Gangue filling amount calculation method based on machine learning

Abstract

The invention discloses a gangue filling amount calculation method based on machine learning, and the method comprises the following steps: 1, measuring a gangue sample, and obtaining a gangue original data set; 2, according to the original gangue data set, preparing a gangue sample and carrying out a compression test to obtain test data; step 3, according to the test data, constructing a gangue filling amount prediction data set; step 4, establishing a machine learning model, and performing optimization according to the filling amount prediction data set to obtain an optimal model; and 5, giving actual design parameters of the goaf, inputting the actual design parameters into the optimal model, and calculating to obtain the gangue filling amount. According to the gangue filling amount calculation method based on machine learning, the problems that in the prior art, the estimation precision of the gangue filling amount is low, and the calculation process depends on empirical parameters are solved.

Description

Technical Field

[0001] This invention belongs to the field of mining engineering and solid backfilling technology, specifically relating to a machine learning-based method for calculating gangue backfilling volume. Background Technology

[0002] Solid backfilling in coal mining is an important technical means to control roof settlement and improve resource recovery rate, using solid waste materials such as gangue to fill the goaf. In engineering practice, the backfilling effect is affected by many factors, including gangue gradation (coarse and fine aggregate ratio), gangue physical properties, and on-site compaction construction parameters. Accurately calculating the actual compaction density after backfilling is the foundation for controlling the backfilling volume, transportation scheduling, and optimizing material proportions. Currently, commonly used gangue backfilling volume calculations are mostly based on empirical coefficients or simple volume-mass conversions, usually assuming a fixed compaction density or estimating the required material mass only based on the overall bulk density, lacking accurate calculation of the compaction density change under the combined effect of admixture ratio and construction compaction strength. Existing methods have the following shortcomings: ① Most methods do not include the coarse and fine aggregate ratio and compaction strength as joint influencing factors in density prediction, resulting in large estimation deviations of compaction density under different admixture and construction conditions; ② Relying solely on statistical experience or single test points is difficult to cover the discreteness and nonlinear coupling relationship of field parameters, making it difficult to meet the needs of rapid engineering calculations and online decision-making. Summary of the Invention

[0003] The purpose of this invention is to provide a machine learning-based method for calculating gangue backfill volume, which solves the problems of low accuracy in gangue backfill volume estimation and reliance on empirical parameters in the existing technology.

[0004] The technical solution adopted in this invention is a machine learning-based method for calculating gangue backfill volume, comprising the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0005] The invention is further characterized by: Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. rStrengthen the strength P}

[0006] Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes.

[0007] In step 2.1, the doping ratio α This refers to the percentage of coarse aggregate by weight. α The value range is 0 to 1.

[0008] In step 2.1, the loading adopts a graded loading method, with a loading pressure range of 0.5 to 10 MPa. Each loading stage is kept stable for 5 minutes, and the corresponding volume change is recorded for stress-strain curves.

[0009] Step 3 specifically involves: adjusting the compaction strength. P Gangue mixing ratio α True density of gangue r As the input variable X, X=[ P， a,r The compaction density obtained from the experiment r c As the output variable Y, a dataset D for predicting gangue filling volume is constructed, D={X,Y}, and the input variables are normalized so that variables of different dimensions are mapped to the interval [0,1].

[0010] Step 4 is as follows: Step 4.1: Establish a compaction density prediction model using the gradient boosting tree algorithm. The model structure is shown in equation (3): (3); in, fThe mapping relationship of the nonlinear regression function learned by the machine learning algorithm; Step 4.2: Train the compaction density prediction model using training samples. The objective function is shown in equation (4). (4); in, These are the model's predicted values. These are experimental values. This is a regularization term for model complexity. This is the penalty coefficient; Step 4.3: Optimize and adjust model parameters, including learning rate, tree depth, subsample ratio, and regularization parameters, through cross-validation and Bayesian algorithm to obtain the optimal model.

[0011] In step 4.1, the machine learning algorithm selected is one of CatBoost, XGBoost, Random Forest, and Deep Neural Network models.

