A method for optimizing mine cemented filling parameters based on predictive modeling and improved NSGA-III algorithm

Abstract

The present invention relates to the technical field of mine filling parameter optimization, and in particular to a mine cementation filling parameter optimization method based on predictive modeling and an improved NSGA-III algorithm. The method comprises: collecting working condition parameters and historical working condition parameters in real-time mine filling operations, performing preprocessing, and obtaining preprocessed working condition parameters and preprocessed historical working condition parameters; using a nested distribution recursive transition graph algorithm to construct a working condition parameter prediction model, performing predictive processing on the preprocessed working condition parameters, and obtaining a working condition parameter prediction vector; constructing a multi-objective optimization model, optimizing the multi-objective optimization model, and obtaining an optimized multi-objective optimization model; optimizing and solving the optimized multi-objective optimization model, and obtaining optimized mine cementation filling parameters. The method solves the technical problems faced by existing technologies in multi-objective optimization, such as uneven solution set distribution, physical inconsistency, and insufficient response capability to future fluctuations.

Description

Technical Field

[0001] The present invention relates to the technical field of mine filling parameter optimization, and in particular to a mine cementation filling parameter optimization method based on predictive modeling and an improved NSGA-III algorithm. Background Art

[0002] Mine filling operations are a critical step in underground mining, widely used for backfilling goafs, improving stability, and comprehensively utilizing mineral resources. The filling process requires real-time monitoring and control of multiple operating parameters, such as slurry concentration, water-cement ratio, cementitious material proportion, pumping pressure, flow rate, and external environmental conditions. These operating parameters not only impact the quality, efficiency, and safety of the filling operation but are also directly linked to multiple objectives, including material cost, construction cycle, pumping energy consumption, and carbon emissions. Therefore, the optimal design of filling parameters is crucial, aiming to achieve a balance between cost, strength, energy efficiency, and environmental protection through a rational combination of parameters.

[0003] Traditional filling parameter optimization methods mostly rely on empirical models or static proportioning methods, and are usually adjusted based on experimental data or manually set rules. Although these methods can provide references in some cases, due to the complexity and dynamic changes of mine filling operation conditions, these static methods often cannot adapt to on-site fluctuations in real time, and lack accurate modeling of complex, multi-objective coupling relationships, making it difficult to dynamically obtain the optimal solution. With the rapid development of data acquisition technology and sensors, the multi-source data generated in the mine filling process has become more abundant, but traditional filling parameter optimization methods have failed to effectively utilize these high-frequency, large-scale real-time data, nor have they taken into account the nonlinearity, time lag effects and complex interdependence between variables in the filling process, resulting in optimization results that often cannot meet the needs of actual construction.

[0004] At the same time, the above-mentioned existing technologies still have the problem that traditional operating parameter optimization methods cannot efficiently adapt to the complex, nonlinear, and highly coupled process characteristics of the site, especially in multi-objective optimization, facing technical problems such as uneven distribution of solution sets, premature convergence, physical inconsistency, and insufficient response to future fluctuations. Summary of the Invention

[0005] In order to solve the technical problems described in the background technology section, the present invention provides a method for optimizing mine cementation filling parameters based on predictive modeling and an improved NSGA-III algorithm.

[0006] The present invention provides a method for optimizing mine cementation filling parameters based on predictive modeling and an improved NSGA-III algorithm, which specifically includes the following technical solutions:

[0007] A method for optimizing mine cementation filling parameters based on predictive modeling and an improved NSGA-III algorithm includes the following steps:

[0008] S1. Collecting and preprocessing operating parameters and historical operating parameters from real-time mine filling operations to obtain preprocessed operating parameters and preprocessed historical operating parameters; constructing an operating parameter prediction model based on the preprocessed historical operating parameters using a nested distribution recursive transition graph algorithm; and using the operating parameter prediction model to predict the preprocessed operating parameters to obtain an operating parameter prediction vector;

[0009] S2. Define the operating condition parameter prediction vector as a decision variable, construct a multi-objective optimization model, and use the dynamic boundary adjustment mechanism and soft constraint guidance mechanism to optimize the multi-objective optimization model to obtain the optimized multi-objective optimization model; by introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism, the optimized multi-objective optimization model is optimized and solved to obtain the optimized filling parameters, that is, the optimized mine cementation filling parameters.

