

Abstract

The invention relates to a heavy metal waste liquid electro-deposition cyclic utilization treatment method which comprises the following steps: establishing an interface charge migration model based on waste liquid ion composition and a polarization curve, and quantifying migration impedance characteristics of Cu < 2 + >, Zn < 2 + > and Ni < 2 + >. Through progressive potential scanning, ions are subjected to gradient deposition according to migration resistance, preferential adsorption and reduction are induced through pulse frequency adjustment, and a metal reduction potential window is matched. A movable electrode structure is adopted to transversely move a deposition reaction area on a microscale, and reverse high-frequency pulses are applied in a deposition intermittent period or an electrode migration period to trigger stripping and crystallization of a metal deposition layer and removal of a passivation layer. And continuously depositing the electrode to form a dynamic circulation closed loop. A micro conductivity / ion selective electrode sensor array is integrated, and the concentration and intensity distribution of residual heavy metal ions in the waste liquid are evaluated. The concentration mode is switched according to the sensing result, and the optimized concentration reaches a redeposition threshold value. The AI scheduling system controls the reflux proportion of the concentrated solution, ensures the concentration matching reflux, improves the treatment efficiency and reduces the waste.

Description

Technical Field

[0001] The invention relates to a heavy metal waste liquid electrodeposition recycling treatment method. Background Art

[0002] The heavy metal waste liquid electroplating treatment method currently used in the industry can, to a certain extent, achieve the recovery of some heavy metal resources and reduce waste liquid emissions. However, from the perspective of the overall process chain, automation control level, resource recovery efficiency, operation stability, and system intelligence, there are still many obvious deficiencies and technical bottlenecks, which limit its promotion and application in high-complexity, high-load industrial scenarios and its ability to maximize resource recovery.

[0003] From the perspective of the process itself, most current electrodeposition methods still use traditional constant current or constant voltage control modes, lacking the ability to dynamically adjust the type of ions, electrode polarization state, and concentration gradient reaction conditions. In actual operation, wastewater often contains multiple heavy metal ions at the same time, such as copper, nickel, zinc, and chromium. The reduction potentials of these ions are close and they are easy to co-deposit, resulting in the presence of metal mixing, severe hydrogen evolution side reactions, and increased electrode passivation in the final deposition product, thereby reducing the quality of metal purification and current utilization efficiency. Especially in scenarios where the impurity concentration fluctuates greatly or the inlet flow rate is unstable, traditional electrodeposition systems are unable to adapt to changes in working conditions in real time, and often require manual shutdown adjustments or frequent replacement of electrodes, resulting in a series of problems such as discontinuous operation, heavy labor burden, and rapid equipment wear. In the wastewater concentration and reflux links, traditional methods mostly rely on a single concentration technology such as vacuum evaporation or reverse osmosis. Their process paths are fixed and have poor adjustability. When faced with high-concentration, high-salinity, or high-temperature and high-conductivity wastewater, membrane blockage, decreased heat transfer efficiency, and metal ion precipitation before deposition often occur. In addition, the traditional process lacks an automatic selection mechanism based on real-time liquid composition, and is unable to intelligently switch the concentration path according to the ion composition, conductivity, pH and temperature parameters of different batches of waste liquid, resulting in uneven energy consumption and unstable efficiency. It may even lead to the embarrassing situation that the concentrated liquid cannot meet the deposition requirements and must be processed or discharged again, which not only wastes resources but also increases operating costs and sewage discharge pressure.

[0004] Currently, waste liquid electrodeposition systems generally use fixed-ratio control or manual experience-based judgment to control reflux, setting the reflux ratio of each batch of concentrate at 70% and the discharge ratio at 30%. This crude control method completely ignores the key influencing factors such as the composition differences of the concentrate itself, the actual deposition efficiency, and the current load of the system. When the metal ion concentration in the concentrate is low or the impurity content is too high, blind reflux not only fails to form effective deposition, but may also increase the burden on the main deposition system, leading to increased energy consumption, increased electrode polarization, and even inducing abnormal shutdown of the deposition system. Maintaining a low reflux ratio during periods of good concentrate quality and strong sedimentation tank absorption capacity wastes resources that should be used efficiently. The fixed-ratio system not only fails to maximize resource utilization efficiency, but also makes the system's ability to respond to sudden load fluctuations and changes in liquid composition almost zero.

[0005] Most systems are either completely blank or in their infancy when it comes to intelligent and predictive control of wastewater treatment. They lack tools for proactively assessing deposition efficiency trends, changes in loading capacity, and fluctuations in ion concentrations. Most systems operate in a passive response mode, adjusting parameters only after problems arise, completely failing to implement proactive control actions based on predictive results. Traditional systems lack the ability to identify and proactively adjust issues as deposition efficiency declines, metal concentrations fluctuate, and impurity accumulation gradually accumulate. This often leads to deposition failure, electrode replacement, or system downtime, severely impacting process continuity and production cadence. Data collection also suffers from significant shortcomings in terms of dimensionality and granularity. Most traditional systems monitor only the total influent concentration and effluent conductivity, failing to selectively detect specific ion species. They lack the ability to accurately capture deposition status on the electrode surface, changes in electrochemical reaction interfaces, or local polarization levels. Furthermore, they lack mechanisms for coupling ion behavior, liquid characteristics, and process operating conditions. This data gap leads to closed control logic and ambiguous judgment criteria, trapping the entire process system in a vicious cycle of high energy consumption, low recovery, high emissions, and heavy reliance on manual labor. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for recycling and treating heavy metal waste liquid by electrodeposition, thereby solving some of the drawbacks and deficiencies pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solution, which includes the following steps:

[0008] Based on the waste liquid ion composition and polarization curve, an interfacial charge migration model was established to quantify Cu 2+ 、Zn 2+ 、Ni 2+Ion migration impedance characteristics; applying a progressive potential scan in the deposition area causes different ions to deposit sequentially according to the migration resistance gradient, forming a transient high charge density band on the electrode surface. By adjusting the pulse frequency, the preferential adsorption and reduction of ions is induced; by adjusting the migration rate to match the metal reduction potential window, the metal is preferentially reduced;

[0009] The deposition reaction zone is laterally moved on a microscale through a movable electrode structure. During the deposition interval or electrode migration period, a reverse high-frequency pulse is applied to trigger partial stripping or recrystallization of the metal deposition layer, removing the passivation layer. One portion of the electrode is scheduled for continuous deposition while the other portion is cleaned and metal stripped, forming a dynamic closed loop.

[0010] An integrated micro-conductivity / ion-selective electrode sensor array evaluates the concentration and intensity distribution of residual heavy metal ions in the waste liquid in real time; switches the concentration mode based on the perception results of low-pressure membrane concentration, electrodialysis or vacuum evaporation to optimize the concentration to reach the re-deposition threshold; uses the AI scheduling system to control the proportion of concentrated liquid to be reintroduced into the deposition area and the proportion to be sent to the waste liquid recovery or discharge unit to ensure that the concentration matches the reflux.

[0011] Furthermore, the method for real-time evaluation of the concentration and intensity distribution of residual heavy metal ions in the wastewater includes:

[0012] A multi-channel microsensor array is deployed at multiple locations in the wastewater treatment device. Each sensor unit includes a conductivity sensor and multiple ion-selective electrodes, which are used to collect the total ion concentration, conductivity, and active concentration of heavy metal ions in the wastewater.

[0013] Data acquisition and processing module, used to collect the original electrical signals of each sensor and perform temperature and pH drift compensation;

[0014] Multi-channel data fusion module, which is used to synchronously process the data of each ion channel through feature extraction and cross-response modeling algorithms, and output the concentration vector and total intensity characteristics of heavy metal ions;

[0015] A spatial distribution modeling module is used to establish a two-dimensional or three-dimensional concentration distribution map of heavy metal ions in the deposition system based on the layout position of the array sensor in the physical space and the waste liquid flow parameters;

[0016] The deposition feedback control module is used to determine whether the current deposition area is in a concentration depletion, polarization or reflux trigger state based on the above concentration and distribution map results, and output control instructions to the electrodeposition main control unit or concentration loop.

