Method and device for predicting concrete degradation in plateau environment based on combinatorial algorithm

Abstract

The invention relates to the technical field of concrete strength detection, in particular to a method and a device for predicting concrete degradation in a plateau environment based on a combinatorial algorithm. The method comprises the following steps: constructing a first data set; decomposing the first data set according to an improved EMD (Empirical Mode Decomposition) model to obtain a second data set; performing single prediction on the second data set to obtain a single prediction value of the second data set; integrating the single predicted values of the second data set to obtain the single predicted values of the first data set; constructing a variable weight combination prediction model according to the single prediction value of the first data set; and verifying the effectiveness of the variable weight combination prediction model according to a model advantage and disadvantage criterion. The concrete degradation in the plateau environment is predicted by using the combinatorial algorithm, and the prediction precision can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of concrete strength testing technology, specifically to a method and apparatus for predicting concrete deterioration in high-altitude environments based on a combined algorithm. Background Technology

[0002] With the rapid development of society and the economy, the scale of buildings is also increasing day by day. Concrete, with its many advantages such as convenient material sourcing, simple construction, and low cost, has become the most widely used building material in various engineering structures, occupying an extremely important position in industrial buildings. Predicting the degree of concrete cracking can provide a reference for predicting the remaining life of concrete and effectively help in formulating maintenance strategies for subsequent buildings.

[0003] The unique environment of high-altitude regions, characterized by low air pressure and large temperature variations, places higher demands on the performance of construction materials. Conventional concrete suffers from drawbacks such as low tensile strength, low ultimate elongation, and brittleness. Fiber-reinforced concrete, a general term for composite materials composed of fibers and cementitious matrix, overcomes these shortcomings by incorporating fibers with high tensile strength, high ultimate elongation, and good alkali resistance. Using fiber-reinforced concrete, robust and durable roads and bridges can be constructed, providing reliable transportation infrastructure and meeting the needs of population mobility and economic development in high-altitude areas. However, research on predicting the degradation of fiber-reinforced concrete in high-altitude environments is limited. For example, Chinese patent application CN201611185963.9 discloses a method for predicting and evaluating the strength degradation of concrete in marine environments; however, this study only predicts the degradation of conventional concrete and only investigates concrete degradation in marine environments. Furthermore, the literature (“Research on Corrosion Deterioration Prediction Model of Fiber-Mechanical Concrete Based on Grey-Neural Network Combination Model”, *Bulletin of the Chinese Ceramic Society*, Rong Zebin et al., 2023-06-13) discloses a prediction model that incorporates 0.3% polyvinyl alcohol (PVA) fiber by volume into concrete with a design strength of C30. This model only predicts the deterioration of fiber-mechanical concrete under different solution immersion conditions and is not applicable to prediction in high-altitude environments. Moreover, there is currently no existing technology with a prediction model specifically suitable for the deterioration degree of fiber-mechanical concrete in my country's high-altitude environments.

[0004] To address the issue of low accuracy in predicting the deterioration of fiber-reinforced concrete in high-altitude environments, this paper proposes a method and apparatus for predicting concrete deterioration in high-altitude environments based on a combined algorithm. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for predicting concrete deterioration in high-altitude environments based on a combined algorithm, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for predicting concrete deterioration in high-altitude environments based on a combined algorithm includes:

[0008] S10. Collect characteristic data on the degree of deterioration of fiber-reinforced concrete in a plateau environment and construct the first dataset.

[0009] Furthermore, the degradation characteristic data includes mass loss rate and relative dynamic elastic modulus; the higher the mass loss rate, the higher the degree of concrete degradation; the lower the relative dynamic elastic modulus, the higher the degree of concrete cracking.

[0010] Furthermore, the formula for calculating the mass loss rate is as follows:

[0011]

[0012] Wherein, P t It is the mass loss rate of concrete at time point t; G0 is the original mass of concrete, G t The mass of the concrete at the end of time point t is the mass of the concrete.

