A layered multi-agent-based mineral development whole-process intelligent decision method and system

Abstract

The application provides a mineral development whole-process intelligent decision-making method and system based on hierarchical multi-agent, which comprises the following steps: constructing a mineral development whole-process hierarchical collaborative decision-making architecture based on three-level intelligent decision-making agents; designing a communication protocol; establishing a quantitative coupling relationship between cross-link coupling constraints and process parameters; constructing state spaces and action spaces of the three-level intelligent decision-making agents; designing a global objective function, a link objective function and a key parameter collaborative optimization model; establishing an adaptive weight optimization mechanism; designing a distributed collaborative optimization mechanism based on an alternating direction multiplier method; implementing an agent collaborative decision-making process of evolutionary game; converting the selected decision-making scheme into specific production instructions and issuing the instructions for execution; monitoring the execution effect and feedback data in real time; and dynamically adjusting and optimizing the strategy. The application realizes multi-objective collaborative optimization of the whole life cycle of mineral development, and promotes the intelligentization, greenization and sustainable development of mineral resource development.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to an intelligent decision-making method and system for the entire process of mineral development based on hierarchical multi-agent systems. Background Technology

[0002] With the continuous growth of global demand for mineral resources and the continuous improvement of the level of intelligence in the mining industry, mineral resource development is gradually developing towards large-scale, complex, and refined operations. Mineral resource development is a typical complex industrial process, encompassing multiple closely related links such as mining, beneficiation, and smelting. Its overall efficiency is closely related to the level of collaborative operation throughout the entire process.

[0003] However, existing mineral resource development decision-making processes face numerous technical bottlenecks, severely restricting development efficiency and sustainability. First, fragmented processes make overall optimization difficult. Most existing optimization systems target a single stage, lacking cross-stage collaborative decision-making mechanisms. When mining departments pursue high output, they may introduce high-impurity ore into the beneficiation stage, leading to a decrease in overall recovery rate and an increase in costs. Second, multiple conflicting objectives lack quantitative balance. Complex trade-offs exist between objectives such as economic benefits, resource recovery, energy consumption, and environmental impact. Existing technologies often employ fixed-weighting methods, failing to dynamically adapt to market changes and technological shifts. Decision-making processes rely heavily on expert experience, resulting in strong subjectivity. Third, static models struggle to cope with dynamic changes. Traditional optimization models are mostly static, unable to respond in real-time to dynamic factors such as changes in ore properties, equipment status fluctuations, and market price fluctuations, leading to poor adaptability of decision-making solutions.

[0004] Therefore, developing an intelligent decision-making method and system for the entire process of mineral resource development, and realizing multi-objective dynamic collaborative decision-making throughout the mining, beneficiation, and metallurgical processes, is key to improving the overall efficiency and sustainability of mineral resource development, and is of great significance to promoting high-quality development of the mining industry. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] Based on this, the present invention provides an intelligent decision-making method and system for the entire process of mineral development based on hierarchical multi-agent, in order to solve the problems mentioned in the background technology, such as the difficulty of global optimization caused by the fragmentation of links and the lack of cross-link collaborative decision-making; the use of fixed weights in multi-objective optimization, which cannot be dynamically changed; and the difficulty of static models to cope with dynamic changes.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, this invention provides an intelligent decision-making method for the entire mineral development process based on hierarchical multi-agent systems, comprising:

[0009] Step S1: Construct a hierarchical collaborative decision-making architecture for the entire mineral development process based on three-level intelligent decision-making entities. The three-level intelligent decision-making entities include: a global coordination intelligent entity, a process collaboration intelligent entity, and a unit execution intelligent entity; design an intelligent entity communication protocol based on the fusion of "material flow-energy flow-information flow"; establish coupling constraints between cross-process material flow and energy flow; and establish quantitative coupling relationships between mining, beneficiation, and metallurgical process parameters.

[0010] Step S2: Construct a three-level intelligent decision-making agent based on the state space of process coupling and the action space based on process constraints;

[0011] Step S3: Design a collaborative optimization model for the global objective function, the component objective function, and key parameters; establish an adaptive weight optimization mechanism to dynamically adjust the global objective weights according to changes in the external environment;

[0012] Step S4: Design a distributed collaborative optimization mechanism based on the Alternating Direction Multiplier Method (ADMM): Under the premise of satisfying the coupling constraints of cross-stage material flow and energy flow, the Alternating Direction Multiplier Method is used to solve the optimal process parameter solution of the local objectives and global constraints of each stage by iteratively interacting the intermediate and dual variables of the intelligent agents in each stage.

[0013] Step S5: Implement the agent collaborative decision-making process of evolutionary game: Based on the solution obtained by ADMM collaborative optimization in step S4, generate an initial policy population, and carry out evolutionary iteration through selection, crossover, mutation and multi-objective fitness evaluation, and output a Pareto optimal policy set containing multiple non-dominated solutions for decision selection.

[0014] Step S6: Transform the selected decision plan into specific production instructions for execution, monitor the actual execution effect in real time, collect execution feedback data, dynamically adjust and optimize strategies, and form a closed-loop optimization mechanism.

[0015] This invention also discloses an intelligent decision-making system for the entire process of mineral development based on hierarchical multi-agent systems, comprising:

[0016] At least one processor; and at least one memory communicatively connected to said processor, wherein:

[0017] The memory stores program instructions that can be executed by the processor, and the processor can execute the method by calling the program instructions.

[0018] (III) Beneficial Effects

[0019] As can be seen from the above technical solution, the beneficial effects of the intelligent decision-making method and system for the entire process of mineral development based on hierarchical multi-agent proposed in this invention are as follows:

[0020] 1. Breaking down inter-process barriers to achieve global optimum: By establishing coupled constraints and a hierarchical decision-making architecture for mining, beneficiation, and smelting, the entire decision-making chain of mining, beneficiation, and smelting is connected, breaking down the "information silos" between processes and avoiding the local optimization trap of "exchanging mining depletion for output, leading to a surge in beneficiation costs." Combining global collaborative variables and Lagrange multiplier updates, the collaborative agents of each process are driven to converge synchronously to the global optimum that satisfies all physical constraints during iteration, rather than a series of suboptimal solutions, effectively breaking down information and decision barriers between processes.

[0021] 2. Embedding Physical and Energy Coupling Constraints to Ensure Decision-Making Safety: By establishing cross-stage material and energy flow balance equations, the process mechanism is transformed into computable coupling constraints and embedded into the global coordination subproblem of ADMM. In each iteration, by solving the constrained global consistency problem, it is ensured that the output decision solution strictly meets the physical feasibility and equipment safety domain of actual production, thereby avoiding the illegal operations that may occur in a purely data-driven model and significantly improving the reliability and feasibility of the decision.

[0022] 3. Dynamic weighting mechanism flexibly responds to external environment and market fluctuations: This invention designs an adaptive weight adjustment mechanism led by a globally coordinating intelligent agent. By sensing external environment vectors in real time (price index, environmental policy intensity, demand prosperity), and combining iterative weight-policy dual-layer optimization operators, the weight coefficients of economic, resource, energy consumption, and environmental objectives are dynamically adjusted. This ensures that the optimization direction throughout the entire process can keep pace with market and policy changes, achieving dynamic balance among multiple objectives and maximizing long-term comprehensive benefits. Attached Figure Description

[0023] The features and advantages of the invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the drawings:

[0024] Figure 1 This is a flowchart of the intelligent decision-making method for the entire mineral development process based on hierarchical multi-agent systems, as described in this invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention provides an intelligent decision-making method for the entire mineral development process based on hierarchical multi-agent systems, including:

[0027] Step S1: Construct a hierarchical collaborative decision-making architecture for the entire mineral development process based on three-level intelligent decision-making entities. The three-level intelligent decision-making entities include: a global coordination intelligent entity, a process collaboration intelligent entity, and a unit execution intelligent entity; design an intelligent entity communication protocol based on the fusion of "material flow-energy flow-information flow"; establish coupling constraints between cross-process material flow and energy flow; and establish quantitative coupling relationships between mining, beneficiation, and metallurgical process parameters.

