Solid waste composite cementitious material for mixing pile and proportion optimization method of solid waste composite cementitious material

Abstract

The invention provides a solid waste composite cementing material for a mixing pile and a proportioning optimization method of the solid waste composite cementing material, and belongs to the technical field of concrete materials. The aluminum mud, the red mud, the building micro powder and the phosphorite tailing powder are used as raw materials and compounded to form the cementing material for the mixing pile, cement can be effectively replaced, the problem that a cement-based material is poor in performance in a peat silty soil scene is solved, and cyclic utilization of solid waste materials can be achieved; and in addition, through a meta learning-multi-objective reinforcement learning fusion algorithm, Pareto optimal proportioning schemes of all components in the cementing material under different environment conditions and different maintenance conditions are rapidly generated, the bottleneck of traditional trial and error research and development is broken through, and the research and development progress is accelerated.

Description

Technical Field

[0001] The present invention belongs to the technical field of concrete materials, and particularly relates to a solid waste composite cementitious material for mixing piles and a method for optimizing its proportioning. Background Art

[0002] As a typical soft soil with high organic matter content, peat silt soil is widely distributed in coastal sedimentary areas such as the Yangtze River Delta and Pearl River Delta in China, as well as in the lake sedimentary zone of the Yunnan-Guizhou Plateau. Its material composition has significant regional characteristics. Such soil usually exhibits special engineering properties such as a water content exceeding 45%, a natural void ratio greater than 1.5, and an organic matter content of 20-60%, which directly leads to a compression coefficient 2-3 times that of conventional soft soil (reaching 1.5-4.0 MPa -1 ), and a permeability coefficient as low as 10 -7 cm / s order of magnitude. In actual engineering, problems such as pile body necking and insufficient bearing capacity of the composite foundation are likely to occur when using traditional mixing piles for solidification. The reasons are as follows: Firstly, organic matter components such as humic acid and fulvic acid in the soil react with the cement hydration product Ca(OH)2 through chelation reaction, inhibiting the formation of C-S-H gel, resulting in a loose flocculent microstructure of the solidified body; Secondly, a high water content environment will accelerate the loss of Ca 2+ and cause the phenomenon of strength retrogression.

[0003] On the other hand, the annual output of industrial solid waste increases. Among them, the annual emissions of aluminum electrolysis waste residue (aluminum mud) and the stockpiles of red mud are large, and the utilization rate of bulk solid wastes such as construction micro-powder and phosphate tailings powder is less than 30%. For these industrial solid wastes, the current disposal mode generally adopts landfill treatment, which not only causes waste of land resources (0.5-1.2 mu of land is required for every 10,000 tons of solid waste), but also poses a risk of heavy metal pollution.

[0004] Moreover, although existing research has confirmed that components such as Fe2O3 and Al2O3 in red mud have potential cementitious activity, and the CaO content in phosphate tailings powder exceeds 40%, technical problems such as unclear multi-solid waste synergistic activation mechanism and unclear heavy metal stabilization path have led to the long-term stagnation of the development of solid waste-based cementitious materials at the laboratory stage.

[0005] Therefore, it is necessary to provide a solid waste composite cementitious material for mixing piles and a method for optimizing its proportioning to solve the above technical problems. Summary of the Invention

[0006] The present invention provides a solid waste composite cementitious material for mixing piles and a method for optimizing its proportion. Using aluminum sludge, red mud, construction micro-powder, and phosphate tailings powder as raw materials, a composite cementitious material for mixing piles is formed, which can effectively replace cement, solve the problem of poor performance of cement-based materials in peat soft soil scenarios, and realize the recycling of solid waste materials. In addition, through the meta-learning multi-objective reinforcement learning fusion algorithm, the Pareto optimal proportioning scheme under different environmental conditions and different curing conditions can be quickly generated, breaking through the bottleneck of traditional trial-and-error research and development, so as to solve at least one technical problem involved in the background technology.

[0007] To solve the above technical problems, the present invention is implemented as follows: A method for optimizing the proportion of a solid waste composite cementitious material for mixing piles, comprising the following steps: Step S1, select the components of the solid waste composite gel material, including aluminum sludge, red mud, construction micro-powder, and phosphate tailings powder, measure the physical and chemical parameters of each component, determine the feasible interval of the proportion of each component based on the interaction mechanism between the components, take equal-spacing values from the feasible interval of the proportion of each component, configure the solid waste composite gel material according to the dosage combination, then add different dosages of peat soft soil to the solid waste composite gel material to form a mixture, add water to the mixture to prepare slurries with different water contents, then prepare the slurries into pile specimens, cure them under different curing conditions, and after the curing is completed, test the performance data of the pile specimens. Define the proportion of each component in the solid waste composite gel material, the dosage of peat soft soil, the water content of the slurry, and the curing conditions as variables, and combine the variables and performance test results in each test process to form a variable-performance test data set; Step S2, fit the relationship between the variables and the performance based on the variable-performance test data set, use the relationship to generate the performance fitting results under different variables, and combine the variables and performance fitting results in each fitting process to form a variable-performance fitting data set, and jointly form a sample data set with the variable-performance test data set and the variable-performance fitting data set; Step S3, construct a performance prediction model, input the sample data set into the performance prediction model for training, after the training is completed, given different variables, use the performance prediction model to predict the performance data of the pile, and combine the variables and performance prediction results in each prediction process to form a variable-performance prediction data set; Step S4, construct a meta-learning policy network, input multiple groups of variable-performance prediction data sets into the meta-learning policy network for training, after the training is completed, set the target threshold of the pile performance data, and generate multiple groups of variable combination candidate schemes with different values through the meta-learning policy network; Step S5, construct a multi-objective reinforcement learning optimization model, and screen out the proportion combination with the optimal comprehensive performance from the candidate schemes based on the state-action-reward feedback mechanism.