[0012] In step 4.2, the performance evaluation of the optimal model prediction results uses the coefficient of determination R. 2 Quantitative analysis was performed on the mean square error (MSE) and mean absolute error (MAE) when R 2 A value of ≥0.9 is considered to indicate that the model training has reached the accuracy required for engineering applications.

[0013] Step 5 specifically involves: Step 5.1: Specify the design parameters for the goaf; Step 5.2: Substitute the design parameters into the optimal model to obtain the predicted compaction density. r c ' The required amount of gangue filling for the goaf is calculated according to formula (5). M' ; (5); in, or The filling rate; P' To solidify the strength; α' The ratio of coarse to fine aggregates; r' The true density of the gangue; V This represents the volume of the goaf.

[0014] The beneficial effects of this invention are: This invention provides a machine learning-based method for calculating gangue backfill volume. Through machine learning algorithms, it achieves accurate prediction of the compaction density of gangue backfill materials, significantly improving prediction accuracy compared to traditional empirical formulas or single-factor regression models. This method effectively characterizes the nonlinear influence between factors such as gangue particle characteristics, blending ratio, and compaction strength, exhibiting strong universality and engineering adaptability. In engineering applications, only relevant parameters need to be input to quickly obtain the calculation results of compaction density and backfill volume, significantly improving computational efficiency and reducing manual intervention and empirical errors. Furthermore, this method can be embedded into intelligent mine backfilling systems to achieve real-time prediction and dynamic optimization of backfill status, providing a scientific basis for backfill volume control and design decisions in goaf areas, and further promoting the digital and intelligent development of solid backfill mining. Attached Figure Description

[0015] Figure 1 This is a flowchart of the machine learning-based method for calculating gangue backfill volume according to the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0018] Example 2 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0019] Example 3 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0020] Example 4 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. In step 2.1, the doping ratio α This refers to the percentage of coarse aggregate by weight. α The value range is 0 to 1; In step 2.1, the loading adopts a graded loading method, with a loading pressure range of 0.5 to 10 MPa. Each loading stage is kept stable for 5 minutes, and the corresponding volume change is recorded for stress-strain curves. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0021] Example 5 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. In step 2.1, the doping ratio α This refers to the percentage of coarse aggregate by weight. α The value range is 0 to 1; In step 2.1, the loading adopts a graded loading method, with a loading pressure range of 0.5 to 10 MPa. Each loading stage is kept stable for 5 minutes, and the corresponding volume change is recorded for stress-strain curves. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 3 specifically involves: adjusting the compaction strength. P Gangue mixing ratio α True density of gangue r As the input variable X, X=[ P， a,r The compaction density obtained from the experiment r c As the output variable Y, a gangue backfilling volume prediction dataset D is constructed, D={X,Y}, and the input variables are normalized so that variables of different dimensions are mapped to the interval [0,1]. Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0022] Example 6 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. In step 2.1, the doping ratio α This refers to the percentage of coarse aggregate by weight. α The value range is 0 to 1; In step 2.1, the loading adopts a graded loading method, with a loading pressure range of 0.5 to 10 MPa. Each loading stage is kept stable for 5 minutes, and the corresponding volume change is recorded for stress-strain curves. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 3 specifically involves: adjusting the compaction strength. P Gangue mixing ratio α True density of gangue r As the input variable X, X=[ P， a,r The compaction density obtained from the experiment r c As the output variable Y, a gangue backfilling volume prediction dataset D is constructed, D={X,Y}, and the input variables are normalized so that variables of different dimensions are mapped to the interval [0,1]. Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 4 is as follows: Step 4.1: Establish a compaction density prediction model using the gradient boosting tree algorithm. The model structure is shown in equation (3): (3); in, f The mapping relationship of the nonlinear regression function learned by the machine learning algorithm; In step 4.1, the machine learning algorithm selected is one of CatBoost, XGBoost, Random Forest, and Deep Neural Network models; Step 4.2: Train the compaction density prediction model using training samples. The objective function is shown in equation (4). (4); in, These are the model's predicted values. These are experimental values. This is a regularization term for model complexity. This is the penalty coefficient; In step 4.2, the performance evaluation of the optimal model prediction results uses the coefficient of determination R. 2 Quantitative analysis was performed on the mean square error (MSE) and mean absolute error (MAE) when R 2 A value ≥0.9 is considered to indicate that the model training has reached the accuracy required for engineering applications. Step 4.3: Optimize and adjust model parameters, including learning rate, tree depth, subsample ratio, and regularization parameters, through cross-validation and Bayesian algorithm to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling.

