Cascading model-based cement strength combined prediction method and system

Abstract

The invention provides a cement strength combined prediction method and system based on a cascade model, and relates to the field of cement strength prediction.The method comprises the steps that initial component information of to-be-predicted cement is obtained; inputting the initial component information into the trained cascade model, and predicting the final cement strength; wherein the cascade model comprises a structure mapping model and a strength prediction model which are cascaded, the structure mapping model is firstly utilized to map structure information through a mapping relation from initial component information to final structure information, and then the strength prediction model is utilized to jointly predict the cement strength in combination with the initial component information and the mapped structure information; according to the method, the influence of the initial cement component information and the final structure information on the cement strength is considered at the same time, the problem of insufficient prediction precision caused by single type of information is avoided, and high-performance cement strength prediction is realized.

Description

Technical Field

[0001] The present invention relates to the field of cement strength prediction, and in particular to a method and system for jointly predicting cement strength based on a cascade model. Background Art

[0002] Cement is an important building material and plays an irreplaceable role in modern infrastructure construction. Cement is not only widely used in engineering projects such as roads, bridges, tunnels, and high-rise buildings, but also shows its key structural support role in many fields such as ocean engineering and water conservancy facilities. Among them, the strength performance of cement materials is the core factor to ensure the safety and durability of buildings and has a decisive impact on the project quality. The strength of cement usually needs to be cured under standard curing conditions for 28 days, and then the strength test is carried out through manual experiments. The long curing time and manual experiment tests will consume a large amount of human and material resources, and at the same time affect the market sales and engineering construction of cement materials. Therefore, more and more research begins to use computer technology to quickly predict the cement strength to improve the detection efficiency of cement strength.

[0003] Currently, the computer technology-based cement strength prediction methods mainly start from two aspects. On the one hand, it is based on the information of the initial composition of cement (abbreviated as "initial state"). The characteristic of this type of technology is that it can directly skip the 28-day standard curing environment of cement to obtain the strength performance, but it cannot consider the influence of the change of cement structure information on the strength performance during 28 days. On the other hand, it is based on the state at 28 days (abbreviated as "final state", where the "final state" does not mean that the cement completely stops hydration or all states no longer change, but refers to the state where the performance of cement basically reaches stability at the 28th day) information for prediction. For example, the strength performance is obtained according to the microscopic structure characteristic information in the final state. The characteristic of this type of technology is that it is directly based on the state at 28 days and can obtain more stable performance prediction results, but the curing time of 28 days and the measurement and characterization costs of information such as microscopic structure are high.

[0004] The strength performance of cement is jointly affected by the initial state and the final state. Lack of consideration of any aspect will affect the prediction result of cement strength performance. If, in order to ensure the prediction accuracy, directly measure the initial state information and the final state information and then make a prediction, it will still require a long curing time and high acquisition cost every time a prediction is made, and it is impossible to ensure the low-cost and rapid prediction of cement strength.

[0005] Therefore, the existing technology still cannot achieve rapid, low-cost, and comprehensive cement strength prediction, and cannot avoid the influence of long curing time and high acquisition cost on the prediction performance. Summary of the Invention

[0006] To solve the above problems, the present invention proposes a method and system for jointly predicting cement strength based on a cascade model, which takes into account the influence of both the initial cement component information and the final structure information on the cement strength, avoids the problem of insufficient prediction accuracy caused by a single type of information, and realizes high-performance prediction of cement strength.

[0007] According to some embodiments, the present invention adopts the following technical solutions: A method for jointly predicting cement strength based on a cascade model, comprising: Obtaining the initial component information of the cement to be predicted; Inputting the initial component information into the trained cascade model to predict the final cement strength; Wherein, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, the structure mapping model is used to map the structure information through the mapping relationship from the initial component information to the final structure information, and then the strength prediction model is used to jointly predict the cement strength by combining the initial component information and the mapped structure information.

[0008] According to some embodiments, the present invention adopts the following technical solutions: A system for jointly predicting cement strength based on a cascade model, comprising: A component acquisition module configured to: obtain the initial component information of the cement to be predicted; A strength prediction module configured to: input the initial component information into the trained cascade model to predict the final cement strength; Wherein, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, the structure mapping model is used to map the structure information through the mapping relationship from the initial component information to the final structure information, and then the strength prediction model is used to jointly predict the cement strength by combining the initial component information and the mapped structure information.

