Machine learning based geopolymer production optimization method and system

Abstract

The application provides a geopolymer preparation optimization method and system based on machine learning. Applied to the field of material science and machine learning technology, the method comprises the following steps: collecting geopolymer preparation experimental data and performing pretreatment; performing nonlinear regression modeling on the geopolymer performance based on four machine learning regression algorithms to construct a geopolymer performance prediction model; calculating and assigning weights according to the mean square error of each machine learning model on the validation set to obtain a performance prediction result; receiving target performance parameters and initial raw material ratio range input by a user, performing performance prediction by using the trained geopolymer performance prediction model, and searching for an optimal ratio combination satisfying the target performance constraint through an optimization algorithm; and preparing a coal gangue-slag-fly ash geopolymer grouting material according to the optimal ratio combination. The application effectively improves the prediction accuracy and model generalization ability, and realizes intelligent recommendation and accurate performance prediction of raw material ratios.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material science and machine learning, and particularly relates to a geopolymer preparation optimization method and system based on machine learning. BACKGROUND

[0002] Geopolymer is a new type of inorganic polymer material formed by polymerization reaction under alkaline activation conditions, using industrial solid waste (such as fly ash, coal gangue, slag, etc.) rich in silicon and aluminum components as the main raw material. Due to its excellent mechanical strength, good thermal stability, corrosion resistance and environmental friendly characteristics, geopolymer has broad application prospects in green building materials, environmental engineering, transportation infrastructure and other fields.

[0003] However, the chemical activity of coal-based solid waste (especially coal gangue, fly ash, etc.) is usually low, and the composition of raw materials varies greatly, resulting in complex process, multiple repeated tests, low efficiency and large performance fluctuations in traditional geopolymer preparation process. Especially in practical engineering, how to quickly predict the performance of geopolymer under complex proportioning conditions and optimize the proportioning parameters is a key problem restricting its wide application.

[0004] The traditional method has the following disadvantages: the proportioning design relies on experience and trial and error: hundreds of orthogonal tests are needed to obtain a feasible formula, and the research and development cycle is long; the performance prediction accuracy is insufficient: the fitting error of empirical formula (such as Bolomey formula) for complex nonlinear relationship is large, and it is difficult to meet the engineering precision requirements; the system optimization efficiency is low: thousands of iterations are needed for multi-objective optimization (such as pursuing strength and fluidity at the same time), and the existing methods (such as genetic algorithm) have slow convergence speed, which limits the engineering application.

[0005] In recent years, with the wide application of machine learning in material science, there are still two challenges in the field of geopolymer: weak generalization ability of single model: traditional support vector machine (SVM) is prone to overfitting when dealing with high-dimensional nonlinear data, and artificial neural network (ANN) is sensitive to hyperparameters and requires a large amount of data; insufficient capture of local features: the performance of geopolymer is sensitive to the proportioning parameters in a segmented manner (such as the performance mutation of sodium silicate modulus in the interval of 1.8-2.2 is significant), and the existing global modeling method is difficult to capture such local rules.

[0006] Therefore, there is an urgent need for a new geopolymer preparation optimization method and system, which can efficiently establish the complex nonlinear mapping relationship between the proportioning and the performance, and realize the intelligent optimization of the whole process from raw material proportioning to performance prediction. SUMMARY

[0007] The application is to solve the problems of relying on experience in ratio design, inaccurate performance prediction and low system optimization efficiency in the preparation process of geopolymer, and proposes a geopolymer preparation optimization method and system based on machine learning, which effectively improves the prediction accuracy and model generalization ability by fusing multiple machine learning models and introducing an adaptive weighting mechanism, and realizes intelligent recommendation of raw material ratio and accurate performance prediction.

[0008] To achieve the above purpose, the following technical solutions are implemented:

[0009] According to the first aspect of the application, a geopolymer preparation optimization method based on machine learning is provided, comprising the following steps:

[0010] S1, collecting geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and performing missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data;

[0011] S2, based on four machine learning regression algorithms, a nonlinear regression model of geopolymer performance is established, a geopolymer performance prediction model fusing four machine learning models is constructed, and a nonlinear mapping relationship between raw material ratio and performance is established, including:

[0012] Support vector regression model based on Rational-Minkowski kernel function;

[0013] Segmented Gaussian process regression model, by dividing the input space into subintervals, modeling each subinterval independently and then performing boundary weighted fusion;

[0014] Regularized neighborhood component analysis model, which learns feature weights by minimizing a loss function containing an L2 regularization term, and defines a feature weighted distance measure;

[0015] Artificial neural network model based on Levenberg-Marquardt algorithm;

[0016] S3, according to the mean square error of each machine learning model on the validation set in the data set, the weight is calculated and distributed, and the performance prediction result is obtained by weighted fusion;

[0017] S4, receiving the target performance parameters and the initial raw material ratio range input by the user, using the trained geopolymer performance prediction model to perform performance prediction, and searching the best ratio combination that meets the target performance constraint through an optimization algorithm;

[0018] S5, according to the best ratio combination, a coal gangue-slag-fly ash geopolymer grouting material is prepared.

[0019] Further, the construction of the support vector regression model based on the Rational-Minkowski kernel function comprises:

[0020] According to the Rational-Minkowski kernel function, an improved kernel function support vector regression model is constructed, wherein the expression of the Rational-Minkowski kernel function is:

[0021]

[0022] wherein, is a sample input vector, and n is the sample quantity; represents the Minkowski distance norm, p∈[1,3], used for adjusting the norm order of the Minkowski distance; τ is a scale parameter and τ>0, used for controlling the width of the kernel function; q is an exponential adjustment parameter and q∈[1,5], used for controlling the decay speed of the kernel function.

[0023] Further, the construction of the piecewise Gaussian process regression model comprises:

[0024] Input space division: based on principal component analysis and K-means clustering or expert knowledge rules, the geopolymer mixture input space is divided into M subintervals, each of which meets the local data stability;

[0025] Local model training: for each subinterval, a Gaussian process regression submodel is independently trained, and the kernel function type and hyperparameters are selected; the kernel function includes a squared exponential kernel function, a Matérn3 / 2 kernel function, or a RationalQuadratic kernel function;

[0026] Segmented prediction integration: for a test sample, its belonging subinterval is determined, and the corresponding submodel is used for prediction; if it is located in a boundary region, the prediction results of adjacent submodels are fused through weighted averaging;

[0027] Hyperparameter optimization: the hyperparameters of each submodel are estimated by maximizing the log marginal likelihood function.

