Solid waste raw material activity prediction method based on oxide composition characteristics

Abstract

The invention discloses a solid waste raw material activity prediction method based on oxide composition characteristics, and the method comprises the steps: firstly, constructing a standard data set, obtaining oxide component data, D50 particle size data and corresponding actually measured activity indexes of silicon-aluminum solid wastes such as fly ash, slag, steel slag and coal gangue, and carrying out the normalization and abnormal value processing to form training data; secondly, training an oxide-particle size and activity nonlinear mapping model by adopting a neural network regression model, analyzing and quantifying the contribution degree of each input feature to the activity in combination with SHAP, and ensuring the interpretability of the model; and finally, inputting the oxide of the solid waste to be detected and D50 data into the trained model, and outputting an activity predicted value. According to the method, the solid waste activity can be rapidly predicted within 10 minutes, the prediction precision R2 is larger than or equal to 0.9, the relative error is smaller than or equal to 10%, the detection cost is greatly reduced, the method can be directly used for solid waste purchase front-end pre-judgment, cement-based material formula design and raw material screening, and efficient technical support is provided for solid waste resource utilization.

Description

Technical Field

[0001] This invention belongs to the field of cement material technology, and in particular relates to a method for predicting the activity of solid waste raw materials based on oxide composition characteristics. Background Technology

[0002] Resource utilization of industrial solid waste is one of the core pathways for the building materials industry to achieve its "dual carbon" goals and promote green and low-carbon development. Coal gangue, calcined kaolin tailings, fly ash, and other siliceous and aluminous solid wastes, due to their potential pozzolanic activity, can be used as auxiliary cementing materials to partially replace cement. This reduces the environmental pressure caused by industrial solid waste stockpiling and lowers energy consumption and carbon emissions during cement production, resulting in significant economic and environmental benefits.

[0003] However, the activity of silica-alumina solid waste is greatly affected by multiple factors, including its formation process, chemical composition, and physical form, resulting in a wide range of fluctuations. For example, the SiO2 and Al2O3 content of coal gangue from different origins can vary by 10% to 20% due to differences in coal seam composition, directly leading to a deviation of over 30% in the activity index. The activity level of solid waste directly determines its safe dosage in cement-based materials: too low activity leads to a decrease in the mechanical properties and durability of concrete; too high activity may cause volume stability problems due to excessively rapid hydration. Therefore, accurately assessing the activity of solid waste is a prerequisite for its resource utilization.

[0004] Currently, the industry's assessment of solid waste activity mainly relies on traditional physicochemical experimental methods, typically including: 1. Strength Activity Index Method: This method is based on GB / T 1596-2017 "Fly Ash Used in Cement and Concrete". It requires the preparation of mortar test blocks and curing for 28 days before compressive strength can be measured and the activity index can be calculated. The testing cycle is as long as 1 month, which cannot meet the production line's need for rapid screening of raw materials. 2. Saturated Limestone Test Method: This method involves measuring the CaO content after the solid waste reacts with a saturated limestone solution. 2+ Although the concentration change method for assessing activity has a shorter cycle than the intensity method (about 7 days), it requires high-precision ion chromatographs and other equipment, and the cost of a single test exceeds 500 yuan. Moreover, the results have a low correlation with the actual activity in cement-based systems. 3. Hydration heat analysis: This method determines activity by monitoring the heat release rate during the hydration process of solid waste. However, it requires high equipment investment (a single calorimeter costs over 500,000 yuan) and is easily affected by factors such as ambient temperature and sample moisture content, resulting in poor repeatability.

[0005] In summary, existing technologies suffer from core problems such as lengthy cycles, high costs, limited accuracy, and lack of interpretability, failing to meet the demand for "front-end, rapid, and accurate" activity assessment in solid waste resource utilization. There is an urgent need to develop a solid waste activity prediction method that integrates chemical composition and physical characteristics, and combines accuracy and interpretability. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the activity of solid waste raw materials based on oxide composition characteristics, so as to solve the problems existing in the prior art.