[0008] As a preferred improvement, the determination of the physical and chemical parameters of each component specifically includes: (i) Using X-ray fluorescence spectrometry to determine the contents of Al2O3 and SiO2 in the aluminum sludge; analyzing the forms of heavy metal elements in the aluminum sludge by sequential extraction method; determining the leaching concentrations of heavy metal elements in the aluminum sludge by toxicity characteristic leaching procedure; (ii) Using X-ray fluorescence spectrometry to determine the content of Fe2O3 in the red mud; analyzing the forms of heavy metal elements in the red mud by sequential extraction method; determining the leaching concentrations of heavy metal elements in the red mud by toxicity characteristic leaching procedure; (iii) Determining the specific surface area of the building micropowder by nitrogen adsorption-desorption isotherm method; calculating the average pore diameter and total pore volume of the building micropowder by BJH method, and combining SEM-EDS analysis to confirm the unit adsorption capacity of the building micropowder for heavy metal ions; (iv) Using XRD quantitative analysis to determine the content of hydroxyapatite in the phosphate rock tailings powder, and carrying out batch experiments to verify the unit chemical fixation amount of the phosphate rock tailings powder for heavy metal ions.

[0009] As a preferred improvement, the preparation of the slurry into a pile specimen specifically includes the following process: drying each component at 105 °C to constant weight and passing through an 80 μm square-hole sieve; freeze-drying the peat silty soil and then crushing it to the same particle size; using a planetary mixer to implement the "dry mixing - wet mixing" two-stage method: first, perform dry mixing, premixing the dry materials of each component at a speed of 60 rpm for 3 min; then, perform wet mixing, injecting different amounts of water into the mixer to obtain slurries with different water contents, and then wet mixing at a speed of 120 rpm for 5 min; after mixing, taking the material to make piles; and then curing under the set curing conditions.

[0010] As a preferred improvement, the curing conditions include: curing temperature, curing humidity, and curing age; the properties include compressive strength and heavy metal ion leaching rate, the compressive strength is obtained by unconfined compressive strength test, and the heavy metal ion leaching rate is obtained by toxicity characteristic leaching procedure.

[0011] As a preferred improvement, the compressive strength is represented by the fitting formula: ; In the formula, represents the comprehensive correction coefficient; represents the dosage of the peat silty soil; represents the ratio of the th component, ; represents the total number of component types, with a value of 4; represents the actual water content of the slurry; Indicates the optimal moisture content of the slurry; Indicates the curing temperature; Indicates the reference temperature, taken as 20 °C; Indicates the relative curing humidity; Indicates the reference humidity, taken as 95%; Indicates the curing age; Indicates the reference age, taken as 7 days; 、 、 、 、 Indicates the material characteristic parameters, obtained by fitting; Heavy metal ion leaching rate The fitting formula is expressed as: ; In the formula, Indicates the initial leaching concentration; Indicates the leaching time; Indicates the time decay coefficient; 、 Are the curing and stabilization efficiency variables, 、 、 Obtained by fitting.

[0012] As a preferred improvement, the performance prediction model is a residual neural network, including an input layer, a hidden layer and an output layer. The input layer includes 9 nodes, and inputs 9-dimensional variables of the ratio of aluminum sludge, the ratio of red mud, the ratio of building micro-powder, the ratio of phosphate tailings powder, the content of peat silt soil, the moisture content of the slurry, the curing temperature, the curing humidity and the curing age respectively. The hidden layer consists of three layers in total. The first layer is provided with 64 nodes, the second layer is provided with 32 nodes, and the third layer is provided with 16 nodes. Each node is provided with a residual block, and each residual block includes two fully connected layers, with a skip connection between the two fully connected layers. The output layer includes 2 nodes, which output the compressive strength of the pile body and the average value of the heavy metal ion leaching rate respectively. The hidden layer is activated by the Swish function, and the output layer is activated by the Sigmoid function.

[0013] As a preferred improvement, the meta-learning strategy network divides the tasks into two types. One type takes the compressive strength as the optimization goal, and the other type takes the heavy metal ion leaching rate as the optimization goal. Each type generates multiple sub-tasks. In the inner loop stage, the initial values of the ratio of the components generated by the current meta-strategy, the content of peat silt soil, the moisture content of the slurry and the curing conditions are used, and the strategy network is optimized by gradient descent to make the ratio of the components generated, the content of peat silt soil, the moisture content of the slurry and the curing conditions meet the requirements of the optimization goal. In the outer loop stage, the inner loop gradients of all tasks are aggregated to update the meta-strategy parameters.

[0014] As a preferred improvement, before the training of each type of task, it is also necessary to calculate the correlation between variables and performance data, specifically including the following process: Both the Pearson correlation coefficient and the Spearman rank correlation coefficient are used to analyze the global correlation between variables and performance data. Among them, the Pearson correlation coefficient is used to measure the linear correlation between variables and performance data, and the Spearman rank correlation coefficient is used to measure the non-linear rank dependence between variables and performance data. By assigning different weights to the Pearson correlation coefficient and the Spearman rank correlation coefficient, the global correlation between variables and performance data is obtained; Then, the kernel principal component analysis algorithm is introduced for non-linear dimensionality reduction. The variables are mapped from the original feature space to the high-dimensional Hilbert space through the kernel function, and the standard PCA analysis is performed in this space to extract multiple principal components for subsequent learning and modeling, so as to compress the input dimension, retain the main variation features and reduce the problem of multicollinearity; The loss function of the meta-learning policy network is expressed as: ; In the formula, , represent the true value and the predicted value of the task related to the compressive strength, , the true value and the predicted value of the task related to the leaching rate of heavy metal ions, , represent the weight coefficients respectively, which are dynamically adjusted according to the global correlation matrix to ensure that the high-correlation targets occupy higher weights in the learning.

