Method and system for predicting shearing strength of rusted RC beam by fusing domain knowledge

Abstract

The invention belongs to the technical field of artificial intelligence, discloses a rusted RC beam shear strength prediction method and system fused with domain knowledge, and solves the problems of low accuracy, insufficient transparency and poor interpretability in rusted RC beam shear strength prediction in the prior art. According to the specific scheme, the rusted RC beam shear strength prediction method fusing domain knowledge comprises the steps that input parameters related to the rusted RC beam shear strength are determined, and the input parameters serve as input to generate a rusted RC beam shear strength explicit calculation formula through symbolic regression analysis on the basis of a gene expression programming algorithm; taking an existing empirical model as domain knowledge, fusing the existing empirical model with an explicit calculation formula based on symbolic regression through a decision tree algorithm, and constructing a rusted RC beam shear strength prediction model based on physical information machine learning; and carrying out visual analysis on the decision-making process of the rusted RC beam shear strength prediction model based on physical information machine learning.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting the shear strength of rusted RC beams by integrating domain knowledge. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Reinforced concrete (RC) beams, as fundamental structural components, play a crucial role in load transfer. Since most existing buildings are directly exposed to the natural environment, the internal steel reinforcement of these beams corrodes over time due to the combined effects of accumulated service life and environmental erosion. Existing research indicates that as the degree of steel corrosion intensifies, the failure mode of corroded RC beams shifts from bending failure to shear failure. Shear failure, a brittle failure, can lead to sudden structural collapse. Therefore, accurately predicting the shear strength of corroded RC beams is of great significance for ensuring structural safety.

[0004] In the study of shear strength of corroded RC beams, experiments are an important means of obtaining real data, usually using a four-point bending loading method, providing irreplaceable support for theoretical research. However, experimental research is time-consuming, requires a lot of manpower and resources, and is difficult to cover complex scenarios with multiple variables and multiple working conditions.

[0005] Therefore, while emphasizing experimental research, existing technologies have analyzed the shear strength of corroded RC beams based on mechanism-driven approaches. In terms of theoretical models, there are prediction models for the shear strength of corroded RC beams based on equivalent truss theory, limit equilibrium theory, and modified compression field theory. Due to the complexity of the shear mechanism of corroded RC beams, empirical models often rely on idealized assumptions and struggle to account for the influence of multiple factors. This limitation leads to significant dispersion in the calculation results, and the prediction accuracy needs improvement. When dealing with multi-factor coupled problems, machine learning (ML) can uncover the nonlinear correlations between different parameters without making idealized assumptions, thus overcoming the limitations of empirical models in solving complex problems. In recent years, the application of ML in civil engineering has become increasingly widespread. In the prediction of the shear strength of corroded RC beams, some models use gradient boosting regression trees (GBRT) to establish shear strength prediction models, while others use artificial neural network algorithms. Both machine learning models achieved good predictive results, but they suffer from insufficient transparency and poor interpretability, classifying them as "black box models." They cannot accurately describe the basic prediction mechanism, nor can they establish an explicit formula between the input and shear strength. Some models use gene expression programming to build predictive models for the shear strength of corroded RC beams. While this improves transparency, it does not fully utilize existing domain knowledge, and the accuracy of the prediction results needs further improvement. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for predicting the shear strength of corroded RC beams that integrates domain knowledge, enabling accurate, efficient, and interpretable prediction of the shear strength of corroded RC beams.