[0017] Furthermore, the ion-selective electrode adopts a customized multi-layer composite membrane structure; the data fusion module uses a principal component analysis algorithm or support vector regression to perform feature dimension reduction and concentration prediction on multiple channel data.

[0018] Furthermore, the spatial distribution modeling module uses interpolation reconstruction methods (including inverse distance weighting, Kriging interpolation and neural network spatial mapping algorithms) based on ion concentration data collected by multiple point sensors to accurately predict and visualize the heavy metal ion concentration in the electrodeposition system in two-dimensional or three-dimensional space;

[0019] The prediction results are input into the deposition feedback control module, which automatically adjusts the system operating parameters based on whether the concentration in each area is below the critical value, including the concentrate reflux ratio, electrode polarity reversal, dynamic current density adjustment, and liquid reflux path switching, to maximize deposition efficiency and optimize resource utilization.

[0020] To calculate the dynamic intensity distribution of heavy metal ions in the deposition reaction zone, a responsive concentration evolution function model was constructed to quantify the nonlinear response relationship between time, space, and concentration. The model is defined as follows:

[0021]

[0022] in:

[0023] Φ(x,y,t): represents the cumulative spatial response function value at the two-dimensional coordinate x,y to time t, which is used to evaluate the deposition capacity trend of the point; ω(x,y,τ): represents the effective activity of metal ions per unit volume of waste liquid at the point at time τ (obtained by the sensor); ψ(x,y,τ): represents the mass transfer rate function of the local deposition area, reflecting the degree of ion migration restriction; λ: system response sensitivity coefficient, set according to flow rate, electrode material and stirring state, 0<λ<1; α: nonlinear response adjustment index, which controls the weighting degree of the model for high-concentration areas, and α>1 indicates a preference for high-concentration deposition. It represents the contribution rate of changes in mass transfer conditions to deposition behavior during the time evolution process; the integral upper limit t is the end point of system sampling in the current control cycle.

[0024] Application and control logic:

[0025] 1. When the Φ(x,y,t) value of a certain area continues to decrease and falls below the set threshold Φ min When , the system automatically determines that the metal ion supply in the area is tending to be exhausted;

[0026] 2. The feedback module immediately performs one or more of the following control actions:

[0027] Start the concentrate reflux pump to guide the liquid in the high concentration area back to that area;

[0028] Switching electrode polarity to break local polarization or anodic passivation;

[0029] Lower the current density to slow down the local deposition rate and wait for the ion supply to be replenished;

[0030] If recovery is not possible for a long time, the system in that area will be switched to a low-efficiency processing mode to avoid wasting resources.

[0031] Furthermore, the deposition feedback control module includes a prediction submodule based on a time series learning algorithm, which uses a recurrent neural network to predict the future trend of metal ion concentration for adjusting the control strategy in advance.

[0032] Furthermore, the method for implementing the concentration matching reflux includes:

[0033] A concentration device switches between three concentration methods: low-pressure membrane concentration, electrodialysis, or vacuum evaporation, based on the ion concentration, conductivity, temperature, and pH sensing results of the heavy metal waste liquid;

[0034] An AI scheduling system that receives data from sensors and performs the following operations:

[0035] (1) Real-time evaluation of the residual concentration and redeposition value of heavy metal ions in the concentrate;

[0036] (2) Construct a feasibility prediction model for redeposition of concentrate based on historical operating data;

[0037] (3) controlling a set proportion of the concentrated liquid to flow back to the electrodeposition unit, and sending the rest to the waste liquid recovery or discharge unit;

[0038] A ratio control unit is used to dynamically adjust the volume ratio of concentrated liquid return and discharge according to the instructions of the AI scheduling system so that the concentration matches the return.

[0039] Furthermore, the concentrating device selects a concentration mode based on real-time feedback parameters according to preset energy consumption and efficiency thresholds, and maintains the metal ion activity of the concentrated liquid not lower than a set value during switching between different modes.

[0040] Furthermore, the concentrate redeposition benefit evaluation model in the AI scheduling system comprehensively analyzes multi-factor data in the concentrate, predicts whether it has sustainable deposition value under current working conditions, and classifies the concentrate into different reuse priority labels, thereby determining whether it should be returned to the deposition area or enter the waste liquid disposal process to maximize resource recovery.

[0041] In order to realize the prediction of sedimentation benefits under multi-parameter coupling, the concentrated liquid benefit density function model is adopted, which is defined as follows:

[0042]

[0043] in:

[0044] Γ(t): represents the comprehensive redeposition benefit score (benefit accumulation function) of the unit volume of the concentrate in the time period [t0, t], and its value is used to judge whether it has the value of reflow; θ1, θ2, θ3: are the system self-learning weight coefficients, representing the ion purity contribution factor, impurity suppression factor and deposition ability enhancement factor respectively, and the value is dynamically adjusted; χ(i t ): The target heavy metal ion concentration in the concentrate constitutes the vector i t =[c Cu ,c Zn ,c Ni ,…] is a weighted evaluation function to measure the sedimentation potential; ν(j t ) is the impurity component influence function, which is based on the vector j consisting of the total amount of non-target ions, colloids, and organic matter in the current liquid. t , modeling the inhibitory effect on deposition behavior; ρ(k t ): represents the absorption capacity function of the upstream deposition unit, which is composed of vector k based on the current load state, electrode state, current density and other parameters t , the higher the value, the more suitable it is for continued deposition; μ(u t ): is the conductivity change function of the concentrated solution, vector u t represents the time-varying conductivity behavior, is the conductivity change rate, reflecting the dynamic response speed of the system deposition.

[0045] Control logic and actual function:

[0046] 1. The AI scheduling system calculates Γ(t) in real time for each batch of concentrate within a given time interval;

[0047] 2. If Γ(t)>Γ th (system-set threshold), the concentrate is judged to have high-efficiency redeposition potential and is preferentially directed back to the electrodeposition area;

[0048] 3. If Γ(t) is in the middle range, part of the wastewater is refluxed and part of the wastewater is sent to the secondary treatment unit (e.g., deep concentration, chemical conditioning).

[0049] 4. If Γ(t)<Γ min , indicating that the metal efficiency of this batch of concentrate is low or the impurity suppression is strong, and it is determined to be a non-redepositable resource and is directed to waste liquid discharge or regeneration treatment.

[0050] Furthermore, the proportional control unit is composed of a variable frequency pump valve, a flow controller or an electromagnetic flow regulator, and adjusts the ratio of reflux and discharge in real time under the control of the AI scheduling system.

[0051] Furthermore, the AI scheduling system further includes a time series prediction module, which uses a learning model based on historical concentrate recovery rate and deposition conversion rate to predict system load fluctuations in future time periods and adjust the reflux ratio in advance to match the deposition efficiency.

[0052] The present invention has significant beneficial effects in multiple dimensions, including technical architecture, intelligent control, resource utilization efficiency, and system stability, which are specifically reflected in the following aspects:

[0053] By integrating selective electrodeposition, concentrated solution reflux control and ion concentration prediction mechanism, the Cu 2+ 、Ni 2+ 、Zn 2+ The efficient separation and step-by-step deposition of multiple heavy metals achieves a metal recovery rate of over 90%, significantly higher than traditional deposition methods (typically 70% to 80%). The introduction of an AI scheduling system and time series prediction module not only identifies changes in wastewater composition in real time, but also predicts deposition load trends and adjusts operating parameters in advance, enabling proactive decision-making and rapid response, effectively avoiding local polarization, passivation, or metal waste.