[0015] As a preferred improvement, the state space of the multi-objective reinforcement learning optimization model is expressed as: ; The action space is a set of micro-adjustment operations of variables, which is expressed as: ; Each action execution corresponds to a local perturbation of the mixing ratio scheme in the current state, thus jumping to a new candidate rate state; To balance the dual performance indicators, the piecewise reward function is defined as follows: ; In the formula, , represent the weight coefficients; represents the threshold value of the compressive strength; Indicates the threshold value of heavy metal ion leaching rate; The reinforcement learning training phase is carried out using a deep Q-network algorithm.

[0016] A solid waste composite cementitious material for mixing piles comprises aluminum mud, red mud, construction micro powder and phosphate tailing powder, wherein the proportion of each component is obtained by the proportion optimization method of the solid waste composite cementitious material for mixing piles.

[0017] The beneficial effects of the present invention are: (1) Through the synergistic excitation effect of aluminum mud and red mud, without cement, aluminum silicate and iron oxide are reacted under alkaline conditions to generate a new CASH gelling network and iron-aluminum composite gel, which makes the 28-day compressive strength of the solidified body jump to more than 3MPa, which is about 10 times higher than the curing efficiency of traditional cement-based materials, and successfully breaks through the technical bottleneck of organic matter inhibiting hydration reaction. This system not only overcomes the defect of insufficient strength of traditional processes, but also embeds Cd, Pb and other ions into the silicon-oxygen skeleton and oxide layer through the lattice sealing effect of the cementitious products on the heavy metals in aluminum mud and red mud, thus achieving the dual goals of "self-purification" and structural strengthening of solid waste. (2) Integrate four types of solid waste, namely aluminum mud, red mud, construction powder and phosphate tailings, to build a "gelation-adsorption-stabilization" functional chain: aluminum mud and red mud provide active gelling components, construction powder takes advantage of porous adsorption, and phosphate tailings release phosphate to achieve chemical precipitation; the annual solid waste disposal volume reaches tens of millions of tons, reducing landfill land, forming a "waste-to-soil" circular economy model, and reversing the dilemma of resource waste and pollution risks coexisting in traditional disposal; (3) Building powder physically absorbs free heavy metals with a network structure with a porosity of 75%; phosphate tailings and Cd 2+ The solubility product of the reaction is as low as 2.5×10 -33 Cadmium phosphate precipitation; aluminum mud and red mud gel products fix heavy metals in the aluminosilicate lattice through ion replacement; under triple protection, the leaching concentrations of Cd and Pb are lower than the strict control values ​​of national standards, eliminating the secondary environmental risks of solid waste reuse; (4) A dynamic response system is established by adopting a meta-learning-multi-objective reinforcement learning fusion model to quickly generate the Pareto optimal ratio scheme of each component in the cementitious material under different environmental conditions and different curing conditions, accurately matching multi-objective constraints such as compressive strength and heavy metal ion leaching rate, and compressing the traditional trial-and-error process that takes several months to three days, thus achieving real-time optimization and adaptive adjustment of the ratio scheme under complex conditions; (5) With local utilization of solid waste and zero addition of cement as the core, the solidification carbon emissions are reduced. The pH of the solidified body is stabilized in the eco-friendly range of 7.5-8.5, ensuring the survival rate of soil microorganisms, and simultaneously meeting the strength requirements of foundation reinforcement and ecological restoration needs, forming a green engineering technology paradigm from pollution control to microenvironment reconstruction. Brief Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings, where: Figure 1 It is the SEM electron microscope scanning result diagram of the mixing pile at 28 days of age provided by the embodiment of the present invention; Figure 2 is Figure 1 the EDS energy spectrum analysis result at point 1; Figure 3 is Figure 1 the EDS energy spectrum analysis result at point 2. Detailed Embodiments

[0019] The following will combine the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0020] Embodiment 1 This embodiment provides a method for optimizing the proportion of solid waste composite cementitious materials for mixing piles, including the following steps: Step S1, select the components of the solid waste composite gel material, including aluminum sludge, red mud, building fine powder and phosphate tailings powder, measure the physical and chemical parameters of each component, determine the feasible interval of each component ratio based on the interaction mechanism between the components, take equal-spacing values from the feasible interval of each component ratio, configure the solid waste composite gel material according to the dosage combination, then add different dosages of peat silt soil to the solid waste composite gel material to form a mixture, add water to the mixture to prepare slurries with different water contents, then prepare the slurries into pile specimens, cure them under different curing conditions, and test the performance data of the pile specimens after curing. Define the ratios of each component in the solid waste composite gel material, the dosage of peat silt soil, the water content of the slurry and the curing conditions as variables, and combine the variables and performance test results in each test process to form a variable-performance test data set.

[0021] Aluminum sludge, red mud, construction fine powder, and phosphate ore tailings powder all belong to industrial solid waste materials. In the solid waste composite gel material provided by the present invention, a composite curing system is constructed with aluminum sludge, red mud, construction fine powder, and phosphate ore tailings powder. Aluminum sludge is rich in Al2O3 and SiO2, and red mud is rich in highly alkaline iron oxides. Al2O3 and SiO2 can be converted into aluminosilicate, and through the alkaline activation of aluminosilicate and iron oxides, a C-A-S-H gel network is generated, which can completely replace cement-based materials to play a role in curing peat silt soil.