[0028] Further, if it is located in a boundary region, the prediction results of adjacent submodels are fused through weighted averaging, comprising:

[0029]

[0030]

[0031] wherein, is a prediction mean value, representing the conditional expectation estimation value of the geopolymer performance index under the given input condition. : Prediction variance, representing the variance of the predicted mean A quantitative measure of uncertainty or confidence level; Input samples for testing; This represents the total number of sub-intervals or sub-models. Indicates the first Sub-model exist The predicted mean at the location; Indicates the first Sub-models in The predicted variance at the location; For the first The spatial weights of each sub-model satisfy the normalization constraint:

[0032]

[0033] Weight according to Calculation of the inverse proportional function of distance to the center of each subinterval:

[0034]

[0035] in, Indicates test input sample With the Center of each sub-interval Euclidean distance.

[0036] Furthermore, the estimation of hyperparameters of each sub-model by maximizing the log-marginal likelihood function includes:

[0037]

[0038] in, The training output vector, i.e., the target variable, includes compressive strength; The training input sample matrix, i.e., the geopolymer sizing parameters, has a total of There are samples, each sample has One input feature; The covariance matrix is ​​defined as follows: , by kernel function structure; : The variance of observation noise, used to characterize unexplainable noise in the data; : Identity matrix, dimension , used to introduce noise.

[0039] Furthermore, the construction of the regularized neighborhood component analysis model includes the following steps:

[0040] Feature-weighted distance definition: Construct the weighted Manhattan distance metric function:

[0041]

[0042] in, : No. The input vector of each sample; : No. In the nth sample The values ​​of each feature; : No. Weight coefficients of each feature; distance metric It is a weighted Manhattan distance form;

[0043] Neighborhood probability calculation: Define sample choose The probability of being a valid neighbor:

[0044]

[0045] : Indicates a sample Will The higher the probability value, the better the probability of it being a "valid neighbor". The more likely to be The predictions have an effect; : No. The smoothing scale parameter for each feature is used to normalize the distance values ​​of different feature dimensions;

[0046] Model training: Feature weights are learned by minimizing a loss function that includes an L2 regularization term;

[0047] Feature selection: By compressing the influence of redundant features through optimized weights, high-dimensional feature selection and dimensionality reduction are achieved.

[0048] Furthermore, the artificial neural network model based on the Levenberg-Marquardt algorithm satisfies the following conditions:

[0049] Network structure: A three-layer feedforward neural network is adopted. The input layer receives the aggregate ratio parameters, including coal gangue content, fly ash content, slag content, water glass modulus, and Na2O content. The number of neurons in the hidden layer is 2-15. The output layer predicts the target performance indicators.

[0050] Weight update formula: The weights are dynamically adjusted using the Levenberg-Marquardt algorithm. The update formula is as follows:

[0051]

[0052] Let be the weight vector for the r-th iteration; This is the Jacobian matrix, containing the partial derivatives of the network output with respect to the weights; This is the error vector; To adjust the parameters; It is the identity matrix;

[0053] Training mechanism: When When →0, the algorithm approximates the Gauss-Newton method to accelerate convergence; when As the value increases, the algorithm degenerates into gradient descent to enhance stability;

[0054] Early stopping mechanism: During training, the validation set error is monitored, and training is terminated when the error does not decrease for a preset number of consecutive times to prevent overfitting.

[0055] Furthermore, in step S3, the weight calculation includes:

[0056] Calculate the mean squared error of each machine learning model on the validation set:

[0057]

[0058] in, For the first The mean squared error of a machine learning model; It is the first The predicted value of each model for the i-th sample; It is the true value of the i-th sample;

[0059] According to the mean square error Calculate the weight using the reciprocal :

[0060]

[0061] in, It is the first The weights of a machine learning model;

[0062] Based on the weights of the four machine learning models The prediction results of the four machine learning models are weighted and averaged to output the final performance prediction value.

[0063] Furthermore, in step S4: the user-defined target performance parameters include compressive strength > 80 MPa and permeability coefficient < 1 × 10⁻⁶. -7 cm / s, and durability indicators including sulfate corrosion resistance cycles and wet-dry cycle mass loss rate;

[0064] The initial raw material proportioning range includes: coal gangue content 60-80 wt%, slag 10-20 wt%, fly ash 10-20 wt%; water glass modulus 1.5-2.5, Na2O content 3-22 wt%; the mass ratio of the activator to the solid raw material is 0.3-0.5;

[0065] S5, according to the optimal proportioning combination, preparing a coal gangue-slag-fly ash geopolymer grouting material, comprising:

[0066] The coal gangue is crushed and ball milled to a specific surface area of 400-600 m² / kg, and the particle size of the slag and the fly ash is controlled to be 0.045-0.075 mm;

[0067] According to the recommended optimal proportioning combination, the coal gangue, the slag and the fly ash are mixed, the activator is added, and the slurry is stirred to a flow degree of 150-200 mm to form a coal gangue-slag-fly ash geopolymer grouting slurry;

[0068] Grouting curing: after the coal gangue-slag-fly ash geopolymer grouting slurry is pre-cured at room temperature for 24 hours, it is transferred to an environment with a temperature of 20±2℃ and a humidity of ≥95% for curing for 28 days to form a coal gangue-slag-fly ash geopolymer grouting material.

[0069] According to the second aspect of the present application, a geopolymer preparation optimization system based on machine learning is also provided, comprising:

[0070] A data acquisition module is configured to acquire geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and to perform missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data;

[0071] A model construction module is configured to perform nonlinear regression modeling on geopolymer performance based on four machine learning regression algorithms, to construct a geopolymer performance prediction model that integrates four machine learning models, and to establish a nonlinear mapping relationship between raw material proportioning and performance, including:

[0072] A support vector regression model based on a Rational-Minkowski kernel function;

[0073] A piecewise Gaussian process regression model that divides the input space into subintervals, independently models each subinterval, and then performs boundary weighted fusion;

[0074] A regularized neighborhood component analysis model that learns feature weights by minimizing a loss function containing an L2 regularization term and defines a feature weighted distance metric;

[0075] An artificial neural network model based on the Levenberg-Marquardt algorithm;

[0076] a prediction output module configured to calculate and assign weights according to the mean square errors of the machine learning models on the validation set in the data set, and to obtain a performance prediction result by weighted fusion;

[0077] a ratio recommendation module configured to receive a target performance parameter input by a user and an initial raw material ratio range, perform performance prediction by using the trained geopolymer performance prediction model, and search for an optimal ratio combination satisfying the target performance constraint by using a reverse search algorithm;

[0078] a grouting preparation module configured to prepare a geopolymer preparation coal gangue-slag-fly ash geopolymer grouting material according to the optimal ratio combination.