[0054] By setting energy consumption and efficiency thresholds and integrating real-time sensor data, the concentrator automatically switches between electrodialysis, low-pressure membrane, and vacuum evaporation, flexibly selecting the optimal mode based on wastewater characteristics. This reduces energy consumption by 10% to 20%, resulting in a corresponding decrease in operating costs. Utilizing a concentrate benefit evaluation model and a ratio control unit, the system dynamically adjusts the return and discharge ratio based on the sedimentation value of each batch of concentrate, achieving dual optimization of efficient reuse and wastewater discharge. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the heavy metal waste liquid electrodeposition recycling treatment method of the present invention.

[0056] Figure 2 This is a flow chart of the method for real-time evaluation of the concentration and intensity distribution of residual heavy metal ions in wastewater according to the present invention.

[0057] Figure 3 Flow chart of the method for realizing concentration matching reflux of the present invention DETAILED DESCRIPTION

[0058] The following is a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings.

[0059] Combined process Figure 1 As shown, the Cu in wastewater is collected by a multi-point online sensor array.2+ 、Zn 2+ 、Ni 2+ The key data of heavy metal ion initial concentration, conductivity, electrode reaction potential, pH value and temperature. These raw data undergo data preprocessing steps, including outlier removal, drift correction, temperature compensation and normalization, to ensure that various parameters can be accurately modeled at the same physical scale. In particular, in the process of obtaining the electrode polarization characteristic curve, the linear sweep voltammetry and step potential method are used to experimentally measure the polarization response behavior of each heavy metal ion, and compared with the known standard electrode potential to obtain the actual migration impedance performance of the ion at a specific concentration and electric field strength. The output of this stage is a data set based on the voltage-current response curve with the ion type as the dimension.

[0060] The model constructed based on the above data is an interfacial charge migration model, which simulates the electrochemical kinetic behavior of different ions when approaching the electrode interface by comparing the ion migration impedance characteristics. The model adopts an improved electrochemical effective circuit method, which includes a variable structure RC network submodule to simulate the response hysteresis of ions in the double layer formation process, and further uses Laplace transform to convert the impedance spectrum into a charge accumulation rate expression in the time domain. At the same time, an adaptive weighting function is introduced to characterize the nonlinear relationship between the electron transfer rate of different ions and the interfacial adsorption tendency. The model training process uses experimental reflux data and simulation data for comparison, and optimizes through minimum residual fitting. The training algorithm used is a quasi-Newton method with regularization constraints to ensure model convergence and generalization ability. A progressive potential scanning control strategy is applied in the electrodeposition operation area, that is, the entire metal reduction range is swept from low potential to high potential, but it is not advanced at a linear rate. Instead, it is combined with the response rate gradient of each ion in the charge migration model for nonlinear adjustment. This strategy can effectively control the rate of change of the electric field intensity on the electrode surface, promoting ions with low migration impedance (such as Cu 2+ ) are preferentially deposited at an earlier stage, while ions with higher migration resistance (such as Zn 2+ 、Ni 2+ ) are deposited sequentially in subsequent stages. This approach avoids metal co-deposition and alloy confusion, significantly improving metal purification results.

[0061] In addition, during the deposition process, the system will form a transient high charge density band on the electrode surface, that is, at certain time points, the local potential is modulated by short-term high-frequency pulses, shortening the ion migration path and forcing it to be close to the electrode surface. During this process, the system dynamically adjusts the pulse frequency and amplitude to match the interface adsorption characteristics and reduction potential window of various metal ions, so that the electron transfer and deposition process can be completed under optimal conditions. In particular, in terms of matching the migration rate and the metal reduction window, the system monitors the ion concentration gradient changes and deposition rate in the deposition area in real time, and by fine-tuning the pulse waveform parameters, different ions are always in their most effective deposition range, ensuring the goal of preferential metal reduction, suppression of hydrogen evolution reaction, improved deposition efficiency and reduced energy consumption.

[0062] In response to industrial bottlenecks such as electrode passivation, decreased deposition efficiency, and frequent electrode replacement, a dynamic deposition system based on a movable electrode structure and an intelligent flow control mechanism is proposed. Through physical structure innovation and algorithm-driven electrode self-cleaning closed loop, long-term, efficient, and automated operation is achieved. The system introduces a movable electrode structure at the physical level. That is, on the basis of the traditional fixed deposition tank, an electrode module with a linear slide or flexible support component is designed to enable the electrode to move laterally on a microscale within the deposition area. This movement is not a macroscopic displacement, but a periodic offset in the lateral direction (perpendicular to the main current direction) with an amplitude of 0.1 to 2 mm. The movement is mainly driven by a servo control unit, and its scheduling frequency is determined by the upper AI controller according to the deposition rate and local ion concentration distribution. The core purpose of this lateral movement design is to break the concentration extremes and polarization phenomena caused by long-term deposition in a certain area of the electrode surface, and to maintain the activity uniformity of the electrode reaction surface.

[0063] During system operation, when it is detected that the deposition behavior on the surface of a certain electrode enters a state of decreased efficiency (usually manifested as a decrease in current density, intensified local potential fluctuations, and increased polarization voltage), the system schedules it to enter a deposition interval. During this period, the electrode will temporarily stop the output of conventional deposition current and receive the excitation of a set of reverse high-frequency pulse signals. These pulse signals are generated by a high-frequency inverter module, and the frequency is usually set between 10 and 100 kHz. The waveform can be a positive and negative symmetrical or asymmetrical rectangular wave, and the amplitude is not higher than 80% of the normal deposition voltage. The function of these reverse high-frequency pulses is to trigger the existing metal deposition layer on the surface of the electrode to produce micro-cracking or recrystallization, so that the surface grains are reorganized and the crystal defects are reduced, thereby breaking the reaction blockage caused by the passivation film or dense deposition layer. In addition, these pulses also have a partial stripping effect, which frees some of the metal particles deposited on the surface into the electrolyte, which is then collected and processed by the downstream filtration or recovery system.

[0064] In the electrode circulation mechanism, the system is configured with multiple deposition units running in parallel, dynamically switching between deposition-migration-cleaning states. The AI scheduling system establishes an electrode operation state matrix based on sensor data (including each electrode's current response, potential change, metal deposition rate, and ion concentration gradient). It then assigns work roles based on real-time analysis results: some electrodes continue to perform deposition tasks, while others are switched to cleaning mode or standby mode, creating a scheduling rhythm similar to the circulating workstations in a production line.

[0065] To achieve precise control, the system integrates a dynamic electrode scheduling optimization model. This model first collects structured data from the sensor network, including the amplitude of deposition current fluctuations per unit time, the trend of electrode surface impedance changes, the mass concentration of deposited metal in the liquid phase, and the average electromigration velocity of metal ions. Before entering the model, the raw data undergoes normalization and filtering, using methods such as bandpass filtering to remove high-frequency interference and a mean sliding window to stabilize the signal curve. The processed data is fed into the model's feature extraction layer, where a convolutional neural network (CNN) is used to identify each electrode's current deposition phase and passivation trend. The scheduling model consists of two main components: an electrode state switching strategy network based on a reinforcement learning algorithm. This network uses rewards and penalties based on historical deposition efficiency and stripping effect feedback to optimize when to switch the state of a particular electrode. A deposition balance discriminator constructs an overall balance scoring matrix based on multi-point spatial deposition rates to determine whether lateral electrode movement is necessary to alleviate localized fouling.

[0066] The system uses micro conductivity sensors and ion selective electrode arrays deployed at key nodes of the electrodeposition device and concentration treatment path to detect target metal ions (such as Cu 2+ 、Zn 2+ 、Ni 2+ ) for real-time online collection of key parameters such as residual concentration, conductivity, and potential difference. The data sampling period is 1 to 10 times per second, with an accuracy of μS / cm. To ensure the stability and anti-interference of sensor data, the system first performs data preprocessing before the data enters the main control algorithm. This includes baseline drift correction, temperature compensation (introducing thermistor readings collected at the same time), outlier removal (by setting upper and lower limits and local sliding median detection), and normalization processing, so that all data are compatible with a unified analysis framework.