[0007] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for predicting the shear strength of corroded RC beams that integrates knowledge in the field, including the following: Step 1: Establish a dataset by collecting shear strength test data of corroded RC beams; Step 2: Select design parameters related to the geometric properties, material strength, and degree of steel corrosion of the corroded RC beam, and determine the input and output parameters of the model; Step 3: Using the input parameters as input, based on the gene expression programming algorithm, generate an explicit calculation formula for the shear strength of the corroded RC beam through symbolic regression analysis; Step 4: Using the existing experience model as domain knowledge, the decision tree algorithm is used to integrate the existing experience model with the explicit calculation formula for the shear strength of the rusted RC beam in Step 3 to construct a PIML-CBSS model for predicting the shear strength of the rusted RC beam based on physical information machine learning, and the hyperparameters of the model are verified and determined. Step 5: Visualize and analyze the decision-making process of the PIML-CBSS model, and predict the shear strength of the corroded RC beam.

[0008] As a further technical solution, the design parameters mentioned in step 2 are obtained from the collected shear strength test data of the rusted RC beams. The shear strength test data of the rusted RC beams includes the concrete compressive strength. Cross-sectional width Cross-sectional height , slit span length Longitudinal reinforcement ratio Stirrup reinforcement ratio longitudinal reinforcement yield strength Yield strength of stirrups Stirrup spacing , shear span ratio Longitudinal reinforcement section loss rate Stirrup section loss rate Effective height of cross section Shear strength .

[0009] As a further technical solution, the input parameters in step 2 include the shear span ratio. Effective height-to-width ratio of the cross section Longitudinal reinforcement index Horizontal reinforcement index Longitudinal reinforcement section loss rate and stirrup section loss rate The output parameter is shear strength. .

[0010] As a further technical solution, in step 3: based on the gene expression programming algorithm, the splitting-to-stretch ratio is determined through symbolic regression analysis. Effective height-to-width ratio of the cross section Longitudinal reinforcement index Horizontal reinforcement index Longitudinal reinforcement section loss rate Stirrup section loss rate With shear strength The relationship between these factors generates an explicit formula for calculating the shear strength of rusted RC beams without dimensionless imbalance, thus achieving efficient transformation from data to knowledge.

[0011] As a further technical solution, step 4 specifically involves: using the calculation results of the empirical formula as the baseline physical feature, introducing the explicit calculation formula of symbolic regression mining as the enhanced feature, constructing a composite feature space, and establishing a PIML-CBSS model based on the decision tree algorithm.

[0012] As a further technical solution, the gene expression programming algorithm processing in step 3 includes the following: Information related to the shear strength of the corroded RC beam is encoded to generate an initial population of candidate solutions; Calculate the fitness of each individual to evaluate its quality. If the termination condition is met, output the optimal result. If not, retain the best individual in the population and generate new individuals through crossover and mutation operations. After the new individuals are verified, they are integrated with the best individuals to form a new population. Then, the next iteration cycle begins, repeating the entire process of fitness evaluation, selection, crossover, mutation, and population renewal. This cycle continues until the termination condition is met.

[0013] As a further technical solution, in step 4, the hyperparameters of the model are determined by Bayesian optimization combined with ten-fold cross-validation.

[0014] Secondly, this invention also discloses a system for predicting the shear strength of rusted RC beams that integrates knowledge in the field, comprising: The dataset creation module is configured to create a dataset by collecting test data on the shear strength of corroded RC beams. The filtering module is configured to filter design parameters related to the geometric properties, material strength, and degree of steel corrosion of the corroded RC beam, and to determine the input and output parameters of the model. The calculation formula generation module is configured to take input parameters as input, and generate an explicit calculation formula for the shear strength of the rusted RC beam based on the gene expression programming algorithm and symbolic regression analysis. The prediction model building module is configured to use the existing experience model as domain knowledge, and through the decision tree algorithm, integrate the existing experience model with the explicit calculation formula of the shear strength of the rusted RC beam in step 3 to build the PIML-CBSS model, and verify and determine the hyperparameters of the model. The visualization and prediction module is configured to visualize and analyze the decision-making process of the PIML-CBSS model and predict the shear strength of the rusted RC beam using the PIML-CBSS model.

[0015] Thirdly, the present invention also provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method for predicting the shear strength of rusted RC beams that integrates field knowledge as described above in this invention.

