Intelligent recommendation method for mine water treatment agent proportioning based on multi-objective optimization

Abstract

The present application relates to the technical field of mine water treatment, and particularly relates to a mine water treatment agent proportioning intelligent recommendation method based on multi-objective optimization, comprising: collecting mine water treatment multi-dimensional dynamic data and preprocessing to obtain standardized engineering benchmark data sequences; constructing a heavy metal and suspended matter competitive adsorption index and a flocculation body hydraulic shear vulnerability index according to the standardized engineering benchmark data sequences; fusing the standardized engineering benchmark data sequences with the heavy metal and suspended matter competitive adsorption index and the flocculation body hydraulic shear vulnerability index, and predicting future water outflow turbidity by using a classification enhanced gradient boosting tree prediction algorithm; constructing a fitness objective function according to the future water outflow turbidity combined with engineering penalty constraints, and optimizing by using a multi-objective particle swarm optimization algorithm. Through the technical scheme of the present application, the solidity of the flocculation body microstructure can be ensured under harsh working conditions, and invalid consumption of the agent can be avoided.

Description

Technical Field

[0001] This invention relates to the field of mine water treatment technology, and in particular to a method for intelligent recommendation of mine water treatment reagent ratios based on multi-objective optimization. Background Technology

[0002] In the gold mining and smelting industry, the resulting mine water not only contains high concentrations of suspended solids, but also heavy metal ions such as copper, lead, and zinc. To treat this complex water, a combined water treatment process is currently used, which involves adding industrial polyaluminum chloride as a coagulant and polyacrylamide as a polymeric flocculant. By adjusting the pH of the fluid, the suspended solids and heavy metals are separated through co-precipitation. However, for this complex mine water, the requirements for the addition and ratio control of the reagents are extremely high.

[0003] In existing water treatment processes, data processing systems for dosing mine water primarily rely on inputting historical water quality monitoring data and historical dosing data into an implicit nonlinear mapping model for training and calculation. Since heavy metal ions and suspended particles in mine water objectively compete for adsorption and complexation sites on the polyacrylamide polymer chains, when the influent pH fluctuates beyond the optimal sedimentation range, the dissolved form of heavy metal ions changes, causing a nonlinear shift in the competitive relationship between heavy metal ions and suspended matter. Existing pure data algorithms cannot characterize this chemical competition, leading to the ineffective consumption of recommended reagents by heavy metal ions and resulting in flocculation failure. This method, which relies solely on historical data for fitting, ignores the underlying physicochemical reaction mechanisms of water treatment. Consequently, when faced with conditions such as sudden changes in water pH, the data processing system not only wastes valuable reagent resources but also renders the entire chemical precipitation process ineffective, seriously threatening the water quality compliance rate.

[0004] Furthermore, existing methods neglect the physical shear failure mechanism under hydraulic dynamic conditions. In actual engineering, the influent flow rate fluctuates dynamically, and a surge in flow rate can generate huge hydraulic shear stress inside the coagulation reaction tank. If the dosage ratio of polyaluminum chloride and polyacrylamide is unbalanced, the generated flocs lack high-density coagulation nuclei and are prone to irreversible physical breakage under extremely high hydraulic shear stress. Existing optimization algorithms do not take the hydraulic dynamic environment as a constraint and penalty condition, resulting in the output optimal ratio having extremely poor shock resistance under dynamic flow conditions. Once a sudden surge in water flow causes an instantaneous increase in influent flow, the data processing system that only pursues theoretical indicators will not be able to provide early warning of the physical fragility of the flocs. Its output ratio instructions cannot guarantee the robustness of the microstructure, ultimately causing the already coagulated impurities to re-disperse, leading to water quality exceeding standards. Summary of the Invention

[0005] To address the technical problem that existing intelligent recommendation technologies fail to consider chemical competitive adsorption and physical shear damage, leading to ineffective reagent consumption and poor shock resistance, this invention provides an intelligent recommendation method for mine water treatment reagent ratios based on multi-objective optimization.