[0067] The processed data is input into the AI module of the ion strength-reflux matching analysis model. The core structure of this module consists of two parts: the first part is the concentration mode switching sub-model, which uses a decision tree structure to determine which concentration treatment path (low-pressure membrane, electrodialysis or vacuum evaporation) is most suitable for the current waste liquid based on real-time perceived liquid parameters (such as current ion concentration, conductivity, membrane pressure difference, and temperature). The model's judgment is based on historical operating data and energy efficiency feedback. If the system detects a high-concentration liquid and the temperature is moderate, the low-pressure membrane method is preferred; if the liquid conductivity is too high and the membrane is blocked, it switches to electrodialysis mode; if the concentration is extremely low but the goal is to minimize the volume of the liquid, vacuum evaporation is selected.

[0068] The second part is the key concentrate redeposition benefit prediction model, which is based on a machine learning regression algorithm and is used to determine whether a batch of concentrate has the value of redeposition. The input parameters of the model include: the concentration vector of the target ion in the concentrate (Cu 2+ 、Zn 2+ ), conductivity, pH, an estimated total impurity level (indirectly inferred from electrode response), the current electrodeposition system load capacity (i.e., the electrode's remaining adsorption capacity), and feedback on the deposition efficiency of this type of concentrate in the previous cycle. The model uses a weighted random forest regressor, combined with actual post-deposition metal recovery rates as training labels, to output a redeposition benefit score (a floating-point value between 0 and 1).

[0069] The model training process consists of two phases: an offline training phase, in which an initial model is constructed using several months of deposition operation records and chemical analysis data from the plant; and an online learning phase, in which the system continuously collects data on ion concentration changes before and after deposition, current utilization, and metal recovery quality, and corrects the model output in real time. After structured processing, the training data is fed into a regressor for residual fitting optimization. Cross-validation is then used to assess generalization capabilities to prevent biased output at critical concentrations.

[0070] Example 1:

[0071] In the copper-nickel alloy electroplating production line of an electronic electroplating factory, about 18 tons of heavy metal waste liquid is discharged daily, of which Cu 2+ The initial concentration is about 120 mg / L, Ni 2+ About 80mg / L, also contains a certain amount of impurity Zn 2+ (25mg / L) and residual organic complexing agent. To achieve resource recovery, the plant deployed a set of electroplating-reflux cycle treatment system. The key of the system includes a multi-channel micro sensor array arranged in three sections of waste liquid treatment path (reaction inlet area, deposition main tank, and before reflux inlet). Each sensor point integrates a high-precision conductivity chip (resolution 0.1μS / cm) and three ion selective electrodes, one for Cu2+ 、Ni 2+ 、Zn 2+ Selective monitoring. The waste liquid flow rate is 0.5m / s and the total volume of the reaction tank is about 2m 3 ,The sampling frequency of each sensor is 2Hz, and the system collects hundreds of data per second.

[0072] Before the waste liquid enters the main sedimentation tank, the first group of sensor arrays (entrances A, B, and C) detects Cu 2+ The concentrations were 118, 116, and 117 mg / L, and Ni 2+ The data were 82, 79, and 80 mg / L, and the conductivity was about 7.5 mS / cm. The system made compensation adjustments based on a temperature of 23.5°C and a pH of 4.3, converted the raw readings into standardized concentration data, and eliminated outliers caused by point A mutations (the conductivity was as high as 15 mS / cm in the previous minute). After data cleaning, it was sent to the multi-channel data fusion module, which eliminated the interference coefficients between different ion electrodes (Cu 2+ Electrode Zn 2+ The influence is +8%), and then the main features are extracted based on PCA dimensionality reduction, and the concentration feature vector of the current waste liquid is output as:

[0073] Cu 2+ =117.4mg / L,Ni 2+ =80.3mg / L,Zn 2+ =26.1mg / L

[0074] The calculated ionic strength index σ_total = 224.8 mg / L. This value is higher than the system's set reflux critical concentration (180 mg / L), preliminarily judging that this batch of waste liquid has recycling value.

[0075] Subsequently, this set of data is fed into the spatial distribution modeling module and combined with the flow field and concentration data obtained from the other four sensor array points (D, E, F, G) in the sedimentation tank: Cu at point D 2+ =98mg / L, point G is only 63mg / L, the system combines the flow path, flow velocity and liquid disturbance data through a three-dimensional interpolation algorithm (based on a custom improved Kriging model) to construct the Cu inside the sedimentation tank 2+ The distribution map shows an obvious gradient of high at the front and low at the back; Zn 2+ The local accumulation at point E reached 37 mg / L, indicating that there was an impurity enrichment problem in some areas. The spatial model predicted that area G would reach the critical depletion state within 7 minutes (Cu 2+ <60mg / L), Ni 2+ Slightly slower (~12 minutes).

[0076] The map is sent to the deposition feedback control module in real time. The system determines that the risk of ion depletion in the G area is approaching and immediately outputs a control instruction to execute the following response: First, the current density of the deposition electrode in the G area is reduced by 15% to slow down the ion consumption rate. At the same time, the opening of the reflux pump is increased by 20% through the adjustable solenoid valve to preferentially inject high Cu ions from the inlet concentration area into the G area. 2+ At the same time, the current density in area A (high concentration area) of the sedimentation tank remains unchanged and is marked by the system as a resource-rich area. Within the next 5 minutes, some sedimentation units will be dispatched for polarity reversal and cleaning as needed.

[0077] In the concentration unit module, the system selects the low-pressure membrane concentration + electrodialysis linkage mode according to the inlet waste liquid conductivity and liquid level parameters. The concentration efficiency is about 60%, and the output concentrated liquid Cu 2+ The concentration increased to 185 mg / L, Ni 2+ The benefit prediction model evaluated this batch of concentrate at a score of 0.86 (out of a maximum of 1.0), placing it in the high sedimentation benefit range. Therefore, the AI scheduling system set 80% of this concentrate to flow into the return pipe and be sent to the main sedimentation tank, with the remaining 20% sent to the deep treatment backup tank for secondary adjustment.

[0078] In the sensor array arranged at points A / B / C at the entrance of the factory deposition system and points D / E / F / G in the deposition main tank, each sensor unit integrates three ion-selective electrodes with different structures, which are used to detect Cu 2+ 、Ni 2+ 、Zn 2+ Heavy metal ion concentration. Traditional ISE (Ion-Selective Electrode) is susceptible to cross-interference and organic complexing agent adsorption problems in complex waste liquid environments, and the response signal is unstable. To solve this problem, the system uses a multi-layer composite membrane structure ISE, each layer of which has different functions: the outermost layer is a hydrophobic microporous protective membrane used to shield large molecular organic impurities; the middle layer is a doped conductive polymer membrane (such as doped PEDOT:PSS), which has good electron transport performance and can improve the response speed; the innermost layer is a customized ion selective membrane that selectively adsorbs Cu 2+ 、Ni 2+ Specific ions, by fixing negatively charged groups or chelating groups to regulate the freedom of ions to enter the membrane layer, and significantly improve the ion selectivity coefficient. 2+ The selective membrane is based on nitrogen-coordinating groups and has a response sensitivity linearity exceeding 95% in the range of 50–180 mg / L.