[0016] Fourthly, the present invention also provides an electronic device, which adopts the following technical solution: An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps in the method for predicting the shear strength of rusted RC beams that integrates field knowledge as described above.

[0017] The beneficial effects of the present invention are as follows: 1) This invention determines the input parameters, generates a formula for calculating the shear strength of rusted RC beams based on gene expression programming algorithm and symbolic regression analysis, and realizes the efficient transformation from data to knowledge; it uses existing empirical models as domain knowledge, and integrates existing empirical models with explicit calculation formulas based on symbolic regression through decision tree algorithm to construct PIML-CBSS model, and visualizes and analyzes the decision process of the model to achieve accurate, efficient and interpretable prediction of the shear strength of rusted RC beams.

[0018] 2) This invention employs a decision tree algorithm to construct the PIML-CBSS model. This model retains the interpretability advantage of the calculation formula (explicit formula) for the shear strength of rusted RC beams. By using visualization and analysis methods such as decision tree splitting rules and node feature importance ranking, the prediction results of the model are no longer a "black box" output, but a reasoning process with clear physical meaning and decision logic. This provides a more reliable and easy-to-understand theoretical basis and technical support for the prediction of the shear strength of rusted reinforced concrete beams in engineering practice.

[0019] 3) The input parameters in this invention include shear span ratio, effective height-to-width ratio of the cross section, longitudinal reinforcement index, transverse reinforcement index, longitudinal reinforcement section loss rate, and stirrup section loss rate. These are combinations of different component design parameters, rather than single design parameters. This avoids the problem of size effect and also avoids the problem of dimensional imbalance in the generated explicit calculation formula. This ensures that the model can maintain stable prediction accuracy in the shear strength prediction task of rusted RC beams of different scales. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart of a method for predicting the shear strength of rusted RC beams that incorporates field knowledge, based on one or more embodiments of the present invention.

[0022] Figure 2 This is a flowchart of the gene expression programming algorithm in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge, according to one or more embodiments of the present invention.

[0023] Figure 3 This is a partial decision-making process of the PIML-CBSS model in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge, according to one or more embodiments of the present invention.

[0024] Figure 4(a) is a comparison of the calculation results of the PIML-CBSS model and the Huo et al. model (hereinafter referred to as the Huo model) in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0025] Figure 4(b) is a comparison of the calculation results of the PIML-CBSS model and the Lu et al. model (hereinafter referred to as the Lu model) in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0026] Figure 4(c) is a comparison of the calculation results of the PIML-CBSS model and the Xue et al. model (hereinafter referred to as the Xue model) in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0027] Figure 4(d) is a comparison of the calculation results of the PIML-CBSS model and Xu et al. (hereinafter referred to as the Xu model) in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0028] Figure 4(e) is a comparison of the calculation results of the PIML-CBSS model and Li et al. (hereinafter referred to as the Li model) in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0029] Figure 5 This is a comparison chart of the comprehensive performance of the PIML-CBSS model and various empirical models in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge, based on one or more embodiments of the present invention.

[0030] Figure 6 This is a comparison chart showing the performance improvement ratio of the PIML-CBSS model and various empirical models in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention.

[0031] Figure 7(a) is a comparison of the calculation results of the PIML-CBSS model and the RF model in a method for predicting the shear strength of a rusted RC beam that integrates domain knowledge according to one or more embodiments of the present invention.

[0032] Figure 7(b) is a comparison of the calculation results of the PIML-CBSS model and the XGBoost model in a method for predicting the shear strength of a rusted RC beam that integrates domain knowledge according to one or more embodiments of the present invention.

[0033] Figure 7(c) is a comparison of the calculation results of the PIML-CBSS model and the GEP model in a method for predicting the shear strength of a rusted RC beam that integrates domain knowledge according to one or more embodiments of the present invention.