[0079] Compared with the prior art, the present application has the following beneficial effects:

[0080] 1. Multi-model fusion improves prediction accuracy and robustness: By integrating support vector regression, segmented Gaussian process regression, regularized neighborhood component analysis and neural network model, a heterogeneous model fusion framework is constructed. Based on the adaptive weighting mechanism of the validation set mean square error (such as the sensitivity of the Rational-Minkowski kernel function to local features and the uncertainty quantification ability of the Gaussian process regression), the model bias and variance are effectively balanced, and the prediction error is effectively reduced compared with single model.

[0081] 2. Local modeling and feature selection enhance generalization ability: The segmented Gaussian process regression (such as dividing the ratio space into 3-5 sub-domains by K-means clustering) is adopted, and the feature weight learning of the regularized neighborhood component analysis (the feature dimension compression rate can reach 60%) is combined, which effectively solves the performance mutation problem caused by the difference of raw material components. In the cross-regional coal gangue sample test, the model generalization error fluctuation range is reduced, and the stability is improved by 2.3 times compared with the traditional global model.

[0082] 3. Reverse optimization drives intelligent ratio design: Based on the back propagation mechanism of neural network of Levenberg-Marquardt algorithm (the convergence speed is improved by 4-8 times compared with gradient descent), the particle swarm optimization algorithm is combined to realize efficient parameter search under multi-objective constraint.

[0083] 4. Process adaptability strengthening and resource conservation: The present application uses coal gangue, slag and fly ash as main raw materials, and prepares a geopolymer grouting material through an alkali activator. By optimizing the coal gangue content (more than 60%), water glass modulus and Na2O content and other parameters, and by optimizing the activator modulus, particle size distribution and other process parameters through characteristic engineering integration, the model can adapt to the activity differences of different solid waste raw materials, and prepare a grouting material with high strength, low permeability and excellent durability, which is suitable for the construction of the impermeable layer of the coal-based solid waste ecological utilization site. The experimental results show that the strength of the material is increased by about 50% compared with the traditional grouting material, the durability is increased by about 40%, and the wettability is increased by about 15%. The technology makes full use of the rich coal gangue resources in Guizhou Province, combines the synergistic effect of slag and fly ash, widens the resource utilization way of coal-based solid waste, reduces the production cost (more than 30%) and environmental risk, and provides a new idea for the research and development of geopolymer materials at home and abroad.

[0084] It should be understood that the content described in the summary section is not intended to limit the key or important features of the embodiments of the present application, nor to limit the scope of the present application. Other features of the present application will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0085] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent by describing in detail preferred embodiments thereof with reference to the attached drawings in which:

[0086] Figure 1 A flowchart of a geopolymer preparation optimization method based on machine learning is shown, which is an embodiment of the present application.

[0087] Figure 2 A module schematic diagram of a geopolymer preparation optimization system based on machine learning is shown, which is an embodiment of the present application. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0089] In addition, the term "and / or" in this document is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the character " / " in this document generally represents an "or" relationship between the front and rear associated objects.

[0090] Figure 1 A flowchart of a machine learning-based geopolymer preparation optimization method is shown. As shown in Figure 1 and Figure 2 The machine learning-based geopolymer preparation optimization method includes the following steps:

[0091] S1, collecting geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and performing missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data;

[0092] S11: Data collection:

[0093] The present application first carries out a large number of geopolymer preparation experiments, collects a plurality of experimental data under different mixing conditions, and constructs a complete input-output data set to provide a reliable basis for subsequent model training. The collected data covers the following aspects:

[0094] Raw material ratio parameters: including the main raw material mixing data of coal-based solid waste, such as coal gangue (60%), fly ash (25%), slag (15%) and the like. These raw materials each have characteristics, and the mixing ratio has an important influence on the structure and performance of the geopolymer.

[0095] Activator parameters: mainly including water glass modulus (SiO2 / Na2O ratio), Na2O content and other alkaline activator parameters. These activators determine the rate and degree of geopolymerization reaction, and are crucial to the final strength and durability.

[0096] Performance output indicators: including compressive strength (MPa), permeability (such as permeability coefficient m / s), durability (such as dry-wet cycle times, sulfuric acid corrosion resistance and the like). These are the prediction targets of the model, which are used to quantify the performance of the geopolymer product in engineering applications.

[0097] S12: Data preprocessing:

[0098] After the data collection is completed, comprehensive data preprocessing is performed, including the following steps:

[0099] Missing value filling and outlier removal: For missing data or error too large abnormal sample in the experiment process, for example, missing value using K nearest neighbor algorithm filling (k=5), abnormal value using Tukey rule to remove (IQR coefficient is 1.5), using interpolation, mean replacement or direct removal way for data cleaning, ensure the reliability and integrity of training data.

[0100] Feature scaling and standardization: In order to improve the convergence speed and prediction accuracy of machine learning model, each input feature is normalized or standardized according to the unified standard. Common methods include Z-score standardization, Min-Max normalization, etc. Different dimensional feature data is mapped to the same numerical interval, avoiding the dominant influence of a certain feature on the learning result in the model training process.

[0101] Feature encoding and mapping: If there are non numerical variables (such as raw material batch classification, activator type, etc.), use one hot encoding or label encoding to convert them into numerical form for model processing.

[0102] Data set division: The whole data set is divided into training set, validation set and test set, for example, according to the proportion of 70%:15%:15% or 80%:10%:10%, to ensure that the model can learn the rule in the training stage, optimize the parameters in the validation stage, and verify the generalization ability in the test stage.