[0007] To achieve the above objectives, this invention provides a method for predicting the activity of solid waste raw materials based on oxide composition characteristics, comprising: Step S1, Database Construction: Obtain oxide composition data, physical particle size data, and corresponding measured activity index of solid waste raw materials to construct a standard dataset for model training; the solid waste raw materials are at least one silica-alumina solid waste selected from fly ash, slag, calcined clay and calcined kaolin tailings, and coal gangue; the oxide composition data includes at least the percentage content of CaO, SiO2, and Al2O3; before constructing the standard dataset, the oxide composition data is preprocessed by normalization. Step S2, Prediction Model Establishment: Based on the standard dataset, a machine learning algorithm is used to train a prediction model for the activity of solid waste raw materials. The model can map the nonlinear relationship between oxide component data and physical particle size data and activity index. Step S3, Activity Prediction: Input the oxide component data and physical particle size data corresponding to the solid waste raw material to be predicted into the trained activity prediction model, and output the activity prediction value of the solid waste raw material to be predicted.

[0008] Optionally, in step S1, the normalization preprocessing adopts the min-max normalization method or the Z-score standardization method. If the min-max normalization method is adopted, the oxide component data is mapped to the [0,1] interval; if the Z-score standardization method is adopted, the oxide component data satisfies the distribution characteristics of mean 0 and standard deviation 1.

[0009] Optionally, the oxide composition data may also include the percentage content of at least one of Fe2O3, MgO, K2O, Na2O, and SO3.

[0010] Optionally, in step S1, the measured activity index includes the strength activity index. The strength activity index is determined according to GB / T 1596-2017 "Fly Ash for Cement and Concrete", specifically including: preparing mortar blocks according to the mortar mass ratio of cement: solid waste: standard sand: water = (1 - substitution rate): substitution rate: 3:0.5, with a substitution rate of 30%~50%, curing for 28 days at a temperature of 20±2℃ and a relative humidity of ≥90%, and measuring the compressive strength of the blocks. The ratio of the compressive strength of the solid waste mortar to the compressive strength of the reference cement mortar is used as the strength activity index. The measured activity index can also be replaced by the saturated limestone test activity index or the heat of hydration activity index.

[0011] Optionally, in step S1, the physical particle size data uses the D50 particle size; the D50 particle size is obtained by acquiring the original particle size distribution data through a particle size testing instrument, then plotting the cumulative distribution curve, and finally reading the particle size corresponding to the 50th percentile from the curve, which is the D50 particle size data, and the average value of three measurements is the D50 particle size data.

[0012] Optionally, in step S1, the data sources for constructing the standard dataset include at least one of publicly published literature, experimental measurement results, and industrial databases; the sample size of the standard dataset is not less than 100 groups, and it covers at least three different types of silicon-aluminate solid waste; when constructing the standard dataset, outlier data is removed using the 3σ principle or box plot method, and missing data is filled using the mean imputation method or linear interpolation method.

[0013] Optionally, in step S2, the machine learning algorithm is a neural network regression model; the neural network regression model includes one input layer, two hidden layers, and one output layer; the number of nodes in the input layer is four, and the data are related to the oxide composition (CaO, SiO2, Al2O3) and physical particle size (D). 50 The total number of features in the data is consistent; the number of nodes in each hidden layer is 16-64, and the activation function of the hidden layer is ReLU; the number of nodes in the output layer is 1, and the activation function of the output layer is linear activation; the model is trained using the Scaled Conjugate Gradient optimization algorithm, with a learning rate of 0.001-0.01, 1000-5000 iterations, and the mean squared error (MSE) loss function; before training, the standard dataset is divided into training, validation, and test sets in a 7:2:1 ratio, and the model hyperparameters are adjusted on the validation set using a Bayesian optimization algorithm, and the model performance is validated on the test set; simultaneously, a set of independent experimental validation sets is used to further evaluate the model's generalization ability, requiring the model's coefficient of determination R on the validation set to be [value missing]. 2 ≥0.9.

[0014] Optionally, in step S2, after the prediction model is established, a model interpretation step is also included: the SHAP value of each input feature is approximately calculated by combining Monte Carlo sampling with the feature marginal contribution method: first, background samples and samples to be interpreted are selected from the training set, and the contribution of each feature to the activity prediction result is obtained after 10 samplings through the method of "feature replacement-model prediction-difference calculation"; then, the average influence intensity of each feature is visualized by the SHAP feature importance bar chart, and the positive and negative contribution distribution of the feature to the prediction result is presented by the SHAP value box plot; finally, the distribution law of SHAP value in the box plot is combined to help analyze the potential nonlinear correlation between a single input feature and the activity prediction value.