[0079] When the waste liquid flows through the sensor array, each channel ISE outputs a different potential signal (measured in mV), and the mixed response superposition of different metal ions, conductivity changes and pH drift cause errors. The system's data acquisition module records the raw electrode potential data of each channel once a second, and simultaneously records the liquid pH value, conductivity, and temperature background parameters. These raw data will first undergo preprocessing operations, including potential value temperature correction (based on the nonlinear correction of the temperature term of the Nernst equation), pH error drift compensation (applying moving median filtering), outlier detection and elimination (based on Z-score standard deviation screening) and standard normalization processing, so that the output signals of different electrodes fall into a unified scale.

[0080] After preprocessing, these data will be fed into the multi-channel data fusion and concentration prediction model. The model structure is designed as a dimensionality reduction + prediction double-layer processing link. The principal component analysis (PCA) algorithm is applied to perform feature dimensionality reduction on the original potential matrix. It is set that each sensor unit outputs 9 electrode potentials (3 ions × 3 points) in each sampling cycle, and 3 principal components are extracted through PCA (which can explain more than 96% of the variance of the original data). This process effectively removes collinearity interference and compresses the data dimension. The principal component vector after dimensionality reduction will be sent as input to the next layer - the support vector regression (SVR) prediction model. The model is designed as a multi-target regression structure that can simultaneously output the Cu 2+ 、Ni 2+ 、Zn 2+ The predicted values of the concentrations of the three ions. The kernel function in the SVR model is the radial basis function (RBF), which has strong nonlinear fitting capabilities.

[0081] The model training process is based on two months of historical plant operation data. The total training set exceeds 6,000 groups of samples, and the label values are the actual ion concentration values verified by ICP-OES spectrometer before deposition. The training process includes five-fold cross-validation and grid search to optimize hyperparameters (C value, ε interval and kernel function γ value). The final model's mean absolute error (MAE) on the test set is: Cu 2+ Prediction error ±3.5mg / L, Ni 2+ ±4.1mg / L, Zn 2+ ±2.7mg / L, the prediction results can meet the requirements of industrial-grade process control.

[0082] Continuing the actual deployment of the plant, the main deposition reactor is a rectangular structure with a length of 3.0 meters, a width of 1.5 meters, and a volume of approximately 3.8 cubic meters. There are 12 ion monitoring points in the tank, numbered from P1 to P12, evenly distributed at different horizontal and vertical coordinate positions of the tank. Each point is equipped with a Cu 2+ 、Ni 2+Dedicated ion selective electrodes and conductivity chips, with a sampling frequency of 2Hz. In one operation cycle (set to 30 minutes), the system collects full-process ion activity data and local mass transfer rate data, with the goal of predicting Cu 2+ The two-dimensional distribution of concentration in the tank is optimized and controlled.

[0083] The system first pre-processes the collected monitoring point data, eliminates abnormal spike readings caused by local bubble interference, adjusts all ion concentrations to the equivalent state of 25°C through the temperature drift correction model, and normalizes them to match the standard input format of the spatial interpolation algorithm. The output of this stage is a two-dimensional coordinate point set P (x, y) and the corresponding activity value ω (x, y, t). With 30 minutes as the current control cycle (t = 1800s), the system starts running the spatial interpolation reconstruction module. The Kriging interpolation method was used for this task. Considering the Cu 2+ In the horizontal and vertical migration restriction characteristics, the non-isotropic variogram model is used, and its covariance weight adjustment coefficient is set to θx = 0.4, θy = 0.6. The interpolation results show that the lower right corner of the tank (coordinates x = 2.8m, y = 1.4m) is Cu 2+ The lowest concentration area is about 53 mg / L, while the central area (x=1.5m, y=0.8m) reaches a peak of 112 mg / L.

[0084] In order to dynamically evaluate the ion supply capacity of the low-concentration area in the lower right corner, the system introduces a responsive concentration evolution function model to calculate the cumulative deposition capacity index Φ(x, y, t) at the target point (x = 2.8, y = 1.4) within the control period. The system selects the following parameters for substitution into the model:

[0085] λ = 0.63: Response sensitivity coefficient. Considering that the electrode in this area is a carbon cloth electrode and the stirring flow rate is low (0.1 m / s), the sensitivity is medium.

[0086] α=1.4: response weighting index, the system tends to weight deposition in high concentration areas;

[0087] ω(x,y,τ): average Cu of sampling values within 1800 seconds 2+ The activity was 53 mg / L (normalized value 0.47);

[0088] The change in mass transfer rate, expressed in mg / L / s, is derived from the inversion of the experimentally recorded current density change and the liquid disturbance frequency.

[0089] Substitute these data into the model for integral approximation calculation:

[0090]

[0091] The minimum response capability threshold set by the system is Φ min =3.0, so it is judged that this area is in a critical supply shortage state. The system then starts the deposition feedback control module and performs the following linkage operations: first, the proportional pump is used to increase the reflux liquid from the concentration area to introduce it into this area, the liquid injection path is adjusted from the center to the corner, and the reflux pump frequency is increased from 18Hz to 26Hz, which is expected to increase the local Cu 2+ The activity was raised to above 70 mg / L. Secondly, the polarity reversal operation was applied to the electrode in the region, with a cycle of once every 10 seconds for 3 minutes to break the formed anodic passivation film. At the same time, the current density of the electrode in the region was increased from the original setting of 6.5 mA / cm 2 Down to 5.2 mA / cm 2 , slow down the electrochemical reaction speed, and wait for the ion supply to be sufficient before returning to the target power.

[0092] The system will continue to track the Φ(x, y, t) curve in the next two control cycles (i.e., the next hour). If it is lower than 2.5 twice in a row, the area will be marked as a low-efficiency deposition area, and the system will switch it to the insulation maintenance state and no longer invest resources to prevent ineffective consumption. In the electrodeposition system, the sensor array collects Cu every 2 seconds. 2+ 、Ni 2+ 、Zn 2+ The local concentration, conductivity, potential, pH and current electrode working status of the electrode are measured. All raw data are synchronized to the prediction submodule in real time to form a multivariate time series input. 2+ For example, in the case of concentration prediction, the system uses the past 30 minutes (900 samples) as the input sequence, with the goal of predicting ion concentration trends over the next 10 minutes (300 seconds). During data preprocessing, the system first normalizes the original sequence, mapping the concentration values to the [0, 1] interval. It then uses a local sliding mean to smooth short-term fluctuations. It then performs lagged variable expansion, constructing an input vector from every 10 historical steps to enhance the ability to express short-term behavior.

[0093] The prediction model structure uses a two-layer stacked recurrent neural network (StackedRNN), where each layer contains 64 hidden units, the activation function uses tanh, and a dropout layer is inserted between the hidden layers to prevent overfitting (the dropout rate is set to 0.2). The input is a three-dimensional tensor of [batch_size, time_steps, features]. Specifically in this example, time_steps = 10 and features = 4 (including concentration, conductivity, electrode current density, and electrode polarity state). The output is a single-step or multi-step Cu 2+Concentration prediction value, the system uses a sliding window method to generate training samples, and each window predicts the concentration change in the next 60 seconds to 300 seconds.

[0094] The model was trained using supervised learning, with labels consisting of laboratory ICP-OES measurements or actual concentrations after sensor array calibration. The loss function was the mean squared error (MSE), and the Adam optimizer was used. The initial learning rate was set to 0.001, the number of training epochs was 200, and the batch size was set to 64. The training data was derived from the plant's operating records over the past two months, covering various concentration fields, electrode states, and operating fluctuations. The total number of training samples was approximately 15,000, and the validation set consisted of measured data from the past week.

[0095] In practical applications, when the system detects Cu 2+ The concentration began to fluctuate and decrease, dropping from 110 mg / L to 92 mg / L at the coordinate (x=2.0, y=1.2). The RNN prediction module determined in real time that its trend would drop to 65±3 mg / L within the next 8 minutes, while the minimum deposition efficiency guarantee threshold set by the plant was 68 mg / L. Based on this prediction, the system did not wait for the concentration to hit bottom before reacting, but instead immediately outputted pre-control instructions: First, it increased the opening of the reflux channel valve in the area from 30% to 60%, ensuring that more concentrated liquid was injected into the area within 4 minutes; second, it lowered the electrode current density in the area from 6.8 mA / cm 2 Adjust to 5.5 mA / cm 2 , to avoid local over-deposition; thirdly, the system marks the point as a low-activity risk point in the future and issues an early warning prompt on the control panel for manual or automated systems to deploy electrode maintenance strategies.