[0028] Based on the physical processes and management levels of mineral resource development, a three-tiered, penetrating, hierarchical collaborative decision-making architecture for the entire mining process, based on technological and physical constraints, is constructed. This architecture maps the material and energy flows of the physical world into information flows (including control and data flows) in the decision-making space. The core objective of this architecture is to achieve global dynamic optimization of the entire mineral development process, rather than local optimization of a single stage, through the collaboration of three intelligent decision-making entities and driven by the fusion of the material, energy, and information flows. Specifically, this includes:

[0029] Step S11: Construct a three-level intelligent decision-making body based on the coupling relationship of mining, beneficiation and metallurgy processes to ensure that the decision-making is deeply adapted to the process logic; the three-level intelligent decision-making body includes: a global coordination intelligent body, a process coordination intelligent body, and a unit execution intelligent body;

[0030] The global coordination agent (first-level agent), as the global strategic layer, is responsible for sensing external market and environmental policies, dynamically setting and adjusting global target weights (including economy, resources, energy consumption, and environment), and issuing resource quotas and constraint instructions to downstream entities. Its inputs are external environmental data (prices, intensity of environmental protection policies, and demand prosperity) and production status feedback from each link. Its outputs are global target weights and macro-constraint instructions from each link (such as the upper limit of mining dilution rate, the lower limit of ore beneficiation recovery rate, and the total energy consumption quota of the entire process).

[0031] The process-coordinating intelligent agents (secondary entities) include a geological and mining collaborative intelligent agent, a mineral processing and separation optimization intelligent agent, and a smelting and extraction control intelligent agent, corresponding to the three major physical processes of mineral development (mining, mineral processing, and smelting), respectively. These agents are responsible for receiving instructions from the overall coordination intelligent agent and coordinating parameter matching among internal process units. Each process-coordinating intelligent agent is responsible for optimization within its own process and collaborates with other process-coordinating intelligent agents to achieve cross-process matching.

[0032] The geological and mining collaborative intelligent agent corresponds to the mining production technology department and dispatch center. Based on geological models and market prices, it dynamically plans mining areas, output, and ore blending schemes, and determines boundary grades. Daily mineral output and ore blending ratio It outputs ore quantity and grade data to the mineral processing stage.

[0033] The mineral processing separation optimization intelligent agent corresponds to the central control room of the mineral processing plant. Based on receiving the ore feedstock, it balances the relationship between grinding fineness and recovery rate, optimizes the reagent system, and determines the grinding particle size. Dosage of flotation reagents The pH value of the flotation pulp is adjusted to deliver concentrate to the smelting process.

[0034] The smelting extraction control agent corresponds to the smelter's production control center. It processes valuable metals and impurities in the concentrate, optimizes smelting reaction conditions, maximizes metal recovery, controls emissions, and determines the smelting furnace temperature. Oxygen concentration and melt residence time .

[0035] The unit execution agent (level three main body) corresponds to a specific process unit (such as blasting unit, crushing unit, grinding and flotation unit, smelting unit), and is responsible for receiving instructions from the link coordination agent and outputting the equipment operation parameters of the process unit.

[0036] The precise execution of the three-level entities provides the foundation for the optimization of the two-level entities; the coordination of the two-level entities ensures that the overall goals of the first-level entities are achieved. Information converges from the bottom up, and instructions are decomposed from the top down.

[0037] Step S12: Design an intelligent agent communication protocol to ensure that the data interaction between intelligent agents conforms to the process logic;

[0038] Intelligent agent communication protocols include: feedforward interaction protocol, feedback interaction protocol, global broadcast protocol, and energy flow cooperative scheduling.

[0039] The feedforward interaction protocol pushes predicted material attribute values ​​from upstream links to downstream links. For example, the geological mining collaborative intelligent agent pushes the three-part data of "feed grade - output quantity - blending ratio" to the mineral processing separation optimization intelligent agent, and the mineral processing separation optimization intelligent agent pushes the data of "concentrate grade - impurity content" to the smelting extraction control intelligent agent, with a transmission delay of ≤10s.

[0040] The feedback interaction protocol is a process in which the downstream links provide feedback on the processing effect and bottleneck constraints to the upstream links. For example, if the smelting and extraction control agent detects that the concentrate impurities are too high, causing the furnace condition to be unstable, it sends an "impurity penalty signal" to the beneficiation and separation optimization agent, prompting the beneficiation and separation optimization agent to adjust the flotation reagent dosage.

[0041] The global broadcast protocol is a protocol in which the global coordinating agent broadcasts the global target weights and environmental parameters to the coordinating agents of each link. For example, when a parameter of a certain link exceeds the upper limit of the process (such as the furnace temperature of smelting ≥1270℃), it will provide feedback to the global coordinating agent in real time, triggering the adjustment of the global target weights.

[0042] Energy flow collaborative scheduling is a global coordinating intelligent agent that summarizes the energy consumption demand of each link, combines peak and valley electricity prices and the potential for cascade utilization of thermal energy, and issues energy consumption optimization instructions to the collaborative intelligent agents of each link.

[0043] The intelligent agent communication protocol maps the material flow and energy flow of the physical world into the information flow (including control flow and data flow) of the decision space, realizing the integration of the three: material flow, energy flow and information flow.

[0044] Step S13: Establish coupling constraints for cross-stage material flow and energy flow;

[0045] Construct cross-stage material flow balance constraints and energy flow constraints to ensure the feasibility of decision-making schemes.

[0046] Material flow constraints include mining balance, beneficiation balance, and metal flow balance.

[0047] The expression for the mining-benefit balance (benefit processing capacity limited by mining output) is as follows:

[0048]

[0049] in, This indicates the daily ore yield at the mining site; Indicates the daily processing capacity of mineral processing; This indicates the impoverishment rate.

[0050] The expression for the smelting-benefit balance (where smelting throughput is limited by concentrate production) is shown below:

[0051]

[0052] in, This indicates the daily output of concentrate; This indicates the daily concentrate processing capacity of the smelter.

[0053] The expression for the metal flow equilibrium constraint is as follows:

[0054]

[0055] in, This indicates the daily output of concentrate; This indicates the daily ore yield at the mining site; Indicates the grade of the ore; Indicates the mineral processing recovery rate; Indicates the grade of the mineral concentrate; This indicates the daily output of metals; This indicates the smelting recovery rate.

[0056] The expression for the energy flow balance constraint is as follows:

[0057]

[0058] in, Indicates total daily energy consumption; Indicates mining energy consumption; Indicates energy consumption in mineral processing; Indicates smelting energy consumption; This indicates the upper limit of total energy consumption.

[0059] Step S14: Establish quantitative coupling relationships and process capability constraints among process parameters in each stage of mining, beneficiation, and smelting;

[0060] Establish a quantitative coupling relationship among process parameters in mining, beneficiation, and smelting, and quantify the sensitivity of upstream parameters to downstream indicators. The expression is as follows:

[0061]

[0062]

[0063] in, This indicates the grinding fineness (i.e., the particle size corresponding to 80% throughput). Indicates the reference value for grinding fineness; Indicates the amount of flotation reagent used; This indicates the baseline value for the dosage of flotation reagents; This indicates the energy consumption per unit of smelting. Indicates the temperature of the smelting furnace; This represents the reference value for smelting furnace temperature; Indicates oxygen concentration; This represents the baseline value for oxygen enrichment concentration; , , , , , All of these represent coupling coefficients, calibrated based on actual measurements.

[0064] Process capacity constraints include: equipment capacity constraints, grinding fineness constraints, smelting temperature safety constraints, and environmental and regulatory constraints, as shown below:

[0065] Equipment capacity constraints: ;in, , , These represent the actual instantaneous throughput, minimum throughput, and maximum throughput of a certain process unit (such as a crusher, ball mill, flotation cell, or smelting furnace) at time t, respectively.

[0066] Grinding fineness constraints: ;

[0067] Smelting temperature safety constraints: ;in, Indicates the temperature of the smelting furnace;

[0068] Environmental and regulatory constraints: ;in, This indicates the emission concentration of sulfur dioxide in flue gas; This indicates the emission concentration limit for sulfur dioxide;

[0069] The specific values ​​in the above formulas can be determined based on the actual production design.