[0103] S2, based on four kinds of machine learning regression algorithm, nonlinear regression modeling of geopolymer performance, building fusion four machine learning model of geopolymer performance prediction model, establish the nonlinear mapping relationship between raw material ratio and performance, including:

[0104] Support vector regression model based on Rational-Minkowski kernel function;

[0105] Segmented Gaussian process regression model, by dividing the input space into sub intervals, modeling each sub interval independently, and then performing boundary weighted fusion;

[0106] Regularized neighborhood component analysis model, by minimizing the loss function containing L2 regularization term to learn feature weight, define feature weighted distance measure;

[0107] Artificial neural network model based on Levenberg-Marquardt algorithm;

[0108] More specifically, S2, based on four kinds of machine learning regression algorithm, nonlinear regression modeling of geopolymer performance, building fusion four machine learning model of geopolymer performance prediction model, establish the nonlinear mapping relationship between raw material ratio and performance, including:

[0109] S21: Constructing support vector regression model based on Rational-Minkowski kernel function:

[0110] Support vector regression is a supervised learning method for solving regression problems, which is particularly suitable for small sample, high dimension and nonlinear problems. In the modeling scenario between the proportion of geopolymer and the target performance, support vector regression can effectively capture the nonlinear mapping relationship between input and output. Support vector regression is a kernel-based machine learning algorithm that solves regression problems by constructing a hyperplane. The key is the kernel function To further improve the fitting ability of the model to complex nonlinear relationships, this paper introduces the Rational-Minkowski kernel function to construct an improved kernel function support vector regression model to improve the prediction accuracy. The expression of Rational-Minkowski kernel function is:

[0111]

[0112] : sample input vector, n is the number of samples; represents the norm of Minkowski distance, p∈[1,3], used to adjust the norm order of Minkowski distance; τ is the scale parameter and τ>0, used to control the width of the kernel function; q is the exponential adjustment parameter and q∈[1,5], used to control the decay rate of the kernel function; when and , the kernel function degenerates into the common RationalQuadratic kernel.

[0113] The Rational-Minkowski kernel function has the characteristics of stronger punishment for distant samples; maintaining a larger kernel value between similar samples is beneficial to capturing local structure; and having good adaptability to non-uniformly distributed data. By combining the adjustable norm of Minkowski distance (p∈[1,3]) with the exponential decay factor of Rational function (q∈[1,5]), its generalization error is significantly lower than that of RBF kernel.

[0114] To ensure the generalization ability and accuracy of the support vector regression model in the performance prediction of geopolymer, the grid search combined with cross-validation technology is used to optimize parameters; and multiple training / test sets are divided (such as ), repeated experiments are conducted to evaluate the stability and robustness of the model.

[0115] The support vector regression model based on the Rational-Minkowski kernel function is introduced in step S21, a composite kernel function form, i.e., the combination of Rational function and Minkowski distance, is introduced to enhance the nonlinear modeling ability of the support vector regression model in processing high-dimensional sparse geopolymer ratio data. Compared with the traditional radial basis kernel function, the kernel function has stronger robustness in fitting complex function relationship and avoiding overfitting, and significantly improves the adaptability and generalization ability of the model in the performance prediction of geopolymer.

[0116] S22: Constructing a segmented Gaussian process regression model:

[0117] Gaussian process regression is a non-parametric Bayesian regression method, which can directly output the prediction value and its uncertainty during modeling. For the highly nonlinear and obvious local fluctuation relationship between geopolymer ratio and performance, the segmented modeling strategy is further adopted to construct a segmented Gaussian process regression model to improve the overall fitting accuracy and model interpretability. The specific process of modeling the segmented Gaussian process regression model includes:

[0118] S221: Input space division:

[0119] Based on principal component analysis and K-means clustering or expert knowledge rules, K-means clustering (elbow rule determines M=4) is performed after PCA dimensionality reduction to 3 dimensions, and the sub-interval boundary is divided by Voronoi diagram, so that the geopolymer ratio space is divided into sub-intervals ; the data in each sub-interval satisfies the local stability and structural consistency;

[0120] S222: Local model training:

[0121] For each sub-interval , an independent Gaussian process regression model is trained; the optimal kernel function (such as square exponential kernel function, Matérn3 / 2 kernel function, RationalQuadratic kernel function) and hyperparameters can be selected for each sub-model;

[0122] S223: Segmented prediction integration:

[0123] For a test sample , first determine which sub-interval it belongs to, and then use the corresponding sub-model for prediction; if is in the boundary area, the prediction outputs of the two sub-models are fused by weighted averaging:

[0124] ​

[0125]

[0126] in, : Predicted mean, representing the mean given input Conditional expected values ​​of geopolymer performance indices under given conditions; : Prediction variance, representing the variance of the predicted mean A quantitative measure of uncertainty or confidence level; : Test input sample; : The total number of sub-intervals or sub-models (usually 2, when Located at the boundary of two sub-intervals); : No. Sub-model exist The predicted mean at the location; : No. Sub-models in The prediction variance at the location (representing uncertainty);

[0127] : No. The spatial weights of each sub-model satisfy the normalization constraint:

[0128]

[0129] Usually according to The inverse distance function to the center (or boundary) of each subinterval is defined, for example:

[0130]

[0131] in, express With the Center of each sub-interval Euclidean distance.

[0132] Sub-models Estimating hyperparameters by maximizing the logarithmic marginal likelihood function:

[0133]

[0134] in, Training output vector (i.e., target variable, such as compressive strength); The training input sample matrix (i.e., geopolymer ratio parameters) has a total of There are samples, each sample has One input feature; The covariance matrix is ​​defined as follows: , by kernel function construct; : variance of observation noise, used to characterize the unexplained noise in data (also called white noise); : identity matrix, dimension , used to introduce noise; : determinant of covariance matrix after adding noise, used to characterize the model complexity; : log marginal likelihood, is the objective function of hyperparameter optimization, using L-BFGS algorithm to maximize the log marginal likelihood, the iteration tolerance is set to 1e-6, maximizing it can get the optimal hyperparameters (such as parameters in kernel function, etc.); : data fitting term, measures the consistency between prediction and true ; : complexity penalty term, used to punish the model that is too complex; : normalization constant term, related to the number of samples .

[0135] Piecewise Gaussian process regression can give the uncertainty of each prediction point , which helps risk control and significantly improves model accuracy and computational efficiency when sample size is large or local mode difference is obvious.

[0136] The piecewise Gaussian process regression model proposed by the above step S22 designs a Gaussian process regression framework for piecewise modeling for non-stationary regions and local abrupt points in the performance curve of geopolymer materials. By fitting sub-models in different intervals and smoothing the boundaries of each segment model, the model has both local accuracy and global consistency, and is particularly suitable for prediction tasks where performance indicators have interval transitions or mutations.

[0137] S23: Construct a regularized neighborhood component analysis model:

[0138] Regularized neighborhood component analysis is an improvement and generalization of neighborhood component analysis in regression problems, aiming to achieve high-dimensional feature selection through distance metric learning. Compared with traditional neighborhood component analysis mainly used for classification tasks, regularized neighborhood component analysis introduces a regularization term on the basis of retaining its nearest neighbor idea, which not only effectively suppresses model overfitting, but also improves the prediction performance on continuous target variables (such as geopolymer compressive strength).