[0023] Example 7 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Measure the gangue samples and obtain the raw gangue dataset; Step 1 specifically involves: measuring the physical parameters of gangue samples from the mining area to obtain the true density of the gangue. r The compaction strength of commonly used hydraulic filling supports under different compaction conditions was recorded. P The original dataset of gangue {true density} was obtained. r Strengthen the strength P}; Step 2: Based on the original gangue dataset, prepare gangue samples and conduct compression tests to obtain test data; Step 2 is as follows: Step 2.1: Select two typical particle sizes of gangue materials and mix them according to different ratios. α Multiple groups of mixed gangue samples were prepared and subjected to a preset compaction strength. P The stress-strain curves and volume changes were recorded step by step, and the volume after compression was measured. V 1. The corresponding compaction density is calculated according to equations (1) and (2). r c : (1); M = m 1+ m 2(2); in, m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate. M This refers to the total mass of the sample. In step 2.1, the doping ratio α This refers to the percentage of coarse aggregate by weight. α The value range is 0 to 1; In step 2.1, the loading adopts a graded loading method, with a loading pressure range of 0.5 to 10 MPa. Each loading stage is kept stable for 5 minutes, and the corresponding volume change is recorded for stress-strain curves. Step 2.2: Repeat the experiment with different mixing ratios and compaction strength combinations, and record the mixing ratios, compaction strengths and the resulting compaction density data for different particle sizes; Step 3: Based on the test data, construct a dataset for predicting the amount of gangue filling; Step 3 specifically involves: adjusting the compaction strength. P Gangue mixing ratio α True density of gangue r As the input variable X, X=[ P， a,r The compaction density obtained from the experiment r c As the output variable Y, a gangue backfilling volume prediction dataset D is constructed, D={X,Y}, and the input variables are normalized so that variables of different dimensions are mapped to the interval [0,1]. Step 4: Build a machine learning model and optimize it based on the filling volume prediction dataset to obtain the optimal model; Step 4 is as follows: Step 4.1: Establish a compaction density prediction model using the gradient boosting tree algorithm. The model structure is shown in equation (3): (3); in, f The mapping relationship of the nonlinear regression function learned by the machine learning algorithm; In step 4.1, the machine learning algorithm selected is one of CatBoost, XGBoost, Random Forest, and Deep Neural Network models; Step 4.2: Train the compaction density prediction model using training samples. The objective function is shown in equation (4). (4); in, These are the model's predicted values. These are experimental values. This is a regularization term for model complexity. This is the penalty coefficient; In step 4.2, the performance evaluation of the optimal model prediction results uses the coefficient of determination R. 2 Quantitative analysis was performed on the mean square error (MSE) and mean absolute error (MAE) when R 2 A value ≥0.9 is considered to indicate that the model training has reached the accuracy required for engineering applications. Step 4.3: Optimize and adjust model parameters, including learning rate, tree depth, subsample ratio, and regularization parameters, through cross-validation and Bayesian algorithm to obtain the optimal model; Step 5: Given the actual design parameters of the goaf, input them into the optimal model to calculate the amount of gangue filling; Step 5 specifically involves: Step 5.1: Specify the design parameters for the goaf; Step 5.2: Substitute the design parameters into the optimal model to obtain the predicted compaction density. r c ' The required amount of gangue filling for the goaf is calculated according to formula (5). M' ; (5); in, or The filling rate; P' To solidify the strength; α' The ratio of coarse to fine aggregates; r' The true density of the gangue; V This represents the volume of the goaf.