[0070] Step S2: Construct a three-level intelligent decision-making agent based on the state space of process coupling and the action space based on process constraints; specifically including:

[0071] Step S21: Construct the state space for each agent;

[0072] The state space of the globally coordinating intelligent agent focuses on the value and comprehensive indicators of the entire process, and perceives the external environment and global state in real time, as defined below:

[0073]

[0074] in, Indicates a price index, and , This indicates the metal price for the day. Indicates the benchmark price; This represents the intensity index of environmental policies, with a value ranging from 0 to 1. The stricter the environmental policies, the larger the value. This indicates the demand prosperity index, and , This indicates the total number of orders placed that day (in tons). Indicates maximum production capacity (tons / day); This represents the net present value (in ten thousand yuan). Indicates the total resource recovery rate (%). Indicates total daily energy consumption; Indicates the environmental impact index;

[0075] The state space of the collaborative intelligent agent is strongly bound to the process parameters.

[0076] The state space of a geological and mining collaborative intelligent agent is defined as follows:

[0077]

[0078] in, Indicates the mineable grade of the ore body; Indicates the mining dilution rate; Indicates energy consumption per unit of mining; Indicates equipment availability;

[0079] The state space of the mineral processing and separation optimization agent is defined as follows:

[0080]

[0081] in, Indicates the grade of the ore feed for mineral processing; Indicates the mineral processing recovery rate; This indicates the unit energy consumption of grinding; Indicates the amount of flotation reagent used; Indicates the pH value of the flotation pulp;

[0082] The state space of the smelting and extraction control agent is defined as follows:

[0083]

[0084] in, Indicates the grade of the concentrate; Indicates the metal recovery rate in smelting; Indicates the temperature of the smelting furnace; This indicates the energy consumption per unit of smelting. This indicates the emission concentration of sulfur dioxide in flue gas;

[0085] Step S22: Construct the action space of each agent based on process constraints;

[0086] The action space of the globally coordinating agent focuses on dynamically adjusting target weights, rather than directly intervening in process operations, and is defined as follows:

[0087]

[0088] in, Indicates the adjustment amount, i.e. This represents the global target weight adjustment amount (±0.05). Specifically, Indicates the amount of economic weighting adjustment. This indicates the amount of resource weight adjustment. This indicates the amount of energy consumption weighting adjustment. This indicates the amount of environmental weight adjustment;

[0089] This invention guides the collaborative optimization direction of the collaborative agents in each stage by using global target weights.

[0090] The action space of the collaborative intelligent agents in each stage directly corresponds to the adjustment of process parameters, and conforms to the equipment operation range and process feasibility.

[0091] The action space of a geological and mining collaborative intelligent agent is defined as follows:

[0092]

[0093] in, This indicates the boundary grade adjustment amount (±0.1%). This indicates the daily ore production adjustment (±100t / d). This indicates the adjustment amount (±5%) in the ore blending ratio.

[0094] The action space of the mineral processing and separation optimization agent is defined as follows:

[0095]

[0096] in, This indicates the grinding fineness adjustment amount (±2%). This indicates the adjustment amount for the flotation reagent dosage (±20g / t). This indicates the amount of pH adjustment for the flotation pulp (±0.3).

[0097] The action space of the smelting and extraction control agent is defined as follows:

[0098]

[0099] in, This indicates the furnace temperature adjustment amount (±5℃). This indicates the adjustment amount for oxygen enrichment concentration (±0.5%). This indicates the adjustment amount for the melt residence time (±5 min).

[0100] Step S3: Design a collaborative optimization model for the global objective function, the component objective function, and key parameters; establish an adaptive weight optimization mechanism to dynamically adjust the global objective weights according to changes in the external environment; specifically including:

[0101] Step S31: Design the global objective function;

[0102] The global objective function is:

[0103]

[0104] in, Represents net present value, and , Indicates metal production. This indicates the metal price for the day. This represents the total production cost. Indicates the tax rate; Indicates the total resource recovery rate, and , Indicates the mining recovery rate. , Indicates the mineral processing recovery rate. Indicates the metal recovery rate in smelting; Indicates total daily energy consumption, and , Indicates energy consumption per unit of mining. This indicates the unit energy consumption of grinding. This indicates the energy consumption per unit of smelting. This indicates the daily output of concentrate; This indicates the daily ore output during mining operations. This indicates the daily output of metals; Indicates the environmental impact index, and , Indicates dust concentration. Indicates COD emissions. express Emissions , , All are standardized to the 0-1 range; Represents the dynamic global target weight. Indicates economic weight. Indicates resource weight, Indicates energy consumption weight. Represents environmental weights, and .

[0105] Step S32: Design the objective function for each stage;

[0106] The objective function of each stage transforms the overall decision obtained by optimizing the global objective function into the objective function of each stage for further optimization. The objective function of each stage can directly constrain and evaluate every decision in the action space of its own stage, as follows:

[0107] Objective function for mining:

[0108]

[0109] in, Indicates the metal price for the day; This indicates the daily ore yield at the mining site; Indicates the grade of the ore; Indicates the mineral processing recovery rate; Indicates the metal recovery rate in smelting; Indicates the tax rate; This indicates mining costs, including costs such as blasting, transportation, and labor, which vary depending on the amount of ore produced and the difficulty of the operation. Indicates the mining dilution rate; Indicates energy consumption per unit of mining; and The coefficient representing the collaborative penalty term, Penalty for low ore grade Seriously deviating from the grade of mineral processing feed , Penalty for mining output Exceeding the maximum daily processing capacity .

[0110] Objective function for mineral processing:

[0111]

[0112] in, This indicates the daily output of concentrate; Indicates the grade of the mineral concentrate; This indicates the energy consumption cost of grinding, which is related to the grinding particle size. The function; The cost of the flotation reagents is a function of the amount of flotation reagents used. This indicates the unit energy consumption of grinding; The coefficient representing the coefficient of the collaborative penalty term, representing the daily output of the penalty concentrate. Exceeding the maximum daily concentrate processing capacity of the smelter .

[0113] Objective function for the smelting process:

[0114]

[0115] in, Indicates the metal recovery rate in smelting; This represents the smelting cost, which includes both energy consumption and refractory material costs. This indicates the energy consumption per unit of smelting. This indicates the environmental impact index of the smelting process.

[0116] Step S33: Design a collaborative optimization model for key parameters;

[0117] To address the convergence difficulties and unreasonable risks arising from the high dimensionality and non-convexity of the whole-process optimization problem, a collaborative optimization model for key parameters based on process physical and chemical mechanisms (including: ore grade optimization function, grinding fineness optimization function, and smelting temperature optimization function) is constructed to provide reasonable basic constraints and search guidance for intelligent decision-making throughout the whole process.

[0118] The ore grade optimization function is shown below:

[0119]

[0120] in, Indicates the metal price for the day; It represents metal production. The function; , , These represent the costs of mining, beneficiation, and smelting, respectively.

[0121] The grinding fineness optimization function is shown below:

[0122]

[0123] in, Indicates the fineness of the grinding process; This indicates the unit energy consumption of grinding. The function; Indicates the mineral processing recovery rate; This represents the tradeoff coefficient.

[0124] The smelting temperature optimization function is shown below:

[0125]

[0126] in, This indicates the energy consumption per unit of smelting. , Indicates the trade-off coefficient; Indicates the metal recovery rate in smelting; This indicates the emission concentration of sulfur dioxide in flue gas.

[0127] Step S34: Design a global target weight adaptive adjustment mechanism;

[0128] To resolve the inherent conflicts between economic, resource, energy consumption, and environmental goals, a dynamic weight determination process led by a global coordinating agent is designed, which determines the global goal weights through a two-level mechanism of "environmental perception + optimization solution".

[0129] First, real-time monitoring of the external environment vector, comprised of price indices, the intensity of environmental policies, and demand sentiment. , The global coordinating agent, based on the external environment vector, uses a pre-defined weighted prior function. And Softmax normalization, to generate prior values ​​for the global target weights. The calculation formula is as follows:

[0130]

[0131] in, It is a 4×3 parameter matrix used to establish the mapping relationship between external environmental variables and global target weights; Indicates the first The external environment variable affects the first The influence coefficients of each global objective weight (including economic, resource, energy consumption, and environmental factors) can be determined using historical operational data or expert experience, for example... This represents the coefficient of influence of the price index on the weight of economic objectives. This represents the coefficient that indicates the impact of the intensity of environmental protection policies on the weight of economic objectives.