[0139] The core idea of regularized neighborhood component analysis is to learn a set of feature weights in all prediction variables, construct a weighted distance metric, so that the output values of similar sample pairs are similar in the new feature space, thereby realizing feature selection and dimension reduction. The specific construction process of the regularized neighborhood component analysis model includes the following steps:

[0140] S231: Definition of Feature-Weighted Distance:

[0141] For the input sample set, construct a weighted Manhattan distance metric function. like:

[0142]

[0143] in, : No. The input vector of each sample; : No. In the nth sample The values ​​of each feature; : No. Weight coefficients of each feature (to be learned); distance metric It is a weighted Manhattan distance form (which can be extended to the Euclidean norm).

[0144] S232: Neighborhood probability calculation:

[0145] By learning appropriate Regularized neighborhood component analysis can compress the weights of irrelevant or redundant features, minimizing their impact on the final distance metric, thereby achieving feature selection. To construct the "proximity" relationship between samples based on this distance, samples are defined... choose The probability of being a valid neighbor:

[0146]

[0147] : Indicates a sample Will The higher the probability value, the better the probability of it being a "valid neighbor". The more likely to be The predictions have an effect; :sample In the Values ​​can be taken in each feature dimension; :sample In the Values ​​can be taken in each feature dimension; Total number of feature dimensions; : No. The learnable weight parameters of each feature are used to measure the importance of that feature in neighborhood determination. Ensure that the weights are non-negative, while enhancing the sensitivity to important features; : No. The smoothing scale parameter is used to normalize the distance values ​​of different feature dimensions, so that features with different dimensions or degrees of variability have a relatively consistent influence in the overall distance metric. Adaptive computation according to characteristic variance: .

[0148] This mechanism ensures that the model can learn a more optimal feature combination within the "neighborhood" to improve overall prediction performance.

[0149] S233: Model training:

[0150] To train the model, the regularized neighborhood component analysis minimizes the following loss function with a regularization term:

[0151]

[0152] where, is the true output; represents the predicted value based on neighbor probability weighting; is a regularization parameter used to penalize excessive feature weights and prevent the model from being too complex, λ is selected from {0.01, 0.1, 1} through cross-validation; is the mean absolute error; is the regularization term.

[0153] This objective function takes into account both prediction accuracy and model simplicity, learning feature weights by minimizing the loss function containing an L2 regularization term, automatically adjusting feature weights , ignoring variables that do not have representative variables, thereby effectively alleviating the influence of multicollinearity.

[0154] S234: Feature selection:

[0155] By optimizing the weight to compress the influence of redundant features, high-dimensional feature selection and dimensionality reduction are achieved.

[0156] In the optimization problem of geopolymer proportioning, there is often a high degree of correlation between features, and direct modeling is prone to redundant interference. Regularized neighborhood component analysis has the following advantages: automatically determines variable importance through feature weight learning without pre-selection; insensitive to multicollinearity, especially suitable for modeling of physical and chemical property-related indicators.

[0157] The regularized neighborhood component analysis model proposed in step S23 is used to further extract the key feature dimensions between raw material proportioning and performance response. The regularized neighborhood component analysis method is introduced to maintain local structure while avoiding dimension redundancy and overfitting problems. This model can significantly improve data separability and information retention rate after dimensionality reduction, providing a more effective feature input space for subsequent prediction models.

[0158] S24: Constructing an artificial neural network model based on the Levenberg-Marquardt algorithm:

[0159] The Levenberg-Marquardt algorithm is used for training, which is efficient when dealing with networks containing hundreds of weights and biases.

[0160] The artificial neural network model based on the Levenberg-Marquardt algorithm satisfies the following conditions:

[0161] Network structure: A typical three-layer feedforward neural network structure is adopted, including an input layer, a hidden layer, and an output layer. The input layer receives multiple gangue proportioning parameters, including coal gangue content, fly ash content, slag content, water glass modulus, Na2O dosage, or NaOH solution concentration, such as: : coal gangue content (%); : fly ash content (%); : slag content (%); : water glass modulus; : dosage (%); The hidden layer uses an adjustable number of neurons (such as ) to mine the complex mapping relationship between features. Specifically, the number of hidden layer neurons can be determined through Bayesian optimization, with the target function being the 5-fold cross-validation error and the search space being [4, 12] integers; the output layer outputs the target performance parameters (such as compressive strength ).

[0162] Weight update: The weights are dynamically adjusted through the Levenberg-Marquardt algorithm, and the update formula is:

[0163]

[0164] where is the weight vector of the rth iteration; is the Jacobian matrix, which contains the partial derivatives of the network output with respect to the weights; is the error vector; : adjustment parameter (initial value is large, gradually decreases with iteration); : identity matrix; : inverse operation; : transpose operation.

[0165] Training mechanism: Levenberg-Marquardt combines two extreme cases:

[0166] When , the algorithm approaches the Gauss-Newton method (fast convergence speed);

[0167] When is large, the algorithm behaves as a gradient descent method (strong stability);

[0168] Continuous adjustment during training , ensure convergence speed and stability.

[0169] Early stopping mechanism: to prevent model overfitting, the model introduces early stopping mechanism to ensure stability and reliability in practical application. Specifically, in the training process, the validation set error is monitored, and when the error does not decrease for a predetermined number of times in succession, the training is terminated to prevent overfitting.

[0170] The artificial neural network weight updating strategy based on Levenberg-Marquardt algorithm proposed by step S24. In view of the problems of slow convergence and easy to fall into local extremum of traditional BP neural network in complex proportion data learning, Levenberg-Marquardt optimization algorithm is adopted to realize network weight updating. The algorithm combines the advantages of gradient descent and Gauss-Newton method, which improves the convergence speed while ensuring global stability, significantly enhances the expression ability of neural network model to complex nonlinear mapping relationship, and is suitable for the prediction modeling task of geopolymer multi-objective performance.

[0171] The four machine learning models established by the application run cooperatively: the Rational-Minkowski kernel SVR captures local nonlinear relationship, the segmented Gaussian process regression model processes input space heterogeneity, the regularized neighborhood component analysis model suppresses high-dimensional noise, and the Levenberg-Marquardt algorithm accelerates ANN convergence, forming a bias-variance balanced fusion framework.

[0172] S3, calculate and distribute weights according to the mean square error of each machine learning model on the validation set of the data set, and obtain the performance prediction result by weighted fusion;

[0173] In actual regression problems, different regression models may show different prediction ability and error characteristics on different input features and data sets. In order to fully utilize the advantages of each model and avoid the limitations of single model, the prediction results of multiple models can be integrated through multi-model weighted fusion, so as to improve the overall prediction performance, reduce the overfitting phenomenon and improve the stability and robustness of the model. In the present application, four different regression models of steps S21 to S24 are used for prediction, and the core idea of weighted fusion is to assign a suitable weight to each regression model, and then weighted average the model output results.