[0022] Industrial solid waste materials generally contain heavy metal elements, such as Cd in aluminum sludge and Pb in red mud. Construction fine powder belongs to a loose and porous material, and its physical adsorption characteristics of porous media can intercept heavy metal ions (such as Pb 2+ , Cd 2+ ), and the calcium-based active components in construction fine powder can also participate in the gelation reaction to promote the formation of the C-A-S-H gel network; in addition, phosphates in phosphate ore tailings powder can form stable precipitates with heavy metals such as Pb3(PO4)2 and Cd3(PO4)2, and then be wrapped by the gel network. Through the "adsorption-precipitation-wrapping" triple stabilization mechanism, in-situ fixation of heavy metals is realized, avoiding the leaching of heavy metal ions to pollute the environmental soil body.

[0023] Determining the physical and chemical parameters of each component specifically includes: (i) Using X-ray fluorescence spectrometry (XRF) to determine the contents of Al2O3 and SiO2 in aluminum sludge; analyzing the forms of heavy metal elements in aluminum sludge by sequential extraction method; determining the leaching concentration of heavy metal ions in aluminum sludge by toxicity characteristic leaching procedure; (ii) Using X-ray fluorescence spectrometry (XRF) to determine the content of Fe2O3 in red mud; analyzing the forms of heavy metal elements in red mud by sequential extraction method; determining the leaching concentration of heavy metal ions in red mud by toxicity characteristic leaching procedure; (iii) Determining the specific surface area of construction fine powder by nitrogen adsorption-desorption isotherm method; calculating the average pore diameter and total pore volume of construction fine powder by BJH method, and combining SEM-EDS analysis to confirm the unit adsorption capacity of construction fine powder for heavy metal ions; (iv) Using XRD quantitative analysis to determine the content of hydroxyapatite in phosphate ore tailings powder, and conducting batch experiments to verify the unit chemical fixation amount of phosphate ore tailings powder for heavy metal ions.

[0024] The formation of the C-A-S-H gel network mainly depends on the alkaline activation of aluminosilicate and iron oxide. Therefore, the contents of Al2O3 and SiO2 in the aluminum sludge and the content of Fe2O3 in the red mud directly affect the formation of the final C-A-S-H gel network. By measuring the contents of Al2O3 and SiO2 in the aluminum sludge and the content of Fe2O3 in the red mud, it can be used as the basis for selecting the dosages of red mud and aluminum sludge. In addition, the fixation of heavy metal ions mainly depends on the construction micro-powder and phosphate tailings powder. Therefore, by measuring the forms and leaching concentrations of heavy metal elements in the aluminum sludge and red mud, as well as the unit adsorption / fixation capacity of the construction micro-powder and phosphate tailings powder for heavy metal ions, it can be used as the basis for selecting the dosages of the construction micro-powder and phosphate tailings powder. By determining the feasible range of the proportions of the basic components in the solid waste materials in the above way, the selection range of the optimal proportion can be narrowed, and thus the optimization process can be accelerated.

[0025] In the actual engineering application scenario, the peat-silt soil is used as the soil structure of the pile-forming stratum. After the gel material is solidified, it forms a bonding effect on the peat-silt soil, enhancing the compressive strength of the peat-silt soil. To simulate the interaction process between the gel material and the peat-silt soil, the present invention directly incorporates the peat-silt soil into the raw materials of the solid waste composite gel material to study the erosion effect of the peat-silt soil on the pile body. By adjusting the dosage of the peat-silt soil, the pile-forming stability of the solid waste composite gel material in different peat-silt soil environments can be evaluated.

[0026] The preparation of the slurry into a pile body specimen specifically includes the following process: drying each component of the solid waste composite gel material at 105 °C to constant weight and passing through an 80 μm square-hole sieve; the peat-silt soil is freeze-dried and then crushed to the same particle size; the "dry mixing - wet mixing" two-stage method is implemented using a planetary mixer: first, dry mixing is carried out, and the dry materials of each component are premixed at a speed of 60 rpm for 3 min; then wet mixing is carried out, injecting different amounts of water into the mixer to obtain slurries with different water contents, and then wet mixing at a speed of 120 rpm for 5 min; after mixing is completed, the material is taken to make a pile; and then it is cured under the set curing conditions.

[0027] The curing conditions include: curing temperature, curing humidity, and curing age.

[0028] The properties of the pile body specimen include compressive strength and heavy metal ion leaching rate. The compressive strength is obtained by testing through an unconfined compressive strength test, and the heavy metal ion leaching rate is obtained by testing through a toxicity characteristic leaching procedure.

[0029] The process of the unconfined compressive strength test is as follows: Use a microcomputer-controlled electronic universal testing machine with an accuracy class of 0.5 to conduct an unconfined compressive strength test on the pile body specimen according to the ASTM D2166 standard. Apply a preload at a loading rate of 0.5 mm / min until the contact stress reaches 5 N, and then switch to a loading rate of 1 mm / min until the sample fails. Collect the stress-strain curve of the sample and calculate the peak strength.

[0030] The process of the toxicity characteristic leaching procedure test is as follows: Follow the US EPA 1311 standard, crush the pile body specimen into particles smaller than 9.5 mm, mix the pile body specimen particles with acetic acid solution with pH = 2.88 ± 0.05 at a solid-liquid ratio of 20:1, and then place it in a rotary shaker for extraction for 18 ± 2 h. The rotation speed of the rotary shaker is 30 ± 2 rpm; after the extraction is completed, filter it through a 0.45 μm filter membrane, and then use ICP-MS (inductively coupled plasma mass spectrometer) to detect the heavy metal isotope signal and calculate the heavy metal ion leaching rate.

[0031] By adjusting the values of different variables, different performance data can be tested to form multiple groups of variable-performance test data sets.

[0032] In step S2, based on the variable-performance test data set, fit the relationship between the variables and the performance, use the relationship to generate the performance fitting results under different variables, and combine the variables and the performance fitting results in each fitting process to form a variable-performance fitting data set. The variable-performance test data set and the variable-performance fitting data set together form a sample data set.