[0015] Optionally, in step S3, the oxide composition data of the solid waste raw material to be predicted is determined by X-ray fluorescence spectroscopy. The measuring equipment is an X-ray fluorescence spectrometer. Before the measurement, the solid waste raw material to be predicted is ground to a particle size ≤0.074mm and dried to constant weight. The sample is prepared by tableting, with a tableting pressure of 20~30MPa and a holding time of 30~60s.

[0016] Optionally, in step S3, the relative error between the output activity prediction value and the measured activity index is ≤10%; if the relative error is >10%, return to step S1 to supplement the standard dataset sample size or optimize the data preprocessing method, and re-execute steps S2~S3.

[0017] The technical effects of this invention are as follows: 1. Extremely short cycle: The traditional 28-day activity evaluation cycle is shortened to within 10 minutes (only the rapid determination of oxides and D50 and model prediction are required), meeting the "instant screening" requirements of the production line and avoiding production stoppages caused by waiting for experimental results; 2. Low cost: The cost of testing a single sample is less than 50 yuan (the cost of consumables for XRF and laser particle size testing), which is only 1 / 4 of the traditional intensity activity index method (about 200 yuan / sample) and 1 / 20 of the hydration heat analysis method (about 1,000 yuan / sample), greatly reducing the testing costs for enterprises; 3. Wide applicability: Applicable to various types of siliceous aluminous solid waste such as fly ash, slag, steel slag, and coal gangue. Through the iterative mechanism of "supplementing samples - optimizing models", it can be adapted to new types of solid waste (such as recycled micro powder from construction waste) without the need to redevelop the model. 4. Pre-production prediction: Activity can be quickly predicted during the solid waste procurement stage (such as when the supplier provides a small sample), avoiding material waste caused by purchasing low-activity solid waste, and providing a basis for pre-production formulation design (such as adjusting the dosage in advance), thereby improving production efficiency. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a graph showing the fitting relationship between the model's predicted values ​​and the measured values ​​in an embodiment of the present invention. Figure 2 This is a sensitivity analysis diagram of the SHAP model in an embodiment of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] The method for predicting the activity of solid waste raw materials based on oxide composition characteristics as described in this invention includes the following steps: Step S1: Database Construction We obtained oxide composition data, physical particle size data, and corresponding measured activity indices of various siliceous aluminate solid waste raw materials. After data preprocessing, we constructed a standard dataset for model training, which specifically includes: Data types and sources: Oxide composition data: including at least the percentage content of CaO, SiO2, and Al2O3, determined by X-ray fluorescence spectrometry (XRF) using a Panaco Axios advanced X-ray fluorescence spectrometer. Before measurement, the solid waste sample was ground to a particle size ≤0.074mm and dried in an oven at 105℃ for 2h to constant weight. The sample was prepared by pressing: 5g of dried sample was weighed and pressed at 20~30MPa for 30~60s to form a disc with a diameter of 30mm. Each sample was measured twice, and the average value was taken as the final oxide composition data. Physical particle size data: The D50 particle size (characterizing the median particle size of the sample and reflecting the concentration trend of particle size distribution) is used. The original particle size distribution data is obtained by using a particle size testing instrument, and then a cumulative distribution curve is plotted. Finally, the particle size corresponding to the 50th percentile is read from the curve, which is the D50 particle size data. The average value of three measurements is the D50 particle size data.

[0021] Measured Activity Index: The Strength Activity Index (SAI) is used as the core indicator, and the procedure is based on GB / T 1596-2017. Specifically, 40mm×40mm×40mm mortar blocks are prepared according to the mortar mass ratio of "cement: solid waste: standard sand: water = (1 - substitution rate): substitution rate: 3:0.5", with a substitution rate set at 30% (general-purpose silica-alumina solid waste). The blocks are cured for 28 days in a standard curing chamber at 20±2℃ and relative humidity ≥90%. The compressive strength is measured using a YES-3000 pressure testing machine (loading rate 2.4kN / s). The Strength Activity Index = (compressive strength of solid waste mortar / compressive strength of reference cement mortar) × 100%. The measured activity index can also be replaced by the saturated limestone activity index (according to JGJ / T 385-2015 "Technical Specification for Application of Recycled Aggregates for Concrete") or the heat of hydration activity index (according to GB / T 1596-2017). (12959-2008 "Method for Determination of Heat of Hydration of Cement"). Data sources include at least one of the following: publicly published literature, laboratory-derived test results, and industrial databases.

