Method and system for predicting volcanic ash activity of aluminosilicate solid waste

Abstract

The invention provides an aluminosilicate solid waste volcanic ash activity prediction method and system, and the method employs XRD diffraction data, XRF spectral analysis data, particle size and other original signal data containing material composition structure information for transformation as input data, and combines an artificial neural network ANN and a convolutional neural network CNN to predict the activity of aluminosilicate solid waste volcanic ash. According to the method, the aluminosilicate solid waste volcanic ash activity performance prediction fusion model is constructed, the performance of the fusion model is improved through hyper-parameter optimization, rapid and accurate prediction of the aluminosilicate solid waste volcanic ash activity is achieved, the method is accurate and stable in prediction, the test period is shortened, the test cost is reduced, and the method is suitable for various types of materials and has wide application prospects.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a method and system for predicting the pozzolanic activity of aluminosilicate solid waste. Background Art

[0002] In China, the annual emissions of aluminosilicate solid wastes such as red mud, coal gangue, fly ash, tailings, and steel slag are huge. This not only occupies a large amount of land resources but also poses a serious environmental pollution risk. Silicate building materials represented by cement and glass play a dominant role in China's basic materials industry. The production process consumes a large number of natural silicon-aluminum raw materials every year, resulting in the increasing scarcity of building material raw materials such as quartz sand, kaolin, and bauxite. Using aluminosilicate solid waste to manufacture building materials can not only effectively treat bulk solid waste but also meet the urgent needs of the sustainable development of the building materials industry.

[0003] Currently, although the generation amount of aluminosilicate solid waste is huge, the basic theoretical research on its resource utilization through synergy is not sufficient, resulting in a low resource utilization rate of solid waste. In the process of solid waste reuse, X-ray diffraction (XRD) technology plays a key role and can accurately identify the mineral composition in materials. The main mineral components in aluminosilicate inorganic solid waste, such as clay minerals like quartz, kaolinite, montmorillonite, and illite, exhibit a certain pozzolanic activity after calcination, and this activity directly affects the dosage of building material raw materials and the performance development of cement concrete. However, the influence of different mineral compositions on pozzolanic activity varies, and their chemical compositions and contents directly determine the mineral composition of aluminosilicate inorganic solid waste. As an important basis for the performance index of mineral admixtures, the pozzolanic activity index needs to be tested in accordance with the national standard "Technical Specification for the Application of Mineral Admixtures" (GBT51003 - 2014) by measuring the 28-day compressive strength of the calcined solid waste. However, this method has a long cycle and large batch volume, reducing the efficiency of aluminosilicate solid waste resource utilization, increasing the uncertainty and risk of engineering construction, and also increasing the difficulty of aluminosilicate solid waste resource utilization.

[0004] At present, domestic and foreign scholars have not conducted sufficient research on the rapid determination of the pozzolanic activity of aluminosilicate inorganic solid waste. However, with the advent of the information age, machine learning technology provides the possibility to solve the limitations faced by experimental and theoretical models. By training complex geological environment data through machine learning, the (non)-linear relationship between variables can be captured to achieve reliable prediction of regression tasks.

[0005] However, the existing machine learning models highly depend on physicochemical experimental conditions, do not fully consider the characteristics of aluminosilicate solid waste, and have a high dependence on the stability of process parameters, with low generalization ability. When facing unknown process parameter fluctuations, it may lead to a serious deviation between performance prediction and reality, and it is impossible to achieve rapid and accurate prediction of the pozzolanic activity of aluminosilicate solid waste. Summary of the Invention

[0006] To solve the above problems, the object of the present invention is to provide a method and system for predicting the pozzolanic activity of aluminosilicate solid waste. By transforming the original signal data containing material composition and structure information such as XRD diffraction data, XRF spectral analysis data, and particle size, as input data, combining artificial neural network (ANN) and convolutional neural network (CNN), a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste is constructed, and then the performance of the fusion model is improved through hyperparameter optimization to achieve rapid and accurate prediction of the pozzolanic activity of aluminosilicate solid waste.