[0034] Figure 8 This is a comparison chart of the comprehensive performance of the PIML-CBSS model and various machine learning models in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge, based on one or more embodiments of the present invention.

[0035] Figure 9This is a comparison chart showing the performance improvement ratio of the PIML-CBSS model and various machine learning models in a method for predicting the shear strength of rusted RC beams that integrates domain knowledge according to one or more embodiments of the present invention. Detailed Implementation

[0036] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0037] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless otherwise expressly indicated by the invention, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. As introduced in the background section, the existing methods for predicting the shear strength of corroded RC beams suffer from low accuracy, insufficient transparency, and poor interpretability. In order to solve the above technical problems, this invention proposes a method for predicting the shear strength of corroded RC beams that integrates domain knowledge.

[0038] A typical embodiment of the present invention is described below. Figure 1 As shown in the figure, this embodiment proposes a method for predicting the shear strength of corroded RC beams by integrating domain knowledge, including the following: 1) Dataset Establishment: Shear test data of corroded RC beams were collected to establish a dataset, providing data support for the establishment of a predictive model for the shear strength of corroded RC beams. The collected shear test data of corroded RC beams includes concrete compressive strength... Cross-sectional width Cross-sectional height , slit span length Longitudinal reinforcement ratio Stirrup reinforcement ratio longitudinal reinforcement yield strength Yield strength of stirrups Stirrup spacing , shear span ratio Longitudinal reinforcement section loss rate Stirrup section loss rate Effective height of cross section Shear strength There are a total of 14 parameters.

[0039] 2) Input Variable Selection: Due to experimental limitations, most of the shear test data for corroded RC beams in the dataset comes from scaled-down models. Directly constructing a machine learning prediction model might introduce size effects. Therefore, a dimensionless input variable is used to establish an explicit calculation formula for the shear strength of corroded RC beams, making the formula independent of changes in the unit system and avoiding dimensional imbalance issues. Since the shear strength of corroded RC beams is related to three aspects: geometric properties, material strength, and the degree of steel corrosion, six dimensionless input parameters are ultimately determined, namely the shear span ratio. Effective height-to-width ratio of the cross section Longitudinal reinforcement index Horizontal reinforcement index Longitudinal reinforcement section loss rate Stirrup section loss rate The selection of these dimensionless parameters takes into account the influence mechanism of the shear strength of reinforced concrete beams. For example, the shear span ratio reflects the influence of the stress mode on the shear strength, the longitudinal reinforcement index and the transverse reinforcement index reflect the contribution of the reinforcement, while the longitudinal reinforcement section loss rate and the stirrup section loss rate quantify the degree of corrosion damage, which can enhance the correlation between the model prediction results and the actual project.

[0040] 3) Data Preprocessing: The 158 sets of dimensionless input parameters were randomly divided into training and testing sets in a 7:3 ratio. 110 sets of dimensionless input parameters were used for training, and 48 sets were used for testing. To improve the accuracy and stability of the model, the data was standardized using z-scores (standard scores) before model training. This method eliminates the influence of differences in dimensions between different features, helping to ensure the accuracy of the results.

[0041] 4) Model establishment: based on shear span ratio Effective height-to-width ratio of the cross section Longitudinal reinforcement index Horizontal reinforcement index Longitudinal reinforcement section loss rate Stirrup section loss rate With six dimensionless parameters as input, based on the gene expression programming (GEP) algorithm, and through symbolic regression analysis, an explicit calculation formula for the shear strength of rusted RC beams is generated, realizing the efficient transformation from data to knowledge; For details, please refer to Figure 2 As shown, the processing steps of the Gene Expression Programming (GEP) algorithm include the following: Information related to the shear strength of the corroded RC beam is encoded to generate an initial population of candidate solutions; The fitness of each individual is calculated to evaluate its quality. If the termination condition is met, the optimal result is output. If not, the optimal individual in the population is retained first, and then new individuals are generated through crossover and mutation operations. After the new individuals are verified, they are integrated with the optimal individuals to form a new population. After updating the evolutionary generation, the fitness calculation and other steps are repeated to continuously iterate until the termination condition is met.