[0013] Furthermore, the formula for calculating the relative dynamic elastic modulus is:

[0014]

[0015] Wherein, the W t The relative dynamic elastic modulus of the concrete at time t is represented; f0 is the original transverse fundamental frequency of the concrete; f t It is the transverse fundamental frequency of the concrete after time point t ends.

[0016] S20. Decompose the first dataset according to the improved EMD decomposition model to obtain the second dataset.

[0017] Furthermore, an improved EMD decomposition model is used to decompose the data in the first dataset; the EMD decomposition model is improved using a stepwise threshold function, and the specific steps include:

[0018] S21. Set the data in the first dataset as the original signal;

[0019] S22. Perform a single-layer improved EMD decomposition on the original signal to obtain the IMF;

[0020] S23. Calculate the standard deviation of the IMF as the initial value for the threshold;

[0021] S24. Calculate the IMF threshold according to the formula of the stepwise threshold function;

[0022] S25. Using the calculated threshold, set the values ​​in the IMF that are less than the threshold to 0 to obtain the threshold component of the IMF;

[0023] S25. Subtract the threshold component from the IMF to obtain the denoised component of the IMF;

[0024] S26. Output the threshold component of the IMF and use the denoised component as the new original signal. Repeat S22 to S25 to obtain the new IMF and denoised component.

[0025] Furthermore, the first dataset contains two columns of data: mass loss rate and relative dynamic elastic modulus. The improved EMD decomposition divides these two columns into P sub-components. The first dataset is then decomposed according to the improved EMD decomposition model to obtain a second dataset containing 2*P columns of data.

[0026] S30. Perform individual predictions on the second dataset to obtain the individual prediction values ​​for the second dataset.

[0027] Furthermore, the second dataset is predicted simultaneously using three individual prediction models: an improved grey prediction model, an improved correlation vector machine model, and an improved BP neural network model. The parameters of each of the three individual prediction models are optimized using the Grey Wolf algorithm. The parameters of the grey prediction model include the development coefficient f; the parameters of the correlation vector machine model include the kernel function width σ. 2 The regularization parameter λ; the parameters of the BP neural network model include the learning rate p, the initial weights q, and the threshold s.

[0028] Furthermore, the steps for optimizing parameters using the Grey Wolf algorithm include:

[0029] S31. Initialize parameters and the gray wolf pack H = (H1, H2, ..., H...) N The position H of each gray wolf j =(H j1 H j2 H jD ) T , (j = 1, 2, ..., N);

[0030] S32. Calculate the fitness value E for each individual gray wolf. i The positions of the top three gray wolf individuals in terms of fitness value are denoted as X. α X β X δ And select the X with the best fitness value α Let this be the optimal solution;

[0031] S33. Calculate the remaining individuals ω and X. α Xβ X δ The distance is used to update the positions of the gray wolf α, β, δ and the prey;

[0032] S34. Update the values ​​of convergence factor a and coefficient factors A and C;

[0033] S35. Determine whether the GWO algorithm has reached the maximum number of iterations k. If it has reached the maximum number of iterations, proceed to step S36; otherwise, return to step S32.

[0034] S36. Output the optimal solution X α The position of gray wolf α is obtained, and the parameters are optimized by the GWO algorithm.

[0035] Furthermore, the Grey Wolf algorithm has the characteristics of strong convergence performance, few parameters, and easy implementation. Optimizing the parameters of the grey prediction model, the correlation vector machine model, and the BP neural network model using the Grey Wolf algorithm can effectively improve the prediction effect.

[0036] S40. Integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset.

[0037] Furthermore, an accumulation method is used to integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset. The predicted value of the j-th individual prediction method in the i-th column of the first dataset can be expressed as:

[0038]

[0039] in This represents the predicted value of the k-th subsequence of the i-th column in the first dataset after improved EMD decomposition, obtained by the j-th single-item prediction method.

[0040] S50. Construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset.

[0041] Furthermore, the variable weight combination prediction model is constructed based on an induced operator and a similarity index; the information integration operator is an induced ordered weighted average operator; and the similarity index is the DICE coefficient.