[0007] According to the first aspect of the present invention, a method for predicting the pozzolanic activity of aluminosilicate solid waste is provided, including the following steps: S1: Data preparation, collect the XRD diffraction data, XRF spectral analysis data, and physical property factors of aluminosilicate solid waste, synchronously collect the pozzolanic activity index of aluminosilicate solid waste as the objective function data, and clean and preprocess the collected XRD diffraction data, XRF spectral analysis data, and physical property factors; S2: Multi-modal model construction, train 2 artificial neural networks ANN and extract features through the preprocessed physical property factors and preprocessed XRF spectral analysis data, train 1 convolutional neural network CNN and extract features through the preprocessed XRD diffraction data, and fuse the outputs of the artificial neural network ANN and the convolutional neural network CNN to construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste; S3: Hyperparameter optimization, use the Bayesian optimization algorithm for hyperparameter optimization to determine the best parameter settings of the fusion model to improve the performance of the fusion model; S4: Model evaluation and correction, use the root mean square error RMSE and the coefficient of determination R 2 to evaluate the performance of the fusion model. When the root mean square error RMSE is within 1.5 and the coefficient of determination R 2 is greater than 1, the fusion model meets the requirements, and the fusion model is applied to predict the pozzolanic activity of aluminosilicate solid waste. Otherwise, the fusion model is corrected until the fusion model meets the requirements.

[0008] Optionally, the multi-modal model construction in step S2 includes the training of sub-networks and the construction of the fusion model, where the training of sub-networks includes 2 ANN sub-networks and 1 CNN sub-network.

[0009] Optionally, one of the two ANN sub-networks starts with an input layer that receives the elemental composition data obtained from XRF spectroscopy. After the input layer, there are two hidden layers responsible for data processing and transformation. The final output layer contains the trained features for providing prediction results. Optionally, the input layer of the other ANN sub-network among the two ANN sub-networks receives the physical property factors of the aluminosilicate solid waste, followed by two hidden layers and an output layer containing the trained features for outputting the prediction results.

[0010] Optionally, in step S4, the coefficient of determination R 2 and the root mean square error RMSE are calculated using the following formulas for various regression metrics, where is the actual pozzolanic activity index, is the predicted pozzolanic activity index, is the average value of the actual values, is the average value of the predicted values, and N is the total number of actual values:

[0011]

[0012] Optionally, the method for correcting the fusion model in step S4 includes: re-checking the data cleaning and preprocessing process in step S1, adjusting the model parameters, or changing the model fusion method.

[0013] Optionally, re-checking the data cleaning and preprocessing process in step S1 includes: checking whether there are mislabeled data or outliers that have not been properly processed; checking whether the normalization or standardization operations of the data are reasonable.

[0014] Optionally, adjusting the model parameters includes: for the ANN sub-network and the CNN sub-network, adjusting the number of network layers and the number of neurons in each layer.

[0015] Optionally, changing the model fusion method includes: adopting different fusion strategies to integrate the outputs of the three sub-networks. Weighted fusion can be used to assign different weights according to the importance of the outputs of each sub-network to the final prediction result.

[0016] According to a second aspect of the present invention, a prediction system for the pozzolanic activity of aluminosilicate solid waste is provided, including: a data preparation module, configured to collect XRD diffraction data, XRF spectral analysis data, and physical property factors of aluminosilicate solid waste, synchronously collect the pozzolanic activity index of aluminosilicate solid waste as objective function data, and clean and preprocess the collected XRD diffraction data, XRF spectral analysis data, and physical property factors; a multi-modal model construction module, configured to train 2 artificial neural networks (ANNs) and extract features through the preprocessed physical property factors and preprocessed XRF spectral analysis data, train 1 convolutional neural network (CNN) and extract features through the preprocessed XRD diffraction data, and fuse the outputs of the artificial neural network (ANN) and the convolutional neural network (CNN) to construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste; a hyperparameter optimization module, configured to perform hyperparameter optimization using the Bayesian optimization algorithm to determine the optimal parameter settings of the fusion model to improve the performance of the fusion model; a model evaluation and correction module, configured to use the root mean square error (RMSE) and the coefficient of determination R 2 to evaluate the performance of the fusion model. When the root mean square error (RMSE) is within 1.5 and the coefficient of determination R 2 is greater than 1, the fusion model meets the requirements, and the fusion model is applied to predict the pozzolanic activity of aluminosilicate solid waste. Otherwise, the fusion model is corrected until the fusion model meets the requirements.

[0017] Adopting the above technical solutions, the present invention has the following beneficial effects:

[0018] By establishing a large number of XRD pattern databases, the accuracy and stability of the prediction are improved. Using the multi-modal prediction model, the pozzolanic activity index of aluminosilicate solid waste can be quickly predicted, avoiding the cumbersome steps and time-consuming operations in the traditional mechanical compressive strength test method, and significantly shortening the test cycle; compared with the traditional mechanical compressive strength test method, it reduces the dependence on expensive test equipment and a large number of samples, and reduces the test cost. This method is applicable to various types of materials, including raw materials such as metals and ceramics that can be tested by XRD patterns, has a wide range of application prospects, provides technical support for realizing the automation and digitization of clinker strength prediction, and is in line with the trend of modern intelligent manufacturing and Industry 4.0. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0020] Figure 1 Schematic diagram of the model construction process for the prediction method of the pozzolanic activity of aluminosilicate solid waste provided by this application;

[0021] Figure 2 Multi-modal model structure for predicting the pozzolanic activity with the XRD pattern, XRF data and physical property factors of aluminosilicate;

[0022] Figure 3 Flowchart of the steps for the prediction method of the pozzolanic activity of aluminosilicate solid waste of this application. Specific implementation manners

[0023] To make the objectives, technical solutions and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some but not all of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the scope of protection of this application.