[0010] Preferably, the S1 specifically includes:

[0011] In the implementation process of the nested distributed recursive transition graph algorithm, the preprocessed historical operating parameters are decomposed into nested state tuples. After constructing the state tuple of each preprocessed historical operating parameter, the state tuple is used as a node to generate a time-dependent parameter transfer graph.

[0012] Preferably, the S1 specifically includes:

[0013] In the implementation process of the nested distributed recursive transition graph algorithm, after the time-dependent parameter transfer graph is constructed, for each target prediction variable, that is, the preprocessed historical operating condition parameter, different influence paths are generated by full-graph search. Each path corresponds to a candidate prediction trajectory, and a set of candidate prediction trajectories is formed.

[0014] Preferably, the S1 specifically includes:

[0015] In the implementation process of the nested distributed recursive transition graph algorithm, the candidate prediction trajectory set is conflict-eliminated, and the retained candidate prediction trajectories are obtained after screening. The retained candidate prediction trajectories are weightedly fused to obtain the trend prediction value of the target prediction variable.

[0016] Preferably, the S1 specifically includes:

[0017] In the implementation process of the nested distribution recursive transition graph algorithm, when the error between the target prediction variable and the trend prediction value exceeds the threshold, the structural migration mechanism is triggered, the time-dependent parameter transfer graph is updated, and the dynamic path structure container is updated at the same time to form a working condition parameter prediction model. The working condition parameter prediction model is used to process the preprocessed working condition parameters to obtain the working condition parameter prediction vector.

[0018] Preferably, the S2 specifically includes:

[0019] The dynamic boundary adjustment mechanism uses the predicted value of each preprocessed operating parameter in the operating parameter prediction vector to fluctuate a certain proportion as the operating parameter boundary, and performs a union operation with the initial inequality constraint condition in the multi-objective optimization model to obtain the dynamic feasible domain of the operating parameter. The floating proportion is described by the prediction boundary floating coefficient; the soft constraint guidance mechanism is achieved by introducing a soft constraint penalty term and adding the soft constraint penalty term to the objective function of the multi-objective optimization model.

[0020] Preferably, the S2 specifically includes:

[0021] An improved NSGA-III algorithm is introduced, and the optimized multi-objective optimization model is optimized and solved by introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism. In the process of implementing the reference point perturbation guidance mechanism, an initial reference point set is constructed, and a Gaussian field perturbation with direction guidance is introduced to each reference point. The reference point is perturbed to obtain the perturbed reference point.

[0022] Preferably, the S2 specifically includes:

[0023] In the process of implementing the target distribution similarity guidance mechanism, the similarity between the current population distribution density and the reference point distribution density is dynamically evaluated to identify the sparsely covered areas in the target space. For each individual in the current generation, that is, the decision variable, the similarity between its corresponding objective function vector and its belonging reference point is calculated.

[0024] Preferably, the S2 specifically includes:

[0025] During the operation of the improved NSGA-III algorithm, an elite retention strategy is adopted. In each generation of iteration, the non-dominated solution set consisting of the current optimal solution is retained and offspring individuals are generated. In each evolutionary cycle, the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism are executed to obtain the non-dominated solution set. Finally, the optimal solution set is selected through the Knee Point strategy to obtain the final Pareto optimal solution set, that is, the optimized mine cementation filling parameters.