[0042] Furthermore, the expression for the induced ordered weighted average operator is:

[0043]

[0044] Among them, F l This is called the induced ordered weighted average operator, μ i Let i = 1, 2, ..., m be called a iThe induced values, μ-index(i) are μ1, μ2, ..., μ m The index of the i-th largest number in the array, arranged in ascending order, L = (l1, l2, ..., l...). m ) T It is a weighted vector that satisfies

[0045] Furthermore, the DICE coefficient is an indicator of the closeness between two sequences, and its expression is as follows:

[0046]

[0047] Wherein, the x t The sequence X = (x1, x2, ..., x...) is... n The t-th number in ) x t ≥0; the y t It is a sequence Y = (y1, y2, ..., y n The t-th number in y) t ≥0, DICE(X,Y) represents the DICE coefficient of sequence X and sequence Y. As can be seen from the formula for the DICE coefficient, the DICE coefficient takes the value [0,1]. The less similar the two sequences are, the closer the DICE coefficient is to 0; the more similar the two sequences are, the closer the DICE coefficient is to 1; in particular, when the two sequences are completely identical, the DICE coefficient takes the value of 1.

[0048] The combined prediction model expression is:

[0049]

[0050] Among them, the For combined predicted values, The x it It is the t-th number in the sequence of predicted values ​​for the i-th single-item prediction method.

[0051] S60. Verify the effectiveness of the combined prediction model based on the error index.

[0052] Furthermore, the verification of the effectiveness of the combined prediction model involves calculating the error index between the combined predicted value and the actual value, and comparing the error index of the combined predicted value with the error index of the individual predicted value; the short-term combined predicted value refers to the prediction of the deterioration characteristic data of fiber-reinforced concrete over the next three cycles.

[0053] Furthermore, the error indices between the combined predicted and actual values ​​include the root mean square error (RMSE) and the root mean square percentage error (RMSE%). The RMSE between the combined predicted and actual values ​​can be expressed as:

[0054]

[0055] Where X represents the actual value of the first dataset; This represents the combined predicted value of the first dataset. The mean squared percentage error between the combined predicted value and the actual value can be expressed as:

[0056]

[0057] Specifically, the root mean square error (RMSE) and mean square percentage error (MSP) of the three individual prediction models and the combined prediction model are compared. If the MSE and MSP of the combined prediction model are less than those of the three individual prediction models, the combined prediction model is considered a superior combination prediction. If the MSE and MSP of the combined prediction model are between those of the three individual prediction models, the combined prediction model is considered a non-inferior combination prediction. If the MSE and MSP of the combined prediction model are greater than those of the three individual prediction models, the combined prediction model is considered an inferior combination prediction.

[0058] A device for predicting concrete deterioration in high-altitude environments based on a combined algorithm, comprising:

[0059] The first construction module is used to construct the first dataset based on the characteristic data of the deterioration degree of fiber-reinforced concrete in the plateau environment;

[0060] The decomposition module is used to decompose the first dataset according to the improved EMD decomposition model to obtain the second dataset;

[0061] The prediction module is used to perform single-item predictions on the second dataset and obtain single-item predicted values ​​for the second dataset.

[0062] An integration module is used to integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset.

[0063] The second construction module is used to construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset.

[0064] The verification module is used to verify the effectiveness of the variable weight combination prediction model based on the error index.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0066] 1. This invention addresses the complexity of characteristic data on the deterioration degree of fiber-reinforced concrete in high-altitude environments by employing an improved EMD model to decompose the characteristic data. The improved EMD decomposition is a fully data-driven adaptive method that extracts nonlinear features from the original data. It divides the characteristic data into multiple subsequences, each representing vibration modes at different scales or frequencies. Compared to the original sequence, the complexity of the decomposed subsequences is reduced. This nonlinear feature extraction makes the analysis and prediction of non-stationary and nonlinear deterioration degree characteristic data more accurate.

[0067] 2. This invention optimizes the parameters in the single-item prediction model by using the Grey Wolf algorithm. The Grey Wolf algorithm has the characteristics of strong convergence performance, few parameters, and easy implementation. Compared with the traditional grey prediction model, correlation vector machine model and BP neural network model, it can reduce the calculation time and implementation difficulty of parameter tuning.