[0033] Compressive strength The fitting formula is expressed as: ; In the formula, represents the comprehensive correction coefficient; represents the content of peat soft soil; represents the proportion of the th component in the solid waste composite gel material, ; represents the total number of components, with a value of 4; represents the actual moisture content of the slurry; represents the optimal moisture content of the slurry; represents the curing temperature; represents the reference temperature, taking 20 °C; represents the relative curing humidity; represents the reference humidity, taking 95%; represents the curing age; represents the reference age, taking 7 days; 、 、 , , represents the material characteristic parameters, which are obtained by fitting; Heavy metal ion leaching rate The fitting formula is expressed as: ; In the formula, represents the initial leaching concentration; represents the leaching time; represents the time decay coefficient; , are the solidification and stabilization efficiency variables, , , which are obtained by fitting.

[0034] After fitting the compressive strength and the heavy metal ion leaching rate expressions, by adjusting the values of different variables, different performance data can be generated to form multiple variable-performance fitting data groups.

[0035] The variable-performance test data group depends on experimental measurement, and the sample data it can provide is extremely limited. It is prone to overfitting during the training process of the subsequent machine learning model. Therefore, in this application, the variable-performance test data group measured by experiments is first used to fit the relational expression, and then the relational expression is used to generate a large number of variable-performance fitting data groups, which plays a role in expanding the samples and is beneficial to the training of the subsequent model.

[0036] In the sample data set, the original data (data measured by experiments) retains the underlying details, and the fitting data provides high-level abstractions. The combination of the two can enhance the feature expression ability; moreover, the fitting data can supplement the distribution areas not fully covered in the original data, improving the model generalization ability; in addition, the fitting data can also reveal the non-linear relationship or hidden pattern between variables, assisting the model to understand complex laws.

[0037] It should be noted that the sample data set also needs to be preprocessed before training, such as normalization processing, enhancement processing, etc. The conventional techniques in the art can be used, and this embodiment will not elaborate on this.

[0038] The sample data set is expressed as: .

[0039] Step S3, construct a performance prediction model, input the sample data set into the performance prediction model for training. After the training is completed, given different variables, use the performance prediction model to predict the performance data of the pile body, and combine the variables and performance prediction results in each prediction process to form a variable-performance prediction data group.

[0040] The performance prediction model is selected as a residual neural network, including an input layer, a hidden layer, and an output layer. The input layer includes 9 nodes, and 9-dimensional variables of the aluminum sludge ratio, red mud ratio, building micro-powder ratio, phosphate ore tailings powder ratio, peat silt soil content, slurry moisture content, curing temperature, curing humidity, and curing age are input respectively. The hidden layer consists of three layers in total. The first layer is provided with 64 nodes, the second layer is provided with 32 nodes, and the third layer is provided with 16 nodes. Each node is provided with a residual block, and each residual block includes two fully connected layers with a skip connection between the two fully connected layers. The output layer includes 2 nodes, which output the 28-day compressive strength of the pile body and the average value of the heavy metal ion leaching rate respectively. The hidden layer is activated by the Swish function, and the output layer is activated by the Sigmoid function.

[0041] Step S4, construct a meta-learning policy network, input multiple groups of variable-performance prediction data groups into the meta-learning policy network for training. After the training is completed, set the target threshold of the pile body performance data, and generate multiple candidate variable combination schemes with different values through the meta-learning policy network.

[0042] The meta-learning policy network divides the tasks into two types. One type takes the compressive strength as the optimization goal, and the other type takes the heavy metal ion leaching rate as the optimization goal. Each type generates multiple subtasks. In the inner loop stage, use the initial values of the variables (the ratios of the components of the solid waste composite gel material, the content of peat silt soil, the slurry moisture content, and the curing conditions) generated by the current meta-policy, and optimize the policy network through the gradient descent method to make the generated variables meet the requirements of the optimization goal. In the outer loop stage, aggregate the inner loop gradients of all tasks and update the meta-policy parameters to make it have the cross-task generalization ability.

[0043] Before the training of each type of task starts, it is also necessary to calculate the correlation between the variables and the performance. The present invention simultaneously uses two statistical methods, the Pearson correlation coefficient and the Spearman rank correlation coefficient, to analyze the global correlation between the variables and the performance. Among them, the Pearson correlation coefficient is used to measure the linear correlation between the variables and the performance, and the Spearman rank correlation coefficient is used to measure the non-linear rank dependence between the variables and the performance. By assigning different weights to the Pearson correlation coefficient and the Spearman rank correlation coefficient, the global correlation between the variables and the performance is obtained. The calculation processes of the Pearson correlation coefficient and the Spearman rank correlation coefficient adopt the conventional techniques in the field, and are not elaborated in this embodiment.

[0044] Then, the Kernel Principal Component Analysis (KPCA) algorithm is introduced for non-linear dimensionality reduction. The variables are mapped from the original feature space to the high-dimensional Hilbert space through the kernel function, and the standard PCA analysis is performed in this space to extract multiple principal components highly correlated with the optimization performance for subsequent learning and modeling, so as to compress the input dimension, retain the main variation features and reduce the problem of multicollinearity.

[0045] Loss function for training the meta-learning policy network It is expressed as: ; In the formula, and respectively represent the predicted value and the true value of the task related to the compressive strength, and respectively represent the predicted value and the true value of the task related to the leaching rate of heavy metal ions, and respectively represent the weight coefficients, which are dynamically adjusted according to the global correlation matrix to ensure that the high-correlation objectives occupy higher weights in the learning.