[0042] Based on a gene expression programming algorithm, a formula for calculating the shear strength of a corroded RC beam is generated through symbolic regression analysis. The formula is as follows: (1) In the formula:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054] In the formula, x 0 represents the shear span ratio. , x 1 represents the effective height-to-width ratio of the cross section. , x 2 represents the longitudinal reinforcement index. , x 3 represents the transverse reinforcement index. , x 4 represents the longitudinal reinforcement section loss rate. , x 5 represents the stirrup section loss rate. , V The value of shear strength for a corroded RC beam is given by the formula above. A , B ,C , D , E , F , G , H , I , J , K , L It simply represents the abbreviation of the corresponding formula and has no special meaning.

[0055] Existing empirical models are used as domain knowledge and fused with explicit calculation formulas based on symbolic regression through decision tree algorithm. Empirical formulas are used as benchmark physical features, as shown in formulas (2) to (6) below. Explicit calculation formulas mined by symbolic regression are introduced as enhanced features to construct a composite feature space. Based on the decision tree algorithm, a PIML-CBSS model is established, and the decision-making process of the model is visualized and analyzed. At the same time, the hyperparameters of the model are determined by Bayesian optimization combined with ten-fold cross-validation. The PIML-CBSS model constructed through physical-data dual-drive can achieve accurate, efficient, and interpretable prediction of the shear strength of rusted RC beams.

[0056] In the formula, and These are two reduction factors that take into account the weight loss of longitudinal reinforcement and stirrups; and These are the shear bearing capacities of concrete due to diagonal tension failure and shear-compression failure, respectively; R is the width of the loading plate, which is considered a constant of 87.2 mm. It is the shear span of the corroded RC beam; j is usually 1 / 1.15; It is the longitudinal steel reinforcement ratio after corrosion; and This is the reduction factor for the stirrups after considering corrosion; and These are the residual width and effective height after corrosion, respectively; and These refer to the thickness of the concrete protective layer on the side of the beam; It is the thickness of the concrete cover for the compression reinforcement of the beam.

[0057] Furthermore, the overall decision-making process of the PIML-CBSS model consists of 8 layers with a total of 65 nodes: 1 root node, 31 internal nodes, and 33 leaf nodes. Each node contains information such as the splitting criteria, mean squared error, number of samples, and predicted value. Figure 3 Only a portion of the decision-making process of the PIML-CBSS model is shown. Figure 3In this context, Li represents the calculated value from the Li et al. model; Xue represents the calculated value from the Xue et al. model; Lu represents the calculated value from the Lu et al. model; and Xu represents the calculated value from the Xue et al. model. Taking the root node as an example, this node is partitioned based on whether the calculated value generated by the explicit calculation formula of the Gene Expression Programming (GEP) algorithm is less than 310.4kN. If the condition is met, the node is partitioned along the left subtree; otherwise, it is partitioned along the right subtree. The root node contains 110 samples, and the average target value of the samples contained in the root node is 118.7kN, which is the predicted value of the sample under this partitioning method.

[0058] In addition, the key evaluation metrics for the PIML-CBSS model are: Mean Absolute Error (MAE) = 11.349 kN, Root Mean Square Error (RMSE) = 15.590 kN, and Mean Absolute Percentage Error (MAP) on the test set. E = 10.46%. These metrics significantly outperform the five empirical models and three traditional machine learning models compared, demonstrating its feasibility and superiority.

[0059] In this embodiment, dimensionless input variables are used as input. Based on the Gene Expression Programming (GEP) algorithm, a symbolic regression analysis is used to generate an explicit calculation formula for the shear strength of the rusted RC beam, realizing an efficient transformation from data to knowledge. By integrating existing empirical models with the explicit calculation formula for the shear strength of the rusted RC beam generated by symbolic regression analysis through decision tree, a PIML-CBSS model is constructed, which can intuitively reveal the decision-making process of each prediction result, making the model's prediction results transparent.