[0006] This invention provides an intelligent recommendation method for mine water treatment reagent ratios based on multi-objective optimization, comprising: collecting multi-dimensional dynamic data of mine water treatment and preprocessing it to obtain a standardized engineering benchmark data sequence; constructing a heavy metal and suspended solids competitive adsorption index and a floc hydraulic shear vulnerability index based on the standardized engineering benchmark data sequence; fusing the standardized engineering benchmark data sequence with the heavy metal and suspended solids competitive adsorption index and the floc hydraulic shear vulnerability index, and using a classification-enhanced gradient boosting tree prediction algorithm to predict the effluent turbidity at future times; constructing a fitness objective function based on the effluent turbidity at future times, combined with engineering penalty constraints, and using a multi-objective particle swarm optimization algorithm to find the optimal solution, outputting the best reagent dosage combination as a recommended ratio instruction to complete the intelligent recommendation of mine water treatment reagent ratios.

[0007] This invention avoids unconstrained pure data dimension combination fitting. It preprocesses the acquired standardized engineering benchmark data sequence, constructs a heavy metal and suspended matter competitive adsorption index and a floc hydraulic shear vulnerability index based on fluid mechanics and chemical logic, and then fuses the standardized engineering benchmark data sequence with these indices. A classification-enhanced gradient boosting tree prediction algorithm is used to predict the effluent turbidity at future times. Finally, based on the future effluent turbidity, a fitness objective function is constructed in conjunction with engineering penalty constraints, and a multi-objective particle swarm optimization algorithm is used for optimization. This ensures that the algorithm optimization process is strictly constrained by objective physical laws, improving the feasibility and computational reliability of the overall data processing method in actual engineering deployment, strictly avoiding ineffective reagent consumption from the source, and enhancing shock resistance.

[0008] Preferably, the process of collecting and preprocessing multidimensional dynamic data of mine water treatment to obtain a standardized engineering benchmark data sequence includes: real-time acquisition of the influent instantaneous flow rate, influent pH value, influent suspended solids concentration, influent total equivalent concentration of heavy metal ions, instantaneous dosage of polyaluminum chloride, and instantaneous dosage of polyacrylamide in the water treatment process to form high-frequency continuous time series data; periodically acquiring the influent total equivalent concentration of heavy metal ions to form low-frequency discrete time series data; smoothing the collected high-frequency continuous time series data using a fixed-length sliding window mean filtering algorithm; and interpolating and filling the time axis of the collected low-frequency discrete time series data using a zero-order hold to asynchronously align it with the sampling rate of the smoothed high-frequency continuous time series data, thereby obtaining a standardized engineering benchmark data sequence.

[0009] By applying a fixed-length sliding window mean filtering algorithm to smooth continuous time series data, transient high-frequency spike noise interference caused by bubbles or solid impurities inevitably obstructing industrial field sensors can be effectively eliminated. This provides a more accurate, smooth, and stable standardized engineering benchmark data sequence as a direct input source for subsequent feature extraction and predictive model inference.

[0010] Preferably, the adsorption indices of the heavy metal and suspended matter compete for adsorption indices, satisfying the following relationship:

[0011] ;

[0012] In the formula, The adsorption index is determined by the competition between heavy metals and suspended solids. This represents the total equivalent concentration of heavy metal ions in the influent. It is an exponential function with the natural constant as its base. This refers to the pH value of the influent. The optimal engineering pH constant for the precipitation of characteristic heavy metal ions into hydroxides. The concentration of suspended solids in the influent. It is the basic background suspended matter concentration constant, and its absolute value is used to calculate the difference between the real-time pH and the optimal engineering pH constant.

[0013] By constructing computational formulas with clear physical dimensions, the chemical deterioration state of reagents in the reaction tank due to ineffective consumption by heavy metals is accurately measured. This helps to avoid unsafe solution regions with low chemical efficiency in advance when searching for optimization in subsequent multi-objective particle swarm optimization algorithms, thereby reducing the ineffective waste of reagents from the source and improving resource utilization.

[0014] Preferably, the hydraulic shear fragility index of the flocs satisfies the following relationship:

[0015] ;

[0016] In the formula, The hydraulic shear fragility index is used to describe the flocculent's vulnerability. It is the natural logarithm function. This represents the instantaneous influent flow rate during the current sampling period. This represents the instantaneous influent flow rate from the previous data collection period. The rated reference volumetric flow rate, This refers to the instantaneous dosage of polyacrylamide. This refers to the instantaneous dosage of polyaluminum chloride. The minimum dosage reference constant is used, and the logarithmic function in the formula consists of a steady-state absolute flow term. With dynamic impact flow term It is composed of superposition.