[0024] The objective of the present invention is to provide a prediction method and system for the pozzolanic activity of aluminosilicate solid waste to improve the accuracy of quickly predicting the pozzolanic activity index of aluminosilicate solid waste. To make the above objectives, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific implementation manners.

[0025] See Figures 1 - 3 , the prediction method for the pozzolanic activity of aluminosilicate solid waste of the present invention includes the following steps:

[0026] S1: Data preparation. Collect the XRD diffraction data, XRF spectral analysis data and physical property factors of aluminosilicate solid waste, synchronously collect the pozzolanic activity index of aluminosilicate solid waste as the objective function data, and clean and preprocess the collected XRD diffraction data, XRF spectral analysis data and physical property factors;

[0027] S2: Multi-modal model construction. Train 2 artificial neural networks ANN with the preprocessed physical property factors and preprocessed XRF spectral analysis data and extract features, train 1 convolutional neural network CNN with the preprocessed XRD diffraction data and extract features, and fuse the outputs of the artificial neural network ANN and the convolutional neural network CNN to construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste;

[0028] It should be understood that the fusion method includes but is not limited to at least one of simple splicing, weighted summation, gated fusion, and multi-layer perceptron fusion;

[0029] S3: Hyperparameter optimization. Use the Bayesian optimization algorithm to optimize the hyperparameters and determine the optimal parameter settings of the fusion model to improve the performance of the fusion model.

[0030] S4: Model evaluation and calibration. Use the root mean square error (RMSE) and the coefficient of determination R 2 to evaluate the performance of the fusion model. When the RMSE is within 1.5 and the coefficient of determination R 2 is greater than 1, the fusion model meets the requirements, and the fusion model is applied to predict the pozzolanic activity of aluminosilicate solid waste. Otherwise, the fusion model is calibrated until it meets the requirements.

[0031] Specifically, S1: Data preparation: Collect a large amount of XRD diffraction data of aluminosilicate solid waste. The XRD diffraction data includes the scanning angle (2θ) and the corresponding peak intensity (I); the physical property factors include particle size, specific surface area, and density; the XRF spectral analysis data includes the composition of the main elements SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, Na2O, TiO2, K2O, P2O5, MnO, ZrO2, and Cl in aluminosilicate solid waste determined by XRF spectral analysis. At the same time, to ensure that there is a basis for subsequent analysis, it is also necessary to synchronously collect the pozzolanic activity index of aluminosilicate solid waste as the objective function data.

[0032] It should be understood that the XRD diffraction data is derived from the XRD pattern and covers the basic information such as mineral composition, lattice parameters, amorphous content, crystallinity, and grain size that determine the pozzolanic activity of aluminosilicate.

[0033] The calculation method of the pozzolanic activity index of aluminosilicate solid waste is calculated according to the test method in GB / T 51003. The specific formula is:

[0034]

[0035] where H is the pozzolanic activity index (%), R is the 28-day compressive strength (MPa) of the test mortar prepared from aluminosilicate, and R0 is the 28-day compressive strength (MPa) of the reference cement mortar.

[0036] Before collecting the XRD diffraction data, it is necessary to accurately calibrate the XRD diffractometer. Use external standards such as alumina and silica to calibrate the XRD diffractometer to ensure that the peak position of the scanning angle is within the range of ±0.0025° to ±0.0125° of its standard value to improve the accuracy and reliability of the data.

[0037] By controlling the range of particle size and the calibration of the diffractometer, ensure the standardization and consistency of the variables in the experimental environment.

[0038] Correlation analysis evaluates the strength of the connection between each pair of variables. The calculation of the correlation coefficient yields a value between -1 and +1, indicating the degree of association between two variables.

[0039] Data preprocessing:

[0040] Use correlation analysis to evaluate the strength of the connection between each data variable collected in step S1. The calculated correlation coefficient is between -1 and +1. Clean and preprocess the collected data to remove data containing missing values or obvious outliers.

[0041] For the XRD spectral analysis data, use the Minmax scaler to adjust the scale of all variables from 0 to 1 for normalization. By using the normalized XRD diffraction data as input, it avoids prediction result deviations caused by missing or garbled data.