[0132] Secondly, a weight-policy two-level optimization solution is performed. Considering that prior weights may not reflect current production constraints and safety boundaries, the global objective weight vector itself is used as a decision variable and incorporated into the optimization model along with the decision variables of each stage:

[0133]

[0134] in, These represent the decision variables at each stage; Represents the global target weight vector; Represents the prior weight vector; Represents the global objective function; Indicates the anti-oscillation regularization term; This represents the regularization strength coefficient, used to control the magnitude of weight adjustment; This represents the entropy regularization coefficient, used to prevent excessive concentration of weights on a single objective and maintain a balance among multiple objectives.

[0135] In the specific solution process, given the external environment vector and prior weights Based on this, an iterative "weight-policy two-layer optimization operator" is adopted, where the inner layer optimizes the current weight. Next, the ADMM distributed collaborative optimization from step S4 is invoked to solve for the decision variables of each stage. The outer layer calculates the corresponding overall performance metrics (i.e., multi-objective fitness values). Adjust through iterative search or heuristic algorithms To maximize long-term overall benefits, updated weights are obtained. When the change in the weight vector is less than the threshold, i.e. (For example And the relative change in the optimization function is less than the threshold, i.e. (For example When convergence is reached, it is determined that the convergence is achieved.

[0136] By periodically running the optimizer, the global target weights fluctuate slowly with price indices, environmental intensity, and demand sentiment, achieving adaptive and online optimization under different scenarios.

[0137] Step S4: Design a distributed collaborative optimization mechanism based on the Alternating Direction Multiplier Method (ADMM): Under the premise of satisfying the coupling constraints of cross-stage material flow and energy flow, the Alternating Direction Multiplier Method is used to solve the optimal process parameter solution of the local objectives and global constraints of each stage by iteratively interacting the intermediate variables and dual variables of the intelligent agents in each stage.

[0138] To address the decision-making complexity caused by coupling constraints between different stages (such as capacity matching and quality transfer), a parallel collaborative solution algorithm is designed. Specifically, it includes:

[0139] Step S41: Distributed problem modeling and decomposition: Based on the decision architecture, state and action space and objective function constructed in steps S1-S3, the whole process optimization problem is decomposed into sub-problems in three stages: mining, beneficiation and smelting, and connected by a set of coupling constraint equations;

[0140] The problem of collaborative optimization of the entire mining, beneficiation, and metallurgical process is constructed in the following distributed form:

[0141] Let the decision variables for the mining, beneficiation, and smelting processes be respectively... The global optimization problem is then:

[0142]

[0143] in, This represents the decision variables at each stage, and each stage's decision variable corresponds to the action space defined by the agent in step S22. In other words... Corresponding representation , Corresponding representation , Corresponding representation ; The global target weight vector is dynamically adjusted in step S34; Corresponds to the material flow and energy flow coupling constraints defined in step S13; The process parameter constraints are defined in step S14.

[0144] Based on the principles of the ADMM algorithm, the global optimization problem is decomposed into subproblems that can be solved in a distributed manner, and a global collaborative variable is introduced. and Lagrange multipliers Among them, global collaborative variables It is a virtual, ideal, fully coupled decision variable, whose core characteristic is that it must strictly satisfy all cross-stage coupling constraints. This represents a mathematically feasible and globally consistent "target operation point." This is achieved by adding consistency constraints. The coupling constraints of the original problem are transformed into a separable form, and an augmented Lagrangian function is constructed:

[0145]

[0146] in, This represents the objective function defined in step S32. This can be transformed into a minimization problem; This represents the penalty parameter.

[0147] Step S42: ADMM iterative solution; specifically including:

[0148] Step S421: Initialization of ADMM based on the key parameter collaborative optimization model;

[0149] Before executing ADMM distributed collaborative optimization, the key parameter collaborative optimization model from step 33 is first invoked to calculate the optimal values ​​of the key parameters under the current external environmental conditions, i.e., the optimal values ​​of the ore grade. Optimal grinding fineness Optimal smelting temperature Build an initial global collaboration scheme :

[0150]

[0151] in, , , , , , All represent the historical average of the corresponding parameter, for example: This represents the historical average daily mineral output. This represents the historical average of the ore blending ratio.

[0152] Decision variables Perform initialization: The Lagrange multipliers are initialized to zero vectors, i.e. .

[0153] Step S422: Perform ADMM iterative solution;

[0154] Based on the three-level agent decision-making architecture established in step S11, the alternating direction multiplier method is used to implement the following distributed iteration. Each agent solves its own optimization problem locally, while a "penalty term" for coupling constraints is introduced, and a global collaborative variable updated by the global coordinating agent is received. The specific steps are as follows:

[0155] (a) Local optimization phase (coordinating the execution of each stage by intelligent agents in parallel)

[0156] The collaborative agents at each stage, given a global collaborative variable and Lagrange multipliers Next, solve the optimization problem for this stage independently:

[0157] Geological and mining collaborative intelligent agents based on the optimization problem:

[0158]

[0159] The solution yields the optimized boundary grade, ore output, and ore blending ratio. (The superscript indicates that...) This represents the transpose of a matrix.

[0160] The mineral processing and separation optimization agent is based on the optimization problem:

[0161]

[0162] The optimal grinding fineness, flotation reagent dosage, and pulp pH value are obtained by solving the problem.

[0163] The smelting and extraction control agent optimizes the following problem:

[0164]

[0165] The optimal smelting furnace temperature, oxygen concentration, and melt residence time are obtained by solving the problem.

[0166] Each agent pushes the optimized intermediate results to the global coordinating agent through the feedforward interaction protocol established in step S12, with a communication delay of ≤10s.

[0167] (b) Global Coordination Phase (Executed by the Global Coordination Agent)

[0168] The global coordinating agent collects data from each stage. Solve the global collaborative variable update problem to ensure that the coupling constraints of matter flow and energy flow are satisfied:

[0169]

[0170] This problem is equivalent to projecting the solutions of each stage onto the feasible set of coupled constraints. The specific solution involves coordinating and adjusting the material flow balance. This ensures that the material flow balance constraint equations are satisfied; energy flow coordinated scheduling is implemented, and energy consumption allocation at each stage is optimized in conjunction with peak-valley electricity pricing to ensure... ;

[0171] (c) Multiplier update and convergence determination

[0172] The global coordinating agent updates the Lagrange multipliers:

[0173]

[0174] The updated version will be broadcast via the global broadcast protocol in step S12. and The broadcast is sent to the intelligent agents at each stage.

[0175] Convergence criterion: Local solution With global co-variables The algorithm converges when the original residuals are sufficiently close, i.e., when the original residuals are sufficiently close. and dual residual If all values ​​are less than a preset threshold, convergence is determined. The original residuals represent the degree of constraint violation. Sufficiently small indicates that the independent decisions of each stage have reached consensus with the global coordinating variables that satisfy all physical coupling constraints; dual residuals Measuring global covariates Its own stability.

[0176] Through multiple iterations of the three stages of "local optimization - global coordination - multiplier update and convergence determination", the decisions of all agents will converge synchronously to a globally feasible solution that satisfies all cross-stage physical constraints, thus achieving true parallel collaboration rather than serial compromise.

[0177] Step S43: Output the collaborative benchmark solution;

[0178] After ADMM converges, it outputs a cooperative benchmark solution that satisfies all physical coupling constraints. This solution also considers the current global objective weights. This solution achieves optimal economic, resource, energy consumption, and environmental outcomes. Using this solution as a high-quality initial population center for subsequent evolutionary games significantly improves evolutionary search efficiency.

[0179] Step S5: Implement the agent collaborative decision-making process of evolutionary game theory: Based on the solution obtained from ADMM collaborative optimization in step S4, generate an initial policy population, and perform evolutionary iterations through selection, crossover, mutation, and multi-objective fitness evaluation to output a Pareto optimal policy set containing multiple non-dominated solutions for decision selection; specifically including:

[0180] Step S51: Initialize the policy population;

[0181] The ADMM collaborative benchmark solution output in step S4 Based on this, an initial strategy population is generated through controlled perturbation. Among them, mining strategy individuals... Nearby, following a normal distribution Sampling, where variance Based on the process range settings of each parameter within the geological and mining collaborative intelligent body (such as...) =0.05%), generating Each mining strategy has a specific individual; similarly, beneficiation strategy individuals generate beneficiation strategy populations, and smelting strategy individuals generate smelting strategy populations. Each complete strategy is a combination of three component strategies (mining strategy individual + beneficiation strategy individual + smelting strategy individual), with an initial population size of [missing information]. .