[0046] After multiple rounds of iteration, the trained meta-policy network has the ability to quickly adapt to unknown proportion tasks. Set the target threshold of the pile body performance data, and generate multiple candidate solutions for the proportions of each component of the solid waste composite gel material, the admixture amount of peat silt soil, the moisture content of the slurry, and the curing conditions through the meta-learning policy network for the next reinforcement learning optimization and screening.

[0047] Step S5, construct a multi-objective reinforcement learning optimization model, and screen out the proportion combination with the optimal comprehensive performance from the candidate solutions based on the state-action-reward feedback mechanism.

[0048] After all variables are normalized and input into the policy network as the environmental state, the state space of the multi-objective reinforcement learning optimization model is expressed as: ; The action space is the set of micro-adjustment operations of the variables, which is expressed as: ; Each action execution corresponds to locally perturbing the proportion plan in the current state, such as increasing the admixture amount of aluminum mud and decreasing the moisture content of the pile body, so as to jump to a new candidate solution state.

[0049] To balance the dual performance indicators, the piecewise reward function is defined as follows: ; In the formula, , represents a weight coefficient; represents the threshold value of compressive strength; represents the threshold value of heavy metal ion leaching rate.

[0050] The reinforcement learning training stage is carried out using the deep Q-network algorithm. The training process belongs to the prior art in this field and will not be elaborated in this embodiment.

[0051] After the training is completed, the reinforcement learning model can output one or more sets of composite ratio results that perform optimally under multiple performance objectives from the candidate ratio schemes, as an important basis for material design and engineering practice.

[0052] Example 2 This embodiment provides a solid waste composite cementitious material for mixing piles, including aluminum sludge, red mud, construction micro-powder, and phosphate tailings powder. The ratio of each component is obtained through the ratio optimization method in Example 1.

[0053] Example 3 In this embodiment, the feasibility of the optimization method provided by the present invention is verified through experiments. In this embodiment, the aluminum sludge is a by-product generated during the production of polyaluminum chloride (PAC) in a certain aluminum salt production company in Central China, mainly composed of aluminum hydroxide and alumina; the red mud is a high-alkali solid waste discharged during the extraction of alumina from bauxite in a certain alumina production enterprise in Central China using the Bayer process, mainly rich in iron oxide, silicon dioxide, alumina, and titanium oxide; the construction micro-powder is a fine-grained mineral material recovered and prepared in the crushing, screening, and ultrafine grinding production lines of a certain construction waste recycling company in Changsha (such as waste concrete, bricks, and mortar), mainly composed of silicate and calcareous materials; the phosphate tailings powder is a powder by-product obtained from a certain phosphochemical enterprise in Central China using the wet-process phosphoric acid production line, which is the tailings generated during the acidolysis of phosphate rock to prepare phosphoric acid, and is obtained after dehydration, drying, and fine grinding, mainly composed of silicon dioxide, calcium oxide, and phosphate residues.

[0054] The physical and chemical parameters of each component are measured as follows: The content range of Al2O3 in the aluminum sludge is 43%, the content of SiO2 is 13%, and the main heavy metal ion in the aluminum sludge is Cd 2+ , and the leaching concentration is 0.12 mg / L; The content of Fe2O3 in the red mud is 38%, and the main heavy metal ion in the aluminum sludge is Pb 2+ , and the leaching concentration is 0.35 mg / L; The specific surface area of the construction waste powder is 218 m² / g, the average pore diameter is 4.7 nm, the total pore volume is 0.68 cm³ / g, and the theoretical adsorption capacity for heavy metal ions is 45 mg / g; The hydroxyapatite content of the phosphate ore tailings powder is 62%. When the dosage is >15%, the removal rate of heavy metal ions reaches 92%.

[0055] Based on the above physical and chemical parameters, the feasible intervals of the component ratios are obtained: The ratio of aluminum sludge is 8.3 - 25%, the ratio of red mud is 8.3 - 25%, the ratio of construction waste micropowder is 25% - 41.7%, and the ratio of phosphate ore tailings powder is 25% - 41.7%; and the ratio of aluminum sludge is the same as that of red mud; the ratio of construction waste micropowder is the same as that of phosphate ore tailings powder.

[0056] Based on the feasible intervals of the above component ratios, the ratio of peat silt soil in the mixture, and the empirical values of the slurry moisture content and curing conditions in engineering practice, an L27(3 9 ) orthogonal array is set up, as shown in Table 1. It should be noted that the ratio of aluminum sludge, the ratio of red mud, the ratio of construction micropowder, and the ratio of phosphate ore tailings powder refer to the ratios of each component in the solid waste composite gel material; the ratio of peat silt soil refers to the ratio of peat silt soil in the mixture.

[0057] Table 1 L27(3 9 ) orthogonal array Prepare pile body specimens according to the above 27 groups of data, and test the performance data of the 27 groups of pile body specimens, as shown in Table 2.

[0058] Table 2 Performance data of 27 groups of pile body specimens The values of each fitting parameter in the expression of the compressive strength fitted from the performance test results are as shown in Table 3 below.

[0059] Table 3 Values of each fitting parameter in the expression of the compressive strength fitting parameter value table The values of each fitting parameter in the expression of the heavy metal ion leaching rate fitted from the test results are as shown in Table 4 below.

[0060] Table 4 Values of each fitting parameter in the expression of the heavy metal ion leaching rate fitting parameter value table The compressive strength is fitted out and the leaching rate of heavy metal ions After the expressions, by adjusting the values of different variables, different performance data can be generated, forming multiple groups of variable-performance fitting data sets. Using the variable-performance test data sets and variable-performance fitting data sets together to form a sample data set to train the performance prediction model. Input the data generated by the performance prediction model into the meta-learning strategy network for training. During the training process, calculate the correlation between the input variables and the performance data.