[0042] The specific processing formula is:

[0043]

[0044] Where X i and X norm,i are the original data and the i-th data of min-max normalization respectively.

[0045] When processing the XRD diffraction data, set the diffraction peak intensity of the strongest peak to 100, normalize the diffraction data at other scanning angles (2θ), and perform segmentation and convolution transformation on the data to form a matrix. X represents 2θ in the XRD spectral data and is set as the input factor of the CNN model.

[0046] Set the diffraction peak intensity of the strongest peak to 100 to standardize the data and simplify subsequent data processing. Using 100 as the benchmark can make the intensities of other peaks relative values with respect to the strongest peak, which helps to compare and identify the characteristics of each peak. Through the normalization method, it helps to maintain the consistency and accuracy of data processing.

[0047] Therefore, adopting this data processing method can not only effectively reduce the computational burden but also maintain sufficient data integrity to improve the prediction accuracy.

[0048] S2: Multi-modal model construction: For the transformed sample composition and strength performance data, construct a multi-modal model for learning and training, that is, construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste. This fusion model combines an artificial neural network (ANN) and a convolutional neural network (CNN).

[0049] Before data input, a series of transformation processes need to be performed. The input of the model is a one-dimensional sequence vector. To inject positional information into the sequence, the scanning angle (2θ) data is combined with the corresponding peak intensity (I) to form a matrix. In addition, in Example 2, to enhance the prediction ability of the model, XRF information is also spliced into the original matrix in a similar manner; in Example 3, physical property factors are further added and spliced into the original matrix.

[0050] S2-1: Training of the sub-network

[0051] Prediction is performed using an artificial neural network (ANN). The ANN model is a bionic simulation of the human brain and neuron system. The ANN model, that is, the artificial neural network simulates the human brain and neuron system and has three main layers: the input layer, the hidden layer, and the output layer. During the prediction process, the ANN uses the Backpropagation Algorithm for prediction. This algorithm minimizes the error between the predicted value and the actual value by adjusting the weights and biases. The ANN model is:

[0052] y(x) = f · ((∑w n ·x n ) + b)

[0053] where W n and b represent the weights and biases respectively, and f is the activation function, which is specifically used to convert the input signal into an output and send it as an input to the next layer in the stack.

[0054] The first ANN sub-network includes an input layer for receiving the XRF element composition, followed by two hidden layers and an output layer with training features.

[0055] The second ANN sub-network includes an input layer for receiving the physical property factor data of aluminosilicate solid waste, followed by two hidden layers and an output layer with training features. The rectified linear unit (ReLU) is selected as the activation function because it has the advantage of setting the small linear component of "x" to zero.

[0056] The purpose of creating a CNN model, i.e., a convolutional neural network, is to estimate the performance of aluminosilicate solid waste, thereby helping to predict the pozzolanic activity index of aluminosilicate solid waste from XRD spectra. A one-dimensional (1D) CNN is constructed because the XRD spectra used for model training are 1D vectors with a fixed x-axis range (corresponding to 2θ). The model includes an input layer, a convolutional layer, and a max pooling layer. Combining the ReLU activation function with dropout, randomly skipping neurons, can double the model performance compared to using the sigmoid to derive the CNN weight parameters in the backpropagation algorithm. Therefore, the ReLU function is selected as the activation function for this model.

[0057] The convolutional layer indicates that the data size input to the network is m×m. The convolutional layer is set to include K convolutional kernels, with the convolutional kernel size being F×F and the padding size If the stride is represented as S, then the size of the output after convolution is [[ID=*]] [[ID=*]]

[0058] For XRD diffraction data, the convolution process is represented by the following formula, where x represents the input sequence, h represents the convolutional kernel, k and j represent the lengths of the data, and * represents the convolution operation; [[ID=*]] [[ID=*]]

[0059] Y(m, n) = x[m, n] * h[m, n] = ∑ j ∑ k x[j, k]h[m - j, n - k] [[ID=*]]

[0060] The output of the activation function ReL is represented as: [[ID=*]] [[ID=*]]

[0061] σ(G(m, n)) = σ(x[m, n]) * h[m, n] = max(0, x[m, n] * h[m, n]) [[ID=*]]

[0062] The pooling layer indicates that the pooling layer compresses each sub-matrix of the input tensor. The pooling area size is set to k×k, and the pooling criterion is max pooling; if the input is m×m dimensional, then the output is represented [[ID=*]] [[ID=*]] [[ID=*]]