[0017] By introducing the nonlinear correlation between the dynamic variation characteristics of the instantaneous influent flow rate and the ratio of instantaneous dosage of polyaluminum chloride and polyacrylamide, the flocculants are accurately predicted to be physically broken under the current operating conditions. This effectively ensures the robustness of the flocculant microstructure and improves the engineering impact resistance under dynamic flow conditions.

[0018] Preferably, the step of using the classification-enhanced gradient boosting tree prediction algorithm to predict the effluent turbidity at future times includes: tensor splicing the standardized engineering benchmark data sequence with the heavy metal and suspended solids competitive adsorption index and the floc hydraulic shear vulnerability index to form an input feature vector; inputting the input feature vector into a pre-constructed classification-enhanced gradient boosting tree prediction algorithm model for forward inference calculation to output the effluent turbidity at future times.

[0019] Preferably, the classification-enhanced gradient boosting tree prediction algorithm model adopts a symmetric decision tree structure, and uses the adsorption index of heavy metals and suspended matter and the hydraulic shear vulnerability index of flocs as the node splitting criterion constraints of the classification-enhanced gradient boosting tree prediction algorithm model.

[0020] Preferably, the step of constructing a fitness objective function based on the effluent turbidity at future times and in conjunction with engineering penalty constraints includes: constructing an operating cost minimization formula and an effluent comprehensive risk minimization formula as the fitness objective function; the operating cost minimization formula is calculated by multiplying the unit purchase cost constant of polyaluminum chloride by the instantaneous dosage of polyaluminum chloride, plus the unit purchase cost constant of polyacrylamide by the instantaneous dosage of polyacrylamide.

[0021] Preferably, the formula for minimizing the overall risk of the effluent is:

[0022] ;

[0023] In the formula, The calculation results of the formula for minimizing the overall risk of effluent are given. For future water turbidity, This is the legally mandated upper limit constant for emission turbidity. The hydraulic shear fragility index is used to describe the flocculent's vulnerability. The adsorption index is the competition between heavy metals and suspended matter.

[0024] By incorporating the hydraulic shear vulnerability index of flocs and the competitive adsorption index between heavy metals and suspended solids as penalty terms into the formula for minimizing the overall risk of effluent, the evolutionary direction of the multi-objective particle swarm optimization algorithm is forced to strictly avoid structurally fragile unsafe solution regions, ensuring stable compliance of effluent and achieving global optimization of economic and environmental benefits.

[0025] Preferably, the step of using a multi-objective particle swarm optimization algorithm to find the optimal combination of drug dosages as a recommended dosage instruction includes: initializing a group of particles within a set upper and lower limit range of drug dosage; mapping the spatial position coordinates of each particle in the group of particles to a set of candidate drug dosage schemes; updating the velocity and position of each particle in the group of particles in the solution space based on the social experience of the group and the cognitive experience of the individual, and calculating the value of the fitness objective function; and after a set maximum number of iterations, outputting a set of mutually non-dominated Pareto optimal solutions.

[0026] Preferably, the step of outputting the optimal combination of reagent dosages as a recommended ratio instruction further includes: calculating the Euclidean distance between all solution vectors in the Pareto optimal solution set and the ideal origin, extracting the compromise solution with the shortest Euclidean distance as the optimal recommended ratio instruction, converting the recommended ratio instruction into a continuous control electrical signal, and sending it to the execution end to complete the closed-loop control of the mine water treatment reagent dosing system.

[0027] The technical solution of the present invention has the following beneficial technical effects:

[0028] This invention avoids unconstrained pure data dimension combination fitting, and constructs feature parameters based on the underlying logic of fluid mechanics and chemical complexation mechanisms. This makes the algorithm optimization process subject to strict constraints of objective physical laws, thereby improving the feasibility and computational reliability of the overall data processing method in practical engineering deployment.

[0029] Furthermore, by directly introducing a mechanism feature index with clear physical penalty significance into the fitness function of the multi-objective particle swarm optimization algorithm, the optimal combination of reagent dosage output as the recommended ratio instruction naturally possesses extremely strong shock resistance properties. Under severe working conditions such as a sudden surge in water inflow or a large jump in pH, it can effectively solve the technical problem of water quality exceeding standards that often occurs when traditional algorithms experience sudden changes in working conditions. Attached Figure Description

[0030] Figure 1 This is a flowchart of an intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization, according to an embodiment of the present invention.