[0009] According to some embodiments, the present invention adopts the following technical solutions: A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, it implements the method for jointly predicting cement strength based on a cascade model as described above.

[0010] According to some embodiments, the present invention adopts the following technical solutions: A non-transitory computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, they implement the method for jointly predicting cement strength based on a cascade model as described above.

[0011] According to some embodiments, the present invention adopts the following technical solutions: An electronic device, comprising: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes a cement strength joint prediction method based on a cascade model as described above.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention simultaneously considers the influence of initial cement component information and final structural information on cement strength, constructs a cement strength joint prediction method, avoids the problem of insufficient prediction accuracy caused by single-type information, solves the influence of long curing time and high acquisition cost on prediction performance, and realizes fast, low-cost, and comprehensive cement strength prediction.

[0013] The present invention establishes a mapping between the initial component information and the final structural information, and combines the initial component information with the mapped structural information and injects them into the strength prediction model together, so that the strength prediction model only needs the initial component information during use to achieve accurate prediction of cement strength. This process avoids the need to collect the cement structural information again after 28 days, and realizes fast and low-cost prediction of cement strength.

[0014] The multi-modal data acquisition of cement is costly, so the quality and diversity of the data used for model training are insufficient, which greatly limits the prediction effect of deep learning-based models. Each sub-model in the cascade model proposed by the present invention adopts a KAN structure, which can still ensure the generalization ability and non-linear modeling ability of cement strength prediction with a small number of samples, and combines a multi-layer perceptron MLP layer to stabilize model training. Considering the sensitivity differences of sub-models at different stages to cement characteristic information, the present invention introduces an attention mechanism into the cascade structure to perform weighted fusion on the outputs of each sub-model, thereby improving the overall prediction performance, especially maintaining a better performance in the case of scarce data. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0016] Figure 1 It is a flowchart of the method for Embodiment 1. Figure 2 It is a schematic diagram of the network structure of the cascade KAN model for Embodiment 1. Figure 3 It is a schematic diagram of the training of the cascade KAN model for Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0018] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0019] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "comprise" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0020] Embodiment 1 An embodiment of the present invention provides a method for jointly predicting the strength of cement based on a cascade model, including: Step 1: Obtain the initial component information of the cement to be predicted; Step 2: Input the initial component information into the trained cascade model to predict the final strength of the cement; Among them, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, use the structure mapping model to map the structure information through the mapping relationship from the initial component information to the final structure information, and then use the strength prediction model to jointly predict the cement strength by combining the initial component information and the mapped structure information.

[0021] As an embodiment, the method for jointly predicting the strength of cement based on a cascade model of the present invention takes into account the influence of both the initial cement component information and the final structure information on the cement strength, avoids the problem of insufficient prediction accuracy caused by a single type of information, and realizes high-performance prediction of cement strength. The following mainly describes the specific implementation process from the perspective of model construction, as Figure 1 shown, including the following steps: S1: Obtain cement component, structure and strength data and perform data processing on them to construct a model data set, including: S11: Prepare cement samples with multiple batches of different components and cure them in a standard curing environment for 28 days.

[0022] Among them, the initial component information of the cement is represented as X = [x1, x2, x3, x4, x5, x6, x7, x8, x9, x 10 , x 11 , x 12, which contains 12 dimensions, namely calcium oxide value x1, aluminum oxide value x2, iron oxide value x3, magnesium oxide value x4, silicon dioxide value x5, potassium oxide value x6, sodium oxide value x7, specific surface area value x8, alkali content value x9, loss on ignition value x 10 , residue value x 11 , sulfur trioxide value x 12 .

[0023] After the cement of each component is prepared, it is first formed in a mold for one day, and then demolded and continued to be cured for 28 days.

[0024] S12: Respectively obtain the microscopic structure image data, porosity, pore size distribution and strength data of the cement sample at 28 days. Specifically, in this example, a micro-computed tomography technique is used to obtain the microscopic structure image of the cement sample; the porosity and pore size distribution are obtained by image analysis. First, the image is subjected to threshold segmentation processing to obtain a binary image of the pores, and then the porosity is calculated; the segmented pores are assumed to be circular, and the equivalent circle diameter is calculated based on the space area they occupy. Finally, the pore size distribution of the microscopic structure is obtained through statistics; finally, a universal testing machine is used to obtain the cement strength data.