[0182] Step S52: Based on the dynamic global target weights from step S34, perform multiple rounds of collaborative evolution iterations (maximum number of iterations). =200); in each iteration, the following loop is executed:

[0183] (a) Multi-objective fitness assessment and game utility calculation;

[0184] First, for each strategy combination in the population A full-process simulation is performed, that is, based on the coupled constraint equations in step S13, the multi-objective fitness calculation under this strategy is performed, as shown below:

[0185]

[0186] The fitness target value in the above formula ( , , , Calculate according to the formula in step S31.

[0187] Then, the game utility is calculated as follows:

[0188] Each agent in each stage evaluates the effectiveness of the strategy from its own perspective:

[0189] The effectiveness of collaborative intelligent agents in geology and mining: ;

[0190] Optimization of mineral processing and separation agent utility: ;

[0191] The effectiveness of intelligent agents in controlling smelting and extraction: ;

[0192] The utility calculation described above uses the function defined in step S32 and also includes a co-penalty term.

[0193] (b) Based on the NSGA-II framework, perform non-dominated sorting and crowding calculation;

[0194] Based on four fitness objectives (see formula) The population is divided into multiple non-dominated fronts (Pareto layers) and subjected to rapid non-dominated ordination. Then, crowding is calculated to determine the crowding distance between individuals in the same non-dominated layer, thus maintaining population diversity.

[0195] (c) Strategy selection, crossover, and variation based on game utility;

[0196] Tournament selection: Randomly select two individuals from the population, compare their non-dominated ranking and crowding, and choose the one with the better ranking as the parent; the game utility of each stage needs to be considered during the comparison (including the utility of the geological and mining collaborative agent). Optimization of mineral processing and separation by intelligent agents , Smelting and extraction control intelligent agent utility This serves as a correction factor for selection preferences. In other words, if two individuals are difficult to distinguish in terms of non-dominant ranking and crowding, the individual with higher utility in the corresponding stage of the current scenario is preferred.

[0197] Cooperative crossover operation: Simulate binary crossover (SBX) between mining strategies, exchanging... , Genes; mineral processing and smelting strategies also perform similar operations, with probability. Perform crossover.

[0198] Adaptive mutation operation: Performs polynomial mutation on the policy, with mutation probability... ( (Number of variables). The mutation intensity is adaptively adjusted based on the current generation and the convergence of the objective; the mutation intensity is increased when population diversity decreases.

[0199] (d) Environmental selection and the formation of new generation populations;

[0200] Merge parent and offspring populations (size 2) Perform non-dominated sorting, select the first Individuals form a new generation of population.

[0201] Step S53: Output the Pareto optimal policy set and decision support;

[0202] When evolution reaches its maximum number of generations Or Pareto frontier improvements less than When the time is reached, the evolution terminates. Output the non-dominated solution set of the final generation. Each solution corresponds to a four-dimensional fitness target vector. These represent its performance in four objectives: economy, resources, energy consumption, and environment.

[0203] Provides an interactive interface for decision-makers: visualizing the Pareto frontier in the form of parallel coordinate graphs, scatter plot matrices, etc., allowing decision-makers to select the final decision-making solution based on real-time preferences (such as current focus on the economy or environmental protection). .

[0204] Step S6: Transform the selected decision plan into specific production instructions for execution, monitor the actual execution effect in real time, collect execution feedback data, dynamically adjust and optimize strategies, and form a closed-loop optimization mechanism; specifically including:

[0205] Step S61: Issuance, execution, and real-time monitoring of decision-making plans;

[0206] Selected strategy This is transformed into specific production instructions for each stage; the instructions are then sent to the execution agents of each unit via the communication protocol in step S12; real-time monitoring is performed, and actual production data (including grade) is collected. ,Yield Energy consumption ,emission (etc.); through the feedback interaction protocol in step S12, the agents in each link feed back the key indicator deviation ∆ (∆=actual value − predicted value) to the global coordinating agent.

[0207] Step S62: Multi-source feedback data fusion and dynamic adjustment, including short-term online correction, medium-term model update, and long-term strategy evolution;

[0208] (a) Short-term online calibration (minute-hour level)

[0209] When a significant deviation is detected (e.g.) Triggering ADMM fast re-optimization: Starting from the current actual state, with fixed weights Unchanged; execute simplified ADMM iterations (up to 10 rounds) to quickly adjust process parameters; issue new instructions to the execution unit to achieve rolling optimization.

[0210] (b) Mid-term learning and model updates (daily-weekly)

[0211] Based on historical data accumulated weekly, including updating global objective weight adaptive model parameters and collecting external environment vectors. With the actual optimal global objective weight The corresponding data is used to update the parameter matrix in step S34 to minimize the error between the model's predicted weights and the actual optimal weights.

[0212]

[0213] in, Indicates the first The optimal global objective weight vector, verified by practice; Indicates the first The external environment vector of the sky; This represents the regularization coefficient, used to prevent overfitting. A parameter matrix representing the influence coefficients of environmental variables on the global objective weights.

[0214] Use actual data (such as actual recovery rate and actual energy consumption) to correct key parameters of the process model (i.e., all process coupling relationships and constraints in steps S13 and S14), such as actual mineral processing recovery rate. Compared with the predicted value The deviation is adjusted; the coefficients in the constraint equations are updated to improve the prediction accuracy.

[0215] (c) Long-term strategy evolution (monthly-quarterly level)

[0216] Incorporate strategies with proven effectiveness into the knowledge base of the evolutionary population; identify different external environments based on historical data. Under the efficient strategy pattern, establish a scenario-policy mapping library.

[0217] The penalty parameters are adaptively adjusted monthly based on the ADMM convergence history. If the average number of iterations increases significantly and the convergence residual increases, it indicates that the current... If the value is not suitable for the current production conditions, it will be adjusted according to the following rules:

[0218]

[0219] The crossover probability is adjusted quarterly based on the efficiency of the evolutionary search. Probability of mutation :

[0220] When population diversity declines too rapidly: increase (e.g., +0.05), and may be slightly improved. This enhances exploration capabilities and helps us break free from local optima.

[0221] When the convergence rate is too slow: appropriately reduce... and We should make better use of existing superior genes to accelerate convergence.

[0222] Step S63: Design the full-cycle closed-loop operation logic. The system will run automatically according to the following time scales:

[0223] Second-minute level: The operating unit executes the intelligent agent and collects data;

[0224] Hourly level: ADMM online correction and rapid re-optimization;

[0225] Daily level: Perform a complete round of evolutionary game and output the production plan for the next day;

[0226] Weekly update: Global target weights and process model parameters;

[0227] Monthly / Quarterly: Strategy knowledge base updates and algorithm parameter tuning;

[0228] Through this closed loop, the system forms a complete intelligent decision-making capability from rapid response to long-term evolution, ensuring that the entire process maintains optimal synergy in dynamically changing ore properties and market environments.