[0096] Example 2:

[0097] Collection Attachment Figure 3 As shown, the factory produces about 4 tons of mixed electroplating waste liquid containing copper, nickel and zinc per hour, with the initial concentration of Cu 2+ About 95 mg / L, Ni 2+ About 72 mg / L, Zn 2+ The waste liquid first enters the sensing module, and the current liquid parameters are obtained through the sensor: the current waste liquid temperature is set to 31.2℃, pH is 4.1, and conductivity is 9.5mS / cm. The sensor array determines the Cu 2+ The concentration is about 93 mg / L, Ni 2+ 70mg / L, Zn 2+ It is 36mg / L.

[0098] Based on the current parameters, the AI system identifies that the wastewater is in a medium concentration and multi-ion mixed state, and combined with historical energy consumption data and concentration conversion rate, selects electrodialysis mode as the current concentration method. The system uses membrane stack number E-72, which can process 800 liters of wastewater per hour, with an expected concentration ratio of 1.6 times and an energy consumption of 0.34kWh / m 3 After the batch treatment, the concentration of the concentrated liquid output by the system (about 300 liters per batch) was measured to be: Cu 2+ Increased to 148 mg / L, Ni 2+ 113 mg / L, Zn 2+ The concentration of 55 mg / L and the conductivity of 14.1 mS / cm were fed into the redeposition feasibility model of the AI scheduling system.

[0099] The prediction model is a multi-factor regression network, and the input parameters include metal ion concentration vector, conductivity, pH, impurity ratio (Zn 2+ / Cu 2+ The system uses a trained XGBoost regression model to calculate the current concentrate's redeposition potential score, which is 0.81 (out of a maximum score of 1.0), with a predicted conversion efficiency of approximately 91%. A score greater than a set threshold (0.75) indicates that the concentrate is worth recirculating for reuse.

[0100] The AI scheduling system then outputs instructions to the proportional control unit, setting the current distribution ratio of the concentrate to reflux: discharge = 85%: 15%. In other words, in this batch of 300 liters of concentrate, 255 liters will be transported to the inlet area of the main sedimentation tank through the reflux pump, and the remaining 45 liters will flow into the discharge or deep treatment module (such as the ion exchange system) through the valve switch. The proportional control unit uses a combination of an intelligent solenoid valve and a variable speed pump to control the flow rate in real time through a feedback flow meter to ensure that the flow rate injected into the sedimentation tank per minute is stable at 8.5 to 9.0 L / min to meet the load balance of the electrode reaction.

[0101] During the actual operation, the system monitors the concentration of Cu in the batch of liquid after it enters the deposition system. 2+ The regional deposition efficiency increased from 87% to 92.5%, and the deposition current fluctuation amplitude decreased by 12%, indicating that the concentration matching strategy effectively alleviated the regional polarization phenomenon. 2+ / Cu 2+ The ratio) is 0.37, which exceeds the set warning value of 0.3. Therefore, the AI system directs it to the deep impurity removal processing unit for subsequent resource recovery.

[0102] In the wastewater treatment system of this plant, the front-end online sensor array collects the wastewater parameters entering the concentration unit once a minute, including Cu 2+ 、Ni 2+ 、Zn 2+ Concentration (mg / L), conductivity (mS / cm), pH value (range 3.8–4.5), liquid temperature (25℃–38℃) and the current number of reflux cycles. The system is equipped with three optional concentration modules, namely low-pressure membrane concentration device (nanofiltration NF), electrodialysis (ED) module and vacuum evaporation system (VE). In order to realize the automation of mode switching, the system is embedded with a set of intelligent judgment model for concentration path. The model uses unit metal recovery efficiency (η) and unit energy consumption (E) as key constraints to determine which concentration mode should be enabled at present, and ensure that the target ions (such as Cu) of the concentrate are not lost during the switching process. 2+ ) The activity concentration is not lower than the set value, ≥135mg / L. The initial threshold set by the system is:

[0103] Recovery efficiency η_min ≥ 75%, that is, at least 75% of the target ions should be recovered per unit of liquid input;

[0104] Energy consumption threshold E_max≤0.5kWh / m 3 ;

[0105] The threshold value of metal ion activity maintenance C_active≥135mg / L (Cu 2+ for example).

[0106] Taking the actual operation batch as an example, the system detects the current waste liquid Cu 2+ =94.2mg / L, Ni 2+ =71.5mg / L, Zn 2+ =30.2mg / L, temperature 31.8℃, conductivity 10.1mS / cm, pH=4.0. After preprocessing including temperature standardization, conductivity drift correction and concentration normalization, the above data are sent to the mode selection judgment model. The model is a heuristic regression classifier based on the decision tree structure. Its input vector is [Cu 2+ ,Ni 2+ ,Zn 2+ ,T,pH,σ], and the output is the concentration mode selection result {NF,ED,VE}.

[0107] The model structure consists of three branch judgment paths, each of which corresponds to a set of comprehensive evaluation functions. For example, in the electrodialysis path, the model evaluates its unit recovery efficiency as η_ED = 0.76 and unit energy consumption as E_ED = 0.43 kWh / m 3 , predict output Cu 2+Active concentration C_ED = 142 mg / L. In contrast, the predicted energy consumption of the vacuum evaporation path is E_VE = 0.65 kWh / m 3 (already exceeded the threshold), although Cu 2+ The concentration was concentrated to 182 mg / L, but its η was only 0.72, which was not advantageous. The final decision was to activate the ED module.

[0108] Using samples collected from the plant's continuous operation for 30 days, about 22,000 sets of historical concentration task records, the training adopted the gradient boosting tree (GBDT) and rule set fusion optimization. The average judgment accuracy of the model training error on the test set reached 94.8%, and the deviation of the predicted metal activity concentration after concentration was less than ±4.5mg / L. During the switching process, in order to ensure that the ion activity is not lower than the threshold, the system introduces an active concentration prediction and feedback mechanism. That is, within the first 5 minutes of the mode switching, the current waste liquid batch is pre-treated and concentrated (for example, 50L test volume), and the Cu 2+ Concentration change curve. If the deviation between the model prediction and the measured concentration is greater than ±5%, the system will pause the mode switch and switch to the transition buffer mode. After slowly concentrating to the target concentration value by using the low-pressure membrane method for a short time, the main concentration path will be switched to ensure that the concentrate C_active in the main process is ≥135mg / L. Taking the current batch as an example, the initial sample Cu 2+ The measured concentration is 94.2 mg / L. After ED treatment with a concentration ratio of 1.5 times, the actual concentration is 143.5 mg / L, with a deviation of +1.5%, which meets the direct switching conditions.

[0109] At the same time, after the switch is completed, the system continues to monitor the concentrated liquid Cu for 30 minutes. 2+ If the concentration fluctuates and its value is found to be below the activity threshold for more than 10 minutes, an early warning is triggered, the reflux ratio of the concentrate is reduced to 70%, and the remaining part is directed to the vacuum evaporation module for deep processing. The system operation report shows that since the deployment of this concentrated intelligent switching control system, Cu 2+ The average activity after concentration was stabilized at 139.3 mg / L, the average redeposition score of the concentrate was 0.88, the energy consumption was reduced by about 14%, and the recovery efficiency was increased by 9%.