[0031] Figure 2 This is a comparison chart of effluent water quality under conditions of drastic changes in water quality and flow rate according to the present invention;

[0032] Figure 3 This is a frontier comparison diagram of the multi-objective optimization of the reagent operating cost and the risk of exceeding the standard in the effluent in this invention. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0034] Figure 1 This is a flowchart of a smart recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to an embodiment of the present invention, referring to... Figure 1 This includes steps S1-S4.

[0035] S1 collects multidimensional dynamic data on mine water treatment and preprocesses it to obtain a standardized engineering benchmark data sequence.

[0036] In one embodiment, an IoT communication module deployed within the programmable logic controller (PLC) data processing system of the mine water treatment workshop acquires multi-dimensional operational status data of the water treatment process in real time. Specifically, an electromagnetic flow meter acquires the instantaneous influent flow rate, measured in cubic meters per hour; an industrial-grade pH meter acquires the influent pH value, dimensionless; an online suspended solids concentration meter acquires the influent suspended solids concentration, measured in milligrams per liter; an online anodic stripping voltammetry heavy metal analyzer acquires the total equivalent concentration of heavy metal ions in the influent, measured in milligrams per liter; and the instantaneous dosage of polyaluminum chloride (PAC) and polyacrylamide (PAA) is calculated using the feedback frequency of the dosing pump frequency converter, both measured in milligrams per liter, thus forming continuous time series data.

[0037] Because the measurement mechanism of online anodic stripping voltammetry heavy metal analyzers involves physicochemical processes such as enrichment and dissolution, its single measurement cycle is relatively long, and the output data exhibits low-frequency step characteristics. Directly mixing this data with high-frequency, second-level sensor data such as flow rate and pH would lead to timing misalignment and filtering distortion. Therefore, this invention employs a zero-order hold mechanism to maintain the current measurement value on the time axis by constant interpolation before the next measurement update from the heavy metal analyzer, thereby achieving precise asynchronous alignment between low-frequency chemical measurements and high-frequency physical measurements in the time domain. Simultaneously, a sliding window mean filter is applied only to high-frequency continuous time series data susceptible to bubble and electromagnetic interference to eliminate high-frequency spike noise. This differentiated preprocessing strategy fully respects the objective physical differences of the underlying industrial sensors, ensuring the authenticity and synchronization of the input data.

[0038] Since industrial field sensors are inevitably blocked by bubbles or solid impurities, resulting in transient high-frequency spike noise, a fixed-length sliding window mean filtering algorithm is used to smooth the collected continuous time series data in order to eliminate the interference of transient high-frequency spike noise. This results in a smooth and stable standardized engineering benchmark data sequence, which is then used as the direct input source for subsequent feature index construction and predictive model inference.

[0039] Thus, by filtering and smoothing the multidimensional dynamic data, random high-frequency interference noise introduced by sensors in the industrial field can be effectively eliminated, providing a more accurate and reliable basic data source for the subsequent extraction of feature indicators and the calculation of algorithm models.

[0040] S2, based on standardized engineering benchmark data sequences, constructs the heavy metal and suspended matter competitive adsorption index and the floc hydraulic shear vulnerability index.

[0041] In one embodiment, based on the objective reaction mechanism of gold mine water, characteristic indicators with clear engineering dimensions and physical significance are constructed: the heavy metal and suspended solids competitive adsorption index and the floc hydraulic shear fragility index. First, the heavy metal and suspended solids competitive adsorption index is constructed. When polyacrylamide plays a bridging adsorption role, suspended particles are the main physical binders. A large number of heavy metal ions will undergo chemical complexation reactions with the amide groups of polyacrylamide, resulting in the consumption of the adsorption effect. When the pH value of the influent deviates from the optimal engineering pH constant for the precipitation of characteristic heavy metal ions into hydroxides, the concentration of free heavy metal ions in the water increases, and the competitive consumption effect intensifies.

[0042] As an optional interpretation of this embodiment, the above-mentioned heavy metals and suspended matter compete for the adsorption index. The construction strictly followed the coordination and complexation mechanism of polymeric flocculants, and the exponential term It accurately maps the nonlinear surge in the solubility of heavy metal ions and their antagonistic effect on the efficacy of competing with colloidal particles for micelle groups when the pH of the water deviates from the isoelectric point.