[0174] To ensure that the contribution of each model can be adaptively adjusted according to its prediction performance, the weight allocation strategy can be dynamically adjusted by the method of error back propagation. Specifically, the weights of the models should be adjusted according to the prediction error of each model on the validation set, and the model with smaller error should obtain larger weight.

[0175] Further, in this step S3, the weight calculation includes:

[0176] S31: Calculate the mean square error of each machine learning model on the validation set:

[0177]

[0178] wherein, is the mean square error of the i-th machine learning model; is the prediction value of the i-th model for the i-th sample; is the true value of the i-th sample; S32: Calculate the weight of the i-th machine learning model according to the inverse of the mean square error

[0179] :

[0180]

[0181] wherein, is the weight of the i-th machine learning model; S33: Weighted average the prediction results of the four machine learning models according to the weights

[0182] of the four machine learning models, and output the final performance prediction value. Through weighted fusion, the prediction results of each model will be effectively combined according to its performance on the validation set. Assuming that the error of a certain model on the validation set is smaller, then the prediction result of this model will have a greater impact on the final fusion output. Through multi-model weighted fusion, the prediction accuracy can be improved, especially in the case of complementarity between different models, the fusion effect is particularly significant.

[0183] For example, in the prediction of compressive strength of geopolymer, some regression models may have higher prediction accuracy in the low proportion range, while other models have better prediction performance in the high proportion range. Through weighted fusion, the advantages of each model can be integrated, and the performance fluctuation of a single model due to changes in data characteristics can be avoided.

[0184]

[0185] ​​​S4, receiving the target performance parameters input by the user and the initial raw material mixing range, performing performance prediction by using the trained geopolymer performance prediction model, and searching for the optimal mixing combination satisfying the target performance constraint by using an optimization algorithm;

[0186] Further, in step S4: the target performance parameters set by the user include compressive strength > 80 MPa, permeability coefficient < 1 x 10 -7 cm / s, and durability indicators including the number of cycles of resistance to sulfate corrosion and the mass loss rate under dry-wet cycles;

[0187] The initial raw material mixing range includes: coal gangue content 60-80 wt%, slag 10-20 wt%, fly ash 10-20 wt%; water glass modulus 1.5-2.5, Na2O content 3-22 wt%; and the mass ratio of activator to solid raw material 0.3-0.5.

[0188] The optimal mixing combination satisfying the target performance constraint is searched by using a reverse reasoning or an optimization algorithm (such as particle swarm optimization or Bayesian optimization).

[0189] S5, preparing the coal gangue-slag-fly ash geopolymer grouting material according to the optimal mixing combination.

[0190] In the prior art, coal gangue is difficult to be directly used as the main raw material of geopolymer due to uneven silicon and aluminum content and insufficient activity. The present application proposes a preparation method for preparing the geopolymer grouting material by taking coal gangue as the main raw material and combining slag and fly ash, realizes efficient resource utilization of coal gangue, and provides high-performance materials for ecological site impervious layers. In some embodiments, step S5 specifically includes:

[0191] Raw material preparation: typical coal gangue in Guizhou Province is selected, crushed and ball milled to a specific surface area of 400-600 m² / kg; slag and fly ash are commercially available products with a particle size controlled at 0.045-0.075 mm.

[0192] Preparation of activator: according to the recommended optimal mixing combination, water glass and Na2O or NaOH solution are mixed in a certain proportion to prepare a composite activator.

[0193] Mixing and stirring: the coal gangue, slag and fly ash are mixed, the activator is added, and the slurry body is stirred to a flow degree of 150-200 mm to form a coal gangue-slag-fly ash geopolymer grouting slurry body.

[0194] Mixing and stirring: the coal gangue, slag and fly ash are mixed, the activator is added, and the slurry body is stirred to a flow degree of 150-200 mm to form a coal gangue-slag-fly ash geopolymer grouting slurry body.

[0195] Grouting maintenance: after the coal gangue-slag-fly ash geopolymer grouting slurry is pre-cured at room temperature for 24 hours, it is transferred to an environment with a temperature of 20±2 DEG C and a humidity of greater than or equal to 95% for curing for 28 days, thereby forming the coal gangue-slag-fly ash geopolymer grouting material.

[0196] Performance verification and feedback: a sample is prepared according to the recommended optimal proportion combination, and the compressive strength, permeability coefficient and durability index are measured, and whether the performance requirements of the compressive strength being greater than 80 MPa, the permeability coefficient being less than 1x10 -7 cm / s, etc. are met; if the measured value deviates from the prediction by more than 10%, the model parameter iteration update is triggered.

[0197] According to the above embodiments of the present application, by establishing a nonlinear regression model, the complex nonlinear relationship between different proportions (including coal gangue, fly ash, slag, water glass modulus, Na2O content, etc.) and the performance of geopolymer (compressive strength, permeability, durability) is modeled, and intelligent recommendation of raw material proportioning is realized through model fusion, which significantly improves the comprehensive performance of geopolymer.

[0198] Figure 2 A module schematic diagram of a geopolymer preparation optimization system based on machine learning is shown. In some embodiments of the present application, in order to realize the rapid application and intelligent proportioning optimization of geopolymer materials in engineering practice, based on the above multi-model fusion prediction mechanism, a geopolymer preparation optimization system 200 based on machine learning is provided. The system 200 can automatically recommend the corresponding raw material proportioning combination according to the target performance requirements input by the user, and provide real-time feedback and performance prediction, which significantly improves the material design efficiency and intelligent level. As shown in Figure 2 the geopolymer preparation optimization system 200 based on machine learning includes:

[0199] The data acquisition module 210 is used for acquiring geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and performing missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data;

[0200] The input received by the system through the user interaction interface includes two types: target performance parameters (user input): compressive strength (MPa), corrosion resistance performance (optional indicators such as salt spray test results, mass loss rate, etc.), impermeability, early strength, etc. (extensible support). Proportioning range constraints (user adjustable): upper / lower limit ranges of fly ash content, slag ratio, activator type and concentration, water-cement ratio, etc. material parameters.