[0024] Example 8 The machine learning-based method for calculating gangue backfill volume proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Data collection and sample testing; A sample of gangue used for backfilling in a mining area in northern Shaanxi was selected, and its true density ρ was measured to be 2.45 g / cm³ using the hydrostatic bottle method. 3 ; The compaction strength range of commonly used filling hydraulic supports was recorded as 0.5–10 MPa, with 5 typical pressure levels selected: 0.5 MPa, 1 MPa, 2 MPa, 5 MPa, and 10 MPa. The obtained experimental raw dataset contains compaction intensity. P With true density of gangue p; Step 2: Sample preparation and compression test; Two typical aggregate sizes were selected: coarse aggregate size range: 20–50 mm, fine aggregate size range: 0–10 mm. Setting the mass ratio of coarse and fine aggregates α = m 1 / ( m 1+ m 2), of which m 1 represents the quality of coarse aggregate. m 2 refers to the quality of fine aggregate; In this embodiment, five doping ratios are designed: α=0.1, 0.3, 0.5, 0.7, 0.9; Total mass prepared under each doping ratio M A mixed sample weighing 5.00 kg was subjected to different compaction strengths on a hydraulic loading testing machine. P The loading rate was controlled at 1 MPa / min, and the pressure was maintained for 2 minutes after reaching the target pressure. The volume after compression was recorded. V 1.

[0025] Calculate compacted density: (1); For example, when P =2MPa α When =0.5, V 1 = 2.06 × 10 -3 m 3 ,but r c =2427kg / m 3 ; Repeated experiments were conducted to obtain compaction density data under different admixture ratios and different compaction strengths, with a total sample size of N=25 groups; Step 3: Dataset Construction and Feature Extraction; Strengthen the compaction P Coarse and fine aggregate ratio α True density of gangue r As an input feature, solid density r c As output features, construct the training dataset:

[0026] All features are normalized and mapped to the interval [0,1].

[0027] Step 4: Machine learning model building and training; The CatBoost algorithm was selected as the regression model. Initial parameters were set as follows: learning rate. l r =0.05; Maximum tree depth d =6; Number of iterations n =500; Loss function adopted: (4); in, =0.1, This is a regularization term for model complexity.

[0028] The optimal structure was finally determined by using five-fold cross-validation and Bayesian optimization to automatically adjust parameters; Step 5: Example of calculating gangue backfill volume; Taking an actual goaf as an example, the volume of the goaf... V =1200m 3 Design filling rate or =0.92; Selected doping ratio α′ =0.6; Compaction strength P′ =3MPa; True density of gangue p′ =2.48g / cm 3 .

[0029] parameter ( P′ , α′ , p′ Inputting the trained model, the predicted compaction density is 2390 kg / m³. 3 .

[0030] Calculate the required amount of gangue backfill: ; That is, approximately 2,630 tons of gangue material; Step 6: Model embedding and engineering applications; The model is embedded in the intelligent mine filling management system, and the real-time compaction strength is input. P and blending ratio α Data, the system automatically predicts compaction density r c and filling volume M This allows for a dynamic adjustment strategy, enabling precise supply and energy efficiency optimization of gangue backfill materials. Furthermore, the model can be updated through transfer learning based on the physical property parameters of gangue from different mining areas, supporting adaptive applications across multiple mining areas.

[0031] Therefore, the machine learning-based method for calculating gangue backfill volume provided by this invention achieves rapid and accurate calculation of gangue backfill volume by establishing a machine learning model relating compaction strength, coarse and fine aggregate ratio, and the relationship between true density and compaction density of gangue. This method effectively reflects the nonlinear characteristic relationships under multi-factor coupling, overcoming the limitations of traditional empirical formulas in terms of applicability and accuracy. By inputting on-site design parameters, the predicted compaction density and required backfill volume can be obtained instantly, providing a scientific basis for the design of backfill schemes in goaf areas.