[0229] Finally, it should be noted that the above control method can be converted into software program instructions. It can be implemented using a control system including a processor and memory, or it can be implemented using computer instructions stored in a non-transitory computer-readable storage medium. The integrated unit implemented as a software functional unit can be stored in a computer-readable storage medium. This software functional unit, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0230] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A hierarchical multi-agent intelligent decision-making method for the entire mineral development process, characterized in that, include: Step S1: Construct a hierarchical collaborative decision-making architecture for the entire mineral development process based on a three-tiered intelligent decision-making body. The three-tiered intelligent decision-making body includes: a global coordination intelligent body, a process coordination intelligent body, and a unit execution intelligent body; design an intelligent body communication protocol based on the fusion of "material flow-energy flow-information flow"; establish coupling constraints for cross-process material flow and energy flow; and establish quantitative coupling relationships between mining, beneficiation, and metallurgical process parameters. Specifically, this includes: Step S11: Construct a three-level intelligent decision-making body based on the coupling relationship of mining, beneficiation and metallurgy processes to ensure that the decision-making is deeply adapted to the process logic; the three-level intelligent decision-making body includes: a global coordination intelligent body, a process coordination intelligent body, and a unit execution intelligent body; As a primary agent, the global coordination agent is responsible for sensing external market and environmental policies, dynamically setting and adjusting global target weights, including economic weights, resource weights, energy consumption weights, and environmental weights, and issuing resource quotas and constraint instructions to downstream entities; its inputs are external environmental data and production status feedback from each stage; its outputs are global target weights and macro-level constraint instructions from each stage. The process-coordinating intelligent agents, as secondary entities, include the geological and mining collaborative intelligent agent, the mineral processing and separation optimization intelligent agent, and the smelting and extraction control intelligent agent, which correspond to the three major physical processes of mineral development: mining, mineral processing, and smelting. They are responsible for receiving instructions from the global coordination intelligent agent and coordinating the parameter matching of each internal process unit. Each process-coordinating intelligent agent is responsible for the optimization within its own process and for collaborating with other process-coordinating intelligent agents to achieve cross-process matching. The geological and mining collaborative intelligent agent corresponds to the mining production technology department and dispatch center. Based on geological models and market prices, it dynamically plans mining areas, output, and ore blending schemes, and determines boundary grades. Daily mineral output and ore blending ratio It outputs ore quantity and grade data to the mineral processing stage; The mineral processing separation optimization intelligent agent corresponds to the central control room of the mineral processing plant. Based on receiving the ore feedstock, it balances the relationship between grinding fineness and recovery rate, optimizes the reagent system, and determines the grinding particle size. Dosage of flotation reagents The pH value of the flotation pulp is adjusted to deliver concentrate to the smelting process. The smelting extraction control agent corresponds to the smelter's production control center. It processes valuable metals and impurities in the concentrate, optimizes smelting reaction conditions, maximizes metal recovery, controls emissions, and determines the smelting furnace temperature. Oxygen concentration and melt residence time ; The unit execution agent acts as a third-level entity, corresponding to a specific process unit. It is responsible for receiving instructions from the link coordination agent and outputting the equipment operation parameters of that process unit. Step S12: Design an intelligent agent communication protocol to ensure that the data interaction between intelligent agents conforms to the process logic; The intelligent agent communication protocol includes: feedforward interaction protocol, feedback interaction protocol, global broadcast protocol, and energy flow collaborative scheduling. The feedforward interaction protocol pushes predicted material properties from upstream to downstream. The feedback interaction protocol provides feedback on processing effects and bottlenecks from downstream to upstream. The global broadcast protocol broadcasts global target weights and environmental parameters from the global coordinating agent to the collaborative agents of each stage. The energy flow collaborative scheduling protocol is where the global coordinating agent summarizes the energy consumption needs of each stage, combines peak and off-peak electricity prices and the potential for cascaded utilization of thermal energy, and issues energy consumption optimization instructions to the collaborative agents of each stage. Step S13: Establish coupling constraints for cross-stage material flow and energy flow; Material flow constraints include mining balance, smelting balance, and metal flow balance; The expression for selection balance is as follows: in, This indicates the daily ore yield at the mining site; Indicates the daily processing capacity of mineral processing; Indicates the impoverishment rate; The expression for the beneficiation equilibrium is as follows: in, This indicates the daily output of concentrate; This indicates the daily concentrate processing capacity of the smelter; The expression for the metal flow equilibrium constraint is as follows: in, This indicates the daily output of concentrate; This indicates the daily ore yield at the mining site; Indicates the grade of the ore; Indicates the mineral processing recovery rate; Indicates the grade of the mineral concentrate; This indicates the daily output of metals; Indicates the smelting recovery rate; The expression for the energy flow balance constraint is as follows: in, Indicates total daily energy consumption; Indicates mining energy consumption; Indicates energy consumption in mineral processing; Indicates smelting energy consumption; Indicates the upper limit of total energy consumption; Step S14: Establish quantitative coupling relationships and process capability constraints among process parameters in each stage of mining, beneficiation, and smelting; Establish the quantitative coupling relationship between process parameters in each stage of mining, mineral processing, and smelting, expressed as: in, Indicates the fineness of the grinding process; Indicates the reference value for grinding fineness; Indicates the amount of flotation reagent used; This indicates the baseline value for the dosage of flotation reagents; This indicates the energy consumption per unit of smelting. Indicates the temperature of the smelting furnace; This represents the reference value for smelting furnace temperature; Indicates oxygen concentration; This represents the baseline value for oxygen enrichment concentration; , , , , , Both represent coupling coefficients; Process capacity constraints include: equipment capacity constraints, grinding fineness constraints, smelting temperature safety constraints, and environmental and regulatory constraints, as shown below: Equipment capacity constraints: ;in, , , These represent the actual instantaneous throughput, minimum throughput limit, and maximum throughput limit of a certain process unit at time t, respectively. Grinding fineness constraints: ; Smelting temperature safety constraints: ;in, Indicates the temperature of the smelting furnace; Environmental and regulatory constraints: ;in, This indicates the emission concentration of sulfur dioxide in flue gas; Indicates the emission concentration limit for sulfur dioxide; Step S2: Construct a three-level intelligent decision-making agent based on the state space of process coupling and the action space based on process constraints; Step S3: Design a collaborative optimization model for the global objective function, the component objective function, and key parameters; establish an adaptive weight optimization mechanism to dynamically adjust the global objective weights according to changes in the external environment; Step S4: Design a distributed collaborative optimization mechanism based on the Alternating Direction Multiplier Method (ADMM): Under the premise of satisfying the coupling constraints of cross-stage material flow and energy flow, the Alternating Direction Multiplier Method is used to solve the optimal process parameter solution of the local objectives and global constraints of each stage by iteratively interacting the intermediate and dual variables of the intelligent agents in each stage. Step S5: Implement the agent collaborative decision-making process of evolutionary game: Based on the solution obtained by ADMM collaborative optimization in step S4, generate an initial policy population, and carry out evolutionary iteration through selection, crossover, mutation and multi-objective fitness evaluation, and output a Pareto optimal policy set containing multiple non-dominated solutions for decision selection. Step S6: Transform the selected decision plan into specific production instructions for execution, monitor the actual execution effect in real time, collect execution feedback data, dynamically adjust and optimize strategies, and form a closed-loop optimization mechanism.

2. The method according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Construct the state space for each agent; The state space of the globally coordinating intelligent agent focuses on the value and comprehensive indicators of the entire process, and perceives the external environment and global state in real time, as defined below: in, Indicates a price index, and , This indicates the metal price for the day. Indicates the benchmark price; This represents the intensity index of environmental policies, with a value ranging from 0 to 1. The stricter the environmental policies, the larger the value. This indicates the demand prosperity index, and , This indicates the total number of orders placed that day. Indicates maximum production capacity; Represents net present value; Indicates the overall resource recovery rate; Indicates total daily energy consumption; Indicates the environmental impact index; The state space of the collaborative intelligent agent is strongly bound to the process parameters; The state space of a geological and mining collaborative intelligent agent is defined as follows: in, Indicates the mineable grade of the ore body; Indicates the mining dilution rate; Indicates energy consumption per unit of mining; Indicates equipment availability; The state space of the mineral processing and separation optimization agent is defined as follows: in, Indicates the grade of the ore feed for mineral processing; Indicates the mineral processing recovery rate; This indicates the unit energy consumption of grinding; Indicates the amount of flotation reagent used; Indicates the pH value of the flotation pulp; The state space of the smelting and extraction control agent is defined as follows: in, Indicates the grade of the concentrate; Indicates the metal recovery rate in smelting; Indicates the temperature of the smelting furnace; This indicates the energy consumption per unit of smelting. This indicates the emission concentration of sulfur dioxide in flue gas; Step S22: Construct the action space of each agent based on process constraints; The action space of the globally coordinating agent focuses on dynamically adjusting target weights, rather than directly intervening in process operations, and is defined as follows: in, Indicates the adjustment amount, i.e. This represents the global target weight adjustment amount, specifically... Indicates the amount of economic weighting adjustment. This indicates the amount of resource weight adjustment. This indicates the amount of energy consumption weighting adjustment. This indicates the amount of environmental weight adjustment; The action space of the collaborative intelligent agents in each link directly corresponds to the adjustment of process parameters, and conforms to the equipment operation range and process feasibility; The action space of a geological and mining collaborative intelligent agent is defined as follows: in, Indicates the boundary grade adjustment amount; This indicates the daily ore output adjustment amount; This indicates the amount of adjustment to the ore blending ratio; The action space of the mineral processing and separation optimization agent is defined as follows: in, Indicates the amount of grinding fineness adjustment; This indicates the adjustment amount for the flotation reagent dosage; This indicates the amount of pH adjustment for the flotation pulp; The action space of the smelting and extraction control agent is defined as follows: in, Indicates the amount of temperature adjustment in the smelting furnace; Indicates the amount of oxygen concentration adjustment; This indicates the amount of adjustment for the melt residence time.