[0026] The beneficial effects of the technical solution of the present invention are:

[0027] 1. The nested distributed recursive transition graph algorithm is adopted, combined with state tuple construction, Granger causal dynamic edge construction, multi-trajectory path generation and conflict elimination mechanism, to achieve the modeling of asymmetric time-dependent relationships between multi-objective prediction variables. Compared with traditional RNN / LSTM-type sequence learning models, this method can better capture the complex physical response mechanisms between variables in the filling system, especially when dealing with periodic fluctuations and on-site interference factors, showing higher prediction accuracy and system generalization ability. In addition, through the structure migration mechanism and the self-update capability of the dynamic path structure container, the working condition parameter prediction model has dynamic self-repair capabilities in the face of actual execution feedback deviations, significantly improving the robustness and real-time adaptability of the working condition parameter prediction model.

[0028] 2. By introducing a dynamic boundary adjustment mechanism and utilizing the operating condition parameter prediction vector to adjust the feasible domain of decision variables in the multi-objective optimization model in real time, the evolution from traditional static optimization to a trend-aware optimization model is achieved. This mechanism effectively avoids the optimization breakpoint problem caused by sudden fluctuations and enhances tolerance to on-site disturbances and process drift. At the same time, by introducing a prediction deviation cost term as an auxiliary objective function, the optimization process is guided towards future trends while maintaining the optimization of the main engineering objectives.

[0029] 3. The reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism are adopted to solve the problem of uneven distribution of solution sets in high-dimensional target space. The Gaussian perturbation field is used to perform direction-guided perturbation on the reference points, which significantly improves the global exploratory power of the initialized population, avoids falling into local optimality, and enhances the comprehensiveness and structural diversity of the solution set. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of a method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm described in the present invention. DETAILED DESCRIPTION

[0031] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0032] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0033] The following describes in detail a specific solution of a mine cementation filling parameter optimization method based on predictive modeling and improved NSGA-III algorithm provided by the present invention with reference to the accompanying drawings.

[0034] Refer to the attached Figure 1 , which shows a flow chart of a method for optimizing mine cementation filling parameters based on predictive modeling and an improved NSGA-III algorithm provided by one embodiment of the present invention, the method comprising the following steps:

[0035] S1. Collecting and preprocessing operating parameters and historical operating parameters from real-time mine filling operations to obtain preprocessed operating parameters and preprocessed historical operating parameters; constructing an operating parameter prediction model based on the preprocessed historical operating parameters using a nested distribution recursive transition graph algorithm; and using the operating parameter prediction model to predict the preprocessed operating parameters to obtain an operating parameter prediction vector;

[0036] The operating parameters of mine filling operations are collected in real time through multi-source sensor equipment such as pressure sensors, flow meters, and radar level gauges. These parameters include the real-time concentration of the slurry, which is obtained by a concentration meter or an online specific gravity detector; the water-cement ratio, which is calculated by the automatic water distribution system and the proportioning signal of the cement silo; the cementitious material content, which is obtained through the flow meter and proportioning records; the pumping pressure, which is collected by the pipeline pressure sensor; the flow rate, which is obtained by the electromagnetic or ultrasonic flow meter; the external environmental parameters, such as temperature and humidity, which are obtained by environmental sensors; the material storage balance and historical proportion records, which are read through the SCADA (Supervisory Control and Data Acquisition) system interface, etc.

[0037] The operating parameters in the mine filling operation are preprocessed by cleaning, anomaly elimination, sliding filtering, standardization and normalization to obtain preprocessed operating parameters. The preprocessing process is a technical means well known to those skilled in the art and will not be described in detail here.

[0038] Historical operating condition parameters are retrieved from an existing database and preprocessed to obtain the preprocessed historical operating condition parameters. Based on the preprocessed historical operating condition parameters, a nested distributed recursive transition graph algorithm is used to construct an operating condition parameter prediction model. The nested distributed recursive transition graph algorithm solves the problem that traditional sequence learning models cannot adapt to highly coupled nonlinear field data through a new structure of nested state tuples + parameter graph dynamic modeling + multi-trajectory generation and screening + distribution weight aggregation + graph structure migration mechanism. The specific process is as follows:

[0039] The pre-processed historical operating parameters are decomposed into nested state tuples to carry richer multi-dimensional historical information. That is, for any i-th pre-processed historical operating parameter, its nested state structure is defined as a state tuple ,in, Represents the value of the i-th preprocessed historical operating condition parameter at the current time t, that is, the i-th target prediction variable; represents the frequency domain transformation characteristics of the i-th pre-processed historical operating condition parameters, which is obtained by discrete Fourier transform (DFT); It is a dynamic path structure container, which is used to save other target prediction variable paths that have a recursive relationship with the target prediction variable (i.e., the historical operating condition parameter after the i-th preprocessing) in history, and is expressed as ; The dynamic path structure container is an empty set at the initial moment, where represents the value of the j-th preprocessed historical operating condition parameter, that is, the j-th target prediction variable; It is the dynamic influence intensity of the historical operating parameters after the i-th preprocessing at the current time t on the historical operating parameters after the j-th preprocessing, which is calculated by the residual ratio of the operating parameter prediction model constructed by the multivariate Granger causality test. The reference value is ; The maximum lag time step represents the dynamic influence of the i-th preprocessed historical operating condition parameter on the j-th preprocessed historical operating condition parameter at the current time t, and the lag time at the minimum error point is obtained by traversing the lag window (for example, from 1 to 5 minutes).

[0040] After constructing the state tuple of each pre-processed historical operating condition parameter, the state tuple is used as a node to generate a time-dependent parameter transfer diagram ,in, , m represents the total number of historical operating parameters after preprocessing; each directed edge It means that the historical operating condition parameters after the i-th preprocessing have a significant dependence on the historical operating condition parameters after the j-th preprocessing in the past time window. The weight of the edge is expressed in a two-dimensional structure: .

[0041] Specifically, if the dynamic influence intensity of the historical operating condition parameter after the i-th preprocessing on the historical operating condition parameter after the j-th preprocessing at the current time t is Higher than the threshold set by expert experience , and the maximum lag time step of the dynamic influence intensity of the i-th pre-processed historical operating condition parameter on the j-th pre-processed historical operating condition parameter at the current time t is less than the upper limit preset according to the expert experience method , then the target predictor variable path is recorded into the dynamic path structure container: , to realize the update of the path in the dynamic path structure container.

[0042] The time-dependent parameter transfer diagram formed through the above process It is an asymmetric directed graph with weight and lag relationship, which provides structural support for the generation of subsequent variable trend paths.

[0043] In the time-dependent parameter transfer diagram After the construction is completed, for each target prediction variable, such as the real-time concentration of slurry, instead of directly using the traditional single-path sequence input for prediction, a full-graph search is used to generate multiple impact paths, each of which corresponds to a candidate prediction trajectory. The candidate prediction trajectory generation process is based on the following logic:

[0044] For the target predictor variable , from all the valid edges The input target predictor variable , collect its hysteresis value , which means the current moment The value of the i-th preprocessed historical operating condition parameter is obtained by a non-parametric function such as a third-order B-spline, a second-order orthogonal polynomial, or a residual compensation function based on the Hilbert spectrum. Mapping generates transfer components , represents the k-th path to the target predictor variable The predicted value at time t+Δ, where Δ represents the prediction time span, which is set based on specific application requirements according to expert experience.

[0045] Due to physical inconsistencies or engineering conflicts in the path, such as the increase in real-time slurry concentration and water-cement ratio, which do not conform to the material properties, the candidate prediction trajectory set is selected using the existing physical constraint map selected according to the expert experience method. ( Represents the total number of paths) to eliminate conflicts and obtain the retained candidate prediction trajectories, and retain the The candidate prediction trajectories are weighted fused to obtain the target prediction variable Trend forecast value ,

[0046] Finally, to ensure that the working condition parameter prediction model has adaptive capabilities, when the error between the target prediction variable and the trend prediction value exceeds the threshold preset according to the expert experience method, such as the slurry real-time concentration prediction error > 1.5%, the structural migration mechanism needs to be triggered to update the time-dependent parameter transfer diagram. The updating process includes: first, the target prediction variable with serious error is updated. , trace back its main dependent edge, if the error exceeds 2 times of the mean for 3 consecutive times, then transfer the edge from the time-dependent parameter graph For the newly added target prediction variable pairs with coupling relationships, weak connection edges are added under the low confidence threshold preset according to the expert experience method through the existing probabilistic graph modeling method; at the same time, the dynamic path structure container is updated to form a self-learning and self-evolving working condition parameter prediction model.