[0022] 2. Data preprocessing: Normalization: The oxide component data were normalized using the min-max normalization method (formula: The data is mapped to the [0,1] interval; if there are extreme outliers in the oxide component data (e.g., CaO content > 50% in a sample), Z-score normalization is used (formula: , μ Where is the mean, σ (Standard deviation), to avoid the interference of extreme values ​​on model training; D50 particle size data does not need to be normalized, but the unit must be uniformly "μm"; Outlier and Missing Value Handling: Outliers are handled using the 3σ rule (removing data exceeding the range of "mean ± 3 × standard deviation") or box plot method (removing data exceeding the range of "interquartile range 1.5 times"). If the proportion of outliers is <5%, they are directly removed; if the proportion is 5%~10%, test data of the same type of solid waste needs to be added to replace the outliers; missing values ​​are filled using mean imputation (suitable for data with a missing rate <3%) or linear interpolation (suitable for data with a missing rate of 3%~5% and the data is linearly distributed). 3. Dataset Requirements: The standard dataset should have at least 100 samples and cover at least 3 different types of silica-alumina solid waste (such as fly ash, coal gangue, and calcined kaolin tailings). The sample size for each type of solid waste should be at least 20 samples to ensure the model's generalization ability. The coefficient of variation (standard deviation / mean) of each feature in the dataset should be >0.1 to avoid the model from overfitting due to overly simplistic features.

[0023] Step S2: Prediction Model Establishment and Interpretation Based on the standard dataset constructed in step S1, a machine learning algorithm is used to train a solid waste raw material activity prediction model, and the contribution of each feature is quantified through interpretability analysis, specifically including: 1. Model selection and structural design: A neural network regression model is adopted (which better fits the nonlinear relationship between oxides, particle size, and activity compared to traditional linear regression and random forests). The model structure is "1 input layer + 2 hidden layers + 1 output layer": the number of nodes in the input layer = the number of oxide components + 1 (D50 particle size). For example, in "CaO+SiO2+Al2O3+D50", the number of input layer nodes is 4, consistent with the total number of features of oxide components (CaO, SiO2, Al2O3) and physical particle size (D50 particle size); the number of nodes in each hidden layer is 16~64; the ReLU function is used as the activation function for the hidden layers (formula: To avoid gradient vanishing; the output layer has 1 node (corresponding to the predicted activity index), and the activation function is a linear activation function (purelin, formula: ), ensuring that the predicted value range is consistent with the measured activity index; 2. Model training parameters: Data partitioning: The standard dataset is randomly divided into a training set (for model parameter fitting), a validation set (for hyperparameter tuning), and a test set (for model performance validation) in a ratio of 7:2:1. Training parameters: The optimizer uses the Scaled Conjugate Gradient (SCG) optimization algorithm with a learning rate of 0.001~0.01; the loss function is the mean squared error (MSE), formula: ,in These are measured values. (Predicted values); iterations 1000-5000 times; adjust model hyperparameters (including the number of nodes in each hidden layer and the learning rate) on the validation set using a Bayesian optimization algorithm; 3. Model performance validation: Key metric: R-coefficient of determination on the test set 2 ≥0.9 (formula: ,in (mean of measured values), relative error (formula: )≤10%, and at the same time, a set of independent experimental validation sets were used to further evaluate the generalization ability of the model; 4. Model interpretability analysis: The SHAP (SHapley Additive exPlanations) analysis method was adopted, and the DeepExplainer interpreter (adapted to the neural network model) was selected to calculate the SHAP value of the prediction result of each input feature (such as CaO, SiO2, D50) for each sample. A positive SHAP value indicates that the feature increases the predicted activity value, and a negative SHAP value indicates that it decreases the predicted value. The larger the absolute value, the stronger the influence. The analysis results were visualized through SHAP summary plots (showing the distribution of feature SHAP values ​​of all samples) and SHAP dependency plots (showing the nonlinear relationship between a single feature and the predicted value). For example, if the SHAP value of SiO2 is mostly positive and the absolute value increases with the increase of content, it indicates that the increase of SiO2 content has a promoting effect on solid waste activity.