[0063] The fully connected layer is represented as: each neuron in the input is fully connected to the neurons in the next layer, and the output neurons represent the probabilities of each category; setting the activation function of the fully connected layer and the number of neurons L in each layer of the fully connected layer, the activation function usually uses the σ = Sigmoid() function, then the output of the connection layer is [[ID=*]] [[ID=*]]

[0064] S = σ(G(m, n)) [[ID=*]]

[0065] Where the Sigmoid function is: [[ID=*]] [[ID=*]]

[0066] [[ID=*]] [[ID=*]] [[ID=*]]

[0067] The physical property factors and XRF data are represented as one-dimensional vectors respectively. The XRD spectrum consists of one-dimensional vectors, reflecting the intensity according to 2θ (0° to 80°). The physical property factors and XRF data are used to train the ANN as a sub-network. The XRD spectrum data is used for the training of the CNN.

[0068] The goal of training the convolutional neural network is to minimize the loss function. The commonly used loss function is the cross-entropy loss function as shown below, where y i represents the true label of sample x i , and h θ (x i ) represents the probability that the sample belongs to the positive class.

[0069]

[0070] The convolutional neural network uses the gradient descent algorithm for iteration. The gradient vector formula is as follows:

[0071]

[0072] Until the loss function converges to a sufficiently small value or the training reaches the maximum number of iterations, the training ends, and a convolutional neural network model that can be used for target recognition is obtained.

[0073] S2-2: Construction of the fusion model

[0074] After training the data and extracting valuable features using the above 3 sub-network modules respectively, their outputs are then fused, combined into a single tensor. A new connection layer is added to the combined tensor, and two dense layers are added after the new connection layer to further process the data and generate the final output fusion model.

[0075] In the final fusion model, the XRD spectrum consists of a 1D vector, reflecting the intensity of the scanning angle 2θ (5 degrees to 20 degrees). The physical property factors and XRF data are presented in the form of 1D vectors containing 3 and 11 variables respectively. The physical property factors and XRF data are used to train 2 ANN sub-networks, and the XRD spectrum data is trained through the CNN.

[0076] S3: Hyperparameter optimization: Hyperparameter optimization is used to determine the best parameter settings to improve the model performance.

[0077] Hyperparameters are parameters whose values should be pre-assigned before training. The hyperparameters, including the number of units, dropout rate, batch size, kernel size, filter size, and pooling size, are optimized using Bayesian optimization to avoid overfitting problems. Bayesian optimization involves a surrogate model that interprets the prior distribution and data generation mechanism, and a loss function used to evaluate the optimality of the query sequence.

[0078] The Bayesian optimization algorithm assumes that there are n independent classifiers. The classifiers are applied to the training set to obtain a confusion matrix CM of c×c dimensions. j :

[0079]

[0080] Among them, the subscript c represents the total number of sample categories, Denotes the j-th classifier D j The training sample W k Class samples are divided into W s The number of classes, CM j The sum of the sth column in is the classifier D j Classify the samples into W s Total number of CM j Then we can get the probability matrix LM j , where the elements Indicates D j Classifier W k Class samples are divided into W s The probability of Because of the mutual independence between classifiers, the true W k The probability is:

[0081]

[0082] Finally, the sample category is determined by taking the larger probability of the true category.

[0083] S4: Model evaluation and calibration: using root mean square error (RMSE) and coefficient of determination (R 2 ) to evaluate the model performance.

[0084] An important step in model interpretation and evaluation of output. It can provide valuable insights into improving the model and understanding how much a specific feature contributes to its overall performance. 2 The following formulas are used to calculate various regression indicators, including is the actual volcanic ash activity index, To predict the volcanic ash activity index, is the average of the actual values, is the average of the predicted values, and N is the total number of actual values.

[0085]

[0086] The correction method described in S4 includes: rechecking the data cleaning and preprocessing process in S1, adjusting model parameters or changing the model fusion method;

[0087] Re-check the data cleaning and preprocessing process in S1:

[0088] (1) Check whether there are data with incorrect labels or outliers that have not been properly processed. For example, when collecting data on the pozzolanic activity index, some data may not match the actual situation due to experimental errors or recording mistakes. These data need to be rechecked and corrected or removed;

[0089] (2) Check whether the normalization or standardization operations of the data are reasonable. For physical property factors such as particle size, specific surface area, and density, their numerical ranges may vary greatly. If the normalization method is inappropriate, it will affect the training effect of the model. Different normalization methods can be tried, such as min-max normalization or Z-score standardization, to see if the model performance can be improved.