[0025] S13: Denoise the microscopic structure image data, and extract the microscopic structure image features based on the gray-level co-occurrence matrix.

[0026] Use a median filter operator to denoise the image, and the filter window is 5×5.

[0027] Subsequently, calculate the gray-level co-occurrence matrix of the image (this matrix contains information such as the gray level and texture of the image), and calculate the characteristic quantities based on the gray-level co-occurrence matrix. The characteristic quantity F includes contrast f1, energy f2, entropy f3, inverse variance f4, and correlation f5. The microscopic structure image features are represented as F = [f1, f2, f3, f4, f5], specifically: For a cement microscopic structure image with a resolution of H×W, its gray level is L (usually L = 256), and the gray-level co-occurrence matrix is defined as: (1) (2) (3) Among them, represents the probability that pixels with gray value i and pixels with gray value j appear simultaneously at a given spatial distance d and direction θ, where i, j = 0, 1, 2, …, L - 1; d is the pixel spacing, θ is the direction (usually 0°, 45°, 90°, 135°), (x, y) is the pixel coordinate, and I(x, y) is the gray value at this coordinate position.

[0028] To reduce the influence of image size and ensure the comparability of eigenvalues, the calculated gray-level co-occurrence matrix is normalized to obtain the normalized gray-level co-occurrence matrix. , (4) Contrast is used to measure the severity of local changes in the microstructure image and reflects the clarity of the texture. Its calculation method is as follows: (5) Energy reflects the uniformity of the gray-level distribution in the microstructure image. Its calculation method is as follows: (6) Entropy reflects the randomness of the texture in the microstructure image. Its calculation method is as follows: (7) Inverse variance reflects the local homogeneity of the microstructure image. Its calculation method is as follows: (8) Correlation reflects the gray-level correlation in the microstructure image. Its calculation method is as follows: (9) (10) (11) (12) (13) If there are gray-level co-occurrence matrices in multiple directions, the final result of the eigenvalues will take their average value.

[0029] S14: Integrate the initial components of cement, the porosity at 28 days, the pore size distribution, the characteristics of the microstructure image, and the cement strength to form a model data set.

[0030] Among them, the cement strength is denoted as CS. The porosity is denoted as s1, the pore size distribution forms data by using the segmented statistical proportion method, and the proportion of different pore sizes is statistically counted in ten intervals, denoted as [s2, s3, s4, s5, s6, s7, s8, s9, s 10 , s 11 , and the pore structure information is denoted as S = [s1, s2, s3, s4, s5, s6, s7, s8, s9, s 10 , s 11 .

[0031] That is to say, taking the initial component information X, pore structure characteristics S, and microstructural image characteristics F as the cement sample data, and CS as the label, a model dataset for training the model is formed, and it is divided into a training set, a validation set, and a test set according to the ratio of 8:1:1.

[0032] S2: Design a cascaded model based on the KAN network architecture, establish a mapping relationship between the initial component information of cement and the cement structure information, inject the mapped structure information as the prediction intermediate value into the model, and jointly predict the cement strength through the initial components of cement and the mapped structure information.

[0033] First, give a brief introduction to KAN: KAN (Kolmogorov-Arnold Network) is a neural network architecture inspired by the Kolmogorov-Arnold Representation Theorem; this theorem states that any multivariate continuous function can be represented as a combination of a finite number of univariate functions; the design goal of KAN is to improve the expressive power and interpretability of neural networks by explicitly using this mathematical theory.

[0034] In this step, design a cascaded model based on the KAN network architecture. Figure 2 This is the structural diagram of the cascaded KAN model provided for this embodiment. Establish a mapping relationship between the initial components of cement and the cement structure information (characterized by pore structure characteristics and microstructural image characteristics), inject the structure information as the prediction intermediate value into the model, and jointly predict the cement strength through the initial components of cement and the mapped structure information, including: S21: Construct a KAN-based pore structure feature mapping model (hereinafter referred to as the "KAN_A model"), use a spline function to implement a learnable non-linear function, this model establishes a connection between the initial component information of cement and the porosity and pore size distribution, the input is the initial component X = [x1, x2,..., x N , and the output is the pore structure information S = [s1, s2,..., s M , the function is shown as follows. (14) Among them, N = 12, corresponding to 12 initial component information, M = 11, corresponding to 11 pore structure characteristics, the input layer dimension of the KAN_A model is 12, and the output layer dimension is 11.