[0060] Table 1 shows the PIML-CBSS model and the empirical model on the test set. R 2 , MAE , RMSE , MAPE In the table, the Huo et al. model corresponds to formula (2) above, and the Lu et al. model corresponds to formula (3) above. Xue et al. model corresponds to the previous formula (4), Xu et al. model corresponds to the previous formula (5), and Li et al. model corresponds to the previous formula (6); Figures 4(a), 4(b), 4(c), 4(d), and 4(e) show the comparison of the prediction results of the PIML-CBSS model and each empirical model on the test set, where x represents the horizontal axis (true value) and y represents the vertical axis (predicted value) in Figures 4(a), 4(b), 4(c), 4(d), and 4(e). Figure 5 The overall performance comparison between the PIML-CBSS model and the empirical model on the test set is shown; Figure 6 The results demonstrate the performance improvement of the PIML-CBSS model compared to other empirical models on the test set.

[0061] Table 1 Evaluation Indicators of PIML-CBSS Model and Empirical Model

[0062] As shown in Figures 4(a), 4(b), 4(c), 4(d), and 4(e), on the test set, the calculation results of the empirical model are mostly distributed outside the 20% error range, while the results of the PIML-CBSS model show a better fitting trend, with the fitted points concentrated within the 20% error range, and the predicted points are basically distributed on the ideal diagonal. From Table 1 and... Figure 5 It can be seen that the PIML-CBSS model has better overall performance on the test set and has higher [performance]. R 2 And lower MAE, RMSE, MAPE; Figure 6 The horizontal axis in the graph represents the various evaluation indicators, and the vertical axis R1 represents the percentage improvement; Figure 6 It can be seen that, compared with various empirical models, the PIML-CBSS model has... R 2 The increases were 34.57%, 38.04%, 45.54%, 67.56%, and 52.76%. MAE The rates decreased by 70.43%, 67.72%, 69.73%, 74.70%, and 76.06%, respectively. RMSE It decreased by 67.53%, 68.54%, 70.29%, 73.56%, and 71.59%. MAPE The figures decreased by 75.94%, 72.50%, 75.22%, 79.26%, and 78.48%, respectively. Table 2 shows the R-values ​​of the PIML-CBSS model on the test set compared to traditional machine learning models such as Random Forest (RF), XGBoost (Extreme Gradient Boosting), and GEP (Gene Expression Programming). 2 MAE, RMSE, MAPE; Figures 7(a), 7(b) and 7(c) show the comparison of prediction results of the PIML-CBSS model and the traditional machine learning model on the test set. In Figures 7(a), 7(b) and 7(c), x represents the horizontal axis (true value) and y represents the vertical axis (predicted value). Figure 8 The overall performance comparison between the PIML-CBSS model and traditional machine learning models on the test set is shown; Figure 9 This demonstrates the performance improvement of the PIML-CBSS model compared to traditional machine learning models on the test set. Figure 9 The vertical axis R1 in the figure represents the percentage increase.

[0063] Table 2 Evaluation metrics for CBSS-PIML and ML models on the test set

[0064] As shown in Figures 7(a), 7(b), and 7(c), on the test set, compared to traditional machine learning models, the fitting points of the PIML-CBSS model are more concentrated within the 20% error range. Table 2. Figure 8 and Figure 9 It can be seen that, compared with the XGBoost model and the GEP model, the PIML-CBSS model proposed in this embodiment has lower performance. MAE , RMSE and higher R 2 , R 2 They increased by 2.00% and 3.97% respectively. MAE They decreased by 8.84% and 24.11% respectively. RMSE Don't reduce it by 22.18% or 28.96%, while the XGBoost model's... MAPE Comparable to the PIML-CBSS model. Compared to the RF model, the PIML-CBSS model... MAE and MAPE Larger, but with higher R 2 and smaller RMSE Furthermore, the PIML-CBSS model integrates domain knowledge, resulting in stronger interpretability and engineering credibility, and thus stronger overall performance and better prediction results.