[0090] Adjust the model parameters: For the ANN sub-network and CNN, the number of network layers of the neural network, the number of neurons in each layer, etc. can be adjusted. For example, if the ANN network is too simple, it may not be able to fully learn the complex relationships in the physical property factors and XRF data. In this case, the number of hidden layers or the number of neurons can be appropriately increased. For the CNN training of XRD spectral data, parameters such as the size and stride of the convolutional kernel can also be adjusted to optimize the extraction of spectral data features.

[0091] Change the model fusion method: Consider using different fusion strategies to integrate the outputs of the three sub-networks. If the previous method was simple concatenation, weighted fusion can be tried, and different weights can be assigned according to the importance of the output of each sub-network to the final prediction result. For example, through experiments, it is found that the features extracted by the CNN from the XRD spectral data are more critical for predicting the pozzolanic activity performance. Then, a higher weight can be assigned to the output of the CNN during fusion.

[0092] The solution of the present invention is further described according to the following embodiments:

[0093] Embodiment 1:

[0094] XRD patterns of more than 500 aluminosilicate solid wastes (including fly ash, Shenzhen humus soil, etc.) were collected, with a scanning angle of 0 to 80°. Among them, the particle size was controlled so that the residue on a 45-μm sieve was ≤ 12%. The strongest peak of the XRD diffraction peak of the test sample was required to be not less than 100. External standards such as alumina and silica were used to calibrate the XRD diffractometer, and the peak position of the scanning angle was within the range of ±0.0025° to ±0.0125° of its standard value. The samples were subjected to XRD tests at a scanning speed of 0.01 to 0.02° / step and 0.1 s / step to 100 s / step, and the residence time for each step needed to satisfy that the strongest counting point of the XRD pattern diffraction peak was not less than 10,000 Counts. At the same time, the 28-day strength performance data of these samples were tested or collected, and the scanning angle (2θ), peak intensity (I) data, and pozzolanic activity index (H) of 500 groups of clinker samples were obtained.

[0095] The collected data were cleaned and preprocessed to remove data containing missing values or obvious outliers. At the same time, with the diffraction peak intensity of the strongest peak of silica (2θ ≈ 26.6°) as 100, the XRD diffraction data were normalized, and the data were segmented and convolution-transformed to form a matrix. The data were divided into a training set and a test set in a ratio of 7:3, and a pozzolanic activity index prediction fusion model was constructed. Optimization was carried out using the natural Bayesian algorithm, and the accuracy of identifying the XRD pattern was improved. The evaluation criteria were that when the root mean square error RMSE < 1.5 Mpa and the coefficient of determination R2 > 1 Mpa, the prediction requirements were met. After a series of trainings, the prediction fusion model could already predict the pozzolanic activity index of actual aluminosilicate solid wastes. When a new XRD pattern of aluminosilicate solid waste was input, the prediction fusion model would analyze and predict the pozzolanic activity index, and the difference between the predicted activity index and the actual activity index was within 1 Mpa.

[0096] Example 2:

[0097] XRD patterns of more than 500 aluminosilicate solid wastes (including fly ash, Beijing humus soil, etc.) were collected at a scanning angle of 0° to 80°. Among them, the particle size was controlled with a residue on a 45-μm sieve ≤ 12%. The strongest peak of the XRD diffraction peak of the test sample was required to be not less than 100. External standards such as alumina and silica were used to calibrate the XRD diffractometer, and the peak position of the scanning angle was within the range of ±0.0025° to ±0.0125° of its standard value. The samples were subjected to XRD tests at a scanning speed of 0.01° / step to 0.02° / step and 0.1 s / step to 100 s / step, and the residence time for each step needed to meet the requirement that the strongest counting point of the XRD pattern diffraction peak was not less than 10,000 Counts. At the same time, the 28-day strength performance data of these samples were tested or collected, and the scanning angle (2θ) and peak intensity (I) data, XRF data, and pozzolanic activity index (H) of 500 groups of clinker samples were obtained.

[0098] The collected data were cleaned and preprocessed to remove data containing missing values or obvious outliers. At the same time, with the diffraction peak intensity of the strongest peak of silica (2θ ≈ 26.6°) as 100, the XRD diffraction data were normalized, and the data were segmented and convolution-transformed to form a matrix. All the data were cleaned and preprocessed and divided into a training set and a test set in a ratio of 7:3 to construct a pozzolanic activity index prediction fusion model. Optimized using the natural Bayesian algorithm, the accuracy of identifying XRD patterns was improved. After a series of trainings, this prediction fusion model can already predict the pozzolanic activity index of actual aluminosilicate solid wastes. When a new XRD pattern of aluminosilicate solid waste is input, the prediction fusion model will analyze and predict the pozzolanic activity index, and the difference between the predicted activity index and the actual activity index is within the range of 0.76 Mpa.