[0068] 3. This invention employs a variable weighted combination algorithm for predicting concrete deterioration. It combines the induced ordered weighted average operator and DICE coefficients to construct a variable weighted combination prediction model. The prediction accuracy of individual prediction models at different periods is calculated, and the combined prediction value is obtained by combining the variable weighted combination prediction model. The variable weighted combination prediction model can combine the advantages of each individual prediction model and take into account the differences in prediction accuracy of each individual prediction model at different periods. Compared with the individual prediction model, the variable weighted combination prediction model has higher prediction accuracy. Attached Figure Description

[0069] Figure 1 A flowchart of a concrete deterioration prediction method based on a combined algorithm in a plateau environment provided in an embodiment of the present invention;

[0070] Figure 2 This is a flowchart of variable weight combination prediction provided in an embodiment of the present invention;

[0071] Figure 3 A flowchart of a concrete deterioration prediction device based on a combined algorithm in a high-altitude environment provided in an embodiment of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figures 1 to 3 This invention provides a method and apparatus for predicting concrete deterioration in high-altitude environments based on a combined algorithm. The technical solution is as follows:

[0074] Specifically, refer to Figure 1 As shown, this invention provides a method for predicting concrete deterioration in high-altitude environments based on a combined algorithm. This method can be executed by a device, which can be implemented by software and / or hardware. In specific implementation, it includes steps S10 to S60.

[0075] Specifically, S10. Collect characteristic data on the degree of deterioration of fiber-reinforced concrete in a plateau environment to construct a first dataset; S20. Decompose the first dataset according to the improved EMD decomposition model to obtain a second dataset; S30. Perform individual predictions on the second dataset to obtain individual predicted values ​​for the second dataset; S40. Integrate the individual predicted values ​​of the second dataset to obtain individual predicted values ​​for the first dataset; S50. Construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset; S60. Verify the effectiveness of the variable weight combination prediction model according to the error index.

[0076] S10. Collect characteristic data on the degree of deterioration of fiber-reinforced concrete in a plateau environment and construct the first dataset.

[0077] Taking the Yunnan-Guizhou Plateau as an example, characteristic data of the degree of deterioration of fiber-reinforced concrete in the Yunnan-Guizhou Plateau environment were collected. The characteristic data of the degree of deterioration included mass loss rate and relative dynamic elastic modulus. The higher the mass loss rate, the higher the degree of concrete deterioration. The lower the relative dynamic elastic modulus, the higher the degree of concrete cracking.

[0078] Specifically, the mass loss rate is an important indicator for measuring changes in the mass of an object or system. It has applications in many fields, including engineering, chemistry, and materials science. By measuring and controlling the mass loss rate, we can assess the durability of materials, predict the lifespan of equipment, and take corresponding measures to reduce or prevent mass loss. The higher the mass loss rate, the higher the degree of concrete deterioration. The relative dynamic elastic modulus refers to the ratio between the strain and stress of a material under dynamic stress loading. It is an important parameter in dynamic elasticity mechanics and is often used to describe the elastic properties of materials under vibration or impact loads. The lower the relative dynamic elastic modulus, the higher the degree of concrete cracking.

[0079] Specifically, the formula for calculating the quality loss rate is as follows:

[0080]

[0081] Wherein, P t It is the mass loss rate of concrete at the end of the t-th cycle; G0 is the original mass of the concrete, G tThis refers to the mass of the concrete at the end of the t-th cycle. The formula for calculating the relative dynamic elastic modulus is:

[0082]

[0083] Wherein, the W t The value represents the mass loss rate of the concrete at the end of the t-th cycle; f0 is the original transverse fundamental frequency of the concrete; f t This is the transverse fundamental frequency of the concrete after the t-th cycle. The transverse fundamental frequency refers to the fundamental frequency of vibration of a system or object in the transverse direction. It is often used to describe the vibration characteristics of structures, mechanical systems, or other vibrating systems in the transverse direction. The transverse fundamental frequency corresponds to the longitudinal fundamental frequency, vertical fundamental frequency, etc., which describe the vibration behavior in different directions. The transverse fundamental frequency depends on the material properties, geometry, and boundary conditions of the structure or system. By analyzing and calculating the structure, the transverse fundamental frequency and related vibration modes can be determined. The transverse fundamental frequency is of great significance for designing and evaluating the safety, stability, and reliability of structures.