3. The method according to claim 2, characterized in that, Step S3 specifically includes: Step S31: Design the global objective function; The global objective function is: in, Represents net present value, and , Indicates metal production. This indicates the metal price for the day. This represents the total production cost. Indicates the tax rate; Indicates the total resource recovery rate, and , Indicates the mining recovery rate. , Indicates the mineral processing recovery rate. Indicates the metal recovery rate in smelting; Indicates total daily energy consumption, and , Indicates energy consumption per unit of mining. This indicates the unit energy consumption of grinding. This indicates the energy consumption per unit of smelting. This indicates the daily output of concentrate; This indicates the daily ore output during mining operations. This indicates the daily output of metals; Indicates the environmental impact index, and , Indicates dust concentration. Indicates COD emissions. express Emissions , , All are standardized to the 0-1 range; Represents the dynamic global target weight. Indicates economic weight. Indicates resource weight, Indicates energy consumption weight. Represents environmental weights, and ; Step S32: Design the objective function for each stage; The objective function of each stage transforms the overall decision obtained by optimizing the global objective function into the objective function of each stage for further optimization. The objective function of each stage can directly constrain and evaluate every decision in the action space of its own stage, as follows: Objective function for mining: in, Indicates the metal price for the day; This indicates the daily ore yield at the mining site; Indicates the grade of the ore; Indicates the mineral processing recovery rate; Indicates the metal recovery rate in smelting; Indicates the tax rate; This represents mining costs, including blasting, transportation, and labor costs that vary depending on the amount of ore produced and the difficulty of the operation. Indicates the mining dilution rate; Indicates energy consumption per unit of mining; and The coefficient representing the collaborative penalty term, Penalty for low ore grade Seriously deviating from the grade of mineral processing feed , Penalty for mining output Exceeding the maximum daily processing capacity ; Objective function for mineral processing: in, This indicates the daily output of concentrate; Indicates the grade of the mineral concentrate; This indicates the energy consumption cost of grinding, which is related to the grinding particle size. The function; The cost of the flotation reagents is a function of the amount of flotation reagents used. This indicates the unit energy consumption of grinding; The coefficient representing the coefficient of the collaborative penalty term, representing the daily output of the penalty concentrate. Exceeding the maximum daily concentrate processing capacity of the smelter ; Objective function for the smelting process: in, Indicates the metal recovery rate in smelting; This represents the smelting cost, which includes both energy consumption and refractory material costs. This indicates the energy consumption per unit of smelting. This indicates the environmental impact index of the smelting process; Step S33: Design a collaborative optimization model for key parameters; Construct a collaborative optimization model for key parameters based on process physics and chemistry mechanisms, including: ore grade optimization function, grinding fineness optimization function, and smelting temperature optimization function; The ore grade optimization function is shown below: in, Indicates the metal price for the day; It represents metal production. The function; , , These represent the costs of mining, beneficiation, and smelting, respectively. The grinding fineness optimization function is shown below: in, Indicates the fineness of the grinding process; This indicates the unit energy consumption of grinding. The function; Indicates the mineral processing recovery rate; Indicates the trade-off coefficient; The smelting temperature optimization function is shown below: in, This indicates the energy consumption per unit of smelting. , Indicates the trade-off coefficient; Indicates the metal recovery rate in smelting; This indicates the emission concentration of sulfur dioxide in flue gas; Step S34: Design a global target weight adaptive adjustment mechanism; To resolve the inherent conflicts between economic, resource, energy consumption, and environmental goals, a dynamic weight determination process led by a global coordinating agent is designed, which determines the global goal weights through a two-level mechanism of "environmental perception + optimization solution". First, real-time monitoring of the external environment vector, comprised of price indices, the intensity of environmental policies, and demand sentiment. , The global coordinating agent, based on the external environment vector, uses a pre-defined weighted prior function. And Softmax normalization, to generate prior values ​​for the global target weights. The calculation formula is as follows: in, It is a 4×3 parameter matrix used to establish the mapping relationship between external environmental variables and global target weights; Indicates the first The external environment variable affects the first The influence coefficient of each global objective weight is calibrated by historical operating data or expert experience; Secondly, a weight-policy two-level optimization solution is performed; the global objective weight vector itself is used as a decision variable and incorporated into the optimization model along with the decision variables of each stage: in, These represent the decision variables at each stage; Represents the global target weight vector; Represents the prior weight vector; Represents the global objective function; Indicates the anti-oscillation regularization term; This represents the regularization strength coefficient, used to control the magnitude of weight adjustment; This represents the entropy regularization coefficient, used to prevent excessive concentration of weights on a single objective and maintain a balance among multiple objectives. In the specific solution process, given the external environment vector and prior weights Based on this, an iterative "weight-policy two-layer optimization operator" is adopted, where the inner layer optimizes the current weight. Next, the ADMM distributed collaborative optimization from step S4 is invoked to solve for the decision variables of each stage. This yields the corresponding overall performance metrics, namely the multi-objective fitness value; the outer layer is used for calculation. Adjust through iterative search or heuristic algorithms To maximize long-term overall benefits, updated weights are obtained. When the change in the weight vector is less than the threshold, i.e. And the relative change in the optimization function is less than the threshold, i.e. When convergence is reached, it is determined that the convergence has occurred. By periodically running the optimizer, the global target weights fluctuate slowly with price indices, environmental intensity, and demand sentiment, achieving adaptive and online optimization under different scenarios.

4. The method according to claim 3, characterized in that, Step S4 specifically includes: Step S41: Distributed problem modeling and decomposition: Based on the decision architecture, state and action space and objective function constructed in steps S1-S3, the whole process optimization problem is decomposed into sub-problems in three stages: mining, beneficiation and smelting, and connected by a set of coupling constraint equations; The problem of collaborative optimization of the entire mining, beneficiation, and metallurgical process is constructed in the following distributed form: Let the decision variables for the mining, beneficiation, and smelting processes be respectively... The global optimization problem is then: in, This represents the decision variables at each stage, and each stage's decision variable corresponds to the action space defined by the agent in step S22. In other words... Corresponding representation , Corresponding representation , Corresponding representation ; The global target weight vector is dynamically adjusted in step S34; Corresponds to the material flow and energy flow coupling constraints defined in step S13; Corresponding to the process parameter constraints defined in step S14; Based on the principles of the ADMM algorithm, the global optimization problem is decomposed into subproblems that can be solved in a distributed manner, and a global collaborative variable is introduced. and Lagrange multipliers Among them, global collaborative variables It is a virtual, ideal, fully coupled decision variable, whose core characteristic is that it must strictly satisfy all cross-stage coupling constraints. By adding consistency constraints The coupling constraints of the original problem are transformed into a separable form, and an augmented Lagrangian function is constructed: in, This represents the objective function defined in step S32. This can be transformed into a minimization problem; Indicates the penalty parameter; Step S42: ADMM iterative solution; specifically including: Step S421: Initialization of ADMM based on the key parameter collaborative optimization model; Before executing ADMM distributed collaborative optimization, the key parameter collaborative optimization model from step 33 is first invoked to calculate the optimal values ​​of the key parameters under the current external environmental conditions, i.e., the optimal values ​​of the ore grade. Optimal grinding fineness Optimal smelting temperature Build an initial global collaboration scheme : in, , , , , , All represent the historical mean of the corresponding parameter; Decision variables Perform initialization: The Lagrange multipliers are initialized to zero vectors, i.e. ; Step S422: Perform ADMM iterative solution; Based on the three-level agent decision-making architecture established in step S11, the alternating direction multiplier method is used to implement the following distributed iteration: each agent solves its own optimization problem locally, while a "penalty term" for coupling constraints is introduced, and a global collaborative variable updated by the global coordinating agent is received; the specific steps are as follows: (a) Local optimization stage The collaborative agents at each stage, given a global collaborative variable and Lagrange multipliers Next, solve the optimization problem for this stage independently: Geological and mining collaborative intelligent agents based on the optimization problem: The solution yields the optimized boundary grade, ore output, and ore blending ratio; where the superscript... Represents the transpose of a matrix; The mineral processing and separation optimization agent is based on the optimization problem: The optimal grinding fineness, flotation reagent dosage, and pulp pH value are obtained by solving the problem. The smelting and extraction control agent optimizes the following problem: The optimal smelting furnace temperature, oxygen concentration, and melt residence time are obtained by solving the problem. Each agent pushes the optimized intermediate results to the global coordinating agent through the feedforward interaction protocol established in step S12, with a communication delay of ≤10s; (b) Global Coordination Phase The global coordinating agent collects data from each stage. Solve the global collaborative variable update problem to ensure that the coupling constraints of matter flow and energy flow are satisfied: This problem is equivalent to projecting the solutions of each stage onto the feasible set of coupled constraints. The specific solution involves coordinating and adjusting the material flow balance. This ensures that the material flow balance constraint equations are satisfied; energy flow coordinated scheduling is implemented, and energy consumption allocation at each stage is optimized in conjunction with peak-valley electricity pricing to ensure... ; (c) Multiplier update and convergence determination The global coordinating agent updates the Lagrange multipliers: The updated version will be broadcast via the global broadcast protocol in step S12. and Broadcast to intelligent agents at each stage; When local solution With global collaborative variables The algorithm converges when the original residuals are sufficiently close, i.e., when the original residuals are sufficiently close. and dual residuals If all values ​​are less than a preset threshold, convergence is determined. The original residuals represent the degree of constraint violation. Sufficiently small indicates that the independent decisions of each stage have reached consensus with the global coordinating variables that satisfy all physical coupling constraints; dual residuals Measuring global covariates Its own stability; Through multiple iterations of the three stages of "local optimization - global coordination - multiplier update and convergence determination", the decisions of all agents will converge synchronously to a globally feasible solution that satisfies all cross-stage physical constraints, thus achieving true parallel collaboration rather than serial compromise. Step S43: Output the collaborative benchmark solution; After ADMM converges, it outputs a cooperative benchmark solution that satisfies all physical coupling constraints. .