[0099] Example 3

[0100] XRD patterns of more than 500 aluminosilicate solid wastes (including fly ash, Chaozhou humus soil, etc.) were collected at a scanning angle of 0° to 80°. Among them, the particle size was controlled with a residue on a 45-μm sieve ≤ 12%. The strongest peak of the XRD diffraction peak of the test sample was required to be not less than 100. External standards such as alumina and silica were used to calibrate the XRD diffractometer, and the peak position of the scanning angle was within the range of ±0.0025° to ±0.0125° of its standard value. The samples were subjected to XRD tests at a scanning speed of 0.01° / step to 0.02° / step and 0.1 s / step to 100 s / step, and the residence time for each step needed to meet the requirement that the strongest counting point of the XRD pattern diffraction peak was not less than 10,000 Counts. At the same time, the 28-day strength performance data of these samples were tested or collected, and the scanning angle (2θ) and peak intensity (I) data, particle size range, XRF data, and pozzolanic activity index (H) of 500 groups of clinker samples were obtained.

[0101] The collected data is cleaned and preprocessed to remove data containing missing values or obvious outliers. At the same time, the XRD diffraction data is normalized with the diffraction peak intensity of the strongest peak of silicon oxide (around 2θ = 26.6°) being 100, and the data is segmented and convolution-transformed to form a matrix. All the data is cleaned and preprocessed, divided into a training set and a test set in a ratio of 7:3, and a prediction fusion model for the pozzolanic activity index is constructed. Optimized using the Naive Bayes algorithm, the accuracy of identifying XRD patterns has been improved; after a series of trainings, this prediction fusion model can already predict the pozzolanic activity index of actual aluminosilicate solid waste. When a new XRD pattern of aluminosilicate solid waste is input, the prediction fusion model will analyze and predict the pozzolanic activity index, and the difference between the predicted activity index and the actual activity index is within the range of 0.43 Mpa.

[0102] Adopting the above technical solution, the present invention has the following beneficial effects:

[0103] Through the establishment of a large number of XRD pattern databases, the present invention improves the accuracy and stability of prediction. Using a multi-modal prediction model, it can achieve rapid prediction of the pozzolanic activity index of aluminosilicate solid waste, avoiding the cumbersome steps and time-consuming operations in traditional mechanical compressive strength testing methods, and significantly shortening the testing cycle; compared with traditional mechanical compressive strength testing methods, it reduces the dependence on expensive testing equipment and a large number of samples, and reduces the testing cost. This method is applicable to various types of materials, including raw materials such as metals and ceramics that can be tested by XRD patterns, has a wide range of application prospects, provides technical support for realizing the automation and digitization of clinker strength prediction, and is in line with the trend of modern intelligent manufacturing and Industry 4.0.

[0104] In addition, it should be noted that the present invention can be a method, a system, a device, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0105] A computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punched card or raised structures in grooves having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed as being a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0106] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to respective computing / processing devices, or can be downloaded to an external computer or an external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include a copper transmission cable, an optical fiber transmission, a wireless transmission, a router, a firewall, a switch, a gateway computer, and / or an edge server. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0107] The computer program instructions for performing the operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine - related instructions, microcode, firmware instructions, state - setting data, or source code or object code written in any combination of one or more programming languages, including object - oriented programming languages such as Smalltalk, C++, Python, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer - readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand - alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or, alternatively, may be connected to an external computer (e.g., through the Internet using an Internet service provider). In some embodiments, by using the state information of the computer - readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field - programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer - readable program instructions to implement various aspects of the present invention.

[0108] Aspects of the present invention are described herein with reference to the flowchart and / or block diagram of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowchart and / or block diagram, and the combinations of blocks in the flowchart and / or block diagram, can be implemented by computer - readable program instructions.

[0109] These computer - readable program instructions can be provided to a processor of a general - purpose computer, a special - purpose computer, or other programmable data - processing apparatus to produce a machine such that the instructions, when executed by the processor of the computer or other programmable data - processing apparatus, create a means for implementing the functions / acts specified in one or more blocks of the flowchart and / or block diagram. These computer - readable program instructions can also be stored in a computer - readable storage medium, which causes a computer, a programmable data - processing apparatus, and / or other devices to operate in a particular manner, so that the computer - readable medium storing the instructions includes a manufacture, which includes instructions for implementing various aspects of the functions / acts specified in one or more blocks of the flowchart and / or block diagram.