[0065] Furthermore, the method for predicting the shear strength of corroded reinforced concrete beams based on physical information machine learning of the present invention can be applied to the performance prediction problems of other materials or components.

[0066] Example 2 This embodiment provides a system for predicting the shear strength of rusted RC beams that integrates domain knowledge, as follows: The dataset creation module is configured to create a dataset by collecting test data on the shear strength of corroded RC beams. The filtering module is configured to filter design parameters related to the geometric properties, material strength, and degree of steel corrosion of the corroded RC beam, and to determine the input and output parameters of the model. The calculation formula generation module is configured to take input parameters as input, and generate an explicit calculation formula for the shear strength of the rusted RC beam based on the gene expression programming algorithm and symbolic regression analysis. The prediction model building module is configured to use existing empirical models as domain knowledge, and through the decision tree algorithm, integrate the existing empirical models with the above-mentioned explicit calculation formula for the shear strength of the rusted RC beam to construct the PIML-CBSS model, and verify and determine the hyperparameters of the model. The visualization and validation module is configured to visualize and analyze the decision-making process of the PIML-CBSS model and predict the shear strength of the rusted RC beam using the PIML-CBSS model.

[0067] The prediction system provided in this embodiment integrates existing domain knowledge, namely the empirical formula and symbolic regression formula for the shear strength of corroded reinforced concrete beams. The calculation results of the empirical formula are used as the baseline physical features, and the explicit physical relationships mined by symbolic regression are introduced as enhancement features. The two are fused together, and a physical information-guided prediction model for the shear strength of corroded reinforced concrete beams is established based on the decision tree algorithm, which can ensure the accuracy and precision of the prediction results.

[0068] Example 3 Embodiment 3 of the present invention provides a computer-readable storage medium.

[0069] This embodiment provides a computer-readable storage medium storing a program that, when executed by a processor, implements the steps in the method for predicting the shear strength of a rusted RC beam by incorporating domain knowledge as described in Embodiment 1 of this invention.

[0070] The detailed steps are the same as those of the method for predicting the shear strength of rusted RC beams that integrates domain knowledge provided in Example 1, and will not be repeated here.

[0071] Example 4 Embodiment 4 of the present invention provides an electronic device.

[0072] This embodiment provides an electronic device, including a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements the steps in the method for predicting the shear strength of a rusted RC beam that integrates domain knowledge as described in Embodiment 1 of this invention.

[0073] The detailed steps are the same as those of the method for predicting the shear strength of rusted RC beams that integrates domain knowledge provided in Example 1, and will not be repeated here.

[0074] This invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this invention can be implemented using various computer languages, such as the object-oriented programming language Java, the interpreted scripting language JavaScript, and the Python language.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0078] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method of predicting the shear strength of a corroded RC beam by fusing domain knowledge, characterized by, As follows: Step 1, a data set is established by collecting corrosion RC beam shear strength test data; Step 2, the design parameters related to the geometric characteristics, material strength and steel corrosion degree of the corrosion RC beam are screened to determine the input parameters and output parameters of the model; Step 3, based on the genetic expression programming algorithm, the shear strength explicit calculation formula of the corrosion RC beam is generated through symbolic regression analysis by taking the input parameters as inputs; Step 4, the existing empirical model is taken as domain knowledge, the existing empirical model and the shear strength explicit calculation formula of the corrosion RC beam in step 3 are fused through the decision tree algorithm, the corrosion RC beam shear strength prediction model based on physical information machine learning is constructed, and the hyperparameters of the model are verified and determined; Step 5, the decision-making process of the corrosion RC beam shear strength prediction model based on physical information machine learning is visualized and analyzed, and the shear strength of the corrosion RC beam is predicted.

2. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 1, wherein, The design parameters described in Step 2 are obtained from the collected corrosion RC beam shear strength test data, including concrete compressive strength , cross-sectional width , cross-sectional height , shear span length , longitudinal reinforcement ratio , stirrup reinforcement ratio , longitudinal reinforcement yield strength , stirrup yield strength , stirrup spacing , shear span ratio , longitudinal reinforcement cross-sectional loss rate , stirrup cross-sectional loss rate , effective cross-sectional height , shear strength .

3. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 2, wherein, The input parameters in step 2 include shear span ratio , cross-section effective height-width ratio , longitudinal reinforcement index , transverse reinforcement index , longitudinal reinforcement cross-section loss rate , and stirrup cross-section loss rate ; and the output parameter is shear strength .

4. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 3, wherein, The step 3 determines the shear span ratio by symbolic regression analysis based on the gene expression programming algorithm , the cross-section effective height-width ratio , the longitudinal reinforcement index , the transverse reinforcement index , the longitudinal reinforcement section loss rate , the stirrup section loss rate and the shear strength , generates a corrosion RC beam shear strength explicit calculation formula without dimension imbalance problem, and realizes efficient conversion from data to knowledge.

5. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 1, wherein, In step 4, the calculation results of the empirical formula are taken as the benchmark physical characteristics, the explicit calculation formula mined by symbolic regression is taken as the enhanced characteristics, a composite feature space is constructed, and a corrosion RC beam shear strength prediction model based on physical information machine learning is established based on the decision tree algorithm.

6. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 1, wherein, In step 3, the processing process of the genetic expression programming algorithm includes the following contents: The information related to the shear strength of the corrosion RC beam is coded to generate an initial population consisting of candidate solutions; The fitness of each individual is calculated to evaluate its advantages and disadvantages, if the termination condition is met, the optimal result is output, if not, the optimal individual in the population is reserved first, and then new individuals are generated through crossover and mutation operations; After verification, the new individuals and the optimal individual are integrated to form a new population, then the next iteration cycle is entered, and the whole process of fitness evaluation, selection, crossover, mutation and population updating is repeated, and the cycle evolves until the termination condition is met.

7. The corrosion RC beam shear strength prediction method with fusion domain knowledge of claim 1, wherein, In step 4, the hyperparameters of the model are determined through Bayesian optimization combined with ten-fold cross-validation.

8. A corrosion RC beam shear strength prediction system fusing domain knowledge, characterized in that, It includes: The data set establishment module is configured to establish a data set by collecting corrosion RC beam shear strength test data; The screening module is configured to screen the design parameters related to the geometric characteristics, material strength and steel corrosion degree of the corrosion RC beam to determine the input parameters and output parameters of the model; The calculation formula generation module is configured to take the input parameters as inputs, based on the genetic expression programming algorithm, to generate the shear strength explicit calculation formula of the corrosion RC beam through symbolic regression analysis; The prediction model establishment module is configured to take the existing empirical model as domain knowledge, fuse the existing empirical model and the shear strength explicit calculation formula of the corrosion RC beam through the decision tree algorithm, construct the corrosion RC beam shear strength prediction model based on physical information machine learning, and verify and determine the hyperparameters of the model; The visualization and verification module is configured to visualize and analyze the decision-making process of the corrosion RC beam shear strength prediction model based on physical information machine learning, and predict the shear strength of the corrosion RC beam through the corrosion RC beam shear strength prediction model.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the corrosion RC beam shear strength prediction method integrating field knowledge according to any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored on the memory and running on the processor, characterized in that, The processor executes the program to implement the steps in the corrosion RC beam shear strength prediction method integrating field knowledge according to any one of claims 1-7.

Images