Claims

1. A method for calculating a waste rock filling amount based on machine learning, characterized by, The method comprises the following steps: Step 1, measuring the gangue sample to obtain a gangue original data set; Step 2, preparing a gangue sample according to the gangue original data set and performing compression testing to obtain testing data; Step 3, constructing a gangue filling amount prediction data set according to the testing data; Step 4, establishing a machine learning model and optimizing the model according to the filling amount prediction data set to obtain an optimal model; Step 5, inputting the actual design parameters of the goaf into the optimal model to calculate the gangue filling amount. 2.The machine learning-based waste filling amount calculation method of claim 1, wherein, The step 1 is specifically: measuring physical parameters of gangue samples in a mining area to obtain true density of the gangue ρ , and recording ramming strength of a commonly used filling hydraulic support under different ramming conditions P , to obtain an original data set of the gangue {true density The step 2 is specifically: , ramming strength P}. 3.The machine learning-based waste filling amount calculation method of claim 2, wherein, ρ Step 2.1, two typical particle size of gangue materials were selected, and different mixing ratios were selected α A plurality of mixed gangue samples were prepared, and a preset compaction strength was loaded P The stress-strain curve and volume change were recorded step by step, and the volume after compression was measured V 1, the corresponding compaction density was calculated according to formulas (1) and (2) Step 2.2, repeating experiments of different mixing ratios and tamping intensities, and recording the data of different particle size mixing ratios, tamping intensities and obtained tamping densities. c : （1）； M = m 1+ m 2（2）； wherein, m 1 is the mass of coarse aggregate, m 2 is the mass of fine aggregate, M is the total mass of the sample; In step 2.1, the loading adopts a hierarchical loading mode, the loading pressure range is 0.5-10 MPa, each level of loading is kept stable for 5 min, and the corresponding volume change is recorded for stress-strain curve. 4.The machine learning-based waste filling amount calculation method of claim 3, wherein, In step 2.1, the mixing ratio of the α The mass ratio of the coarse aggregate is, α The value range is 0~1. 5.The machine learning-based waste filling amount calculation method of claim 4, wherein, ρ 6.The machine learning-based waste filling amount calculation method of claim 5, wherein, The step 3 is specifically: taking the tamping strength P , the gangue mixing ratio α and the true density of the gangue P, α, ρ as input variables X, the X=[ ρ ], taking the tamping density The step 4 is specifically: c obtained by experiment as output variable Y, constructing a gangue filling amount prediction dataset D, the D={X, Y}, and normalizing the input variables to map different dimension variables to the interval [0, 1]. 7.The machine learning-based waste filling amount calculation method of claim 6, wherein, Step 4.1, establishing a tamping density prediction model by using a gradient boosting tree algorithm, and the model structure is shown in formula (3): Step 4.2, training the tamping density prediction model by using training samples, and the objective function is shown in formula (4); （3）； wherein, f a nonlinear regression function mapping relationship learned by the machine learning algorithm; Step 4.3, optimizing and adjusting the model parameters by cross-validation and Bayesian algorithm, including learning rate, tree depth, sub-sample ratio and regularization parameter, to obtain the optimal model. （4）； wherein, is the model prediction, is the experimental value, is the model complexity regularizer, is the penalty coefficient; In step 4.1, the machine learning algorithm is selected from one of CatBoost, XGBoost, random forest and deep neural network model. 8.The machine learning-based waste filling amount calculation method of claim 7, wherein, The step 5 is specifically: 9.The machine learning-based waste filling amount calculation method of claim 8, wherein, In step 4.2, the performance evaluation of the optimal model prediction result is quantitatively analyzed by using the coefficient of determination R 2 , the mean square error MSE and the mean absolute error MAE, and when R 2 ≥ 0.9, the model training is considered to reach the engineering application accuracy. 10.The machine learning-based waste filling amount calculation method of claim 9, wherein, Step 5.1, inputting the design parameters of the goaf; ρ Step 5.2, substitute the design parameters into the optimum model to obtain the predicted ramming density η c ' , calculate the required gangue filling quantity of the goaf according to formula (5) M' ; （5）； wherein, ρ' is the filling rate; P' is the ramming strength; α' is the coarse-fine aggregate mixing ratio; ​ is the true density of the gangue; V is the goaf volume.

Images