[0110] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices, causing a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other devices to generate a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other devices implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0111] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two consecutive blocks may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or by combinations of special purpose hardware and computer instructions. As will be apparent to those of ordinary skill in the art, implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0112] The embodiments of the present invention have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A method for predicting the pozzolanic activity of aluminosilicate solid waste, characterized in that, The steps are as follows: S1: Data preparation, collect the XRD diffraction data, XRF spectral analysis data and physical property factors of aluminosilicate solid waste, synchronously collect the pozzolanic activity index of aluminosilicate solid waste as the objective function data, and clean and preprocess the collected XRD diffraction data, XRF spectral analysis data and physical property factors; S2: Multimodal model construction, train 2 artificial neural networks ANN and extract features through the preprocessed physical property factors and preprocessed XRF spectral analysis data, train 1 convolutional neural network CNN and extract features through the preprocessed XRD diffraction data, and fuse the outputs of the artificial neural network ANN and the convolutional neural network CNN to construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste; S3: Hyperparameter optimization, use the Bayesian optimization algorithm for hyperparameter optimization to determine the best parameter settings of the fusion model to improve the performance of the fusion model; S4: Model evaluation and calibration, using the root mean square error RMSE and the coefficient of determination R 2 to evaluate the performance of the fusion model. When the root mean square error RMSE is within 1.5 and the coefficient of determination R 2 is greater than 1, the fusion model meets the requirements, and the fusion model is applied to predict the pozzolanic activity of aluminosilicate solid waste. Otherwise, the fusion model is calibrated until the fusion model meets the requirements.

2. The method according to claim 1, wherein The multimodal model construction in step S2 includes the training of sub-networks and the construction of the fusion model, and the training of sub-networks includes 2 ANN sub-networks and 1 CNN sub-network.

3. The method according to claim 2, wherein One of the 2 ANN sub-networks starts with an input layer that receives the elemental composition data obtained from XRF spectral analysis. After the input layer, there are two hidden layers responsible for data processing and transformation. The final output layer contains the trained features for providing the prediction result.

4. The method according to claim 2, wherein The input layer of the other ANN sub-network among the 2 ANN sub-networks receives the physical property factors of aluminosilicate solid waste, followed by two hidden layers and an output layer containing the trained features for outputting the prediction result.

5. The method according to claim 1, characterized in that, Coefficient of determination R in step S4 2 and root mean square error RMSE are calculated using the following formulas for various regression metrics, where is the actual pozzolanic activity index, is the predicted pozzolanic activity index, is the mean of the actual values, is the mean of the predicted values, and N is the total number of actual values:

6. The method according to claim 1, characterized in that, The method for correcting the fusion model in step S4 includes: Re-check the data cleaning and preprocessing process in step S1, adjust the model parameters or change the model fusion method.

7. The method according to claim 6, wherein The re-check of the data cleaning and preprocessing process in step S1 includes: Check whether there are mislabeled data or outliers that have not been properly processed; Check whether the normalization or standardization operation of the data is reasonable.

8. The method according to claim 6, characterized in that, Adjusting the model parameters includes: For the ANN sub-network and the CNN sub-network, adjust the number of network layers and the number of neurons in each layer.

9. The method according to claim 6, wherein Changing the model fusion method includes: Adopt different fusion strategies to integrate the outputs of the three sub-networks. Weighted fusion can be used to assign different weights according to the importance of the output of each sub-network to the final prediction result.

10. A prediction system for the pozzolanic activity of aluminosilicate solid waste, characterized in that, It includes: A data preparation module for collecting the XRD diffraction data, XRF spectral analysis data and physical property factors of aluminosilicate solid waste, synchronously collecting the pozzolanic activity index of aluminosilicate solid waste as the objective function data, and cleaning and preprocessing the collected XRD diffraction data, XRF spectral analysis data and physical property factors; The multimodal model construction module is used to train two artificial neural networks (ANNs) and extract features through the preprocessed physical property factors and preprocessed XRF spectral analysis data, train one convolutional neural network (CNN) and extract features through the preprocessed XRD diffraction data, and fuse the outputs of the artificial neural network (ANN) and the convolutional neural network (CNN) to construct a prediction fusion model for the pozzolanic activity performance of aluminosilicate solid waste; The hyperparameter optimization module is used to perform hyperparameter optimization using the Bayesian optimization algorithm to determine the optimal parameter settings of the fusion model to improve the performance of the fusion model; Model evaluation and calibration module, which is used to use the root mean square error RMSE and the coefficient of determination R 2 to evaluate the performance of the fusion model. When the root mean square error RMSE is within 1.5 and the coefficient of determination R 2 is greater than 1, the fusion model meets the requirements, and the fusion model is applied to predict the pozzolanic activity of aluminosilicate solid waste. Otherwise, the fusion model is calibrated until the fusion model meets the requirements.

Images