[0035] In addition, the number of KAN layers in the KAN_A model is 3, each layer uses a B-spline function for fitting, the degree in the B-spline is set to 3, and a layer of MLP network is added before the output layer of the model to enhance the training stability.

[0036] S22: Construct a microscopic structure image feature mapping model based on KAN (hereinafter referred to as the "KAN_B model"), and use a spline function to implement a learnable non-linear function. This model establishes a connection between the initial cement component information and the microscopic structure image features. The input is the initial component X, and the output is the microscopic structure image features F = [f1, f2, …, f T , and the function is shown as follows: (15) Among them, T = 5, corresponding to 5 image feature values. The input layer dimension of the KAN_B model is 12, and the output layer dimension is 5.

[0037] In addition, the number of KAN layers in the KAN_B model is 3, and each layer uses a B-spline function for fitting. The degree in the B-spline is set to 3. Similarly, an MLP network is added before the output layer of the model.

[0038] S23: Construct a strength prediction model based on KAN and the attention mechanism (hereinafter referred to as the "KAN_C model"), and use a spline function to implement a learnable non-linear function. Combine the prediction information S and F output by the KAN_A model and the KAN_B model with the initial component information X and jointly inject them into the KAN_C model, and use the attention mechanism to process the joint input to achieve the joint prediction of the cement strength CS. The function is shown as follows: (16) Among them, the input layer dimension of KAN_C is 28, which comes from 12 initial component information, 5 microscopic structure image feature values, and 11 pore structure information respectively. The output layer dimension is 1, corresponding to the cement strength. The number of KAN layers in the KAN_C model is 3, and each layer uses a B-spline function for fitting. The degree in the B-spline is set to 3.

[0039] An attention layer is added after the input layer of the model to make the model pay more attention to the parts that are more useful for the current sample, and an MLP network is added before the output layer of the model.

[0040] S24: Through the cascading of the KAN_A model, the KAN_B model, and the KAN_C model, realize the joint prediction of the cement strength based on the cascaded KAN model. The input of the cascaded KAN model is only the initial cement component information X , and the function is shown as follows: (17) S3: Construct a joint loss function, and optimize the weight parameters of the model network through the backpropagation algorithm; through the iterative training of the cascaded model, obtain a training model with the best prediction effect, including: S31: Figure 3Schematic diagram of model training provided in this embodiment. The KAN_A model, KAN_B model, and KAN_C in the cascade model are trained and optimized. The overall loss function comes from the losses of the three KAN models, namely the loss function of the KAN_A model loss_A , the loss function of the KAN_B model loss_B , and the loss function of the KAN_C model loss_C . Each loss function has a separate weight coefficient a, b, c . The overall loss function is expressed as: (18) where loss_A represents the difference degree between the true pore feature value r_S and the predicted pore feature value p_S , and is measured by the mean square error, expressed as: (19) loss_B represents the difference degree between the true image feature value r_F and the predicted image feature value p_F , and is measured by the mean square error, expressed as: (20) loss_C represents the difference degree between the true cement strength value r_CS and the predicted cement strength value p_CS , and is measured by the mean square error, expressed as: (21) In this embodiment, the weight coefficients of the three loss functions are respectively set as a = 1, b = 1, and c = 1.

[0041] S32: Input the training data in the model dataset into the model, sequentially execute the forward propagation process of each stage model, obtain the intermediate prediction result and the final output, calculate the joint loss function and execute the backpropagation algorithm to update the model weights.

[0042] Iteratively train the model on the training set, evaluate the prediction performance metrics on the validation set, record the performance of each round and save the model parameter weights corresponding to the best prediction performance, and finally test on the test set. In this embodiment, the prediction performance metric is the mean absolute error.

[0043] During the training process, the Adam optimizer is used to update the network weights, and the learning rate is set to 0.01. loss_A The weights of the KAN_A model are optimized. loss_B The weights of the KAN_B model are optimized. loss_C The weights of the KAN_C model are optimized.

[0044] S4: Based on the initial component information of the cement, use the trained cascade model to predict the cement strength.