[0061] After calculation, for the compressive strength, the Pearson correlation coefficient is 0.7, the Spearman rank correlation coefficient is 0.6, and the assigned weights are 0.6 and 0.4 respectively, obtaining a global correlation of 0.66; for the leaching concentration of heavy metal ions, the Pearson correlation coefficient is -0.8, the Spearman rank correlation coefficient is -0.7, and the assigned weights are 0.5 and 0.5 respectively, obtaining a global correlation of -0.75.

[0062] After non-linear dimensionality reduction by the kernel principal component analysis algorithm, the target threshold is set as: compressive strength ≥ 2 MPa, and the heavy metal leaching rate is lower than 0.05 mg∙L -1 , according to the actual construction conditions, set the dosage of peat-silt soil, the water content of the slurry and the curing conditions, and use the meta-learning strategy network to generate 3 groups of candidate solutions for the component ratios in the solid waste composite gel material, as shown in Table 5: Table 5 Candidate solution table Input the above candidate solutions into the multi-objective reinforcement learning optimization model for screening. In the action space, slightly adjust the dosage of aluminum sludge (increase by 1%) to obtain a new candidate solution state. Train through the deep Q-network algorithm, and calculate the reward value according to the defined piecewise reward function. For the above 3 groups of candidate solutions, the calculated piecewise reward values are 0.85, 0.78, and 0.82 respectively.

[0063] By comparing the reward values, select the candidate solution with the highest reward value as the ratio combination with the optimal comprehensive performance. In this example, the reward value of candidate solution 1 is the highest and is determined as the final optimal ratio solution, which can be used as a reference parameter for material design and engineering practice. Carry out qualitative and quantitative microscopic characterization analysis based on SEM-EDS on the pile body specimens under candidate solution 1, as Figures 1-3 shown. Due to the complementary chemical compositions of each solid waste, fine solid waste particles dissolve (i.e., Na, Ca, Si, and Al components dissolve into the hardening system), and then polymerization and hydration reactions occur synergistically. As Figure 1As shown, the matrix structure of the cementitious material is well-developed and dense, while there are many typical polygonal prismatic and clustered structures attached to the surface. EDS quantitative analysis was carried out on its matrix structure and the attached polygonal prismatic structure respectively, and the results are as Figures 2-3 shown. It can be seen that the hardened matrix of the cementitious material is a dense structure characterized by the coexistence of multi-phase gels (calcium silicoaluminate / sodium hydrate, i.e., (N, C)-A-S-H complex), and the product detected on its surface is calcium silicoaluminate hydrate, i.e., C-(A)-S-H gel. These gels act synergistically to fill the matrix defects and provide nucleation sites for subsequent products.

[0064] The embodiments of the present invention have been described above. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them belong to the protection scope of the present invention.

Claims

1. A method for optimizing the proportion of solid waste composite cementitious material for mixing piles, characterized in that, It includes the following steps: Step S1: Select the components of the solid waste composite gel material, including aluminum sludge, red mud, construction micro-powder, and phosphate tailings powder. Measure the physical and chemical parameters of each component. Based on the interaction mechanism between the components, determine the feasible range of each component ratio. Take equally spaced values from the feasible range of each component ratio, configure the solid waste composite gel material according to the dosage combination. Then, add peat sludge soil with different dosages to the solid waste composite gel material to form a mixture. Add water to the mixture to prepare slurries with different water contents. Then, prepare the slurries into pile specimens and cure them under different curing conditions. After curing, test the performance data of the pile specimens. Define the ratios of each component in the solid waste composite gel material, the dosage of peat sludge soil, the water content of the slurry, and the curing conditions as variables. Combine the variables and the performance test results in each test process to form a variable-performance test data set; Step S2: Based on the variable-performance test data set, fit the relationship between variables and performance, and use the relationship to generate performance fitting results under different variables. Combine the variables and the performance fitting results in each fitting process to form a variable-performance fitting data set, and jointly form a sample data set with the variable-performance test data set; Step S3: Construct a performance prediction model, input the sample data set into the performance prediction model for training. After training, given different variables, use the performance prediction model to predict the performance data of the pile. Combine the variables and the performance prediction results in each prediction process to form a variable-performance prediction data set; Step S4: Construct a meta-learning strategy network, input multiple groups of variable-performance prediction data sets into the meta-learning strategy network for training. After training, set the target threshold of the pile performance data, and generate multiple groups of variable combination candidate solutions with different values through the meta-learning strategy network; Step S5: Construct a multi-objective reinforcement learning optimization model, and screen out the ratio combination with the optimal comprehensive performance from the candidate solutions based on the state-action-reward feedback mechanism.

2. The method for optimizing the proportion of solid waste composite cementitious material for mixing piles according to claim 1, characterized in that Specifically, measuring the physical and chemical parameters of each component includes: (i) Use X-ray fluorescence spectrometry to measure the contents of Al2O3 and SiO2 in aluminum sludge; analyze the forms of heavy metal elements in aluminum sludge by sequential extraction method; measure the leaching concentrations of heavy metal elements in aluminum sludge by toxicity characteristic leaching procedure; (ii) Use X-ray fluorescence spectrometry to measure the content of Fe2O3 in red mud; analyze the forms of heavy metal elements in red mud by sequential extraction method; measure the leaching concentrations of heavy metal elements in red mud by toxicity characteristic leaching procedure; (iii) Measure the specific surface area of construction micro-powder by nitrogen adsorption-desorption isotherm method; calculate the average pore diameter and total pore volume of construction micro-powder by BJH method, and combine SEM-EDS analysis to confirm the unit adsorption capacity of construction micro-powder for heavy metal ions; (iv) Use XRD quantitative analysis to measure the content of hydroxyapatite in phosphate tailings powder, and conduct batch experiments to verify the unit chemical fixation amount of phosphate tailings powder for heavy metal ions.

3. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 1, characterized in that, The preparation of the slurry to form the pile sample specifically includes the following process: drying each component at 105 °C to a constant weight and passing it through an 80 μm square-hole sieve; freeze-drying the peat silt soil and then crushing it to the same particle size; using a planetary mixer to implement the "dry mixing - wet mixing" two-stage method: first, perform dry mixing, premixing the dry materials of each component at a speed of 60 rpm for 3 minutes; then, perform wet mixing, injecting different amounts of water into the mixer to obtain slurries with different water contents, and then wet mixing at a speed of 120 rpm for 5 minutes; after mixing, take the material to form a pile; and then cure it under the set curing conditions.

4. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 3, characterized in that, The curing conditions include: curing temperature, curing humidity, and curing age; the properties include compressive strength and heavy metal ion leaching rate. The compressive strength is obtained through an unconfined compressive strength test, and the heavy metal ion leaching rate is obtained through a toxicity characteristic leaching procedure test.

5. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 4, characterized in that, Compressive strength The fitting formula is expressed as: ； In the formula, represents the comprehensive correction coefficient; represents the content of peat-silt soil; represents the proportion of the th component; represents the total number of component types, with a value of 4; represents the actual moisture content of the slurry; represents the optimal moisture content of the slurry; represents the curing temperature; represents the reference temperature, taking 20°C; represents the relative curing humidity; represents the reference humidity, taking 95%; represents the curing age; represents the reference age, taking 7 days; , , , , represent the material characteristic parameters, obtained by fitting; Leaching rate of heavy metal ions The fitting formula is expressed as: ； In the formula, represents the initial leaching concentration; represents the leaching time; represents the time decay coefficient; , are variables of solidification and stabilization efficiency, , , are obtained by fitting.

6. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 1, characterized in that, The performance prediction model is a residual neural network, including an input layer, a hidden layer, and an output layer. The input layer includes 9 nodes, respectively inputting 9-dimensional variables of the ratio of aluminum mud, the ratio of red mud, the ratio of building micro-powder, the ratio of phosphate tailings powder, the content of peat silt soil, the water content of the slurry, the curing temperature, the curing humidity, and the curing age; the hidden layer includes a total of three layers. The first layer is set with 64 nodes, the second layer is set with 32 nodes, and the third layer is set with 16 nodes. Each node is provided with a residual block, and each residual block includes two fully connected layers, with a skip connection between the two fully connected layers; the output layer includes 2 nodes, respectively outputting the average compressive strength of the pile body and the heavy metal ion leaching rate. The hidden layer is activated by the Swish function, and the output layer is activated by the Sigmoid function.

7. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 5, characterized in that, The meta-learning strategy network divides the tasks into two types. One type takes the compressive strength as the optimization target, and the other type takes the heavy metal ion leaching rate as the optimization target. Each type generates multiple sub-tasks. In the inner loop stage, use the current meta-strategy to generate the initial values of the component ratio, the content of peat silt soil, the water content of the slurry, and the curing conditions, and optimize the strategy network through gradient descent to make the generated component ratio, the content of peat silt soil, the water content of the slurry, and the curing conditions meet the requirements of the optimization target; in the outer loop stage, aggregate the inner loop gradients of all tasks and update the meta-strategy parameters.

8. The optimization method for the proportioning of the solid waste composite cementitious material for mixing piles according to claim 7, characterized in that, Before the training of each type of task starts, it is also necessary to calculate the correlation between the variables and the performance data, specifically including the following process: Both the Pearson correlation coefficient and the Spearman rank correlation coefficient are used to analyze the global correlation between the variables and the performance data. Among them, the Pearson correlation coefficient is used to measure the linear correlation between the variables and the performance data, and the Spearman rank correlation coefficient is used to measure the non-linear rank dependence between the variables and the performance data. By assigning different weights to the Pearson correlation coefficient and the Spearman rank correlation coefficient, the global correlation between the variables and the performance data is obtained; Then, the kernel principal component analysis algorithm is introduced for non-linear dimensionality reduction. The variables are mapped from the original feature space to the high-dimensional Hilbert space through the kernel function, and the standard PCA analysis is performed in this space to extract multiple principal components for subsequent learning and modeling, so as to compress the input dimension, retain the main variation features and reduce the problem of multicollinearity; Loss function for training of meta - learning strategy network It is expressed as: ； In the formula, , represent the true value and predicted value of the task related to the compressive strength, , represent the true value and predicted value of the task related to the heavy metal ion leaching rate, , respectively represent the weight coefficients, which are dynamically adjusted according to the global correlation matrix to ensure that the high-correlation targets occupy higher weights in learning.

9. The method for optimizing the proportion of the solid waste composite cementitious material for mixing piles according to claim 8, wherein State Space of the Multi-Objective Reinforcement Learning Optimization Model It is expressed as: ； The action space is a set of micro-adjustment operations of variables, which is expressed as: ； Each action execution corresponds to a local perturbation of the mixing ratio scheme in the current state, thus jumping to a new candidate ratio state; To balance the dual performance indicators, the piecewise reward function is defined as follows: ； In the formula, , represent weight coefficients; represents the threshold value of compressive strength; represents the threshold value of heavy metal ion leaching rate; The deep Q-network algorithm is used in the reinforcement learning training stage.

10. A solid waste composite cementitious material for mixing piles, characterized in that, It includes aluminum sludge, red mud, building micro-powder and phosphate ore tailings powder, and the mixing ratio of each component is obtained by the mixing ratio optimization method of the solid waste composite cementitious material for mixing piles described in any one of claims 1-9.

Images