[0047] Furthermore, the working condition parameter prediction model obtained by the above process is used to process the preprocessed working condition parameters to obtain the working condition parameter prediction vector .

[0048] S2. Define the operating condition parameter prediction vector as a decision variable, construct a multi-objective optimization model, and use the dynamic boundary adjustment mechanism and soft constraint guidance mechanism to optimize the multi-objective optimization model to obtain the optimized multi-objective optimization model; by introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism, the optimized multi-objective optimization model is optimized and solved to obtain the optimized filling parameters, that is, the optimized mine cementation filling parameters.

[0049] Define the operating condition parameter prediction vector as the decision variable , where any element can be Indicates that The decision variable, The working condition parameter variables are also the filling parameters to be optimized and serve as the input of the multi-objective optimization model. Based on the decision variables, the multi-objective optimization model is constructed as follows:

[0050]

[0051] in, is the objective function vector, containing independent or related optimization objectives; It is objective functions, such as cost, compressive strength, energy consumption, carbon emissions, etc.; It is Inequality constraints, such as the pump pressure must not exceed 2.5 MPa, p is the number of inequality constraints; It is Equality constraints, such as the sum of the dosage ratios must be 1, and q is the number of equality constraints; is the feasible solution space, which is defined by the variable bounds and constraints.

[0052] Furthermore, in order to improve the responsiveness of filling parameters to site fluctuations and their adaptability to engineering projects, the multi-objective optimization model was optimized using a dynamic boundary adjustment mechanism and a soft constraint guidance mechanism based on the working condition parameter prediction vector. Specifically:

[0053] For the dynamic boundary adjustment mechanism, the predicted value of each preprocessed operating parameter in the operating parameter prediction vector is fluctuated up and down by a certain proportion as the operating parameter boundary, and a union operation is performed with the initial inequality constraint condition to obtain the dynamic feasible domain of the operating parameter. The said floating ratio is described by the prediction boundary floating coefficient, and the prediction boundary floating coefficient is determined according to the expert experience method, which can effectively compare the breakpoints caused by sudden disturbances to the multi-objective optimization model.

[0054] In view of the soft constraint guidance mechanism, a soft constraint penalty term is introduced and added to the objective function of the multi-objective optimization model, namely:

[0055]

[0056] in, Represents the prediction deviation cost term, as the first objective function; is the variable weight coefficient, which is determined according to expert experience, and the reference value is ; It represents the soft constraint penalty term, that is, the difference between the decision variable and the predicted vector of the operating parameters, which serves as the input of the objective function in the multi-objective optimization model.

[0057] After the above process, the optimized multi-objective optimization model is obtained. Furthermore, in order to solve the problem that the initial population of the traditional NSGA-III algorithm is concentrated in the direction of some reference points in the initialization reference point and hierarchical selection mechanism, which leads to the inability to achieve uniform Pareto front exploration in the high-dimensional target space, the improved NSGA-III algorithm is introduced. By introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism, the optimized multi-objective optimization model is optimized and solved, and the optimized filling parameters, namely the optimized mine cementation filling parameters, are obtained. Specifically:

[0058] The reference point perturbation guidance mechanism is implemented by first constructing the initial reference point set using the Das and Dennis construction method proposed by Deb et al. , where each reference point is the coordinate vector in the normalized target space, representing the ideal uniform distribution direction in the target space.