[0201] The model construction module 220 is configured to construct a geopolymer performance prediction model by fusing four machine learning models, and to establish a nonlinear mapping relationship between raw material ratios and performance, including:

[0202] a support vector regression model based on a Rational-Minkowski kernel function;

[0203] a segmented Gaussian process regression model, which divides the input space into subintervals, independently models each subinterval, and then performs boundary weighted fusion;

[0204] a regularized neighborhood component analysis model, which learns feature weights by minimizing a loss function containing an L2 regularization term, and defines a feature weighted distance metric;

[0205] an artificial neural network model based on a Levenberg-Marquardt algorithm;

[0206] The prediction output module 230 is configured to calculate and assign weights according to the mean square errors of each machine learning model on the validation set in the data set, and to obtain a performance prediction result by weighted fusion.

[0207] The ratio recommendation module 240 is configured to receive target performance parameters and initial raw material ratio ranges input by a user, to perform performance prediction using the trained geopolymer performance prediction model, and to search for an optimal ratio combination that satisfies target performance constraints through an optimization algorithm.

[0208] The ratio recommendation module 240 is driven by performance, and realizes the mapping from “target performance” to “ratio suggestion” through the following steps: the user inputs a performance target, such as a compressive strength > 80 MPa. Based on the fusion model, multiple sets of ratio initial values that satisfy the performance requirement are searched (which can be generated by an optimization algorithm). The fusion model predicts the performance indicators of these ratio combinations, and selects the optimal solution or Pareto optimal solution (if multi-objective optimization is involved). The system outputs the recommended raw material ratio combination (including the proportion and amount of each material), the predicted performance indicators and their confidence ranges, and multiple sets of alternative solutions for the user to select or further adjust. The results are presented in the form of charts and tables for the user to further screen or adjust the ratio parameters.

[0209] The grouting preparation module 250 is configured to prepare a geopolymer preparation coal gangue-slag-fly ash geopolymer grouting material according to the optimal ratio combination.

[0210] According to the above system, the user interaction interface mainly includes the following functional sub-modules: input area: for inputting target performance values, acceptable error range, raw material types and proportion constraints, etc. Recommendation area: display recommended proportioning schemes in list form, support export and save functions. Performance prediction graph: show the trend of each performance parameter under different proportions, support two-dimensional / three-dimensional visualization. Interactive adjustment area: users can drag the slider to adjust each material parameter, and the system automatically updates the prediction results. The interface can be implemented through Web technology (such as Flask+React) or desktop software (such as PyQt), with good scalability and compatibility, suitable for laboratory formula design and engineering application scenarios. The system supports user manual modification of proportioning parameters, and the system updates and re-predicts the corresponding performance indicators in real time, realizing human-computer interaction closed loop.

[0211] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described modules can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0212] Further, the embodiments of the present application also provide an electronic device, comprising: a processor, a memory, a system bus; the processor and the memory are connected through the system bus; the memory is used to store one or more programs, the one or more programs include instructions, the instructions make the processor execute any one of the above methods when executed by the processor.

[0213] Further, the embodiments of the present application also provide a computer program product, which, when running on a terminal device, causes the terminal device to execute any one of the above methods.

[0214] From the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the above embodiment methods can be implemented by means of software plus the necessary general hardware platforms. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product in essence or in the form of a part of the prior art that makes a contribution. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the method described in each embodiment or some part of the embodiments of the present application.

[0215] It should be noted that the various embodiments described in the specification can be described with reference to a specific sequence or order. Unless otherwise specified, however, the disclosure includes each and every possible combination of the various embodiments described herein. The order or sequence of any process or method can be changed or re-sequenced without departing from the general nature of the application. Unless otherwise specified, the disclosure encompasses each individual embodiment well as mixtures thereof.

[0216] It should also be noted that, in the specification, relational terms such as first and second, and the like can be used solely to distinguish one from another entity or action without necessarily implying a required order of such entities or actions. Moreover, the terms "comprising," "including," containing," or any other similar reference used herein are intended to be equivalent to the term "consisting of" in that any process, method, composition, or apparatus that includes a series of elements is not necessarily limited to those elements but can include other elements not expressly listed or inherent to such process, method, composition, or apparatus. In the specification, the use of the term "about" in relation to a numerical value means that the value is approximately correct and can vary by up to ±5% of the stated value.

[0217] The above description discloses only a few embodiments of the disclosure and the full scope of the disclosure will be pointed out by the appended claims. Those skilled in the art will understand that the embodiments described on the specification illustrate different specific implementations of the technical solutions described in the specification. It will be apparent to those skilled in the art that variations and modifications of the present application can be made without departing from the spirit or scope of the application. Accordingly, it is the intent that all such variations and modifications be considered as within the scope of the application.

Claims

1. A machine learning based geopolymer production optimization method, characterized by, The method comprises the following steps: S1, collecting geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and performing missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data; S2, based on four machine learning regression algorithms, a nonlinear regression model for geopolymer performance is established, a geopolymer performance prediction model is constructed by fusing four machine learning models, and a nonlinear mapping relationship between raw material ratio and performance is established, including: According to the Rational-Minkowski kernel function, an improved kernel function support vector regression model is constructed, wherein the expression of the Rational-Minkowski kernel function is: ； in, Let n be the sample input vector, and n be the number of samples. Minkowski distance The order norm, p∈[1,3], is used to adjust the norm order of the Minkowski distance; τ is the scaling parameter and τ>0, used to control the width of the kernel function; q is the exponential adjustment parameter and q∈[1,5], used to control the decay rate of the kernel function; A segmented Gaussian process regression model is constructed, including: Input space division: based on principal component analysis and K-means clustering or expert knowledge rules, the geopolymer ratio input space is divided into M subintervals, each subinterval satisfies local data stability; Local model training: train a Gaussian process regression submodel for each subinterval, and select a kernel function type and hyperparameters for the submodel; the kernel function includes a square exponential kernel function, a Matérn 3 / 2 kernel function or a RationalQuadratic kernel function; Segmented prediction integration: for a test sample, determine the subinterval to which the test sample belongs, and use the corresponding submodel to make a prediction; if the test sample is located in a boundary region, the prediction results of adjacent submodels are fused by weighted averaging, including: ； ； wherein, : predicted mean representing the conditional expectation estimate of the polymer property index given the input ; : predicted variance representing a quantitative measure of uncertainty or confidence in the predicted mean ; is a test input sample; is the total number of sub-intervals or sub-models; represents the predicted mean of the th sub-model at ; represents the predicted mean of the th sub-model at ; is the predicted variance of the th sub-model, satisfying the normalization constraint: ； Weights According to Inverse function of the distance to the center of each sub-interval ； wherein, represents the Euclidean distance between the test input sample and the center of the th sub-interval . Hyperparameter optimization: estimate the hyperparameters of each submodel by maximizing the log marginal likelihood function; Regularized neighborhood component analysis model, learn feature weights by minimizing a loss function containing an L2 regularization term, and define a feature weighted distance measure; Artificial neural network model based on Levenberg-Marquardt algorithm; S3, calculate and assign weights according to the mean square errors of each machine learning model on the validation set in the data set, and obtain the performance prediction result by weighted fusion; S4, receiving the target performance parameters and the initial raw material ratio range input by the user, using the trained geopolymer performance prediction model to perform performance prediction, and searching the best ratio combination satisfying the target performance constraint through an optimization algorithm; S5, according to the best ratio combination, preparing a coal gangue-slag-fly ash geopolymer grouting material.