[0084] Furthermore, the mass loss rate and relative dynamic elastic modulus of fiber-reinforced concrete were collected over 200 days to construct the first dataset.

[0085] S20. Decompose the first dataset according to the improved EMD decomposition model to obtain the second dataset.

[0086] Specifically, the EMD decomposition model is improved using a stepwise threshold function. The specific steps include:

[0087] S21. Set the data in the first dataset as the original signal;

[0088] S22. Perform a single-layer EMD decomposition on the original signal to obtain the IMF;

[0089] S23. Calculate the standard deviation of the IMF as the initial value for the threshold;

[0090] S24. Calculate the IMF threshold according to the formula of the stepwise threshold function;

[0091] S25. Using the calculated threshold, set the values ​​in the IMF that are less than the threshold to 0 to obtain the threshold component of the IMF;

[0092] S25. Subtract the threshold component from the IMF to obtain the denoised component of the IMF;

[0093] S26. Output the threshold component of the IMF and use the denoised component as the new original signal. Repeat S22 to S25 to obtain the new IMF and denoised component.

[0094] Furthermore, the EMD decomposition model was improved by dividing the fiber-reinforced concrete mass loss rate sequence and the fiber-reinforced concrete relative dynamic elastic modulus sequence into 7 components, and constructing a second dataset based on the 2*7 components.

[0095] In this embodiment, to address the complexity of the characteristic data on the deterioration degree of fiber-reinforced concrete in high-altitude environments, an improved EMD model is used to decompose the characteristic data. The improved EMD decomposition is a fully data-driven adaptive method that can extract nonlinear features from the original data. The improved EMD decomposition divides the characteristic data into multiple subsequences, each representing vibration modes at different scales or frequencies. Compared to the original sequence, the complexity of the decomposed subsequences is reduced. This nonlinear feature extraction makes the analysis and prediction of non-stationary and nonlinear deterioration degree characteristic data more accurate.

[0096] S30. Perform individual predictions on the inferior second dataset to obtain individual prediction values ​​for the second dataset.

[0097] Specifically, 200 data points were divided into a training set and a test set in a 7:3 ratio. Three individual prediction models were used simultaneously to predict the second dataset. These three models were an improved grey prediction model, an improved correlation vector machine model, and an improved backpropagation neural network model, all capable of predicting small sample sizes. The parameters of each model were optimized using the Grey Wolf algorithm. Due to the limited data sample size, rolling prediction was employed to ensure accuracy. The predicted values ​​for the second dataset in the test set were calculated by substituting known data into the individual prediction models.

[0098] Specifically, grey prediction models are a prediction method that uses limited and incomplete information to build a mathematical model and provide predictions. Currently used prediction methods require large samples; if the sample size is small, it can lead to significant errors and cause the prediction target to fail. Grey prediction models require less modeling information, are computationally convenient, and have high modeling accuracy. They are widely used in various prediction fields and are an effective tool for handling small-sample prediction problems. The Grey Wolf algorithm has advantages in parameter optimization, including strong global search capability, fast convergence speed, simple implementation, and strong robustness. The Grey Wolf algorithm is used to optimize the evolution coefficient f of the grey prediction model. The optimization process is as follows:

[0099] S301. Initialize the development coefficient f of the grey prediction model and the grey wolf pack. The location of each gray wolf

[0100] S302. Calculate the fitness value of each individual gray wolf. The positions of the top three gray wolf individuals with the highest fitness values ​​are denoted as follows: and the one with the best fitness value Let this be the optimal solution;

[0101] S303. Calculate the remaining individuals ω and The distance is used to update the positions of the gray wolf α1, β1, δ1 and the prey;

[0102] S304. Update the values ​​of convergence factor a1 and coefficient factors A1 and C1;

[0103] S305. Determine whether the Grey Wolf Algorithm has reached the maximum number of iterations k1. If it has reached the maximum number of iterations, execute S506; otherwise, return to S502.