[0059] Then, a Gaussian field perturbation with direction guidance is introduced to each reference point to improve the spatial distribution breadth of the reference point in the early stage. Based on the multi-scale local density estimation method, the expression of the reference point after perturbation is obtained as follows:

[0060]

[0061] in, After the disturbance reference points; It is The original coordinate vectors of the reference points represent the ideal target position direction in the target space and are evenly distributed on the hyperplane or hypercube based on the Das and Dennis construction method; δ is the global disturbance control coefficient, which is used to control the overall disturbance amplitude and is a scaling parameter determined by expert experience. The reference value range is ; is the total number of target dimensions to be perturbed; It is The disturbance weight on the target dimension is used to control the direction in which the disturbance expands. It is determined according to the expert experience method, and the reference value range is , and the weighted sum is 1; It is Reference points in The density-oriented factor on the target dimension points to the sparse area and is calculated based on the inverse of the non-dominated solution density of the current population in the direction of the reference point dimension. The mean is 0 and the variance is The standard normally distributed noise.

[0062] The above perturbation process ensures that each reference point expands outward from its original position, thereby generating a slight random movement, so that the initialization population covers a wider range in the target space. Furthermore, in order to prevent the reference point from crossing the boundary after perturbation, Any amount exceeding The reference point component of the interval is clipped, that is, when , then set it to 0, if , then set it to 1.

[0063] The target distribution similarity guidance mechanism is implemented by dynamically evaluating the similarity between the current population distribution density and the reference point distribution density, thereby identifying the sparsely covered areas in the target space and enhancing the survival probability of individuals in the area. It is implemented as follows: before each selection operation, for each individual in the current generation , that is, the decision variable, calculates the similarity between its corresponding objective function vector and its reference point , the specific formula is:

[0064]

[0065] in, Indicates the The objective function vector corresponding to the individual is The similarity between the attribute reference points; is an individual The number of objective functions; is the weight coefficient of the lth objective function, which is used to highlight the priority target and is determined according to the expert experience method. The reference value range is , and the weighted sum is 1; Is the first in the population The normalized function value of each individual at the lth objective function is calculated using the Min-Max normalization method; Indicates the The reference point after disturbance; γ is the density penalty attenuation factor, which is used to control the intensity of the penalty in dense areas. It is determined according to the expert experience method, and the reference value range is ; It is The crowding density index of the reference point after the disturbance is determined by calculating the number of individuals after demarcating the area using the expert experience method; is the weighted point integral sub-item, which represents the difference between the current individual and the The weighted dot product of the reference points in each target direction is the main item for measuring directional consistency; and is the modulus normalization term, which is the weighted Euclidean modulus of the individual and the reference point, and is used to normalize the weighted point integral sub-term to form the calculation structure of the weighted cosine angle; is the density penalty exponential term, which means that when the reference point is surrounded by dense individuals ( When is large, the overall similarity decays exponentially, which is used to reduce the survival probability of individuals in the area and avoid excessive concentration in certain areas of the target space.

[0066] Furthermore, if the average value of all similarities corresponding to a reference point is lower than a preset threshold value based on expert experience, such as 0.65, its crowding density index is automatically adjusted to increase the survival priority of individuals in the above area in the next generation of non-dominated sorting, thereby promoting the expansion of the population to the said area.

[0067] The improved NSGA-III algorithm also employs an elite retention strategy during operation, retaining the non-dominated solution set consisting of the current optimal solution in each iteration. Simulated binary crossover (SBX) and polynomial mutation operations are then used to generate offspring individuals. Each evolutionary cycle executes the two aforementioned improvement mechanisms—the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism—until the number of iterations is met or the variance of all objective function values ​​converges to a threshold preset based on expert experience. This results in a non-dominated solution set, or an approximate approximation of the Pareto solution set. Finally, the optimal solution set is selected using the Knee Point strategy, resulting in the final Pareto optimal solution set, representing the optimized mine cementation filling parameters.

[0068] In summary, a method for optimizing mine cemented filling parameters based on predictive modeling and improved NSGA-III algorithm was completed.