2. The machine learning based geopolymer production optimization method of claim 1, wherein, Wherein, The estimation of the hyperparameters of each submodel by maximizing the log marginal likelihood function includes: ； wherein, : training output value vector, i.e. target variable, including compressive strength; : training input sample matrix, i.e. geopolymer formulation parameters, total samples, each sample has input features; : covariance matrix, defined as constructed by kernel function ; : variance of observation noise, used to characterize the unexplained noise in the data; : identity matrix, dimension , used to introduce noise.

3. The machine learning based geopolymer production optimization method of claim 2, wherein, Wherein, The construction of the regularized neighborhood component analysis model includes the following steps: Feature weighted distance definition: construct a weighted Manhattan distance measure function: ； wherein, : input vector of the th sample; : value of the th feature in the th sample; : weight coefficient of the th feature; distance metric is a weighted Manhattan distance form; Neighborhood probability computation: defining samples Selecting Probability of being an effective neighbor: ； : Indicates a sample Will The higher the probability value, the better the probability of it being a valid neighbor. The more likely to be The predictions have an effect; : No. The smoothing scale parameter for each feature is used to normalize the distance values ​​of different feature dimensions; Model training: learn feature weights by minimizing a loss function containing an L2 regularization term; Feature selection: compress the influence of redundant features through the optimized weights, realize high-dimensional feature selection and dimension reduction.

4. The machine learning based geopolymer production optimization method of claim 3, wherein, Wherein, The artificial neural network model based on the Levenberg-Marquardt algorithm satisfies the following conditions: Network structure: a three-layer feedforward neural network is adopted, the input layer receives the geopolymer ratio parameters, including coal gangue content, fly ash content, slag content, water glass modulus, Na2O dosage, the number of hidden layer neurons is 2-15, and the output layer predicts the target performance index; Weight update formula: the weights are dynamically adjusted through the Levenberg-Marquardt algorithm, and the update formula is: ； weight vector for the rth iteration; Jacobian matrix containing the partial derivatives of the network output with respect to the weights; error vector; adjustment parameter; identity matrix; Training mechanism: when → 0, the algorithm approximates the Gauss-Newton method to accelerate convergence; when increases, the algorithm degenerates into the gradient descent method to enhance stability; Early stopping mechanism: the validation set error is monitored during the training process, and the training is terminated when the error does not decrease for a preset number of consecutive times to prevent overfitting.

5. The machine learning based geopolymer production optimization method of claim 4, wherein, Wherein, In step S3, the weight calculation includes: Calculate the mean square error of each machine learning model on the validation set: ； in, For the first The mean squared error of a machine learning model; It is the first The predicted value of each model for the i-th sample; It is the true value of the i-th sample; According to the inverse of the mean square error of the weight : ； wherein, is a weight of the th machine learning model. weighting according to the weights of the four machine learning models performing a weighted average of the prediction results of the four machine learning models, outputting a final performance prediction value.

6. The machine learning based geopolymer production optimization method of claim 4, wherein, Wherein, In step S4: wherein the target performance parameters set by the user include compressive strength > 80 MPa, permeability coefficient < 1 x 10 -7 cm / s, and durability indicators including the number of cycles of resistance to sulfate corrosion, mass loss rate of dry-wet cycle; The initial raw material ratio range includes: coal gangue dosage 60-80 wt%, slag 10-20 wt%, fly ash 10-20 wt%; water glass modulus 1.5-2.5, Na2O dosage 3-22 wt%; activator to solid raw material mass ratio 0.3-0.5; S5, according to the optimal ratio combination, the coal gangue-slag-fly ash geopolymer grouting material is prepared, which includes: The coal gangue is crushed and ball milled to a specific surface area of 400-600 m² / kg, and the particle size of the slag and fly ash is controlled to be 0.045-0.075 mm; According to the recommended optimal ratio combination, the coal gangue, slag and fly ash are mixed, the activator is added, and the slurry is stirred to a flow degree of 150-200 mm to form a coal gangue-slag-fly ash geopolymer grouting slurry; Grouting and curing: after the coal gangue-slag-fly ash geopolymer grouting slurry is pre-cured at room temperature for 24 hours, it is transferred to a 20±2℃, humidity≥95% environment for 28 days, forming a coal gangue-slag-fly ash geopolymer grouting material.

7. A machine learning based geopolymer production optimization system for implementing the machine learning based geopolymer production optimization method of any one of claims 1-6, characterized in that, Including: A data acquisition module for acquiring geopolymer preparation experimental data, including solid raw material parameters, activator parameters and performance output indicators, and performing missing value filling, outlier removal, feature scaling, standardization, encoding and data set division on the experimental data; A model construction module for nonlinear regression modeling of geopolymer performance based on four machine learning regression algorithms, constructing a geopolymer performance prediction model integrating four machine learning models, establishing a nonlinear mapping relationship between raw material ratio and performance, including: A support vector regression model based on a Rational-Minkowski kernel function; A piecewise Gaussian process regression model that divides the input space into subintervals, independently models each subinterval, and then performs boundary weighted fusion; A regularized neighborhood component analysis model that learns feature weights by minimizing a loss function containing an L2 regularization term and defines a feature weighted distance measure; An artificial neural network model based on the Levenberg-Marquardt algorithm; A prediction output module for calculating and assigning weights according to the mean square error of each machine learning model on the validation set in the data set, and obtaining a performance prediction result through weighted fusion; The ratio recommendation module is configured to receive a target performance parameter input by a user and an initial raw material ratio range, perform performance prediction by using a trained geopolymer performance prediction model, and search for an optimal ratio combination satisfying a target performance constraint by using a reverse search optimization algorithm. The grouting preparation module is configured to prepare a coal gangue-slag-fly ash geopolymer grouting material according to the optimal ratio combination.

Images