[0043] To accurately measure the deterioration state of the reagent being ineffectively consumed by heavy metals, the chemical competitive adsorption mechanism is transformed into mathematical constraints, and a competitive adsorption index between heavy metals and suspended solids is constructed, satisfying the following relationship:

[0044] ;

[0045] In the formula, The adsorption index is determined by the competition between heavy metals and suspended solids. This represents the total equivalent concentration of heavy metal ions in the influent. It is an exponential function with the natural constant as its base. This refers to the pH value of the influent. The optimal engineering pH constant for the precipitation of characteristic heavy metal ions into hydroxides. The concentration of suspended solids in the influent. It is the basic background suspended solids concentration constant, which can avoid the calculation error of zero denominator under extreme clear water conditions. The absolute value symbol is used to calculate the difference between real-time pH and the optimal engineering pH constant. The concentration unit in the numerator is milligrams per liter, the exponent term is dimensionless, the concentration unit in the denominator term is milligrams per liter, and the final ratio calculation result is a dimensionless relative evaluation index.

[0046] For example, assuming the total equivalent concentration of heavy metal ions in the influent is 10 mg / L at a certain moment, the pH value of the influent is 6, the optimal engineering pH constant for the precipitation of characteristic heavy metal ions into hydroxides is 8, the concentration of suspended solids in the influent is 100 mg / L, and the background suspended solids concentration constant is 10 mg / L, substituting the above values ​​into the formula for calculating the competitive adsorption index between heavy metals and suspended solids, the calculation process is as follows:

[0047] ;

[0048] The final calculated heavy metal and suspended matter competitive adsorption index was 0.67. This heavy metal and suspended matter competitive adsorption index can measure the deterioration state of the agent being ineffectively consumed by heavy metals.

[0049] Secondly, a hydraulic shear vulnerability index for flocs is constructed. The shear strength of flocs depends on the synergistic ratio of the formation of coagulation nuclei by polyaluminum chloride and the network pulling effect of polyacrylamide. When the instantaneous flow rate of the influent surges dynamically, huge hydraulic shear stress will be generated inside the water treatment reaction tank. At the same time, even if the flow rate remains stable, as long as the absolute flow rate is large, there will still be a continuous strong shear field in the reaction tank. If the instantaneous dosage ratio of polyaluminum chloride to polyacrylamide is unbalanced, the generated flocs will lack high-density coagulation nuclei and are prone to irreversible physical breakage under huge hydraulic shear stress.

[0050] As an optional interpretation of this embodiment, the above-mentioned floc hydraulic shear fragility index The construction integrates unsteady hydrodynamics and fractal geometry theory, and addresses several terms. This characterizes the macroscopic hydraulic shock shear force generated by the sudden change in influent flow rate, while the square ratio of reagent concentration... This reflects the structural strength relationship between the hard, dense core formed by inorganic polymer coagulants and the flexible net-like outer shell formed by organic polymer flocculants.

[0051] To accurately predict the fragility of flocs under current operating conditions, the physical shear failure mechanism is transformed into mathematical constraints, and a floc hydraulic shear fragility index is constructed, satisfying the following relationship:

[0052] ;

[0053] In the formula, The hydraulic shear fragility index is used to describe the flocculant's vulnerability. It is the natural logarithm function. This represents the instantaneous influent flow rate during the current sampling period. The data represents the instantaneous inflow flow rate from the previous data collection cycle. Since the data acquired is continuous time-series data already in operation, the data processing system's buffer always stores baseline data from the previous historical moment. There is always a value to avoid the problem of initial calculation exceeding the boundary. The rated reference volumetric flow rate, This refers to the instantaneous dosage of polyacrylamide. This refers to the instantaneous dosage of polyaluminum chloride. The minimum dosage reference constant is used, and the logarithmic function in the formula consists of a steady-state absolute flow term. With dynamic impact flow term The superposition structure, with the addition of the constant 1, ensures that the independent variable of the natural logarithm function is always greater than or equal to 1, thus guaranteeing that the logarithmic calculation result is a non-negative real number. The square term of the minimum dosage benchmark constant ensures that the denominator is always greater than zero, preventing division by zero errors.

[0054] For example, the instantaneous influent flow rate in the current sampling period is 150 cubic meters per hour, the absolute value of the difference in instantaneous influent flow rate from the previous sampling period is 50 cubic meters per hour, the rated reference volumetric flow rate is 100 cubic meters per hour, the instantaneous dosage of polyacrylamide is 2 mg / L, the instantaneous dosage of polyaluminum chloride is 5 mg / L, and the minimum dosage reference constant is 1 mg / L. Substituting the above values ​​into the calculation formula for the hydraulic shear fragility index of flocs, the calculation process is as follows:

[0055] ;

[0056] Thus, by rigorously deriving and constructing the heavy metal and suspended matter competitive adsorption index and the floc hydraulic shear vulnerability index from the dimensions of chemical competitive adsorption and physical hydraulic shear, we can objectively assess the chemical competition and physical shear risks in complex aquatic environments and establish accurate and effective a priori objective boundary constraints.