[0069] The order in which the embodiments of the invention are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0070] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm, characterized in that: The following steps are involved: S1. Collecting real-time operating parameters and historical operating parameters in mine filling operations, and preprocessing them to obtain preprocessed operating parameters and preprocessed historical operating parameters; Based on the pre-processed historical operating parameters, a nested distribution recursive transition graph algorithm is used to build an operating parameter prediction model, and the pre-processed operating parameters are predicted using the operating parameter prediction model to obtain an operating parameter prediction vector; In the implementation of the nested distributed recursive transition graph algorithm, the preprocessed historical operating condition parameters are decomposed into nested state tuples. After constructing the state tuple of each preprocessed historical operating condition parameter, the state tuple is used as a node to generate a time-dependent parameter transition graph. After the time-dependent parameter transition graph is constructed, for each target prediction variable, that is, the preprocessed historical operating condition parameter, a different influencing path is generated by full-graph search. Each path corresponds to a candidate prediction trajectory, and a set of candidate prediction trajectories is formed. The candidate prediction trajectory set is conflict-eliminated, and the retained candidate prediction trajectories are obtained after screening. The retained candidate prediction trajectories are weighted and fused to obtain the trend prediction value of the target prediction variable. When the error between the target prediction variable and the trend prediction value exceeds the threshold, the structure migration mechanism is triggered to update the time-dependent parameter transfer diagram and the dynamic path structure container to form a working condition parameter prediction model; S2. Define the operating condition parameter prediction vector as the decision variable, construct a multi-objective optimization model, and use the dynamic boundary adjustment mechanism and soft constraint guidance mechanism to optimize the multi-objective optimization model to obtain the optimized multi-objective optimization model; introduce the improved NSGA-III algorithm, and optimize and solve the optimized multi-objective optimization model by introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism to obtain the optimized filling parameters, that is, the optimized mine cementation filling parameters.

2. The method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm according to claim 1, characterized in that: Said S2 specifically includes: The dynamic boundary adjustment mechanism uses the predicted value of each preprocessed operating parameter in the operating parameter prediction vector to fluctuate a certain proportion as the operating parameter boundary, and performs a union operation with the initial inequality constraint condition in the multi-objective optimization model to obtain the dynamic feasible domain of the operating parameter. The floating proportion is described by the prediction boundary floating coefficient; the soft constraint guidance mechanism is achieved by introducing a soft constraint penalty term and adding the soft constraint penalty term to the objective function of the multi-objective optimization model.

3. The method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm according to claim 1, characterized in that: Said S2 specifically includes: An improved NSGA-III algorithm is introduced, and the optimized multi-objective optimization model is optimized and solved by introducing the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism. In the process of implementing the reference point perturbation guidance mechanism, an initial reference point set is constructed, and a Gaussian field perturbation with direction guidance is introduced to each reference point. The reference point is perturbed to obtain the perturbed reference point.

4. The method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm according to claim 3, characterized in that: Said S2 specifically includes: In the process of implementing the target distribution similarity guidance mechanism, the similarity between the current population distribution density and the reference point distribution density is dynamically evaluated to identify the sparsely covered areas in the target space. For each individual in the current generation, that is, the decision variable, the similarity between its corresponding objective function vector and its belonging reference point is calculated.

5. The method for optimizing mine cementation filling parameters based on predictive modeling and improved NSGA-III algorithm according to claim 3, characterized in that: Said S2 specifically includes: During the operation of the improved NSGA-III algorithm, an elite retention strategy is adopted. In each generation of iteration, the non-dominated solution set consisting of the current optimal solution is retained and offspring individuals are generated. In each evolutionary cycle, the reference point perturbation guidance mechanism and the target distribution similarity guidance mechanism are executed to obtain the non-dominated solution set. Finally, the optimal solution set is selected through the Knee Point strategy to obtain the final Pareto optimal solution set, that is, the optimized mine cementation filling parameters.

Images