[0110] In this factory, after front-end processing and mid-stage electrodialysis concentration equipment, the system outputs about 15 to 20 batches of concentrate per day, with each batch of liquid volume being about 200 liters. In each batch of concentrate, the heavy metal ion concentration, impurity content, conductivity change, and sedimentation unit load factors vary significantly due to fluctuations in operating conditions. Therefore, whether it can be effectively refluxed must be determined by an intelligent system. The prediction module in the AI scheduling system calculates the redeposition benefit integral value Γ(t) of the concentrate in real time to classify the priority of the liquid. Taking the 45th batch of concentrate (No. 2024-12-B045) as an example, the entire calculation and control process is explained.

[0111] First, the system monitors the batch of liquid online, and the sensor collects the following key parameters:

[0112] Target metal ion concentration: c Cu =142mg / L,c Ni =96mg / L,c Zn =31mg / L

[0113] Non-target impurity concentration vector: j t =[Ca 2+ =19mg / L, Na + =58mg / L, TOC=43mg / L]

[0114] Conductivity curve changes: from μ(0)=10.5mS / cm to μ(15min)=13.2mS / cm, the rate of change is

[0115] Current sedimentation tank load: electrode utilization rate is 78%, current density is 6.4mA / cm 2 , the load state index is set as ρ(k t )=0.83

[0116] The AI system feeds the above data into the model, using the following coefficient ranges:

[0117] θ1=0.7:Ion concentration contribution coefficient (prioritizes target metal activity)

[0118] θ2=0.5:Impurity suppression coefficient (the higher the value, the more sensitive it is to impurities)

[0119] θ3 = 0.9: Deposition response coefficient (the higher the coefficient, the faster the deposition response)

[0120] The functions in the model are explained as follows:

[0121] χ(i t )=0.8·c Cu +0.6·c Ni -0.2·c Zn =0.8·142+0.6·96-0.2·31=113.6+57.6-6.2=165.0

[0122] ν(j t )=0.3·19+0.2·58+0.5·43=5.7+11.6+21.5=38.8

[0123] Substituting these data into the integral function formula (simplified to the average approximation):

[0124]

[0125] The redeposition classification threshold of the system is defined as:

[0126] Γ th =1200: High-efficiency redeposition potential threshold

[0127] Γ min =800: Minimum acceptable sedimentation benefit limit

[0128] Since the concentrate of this batch is Γ=1441.5>Γ th The system classifies it as a first-level high-deposition-value liquid and sets the reflux ratio to 100% through the proportional control unit. The dispatch pump introduces all of it into the main inlet area of the electrodeposition reaction tank. The system further increases the electrode operating current by 5% to match the enhanced deposition rate brought by the high-concentration solution.

[0129] Next, the system evaluated the next batch of liquid, numbered 2024-12-B046, and the analysis data was as follows:

[0130] c Cu =102mg / L,c Ni =58mg / L,c Zn =35mg / L

[0131] Impurities: TOC68mg / L, Na + 66mg / L, Ca 2+ 25mg / L

[0132] The conductivity increases slowly, with a rate of change of about

[0133] Current deposition load: electrode fatigue value is 0.52, the system is adjusted to ρ(k t )=0.65

[0134] Estimate:

[0135] χ(i t )=0.8·102+0.6·58-0.2·35=81.6+34.8-7.0=109.4

[0136] ν(j t )=0.3·25+0.2·66+0.5·68=7.5+13.2+34.0=54.7

[0137] Substitute into the model:

[0138]

[0139] This value is lower than Γ min=800, the system identifies this liquid as having poor sedimentation efficiency and directs 90% of it to the deep impurity removal module, leaving only 10% for small-scale mixed reflux. The system issues a warning: The TOC content of this batch of liquid is high, and chemical oxidation pretreatment is recommended.

[0140] In the plant's closed-loop waste liquid treatment system, the proportional control unit is deployed at the main pipeline bifurcation node at the concentrate output end, connecting two paths: one reflux branch leading to the main reaction zone of electrodeposition, and the other leading to the deep processing module or safe discharge unit. The control unit consists of a variable frequency screw reflux pump, a set of high-precision electromagnetic flow control valves (CV series) and a group of electromagnetic flow regulators, and the whole is connected to the AI scheduling master PLC via industrial Ethernet. In terms of control logic, the AI system receives the score output Γ(t) of the redeposition benefit model every 10 seconds, and sends the reflux-discharge ratio instruction to the control unit according to the interval to which its value belongs, with an execution target accuracy of ±2%.

[0141] In the aforementioned batch number 2024-12-B045, the system calculated the benefit score of the concentrate to be Γ(t)=1441.5, which is higher than the threshold Γ th =1200, the system strategy is full reflux. The control unit immediately increases the frequency of the variable frequency pump to 60Hz, opens solenoid valve CV-A (to the sedimentation area) to 100% flow, and sets the flow rate to 12L / min. CV-B (to the discharge channel) is closed. The system verifies the flow rate with real-time feedback using a built-in thermal mass flow meter with an accuracy of ±0.15L / min, ensuring precise execution.

[0142] When batch number 2024-12-B046 (score 738.75) was judged to be an inefficient concentrate, the control strategy was switched to 10% reflux and 90% discharge. The system adjusted the pump frequency to 48Hz, reduced the CV-A opening ratio to 10%, and set CV-B to 90%, with target flow rates of 1.5L / min and 13.5L / min, respectively. The control instruction is calculated by the PID controller in the main control program and sent to each flow regulator module, and fine-tuned every second based on the deviation between the actual flow rate and the set value. The feedback curve shows that 5 seconds after the parameter switching, the system reached a steady state and the deviation was controlled at ±0.8%.

[0143] The AI instruction parsing module used in the control system receives the following input data streams: ① target reflux ratio (output of the AI decision module); ② real-time concentrate flow rate, pressure, temperature, and viscosity (collected by sensors); ③ dynamic model of flow resistance (used to predict pipeline pressure loss). Based on these inputs, the system builds a set of proportional adjustment function libraries. Its core logic is to quickly calculate the required pump speed and valve opening under different reflux settings. For example, if the current liquid viscosity rises to 1.8mPa·s and the flow resistance increases, the system will use the prediction model output to increase the pump frequency in advance to ensure that the reflux pipeline can still maintain the target flow rate under the premise of unchanged distribution ratio.

[0144] The control module was initially trained offline in an experimental system using a large amount of backflow test data. The training data structure includes: pump frequency-flow response curve, valve position-diversion ratio mapping table, and liquid viscosity-response delay compensation time. The model structure uses a shallow neural network (3 layers, 16×16×8 units, ReLU activation), with the label being the deviation error between actual flow and target. The optimization goal is to reduce dynamic response overshoot and improve stability. Training uses the Adam optimizer with a learning rate of 0.0005. After 50 rounds of training, the model achieved an accuracy of over 96% in predicting the match between backflow and discharge target flow.

[0145] Later, during deployment under real-world conditions, the system supported adaptive control. When three consecutive batches of scores fluctuated near a threshold (1180-1220), the AI model switched to a mixed distribution strategy, setting 80% return flow and 20% discharge. The pump speed and valve position were also set to a slow-changing mode to prevent equipment wear caused by frequent switching. Data shows that under this control strategy, the system's daily response times decreased by 15%, pump temperature rise was reduced by 2°C, and overall energy consumption dropped by 4.1%.

[0146] In actual operation at the plant, the main deposition reactor has six independent electrode sections, each with its own current control capability and load sensing feedback. The system collects data once a minute to form a time series sample, including the following key indicators: ① Unit concentrate recovery rate (referring to the effective reduced metal mass in every 100L of concentrate, measured in kg); ② Deposition conversion rate (the ratio of the total amount of deposited metal to the total amount of metal fed); ③ Deposition load index (the ratio of the amount of metal deposited in the deposition area per unit time to the electrode reaction capacity); ④ Unit energy consumption (kWh / kg); ⑤ Reflux ratio set value (%). These data are extracted from production history records, and the system records them continuously for three months, resulting in a total of approximately 13,000 sets of time series samples.