[0024] Step S3, Activity Prediction 1. Preparation of data to be predicted: Determine the oxide composition data (at least CaO, SiO2, Al2O3) and D50 particle size data of the solid waste to be predicted according to the method in step S1, and ensure that the measurement method and units are consistent with the standard dataset. 2. Data preprocessing: The same normalization method (such as min-max or Z-score) as in step S1 is used to process the oxide composition of the data to be predicted; 3. Prediction and Result Verification: Input the preprocessed data to be predicted into the trained prediction model and output the activity prediction value. If the relative error between the predicted value and the subsequent measured value (according to the measured activity index method in step S1) is >10%, return to step S1: supplement the sample data of this type of solid waste (at least 10 sets), optimize the data preprocessing method (such as changing the normalization method) or adjust the model structure (such as increasing the number of hidden layer nodes), and re-execute steps S2~S3.

[0025] like Figure 1-2 As shown, this embodiment provides a method for predicting the activity of solid waste raw materials based on oxide composition characteristics: Example 1: This embodiment focuses on calcined kaolin tailings as the main research object, with the strength activity index (SAI) as the target performance.

[0026] Taking the activity prediction of three existing calcined kaolin tailings as examples: (1) Obtain the oxide composition and D50 particle size of different solid waste materials and their corresponding strength and activity indices through literature retrieval to form a standard dataset; (2) Based on the dataset, a predictive model for the performance of each phase oxide was established using a neural network. The model results showed an R² of over 0.90, such as... Figure 1As shown. Furthermore, the SHAP analysis method is used to perform interpretability analysis on the prediction model to quantify the contribution of each input feature to the prediction result, such as... Figure 2 As shown.

[0027] (3) Input the oxide component data and physical particle size data corresponding to the three types of calcined kaolin tailings to be predicted into the trained activity prediction model, and output the activity prediction value of the solid waste raw material to be predicted, as shown in Table 1.

[0028] Table 1. Composition and Prediction of Calcined Kaolin Tailings

[0029] Actual SAI strength activity tests were conducted on the materials in Table 1, and the actual compressive strengths were obtained as shown in Table 1. R² was 0.9350, verifying the feasibility of the prediction model.

[0030] Example 2 Similar to Example 1, except that the raw materials to be tested were three kinds of calcined coal gangue, as shown in Table 2.

[0031] Table 2 Composition and Prediction of Calcined Coal Gangue

[0032] Actual strength activity tests were conducted on the materials according to Table 2, and the actual compressive strength was obtained as shown in Table 2. R2 was 0.9755, which verified the feasibility of the prediction model.

[0033] Taking this as an example, the activity of any solid waste raw material can be predicted, solving the problems of time-consuming, labor-intensive, and slow testing in existing activity testing methods. Through this method, designers can achieve accurate prediction and non-destructive evaluation of the activity of solid waste raw materials based on readily available oxide content and particle size characteristics.

[0034] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the activity of solid waste raw materials based on oxide composition characteristics, characterized in that, Includes the following steps: Step S1, Database Construction: Obtain oxide composition data, physical particle size data, and corresponding measured activity index of solid waste raw materials to construct a standard dataset for model training; the solid waste raw materials are at least one silica-alumina solid waste selected from fly ash, slag, calcined clay and calcined kaolin tailings, and coal gangue; the oxide composition data includes at least the percentage content of CaO, SiO2, and Al2O3; before constructing the standard dataset, the oxide composition data is preprocessed by normalization. Step S2, Prediction Model Establishment: Based on the standard dataset, a machine learning algorithm is used to train a prediction model for the activity of solid waste raw materials. The model can map the nonlinear relationship between oxide component data and physical particle size data and activity index. Step S3, Activity Prediction: Input the oxide component data and physical particle size data corresponding to the solid waste raw material to be predicted into the trained activity prediction model, and output the activity prediction value of the solid waste raw material to be predicted.

2. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S1, the normalization preprocessing adopts the min-max normalization method or the Z-score standardization method. If the min-max normalization method is adopted, the oxide component data is mapped to the [0,1] interval; if the Z-score standardization method is adopted, the oxide component data satisfies the distribution characteristics of mean 0 and standard deviation 1.

3. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 2, characterized in that, The oxide composition data also includes the percentage content of at least one of Fe2O3, MgO, K2O, Na2O, and SO3.

4. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S1, the measured activity index includes the strength activity index. The strength activity index is determined according to GB / T1596-2017 "Fly Ash for Cement and Concrete", specifically including: preparing mortar blocks according to the mortar mass ratio of cement: solid waste: standard sand: water = (1 - substitution rate): substitution rate: 3:0.5, with a substitution rate of 30%~50%, curing for 28 days at a temperature of 20±2℃ and a relative humidity of ≥90%, and measuring the compressive strength of the blocks. The ratio of the compressive strength of the solid waste mortar to the compressive strength of the reference cement mortar is used as the strength activity index. The measured activity index can also be replaced by the saturated limestone test activity index or the heat of hydration activity index.

5. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S1, the physical particle size data uses the D50 particle size. The D50 particle size is obtained by using a particle size testing instrument to obtain the original particle size distribution data, then plotting the cumulative distribution curve, and finally reading the particle size corresponding to the 50th percentile from the curve, which is the D50 particle size data. The average value of three measurements is the D50 particle size data.

6. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S1, the data sources for constructing the standard dataset include at least one of publicly published literature, experimental measurement results, and industrial databases; the sample size of the standard dataset is not less than 100 groups and covers at least 3 different types of silicon-aluminum solid waste; when constructing the standard dataset, outlier data is removed by using the 3σ principle or box plot method, and missing data is filled by using the mean imputation method or linear interpolation method.

7. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S2, the machine learning algorithm is a neural network regression model; the neural network regression model includes one input layer, two hidden layers, and one output layer; the number of nodes in the input layer is four, consistent with the total number of features of the oxide composition data and physical granularity data; the number of nodes in each hidden layer is 16-64, and the activation function of the hidden layer is the ReLU function; the number of nodes in the output layer is one, and the activation function of the output layer is a linear activation function; the model training uses the Scaled Conjugate Gradient optimization algorithm, with a learning rate of 0.001-0.01, an iteration count of 1000-5000 rounds, and a mean squared error function as the loss function; Before training, the standard dataset was divided into training, validation, and test sets in a 7:2:1 ratio. The model's hyperparameters were tuned on the validation set using a Bayesian optimization algorithm, and the model's performance was validated on the test set. Simultaneously, an independent experimental validation set was used to further evaluate the model's generalization ability, requiring the model's coefficient of determination R0 on the validation set to be [value missing]. 2 ≥0.

9.

8. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S2, after the prediction model is established, a model interpretation step is also included: the SHAP value of each input feature is approximately calculated by combining Monte Carlo sampling with the feature marginal contribution method: first, background samples and samples to be interpreted are selected from the training set, and the contribution of each feature to the activity prediction result is obtained after 10 samplings through the method of "feature replacement-model prediction-difference calculation"; then, the average influence intensity of each feature is visualized by the SHAP feature importance bar chart, and the positive and negative contribution distribution of the feature to the prediction result is presented by the SHAP value box plot; finally, the distribution law of SHAP value in the box plot is combined to help analyze the potential nonlinear correlation between a single input feature and the activity prediction value.

9. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S3, the oxide composition data of the solid waste raw material to be predicted is determined by X-ray fluorescence spectroscopy. The measuring equipment is an X-ray fluorescence spectrometer. Before the measurement, the solid waste raw material to be predicted is ground to a particle size ≤0.074mm and dried to constant weight. The sample is prepared by tableting, with a tableting pressure of 20~30MPa and a holding time of 30~60s.

10. The method for predicting the activity of solid waste raw materials based on oxide composition characteristics according to claim 1, characterized in that, In step S3, the relative error between the output activity prediction value and the measured activity index is ≤10%; if the relative error is >10%, return to step S1 to supplement the standard dataset sample size or optimize the data preprocessing method, and re-execute steps S2~S3.

Images