[0057] S3 integrates standardized engineering benchmark data sequences with the heavy metal and suspended solids competitive adsorption index and the floc hydraulic shear vulnerability index, and uses a classification-enhanced gradient boosting tree prediction algorithm to predict the effluent turbidity at future moments.

[0058] In one embodiment, the acquired standardized engineering benchmark data sequence is tensor-concatenated with the calculated heavy metal and suspended matter competitive adsorption index and the floc hydraulic shear vulnerability index to form an input feature vector. Subsequently, the input feature vector is input into a pre-constructed classification-enhanced gradient boosting tree prediction algorithm model for forward inference calculation to output the future effluent turbidity of water quality indicators, with the unit of future effluent turbidity being NTU. Since industrial field data is often accompanied by non-Gaussian distributed heterogeneous noise, the classification-enhanced gradient boosting tree prediction algorithm model adopts a symmetric decision tree structure to resist the risk of overfitting due to gradient bias between feature variables. The heavy metal and suspended matter competitive adsorption index and the floc hydraulic shear vulnerability index are forcibly introduced into the node splitting criterion constraint of the classification-enhanced gradient boosting tree prediction algorithm model, so that when predicting the future effluent turbidity, the classification-enhanced gradient boosting tree prediction algorithm model must prioritize satisfying the objective engineering boundary constraints of chemical competitive adsorption and physical hydraulic shear.

[0059] Thus, by comprehensively injecting the feature parameters with rigorous significance into the computation nodes of the classification augmentation gradient boosting tree prediction algorithm, the overfitting defects caused by pure data-driven approaches can be significantly suppressed, and the generalization ability and data prediction accuracy of the classification augmentation gradient boosting tree prediction algorithm model under complex and harsh working conditions can be greatly improved.

[0060] S4 constructs a fitness objective function based on the turbidity of the effluent at future times and combined with engineering penalty constraints. It then uses a multi-objective particle swarm optimization algorithm to find the optimal combination of reagent dosages and outputs the recommended ratio instruction to complete the intelligent recommendation of reagent ratios for mine water treatment.

[0061] In one embodiment, after obtaining the effluent turbidity at future moments, a multi-objective particle swarm optimization algorithm is used to search for the optimal combination of reagent dosages in the global continuous solution space. Specifically, a fitness objective function is first constructed, which includes an operating cost minimization equation and an effluent comprehensive risk minimization equation. The operating cost minimization equation is calculated by multiplying the unit purchase cost constant of polyaluminum chloride by the instantaneous dosage of polyaluminum chloride, and then adding the unit purchase cost constant of polyacrylamide by the instantaneous dosage of polyacrylamide. The effluent comprehensive risk minimization equation is as follows:

[0062] ;

[0063] In the formula, To calculate the relationship for minimizing the overall risk of effluent, a rigorous dimensionless normalization was applied to the turbidity of the effluent at future times. , legal upper limit constant for emission turbidity Hydraulic shear susceptibility index of flocs Competition for adsorption index with heavy metals and suspended solids By unifying the dosing strategies within the same mathematical topological space, the particle swarm optimization algorithm can accurately pinpoint the Pareto optimal dosing strategy under extreme water flow conditions.

[0064] By introducing the hydraulic shear vulnerability index of flocs and the competitive adsorption index between heavy metals and suspended solids as weighted product penalty terms into the effluent comprehensive risk minimization formula, the evolution direction of the multi-objective particle swarm optimization algorithm is forced to strictly avoid structurally fragile unsafe solution regions. Furthermore, when chemical competition worsens, the risk is amplified by multiplication, forcing the algorithm to tend towards safe and conservative ratios, ensuring stable compliance of effluent and achieving global optimization of economic and environmental benefits.