[0147] Data preprocessing includes missing value interpolation (linear extrapolation + median filling), outlier identification (elimination using 3σ control limits), and normalization (z-score normalization) to ensure that the time series input features are uniformly scaled for model training. The time granularity is uniformly set to every 10 minutes (i.e., 6 sets of data per hour), the time span is set to predict the next hour from the past 8 hours, the input dimension is 6×48, and the output is a series of 6 sets of system load indices for the future. The model architecture uses a two-layer stacked LSTM (Long Short-Term Memory) network, which is suitable for capturing long-term dependencies. Each LSTM layer contains 64 units, the activation function is tanh, and a dropout mechanism (p=0.2) is used to prevent overfitting. The model outputs a prediction sequence of length 6, corresponding to the load fluctuation trend for the next 60 minutes. The loss function is the mean squared error (MSE), the optimizer is Adam, and the initial learning rate is set to 0.001.

[0148] The system uses 80% of historical samples as a training set and 20% as a validation set, employing an early stopping strategy to prevent overtraining. The model converges quickly during training, with an average prediction error within ±0.03 (normalized load units). After model deployment, the system provides an hourly rolling forecast of the sediment load change for the next hour and calculates the optimal recirculation ratio based on the current recirculated concentrate volume.

[0149] In the forecast for December 16, 2024, at 5:00 PM, the model outputs a future load trend curve that fluctuates, with peaks initially and then declines. The specific predicted values are as follows (in standardized load units): [0.78, 0.83, 0.87, 0.71, 0.65, 0.62], clearly indicating that the peak load will be reached in the next 30 minutes. Based on this forecast, the system immediately increases the current reflux ratio from the original 75% to 90% to ensure sufficient metal ion supply for the upcoming deposition load peak. Simultaneously, the AI scheduling module issues an early warning recommendation to the electrode control system, temporarily increasing the current density limit by 10% during the peak window (t+10 minutes to t+30 minutes) to improve the deposition response rate. When the predicted trough period (after t+40 minutes) is approaching, the system slowly reduces the reflux ratio to 60% to avoid excess supply and waste of concentrate.

[0150] After deploying this prediction mechanism, reflux resource utilization increased by approximately 8%, sediment load offset volatility decreased by 18%, reflux fluid waste decreased by over 11%, and system stability significantly improved. Notably, when the system continuously detects a prediction error exceeding ±0.08, it triggers an online retraining process, extracting high-confidence samples from the most recent 1,000 time series to update model parameters and ensure the LSTM network can adapt to the new production rhythm.

[0151] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for recycling heavy metal waste liquid by electrodeposition, characterized in that The following steps are involved: Based on the waste liquid ion composition and polarization curve, an interfacial charge migration model was established to quantify Cu 2+ 、Zn 2+ 、Ni 2+ Ion migration impedance characteristics; A progressive potential scan is applied to the deposition area, causing different ions to deposit sequentially according to the migration resistance gradient, forming a transient high charge density band on the electrode surface. By adjusting the pulse frequency, the preferential adsorption and reduction of ions is induced; by adjusting the migration rate to match the metal reduction potential window, the metal is preferentially reduced. The deposition reaction zone is laterally moved on a microscale through a movable electrode structure. During the deposition interval or electrode migration period, a reverse high-frequency pulse is applied to trigger partial stripping or recrystallization of the metal deposition layer, removing the passivation layer. One portion of the electrode is scheduled for continuous deposition while the other portion is cleaned and metal stripped, forming a dynamic closed loop. An integrated micro-conductivity / ion-selective electrode sensor array evaluates the concentration and intensity distribution of residual heavy metal ions in the waste liquid in real time; switches the concentration mode based on the perception results of low-pressure membrane concentration, electrodialysis or vacuum evaporation to optimize the concentration to reach the re-deposition threshold; uses the AI scheduling system to control the proportion of concentrated liquid to be reintroduced into the deposition area and the proportion to be sent to the waste liquid recovery or discharge unit to ensure that the concentration matches the reflux.

2. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 1, characterized in that The method for real-time evaluation of the concentration and intensity distribution of residual heavy metal ions in wastewater comprises: A multi-channel microsensor array is deployed at multiple locations in the wastewater treatment device. Each sensor unit includes a conductivity sensor and multiple ion-selective electrodes, which are used to collect the total ion concentration, conductivity, and active concentration of heavy metal ions in the wastewater. Data acquisition and processing module, used to collect the original electrical signals of each sensor and perform temperature and pH drift compensation; Multi-channel data fusion module, which is used to synchronously process the data of each ion channel through feature extraction and cross-response modeling algorithms, and output the concentration vector and total intensity characteristics of heavy metal ions; A spatial distribution modeling module is used to establish a two-dimensional or three-dimensional concentration distribution map of heavy metal ions in the deposition system based on the layout position of the array sensor in the physical space and the waste liquid flow parameters; The deposition feedback control module is used to determine whether the current deposition area is in a concentration depletion, polarization or reflux trigger state based on the above concentration and distribution map results, and output control instructions to the electrodeposition main control unit or concentration loop.

3. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 2, characterized in that The ion-selective electrode adopts a customized multi-layer composite membrane structure; the data fusion module uses a principal component analysis algorithm or support vector regression to perform feature dimension reduction and concentration prediction on multiple channel data.

4. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 3, characterized in that The spatial distribution modeling module uses an interpolation reconstruction method selected from inverse distance weighting, Kriging interpolation or a neural network-based spatial mapping model; the deposition feedback control module switches the process state according to the ion concentration threshold setting, including adjusting the concentrate ratio, reversing the electrode polarity, reducing the current density or starting the reflux pump.

5. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 4, characterized in that The deposition feedback control module includes a prediction submodule based on a time series learning algorithm, which uses a recurrent neural network to predict future metal ion concentration trends for early adjustment of the control strategy.

6. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 1, characterized in that The method for implementing the concentration matching reflux includes: A concentration device switches between three concentration methods: low-pressure membrane concentration, electrodialysis, or vacuum evaporation, based on the ion concentration, conductivity, temperature, and pH sensing results of the heavy metal waste liquid; An AI scheduling system that receives data from sensors and performs the following operations: (1) Real-time evaluation of the residual concentration and redeposition value of heavy metal ions in the concentrate; (2) Construct a feasibility prediction model for redeposition of concentrate based on historical operating data; (3) controlling a set proportion of the concentrated liquid to flow back to the electrodeposition unit, and sending the rest to the waste liquid recovery or discharge unit; A ratio control unit is used to dynamically adjust the volume ratio of concentrated liquid return and discharge according to the instructions of the AI scheduling system so that the concentration matches the return.

7. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 6, characterized in that The concentrating device selects a concentration mode based on real-time feedback parameters through preset energy consumption and efficiency thresholds, and maintains the metal ion activity of the concentrated liquid not lower than the set value during the switching process between different modes.

8. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 7, characterized in that The AI scheduling system includes a concentrate redeposition benefit evaluation model, which makes a comprehensive judgment based on the ion type, concentration, conductivity, impurity ratio and absorption capacity of the upstream electrodeposition unit in the concentrate, and assigns a reuse priority label to the concentrate.

9. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 8, characterized in that The proportional control unit is composed of a variable frequency pump valve, a flow controller or an electromagnetic flow regulator, and adjusts the ratio of reflux and discharge in real time under the control of the AI scheduling system.

10. The heavy metal waste liquid electrodeposition recycling treatment method according to claim 9, characterized in that The AI scheduling system further includes a time series prediction module, which uses a learning model based on historical concentrate recovery rate and deposition conversion rate to predict system load fluctuations in future time periods and adjust the reflux ratio in advance to match the deposition efficiency.

Images