[0045] Example 2 In one embodiment of the present invention, a cement strength joint prediction system based on a cascade model is provided, including: A component acquisition module, configured to: acquire the initial component information of the cement to be predicted; A strength prediction module, configured to: input the initial component information into the trained cascade model to predict the final cement strength; Wherein, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, use the structure mapping model to map the structure information through the mapping relationship from the initial component information to the final structure information, and then use the strength prediction model to jointly predict the cement strength by combining the initial component information and the mapped structure information.

[0046] Example 3 In one embodiment of the present invention, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the above-mentioned cement strength joint prediction method based on a cascade model.

[0047] Example 4 In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, and the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the above-mentioned cement strength joint prediction method based on a cascade model is implemented.

[0048] Example 5 In one embodiment of the present invention, an electronic device is provided, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes and implements the above-mentioned cement strength joint prediction method based on a cascade model.

[0049] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to this embodiment. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processing machine, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1one or more processes and / or blocks Figure 1 means for the functions specified in one or more blocks

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the functions specified in one or more processes Figure 1 one or more processes and / or blocks Figure 1 steps for the functions specified in one or more blocks

[0051] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or variations that can be made by those skilled in the art without creative efforts are still within the scope of protection of the present invention.

Claims

1. A combined prediction method for cement strength based on a cascade model, characterized in that Including: Obtain the initial component information of the cement to be predicted; Input the initial component information into the trained cascade model to predict the final cement strength; Wherein, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, use the structure mapping model to map the structure information through the mapping relationship from the initial component information to the final structure information, and then use the strength prediction model to jointly predict the cement strength by combining the initial component information and the mapped structure information.

2. The cement strength combined prediction method based on a cascade model according to claim 1, characterized in that The structure information includes pore structure characteristics and microstructural image characteristics. Among them, the pore structure characteristics are characterized by porosity and pore size distribution, and the microstructural image characteristics are characterized by the gray-scale characteristics and texture characteristics of the image.

3. The cement strength joint prediction method based on a cascade model according to claim 2, wherein, The structure mapping model includes two sub-models: a pore structure feature mapping model and a microstructural image feature mapping model; Among them, the pore structure feature mapping model takes the initial component information as the input and the pore structure characteristics as the output, and the microstructural image feature mapping model takes the initial component information as the input and the microstructural image characteristics as the output.

4. The cement strength combined prediction method based on a cascade model according to claim 2, wherein The strength prediction model takes the initial component information, pore structure characteristics and microstructural image characteristics as the input and the cement strength as the output.

5. The combined prediction method for cement strength based on a cascade model according to claim 1, wherein The cascade model is trained through the constructed model data set. The construction process of the model data set is as follows: Prepare cement samples with different components in multiple batches and cure them in a standard curing environment for 28 days; Respectively obtain the microstructural images, porosity, pore size distribution and strength data of the cement samples at 28 days; Preprocess the microstructural images and extract the gray-scale characteristics and texture characteristics as the microstructural image characteristics; Integrate the initial component information, porosity, pore size distribution, microstructural image characteristics and cement strength to form a model data set.

6. The cement strength combined prediction method based on a cascade model according to claim 3, wherein, The pore structure feature mapping model, the microstructural image feature mapping model and the strength prediction model all adopt the improved KAN network. The improvement is to add a layer of MLP network before the output layer of the KAN network. The strength prediction model also introduces an attention mechanism after the input layer of the KAN network to perform weighted fusion on multiple inputs.

7. A cement strength joint prediction system based on a cascade model, characterized in that, Including: A component acquisition module configured to: obtain the initial component information of the cement to be predicted; A strength prediction module configured to: input the initial component information into the trained cascade model to predict the final cement strength; Wherein, the cascade model includes a cascaded structure mapping model and a strength prediction model. First, use the structure mapping model to map the structure information through the mapping relationship from the initial component information to the final structure information, and then use the strength prediction model to jointly predict the cement strength by combining the initial component information and the mapped structure information.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements a method for jointly predicting cement strength based on a cascade model according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, it implements a method for jointly predicting cement strength based on a cascade model according to any one of claims 1-6.

10. An electronic device, characterized in that, Including: A processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes a method for jointly predicting the strength of cement based on a cascade model as described in any one of claims 1-6.

Images