[0065] For example, assuming the effluent turbidity is 10 NTU in the future, the legal upper limit constant for effluent turbidity is 20 NTU, the hydraulic shear vulnerability index of flocs is 0.062, and the competition adsorption index between heavy metals and suspended solids is 0.67, substituting these values ​​into the formula for minimizing the overall risk of effluent, the calculation process is as follows:

[0066] ;

[0067] The calculation result of the comprehensive risk minimization relation for effluent is 0.887. When the hydraulic shear is extremely weak or the chemical competition is intense, the basic compliance risk will be amplified as a multiplicative factor, strictly guiding the multi-objective particle swarm optimization algorithm to avoid unsafe solutions.

[0068] A swarm of particles is initialized within the set upper and lower limits of the allowable dosage of the reagent. The spatial coordinates of each particle are mapped to a set of candidate dosing ratio schemes. The particles iteratively update their velocity and position in the solution space based on the social experience of the group and the cognitive experience of the individual, and calculate the value of the fitness objective function. After a set maximum number of iterations, a set of mutually non-dominated Pareto optimal solutions is output. The Euclidean distance of all solution vectors in the Pareto optimal solution set from the ideal origin position is calculated. The compromise solution with the shortest Euclidean distance is extracted as the optimal recommended ratio instruction. The recommended ratio instruction is converted into a continuous control electrical signal and sent to the execution end to complete the closed-loop control of the mine water treatment reagent dosing system.

[0069] Thus, by combining a multi-objective particle swarm optimization algorithm with a fitness objective function that has objective penalty constraints for optimization, it is possible to achieve synergistic optimization recommendations for reagent operating costs and water quality compliance risks while ensuring the engineering shock resistance and stability of the data processing system.

[0070] It should be noted that the specific values ​​of the optimal engineering pH constant, background suspended solids concentration constant, rated reference volumetric flow rate, minimum dosage reference constant, legal upper limit constant for emission turbidity, and unit procurement cost constant involved in the above formulas of this invention are not set blindly, but are known basic parameters that are pre-calibrated and stored in the storage unit of the data processing system controller based on the physical equipment nameplate parameters of specific mine water treatment plants, historical prior data from laboratory small-scale titration tests, and relevant national and local legal environmental emission standards.

[0071] like Figure 2 As shown, the horizontal axis corresponds to the forward process of the continuous time series data collected and processed by this invention, and the vertical axis corresponds to the predicted value of effluent turbidity at future moments output by the model. During the hydraulic impact stage of a sudden water inrush in mine water, the control response of this invention is rapid. Relying on the feedforward penalty interception mechanism of characteristic indicators, the effluent turbidity curve exhibits a smooth fluctuation and is always stably suppressed below the legal emission upper limit threshold, verifying the efficient and realistic adaptability of this solution to harsh working conditions.

[0072] like Figure 3 As shown, the horizontal axis corresponds to the calculation result of the operating cost minimization relation in the fitness objective function of this invention. During the calculation process, the unit procurement cost constant is used as a relative price weight in the calculation, so that the calculation result of the horizontal axis is a dimensionless relative cost target value. The vertical axis corresponds to the calculation result of the effluent comprehensive risk minimization relation in the fitness objective function of this invention. This calculation result is a unified dimensionless relative risk target value. In the Pareto solution set space distribution for multi-objective search, the solution set scatter group of this invention has undergone a significant translational convergence towards the origin of the coordinate system in the lower left of the chart. The inflection point of the intelligent recommendation of the optimal ratio in the figure verifies that this scheme has successfully and reasonably achieved global synchronous optimization of engineering operating cost reduction and safety compliance risk control at the mathematical optimization level.

[0073] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for intelligent recommendation of mine water treatment reagent ratios based on multi-objective optimization, characterized in that, include: Collect multidimensional dynamic data on mine water treatment and preprocess it to obtain a standardized engineering benchmark data sequence; Based on standardized engineering benchmark data sequences, a competition adsorption index between heavy metals and suspended solids, as well as a hydraulic shear vulnerability index of flocs, were constructed. for: ； This represents the total equivalent concentration of heavy metal ions in the influent. It is an exponential function with the natural constant as its base. This refers to the pH value of the influent. The optimal engineering pH constant for the precipitation of characteristic heavy metal ions into hydroxides. The concentration of suspended solids in the influent. It is the basic background suspended matter concentration constant, and the absolute value sign is used to calculate the difference between the real-time pH and the optimal engineering pH constant; Floc hydraulic shear susceptibility index for: ； It is the natural logarithm function. This represents the instantaneous influent flow rate during the current sampling period. This represents the instantaneous influent flow rate from the previous data collection period. The rated reference volumetric flow rate, This refers to the instantaneous dosage of polyacrylamide. This refers to the instantaneous dosage of polyaluminum chloride. The minimum dosage reference constant is used, and the logarithmic function in the formula consists of a steady-state absolute flow term. With dynamic impact flow term Superimposed structure; The standardized engineering benchmark data sequence is fused with the heavy metal and suspended solids competitive adsorption index and the floc hydraulic shear vulnerability index, and the classification-enhanced gradient boosting tree prediction algorithm is used to predict the effluent turbidity at future moments. The classification-enhanced gradient boosting tree prediction algorithm model adopts a symmetric decision tree structure, and uses the adsorption index of heavy metals and suspended matter and the hydraulic shear vulnerability index of flocs as the node splitting criterion constraints of the classification-enhanced gradient boosting tree prediction algorithm model. Based on the turbidity of the effluent at future times, a fitness objective function is constructed in conjunction with engineering penalty constraints. A multi-objective particle swarm optimization algorithm is used to find the optimal solution and output the best combination of reagent dosages as a recommended ratio instruction to complete the intelligent recommendation of reagent ratios for mine water treatment.

2. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 1, characterized in that, The process involves collecting and preprocessing multidimensional dynamic data on mine water treatment to obtain standardized engineering benchmark data sequences, including: real-time acquisition of influent instantaneous flow rate, influent pH value, influent suspended solids concentration, influent total equivalent concentration of heavy metal ions, instantaneous dosage of polyaluminum chloride, and instantaneous dosage of polyacrylamide in the water treatment process, to form high-frequency continuous time series data; and periodically acquiring the total equivalent concentration of heavy metal ions in the influent, to form low-frequency discrete time series data. The high-frequency continuous time series data is smoothed using a fixed-length sliding window mean filtering algorithm, and the low-frequency discrete time series data is filled with time axis interpolation using a zero-order hold, so that it is asynchronously aligned with the sampling rate of the smoothed high-frequency continuous time series data, thereby obtaining a standardized engineering reference data sequence.

3. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 1, characterized in that, The method of using a classification-enhanced gradient boosting tree prediction algorithm to predict future effluent turbidity includes: The standardized engineering benchmark data sequence is tensor-concatenated with the heavy metal and suspended matter competitive adsorption index and the floc hydraulic shear vulnerability index to form an input feature vector. The input feature vector is fed into a pre-built classification enhancement gradient boosting tree prediction algorithm model for forward inference calculation to output the effluent turbidity at the future time.

4. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 1, characterized in that, The fitness objective function, constructed based on the future effluent turbidity and combined with engineering penalty constraints, includes: The relationships for minimizing operating costs and minimizing comprehensive effluent risk are constructed as the fitness objective functions. The formula for minimizing operating costs is calculated by multiplying the unit purchase cost constant of polyaluminum chloride by the instantaneous dosage of polyaluminum chloride, and then adding the unit purchase cost constant of polyacrylamide by the instantaneous dosage of polyacrylamide.

5. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 4, characterized in that, The formula for minimizing the overall risk of the effluent is: ； In the formula, The calculation results of the formula for minimizing the overall risk of effluent are given. For future water turbidity, This is the legally mandated upper limit constant for emission turbidity. The hydraulic shear fragility index is used to describe the flocculant's vulnerability. The adsorption index is the competition between heavy metals and suspended matter.

6. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 1, characterized in that, The process employs a multi-objective particle swarm optimization algorithm to find the optimal combination of drug dosages, outputting the best dosage as a recommended ratio instruction. This includes: A swarm of particles is initialized within the upper and lower limits of the set drug dosage. The spatial coordinates of each particle in the swarm are mapped to a set of candidate drug dosing schemes. Each particle in the swarm updates its velocity and position in the solution space based on the social experience of the group and the cognitive experience of the individual, and calculates the value of the fitness objective function. After a set maximum number of iterations, a set of mutually non-dominated Pareto optimal solutions is output.

7. The intelligent recommendation method for mine water treatment reagent ratio based on multi-objective optimization according to claim 6, characterized in that, The optimal combination of drug dosages, as a recommended formulation instruction, also includes: Calculate the Euclidean distance of all solution vectors in the Pareto optimal solution set from the ideal origin, extract the compromise solution with the shortest Euclidean distance as the optimal recommended ratio instruction, convert the recommended ratio instruction into a continuous control electrical signal, and send it to the execution end to complete the closed-loop control of the mine water treatment agent dosing system.

Images