5. The method according to claim 4, characterized in that, Step S5 specifically includes: Step S51: Initialize the policy population; The ADMM collaborative benchmark solution output in step S4 Based on this, an initial strategy population is generated through controlled perturbation; among them, mining strategy individuals are... Nearby, following a normal distribution Sampling, where variance Based on the parameter process range settings within the geological and mining collaborative intelligent system, the following is generated: There are three mining strategies; individual mineral processing strategies generate a mineral processing strategy population, and individual smelting strategies generate a smelting strategy population; each complete strategy is a combination of three strategies: mining strategy individuals + mineral processing strategy individuals + smelting strategy individuals, with an initial population size of [missing information]. ; Step S52: Based on the dynamic global target weights from step S34, perform multiple rounds of collaborative evolution iterations; in each iteration, execute the following loop: (a) Multi-objective fitness assessment and game utility calculation; First, for each strategy combination in the population A full-process simulation is performed, that is, based on the coupled constraint equations in step S13, the multi-objective fitness calculation under this strategy is performed, as shown below: The fitness target value in the above formula is, , , , Calculate according to the formula in step S31; Then, the game utility is calculated as follows: Each agent in each stage evaluates the effectiveness of the strategy from its own perspective: The effectiveness of collaborative intelligent agents in geology and mining: ; Optimization of mineral processing and separation agent utility: ; The effectiveness of intelligent agents in controlling smelting and extraction: ; The above utility calculation uses the function defined in step S32; (b) Based on the NSGA-II framework, perform non-dominated sorting and crowding calculation; Based on four fitness objectives, the population is divided into multiple non-dominated frontiers and a rapid non-dominated ordination is performed. Then, crowding is calculated to determine the crowding distance between individuals in the same non-dominated layer, thus maintaining population diversity. (c) Strategy selection, crossover, and variation based on game utility; Tournament selection: Randomly select two individuals from the population, compare their non-dominated ranking and crowding, and select the better one as the parent; when comparing, the game utility of each stage needs to be considered as a correction factor for selection preference; that is, if two individuals are difficult to distinguish in terms of non-dominated ranking and crowding, the individual with higher utility in the corresponding stage in the current scenario is preferred. Cooperative crossover operation: Simulate binary crossover between mining strategies, swapping... , Genes; mineral processing strategies and smelting strategies are also based on probability. Perform crossover; Adaptive mutation operation: Performs polynomial mutation on the policy, with mutation probability... , The number of variables; the mutation intensity is adaptively adjusted based on the current generation and the convergence of the objective; the mutation intensity is increased when the population diversity decreases. (d) Environmental selection and the formation of new generation populations; Merge the parent and offspring populations, with a size of 2 Perform non-dominated sorting and select the first... Individuals form a new generation of population; Step S53: Output the Pareto optimal policy set and decision support; When evolution reaches its maximum number of generations Or Pareto frontier improvements less than When the time comes, terminate the evolution; output the non-dominated solution set of the final generation. Each solution corresponds to a four-dimensional fitness target vector. These represent its performance in four objectives: economy, resources, energy consumption, and environment. Provides an interactive interface for decision-makers: visualizing the Pareto frontier in the form of parallel coordinate graphs and scatter plot matrices, allowing decision-makers to select the final decision scheme based on real-time preferences. .

6. The method according to claim 5, characterized in that, Step S6 specifically includes: Step S61: Issuance, execution, and real-time monitoring of decision-making plans; Selected strategy The instructions are converted into specific production instructions for each stage; the instructions are sent to the execution agents of each unit through the communication protocol of step S12; real-time monitoring is performed to collect actual production data; through the feedback interaction protocol of step S12, the agents of each stage report the deviations of key indicators to the global coordinating agent. Step S62: Multi-source feedback data fusion and dynamic adjustment, including short-term online correction, medium-term model update, and long-term strategy evolution; (a) Short-term online correction Short-term online corrections are performed on a minute-to-hour basis; when a significant deviation is detected, ADMM rapid re-optimization is triggered: starting from the current actual state, with fixed weights. Unchanged; execute simplified ADMM iteration to quickly adjust process parameters; issue new instructions to the execution unit to achieve rolling optimization; (b) Mid-term learning and model update Mid-term learning and model updates are performed on a daily-weekly basis; based on historical data accumulated weekly, this includes updating global objective weights to adapt model parameters and collecting external environment vectors. With the actual optimal global objective weight The corresponding data is used to update the parameter matrix in step S34 to minimize the error between the model's predicted weights and the actual optimal weights. in, Indicates the first The optimal global objective weight vector, verified by practice; Indicates the first The external environment vector of the sky; This represents the regularization coefficient, used to prevent overfitting. A parameter matrix representing the influence coefficients of environmental variables on the global objective weights; The process model is corrected using actual data, i.e., the key parameters of all process coupling relationships and constraints in steps S13 and S14 are updated; the coefficients in the constraint equations are updated. (c) Long-term strategy evolution Long-term strategies evolve to monthly-quarterly levels; strategies with excellent actual performance are added to the evolutionary population's knowledge base; and different external environments are identified based on historical data. Under the efficient strategy pattern, establish a scenario-policy mapping library; The penalty parameters are adaptively adjusted monthly based on the ADMM convergence history. If the average number of iterations increases significantly and the convergence residual increases, it indicates that the current... If the value is not suitable for the current production conditions, it will be adjusted according to the following rules: The crossover probability is adjusted quarterly based on the efficiency of the evolutionary search. Probability of mutation : When population diversity declines too rapidly: increase and improve To enhance exploration capabilities and break free from local optima; When the convergence rate is too slow: reduce... and ; Step S63: Design the full-cycle closed-loop operation logic. The system will run automatically according to the following time scales: Second-minute level: The operating unit executes the intelligent agent and collects data; Hourly level: ADMM online correction and rapid re-optimization; Daily level: Perform a complete round of evolutionary game and output the production plan for the next day; Weekly update: Global target weights and process model parameters; Monthly / Quarterly: Strategy knowledge base updates and algorithm parameter tuning.

7. A hierarchical multi-agent intelligent decision-making system for the entire process of mineral development, characterized in that, include: At least one processor; and at least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor can execute the method as described in any one of claims 1 to 6 by calling the program instructions.

Images