[0104] S306. Output the optimal solution That is, the position of the gray wolf α1 is used to obtain the phase development coefficient f optimized by the gray wolf algorithm.

[0105] The optimized development coefficient is then input into the grey prediction model, and the predicted value of the test set is calculated in combination with the degradation degree feature data. Next, the model is tested by calculating the mean square percentage error between the predicted value and the actual value of the second feature data in each period. If the mean square percentage error is less than 0.2, the model is considered feasible.

[0106] Specifically, the Relevance Vector Machine (RVM) algorithm is a machine learning model based on the Support Vector Machine (SVM) and a Bayesian framework. It obtains the relevance vectors and weights by maximizing the marginal likelihood. Compared to SVM, RVM has the following advantages: the number of relevance vectors is much smaller than that of support vectors, resulting in high sparsity; the kernel function does not need to satisfy the Mercer condition, increasing the flexibility of kernel function selection. The Grey Wolf algorithm has advantages in parameter optimization, including strong global search capability, fast convergence speed, simple implementation, and strong robustness. The Grey Wolf algorithm is used to optimize the kernel width σ of the RVM. 2 The optimization process for the regularization parameter λ is as follows:

[0107] S311. Initialize the kernel width σ of the correlation vector machine. 2 and regularization parameter λ and gray wolf pack The location of each gray wolf

[0108] S312. Calculate the fitness value of each individual gray wolf. The positions of the top three gray wolf individuals with the highest fitness values ​​are denoted as follows: and the one with the best fitness value Let this be the optimal solution;

[0109] S313. Calculate the remaining individuals ω and The distance is used to update the positions of the gray wolf α2, β2, δ2 and the prey;

[0110] S314. Update the values ​​of convergence factor a2 and coefficient factors A2 and C2;

[0111] S315. Determine whether the Grey Wolf Algorithm has reached the maximum number of iterations k2. If it has reached the maximum number of iterations, execute S516; otherwise, return to S512.

[0112] S316. Output the optimal solution That is, the position of the gray wolf α2, and the kernel function width σ optimized by the gray wolf algorithm. 2 and the regularization parameter λ.

[0113] Based on the optimized kernel function width σ 2 A modified vector machine model is constructed using the regularization parameter λ to obtain the predicted values ​​of the degradation degree feature data for the test set. Then, the model is tested by calculating the mean square percentage error between the predicted and actual values ​​of the degradation degree feature data for each cycle. If the mean square percentage error is less than 0.2, the model is considered feasible.

[0114] Specifically, a backpropagation (BP) neural network is a common type of artificial neural network that can be used for time series forecasting. First, input and output variables are selected. Next, the second dataset is divided into a training set and a test set. The training set is used to train the BP neural network, while the test set is used to validate the model's performance. Then, the degradation feature data is preprocessed, and the data is normalized using min-max normalization, scaling the data to between 0 and 1. Next, the BP neural network model is constructed, with the input and output layers set to single neurons and the hidden layers set to multiple neurons. Then, an appropriate activation function is selected; in this embodiment, the sigmoid activation function is used, mapping any real number to between 0 and 1. The learning rate p, initial weights q, and threshold s are optimized using the Grey Wolf algorithm to improve prediction accuracy. Finally, the root mean square error between the predicted and actual values ​​is calculated to measure the model's performance.

[0115] The steps of the Grey Wolf algorithm to optimize the BP neural network model are as follows:

[0116] S321. Initialize the learning rate p, weights q, threshold s, and wolf pack configuration of the BP neural network. The location of each gray wolf

[0117] S322. Calculate the fitness value of each individual gray wolf. The positions of the top three gray wolf individuals with the highest fitness values ​​are denoted as follows: X with the best fitness value α3 Let this be the optimal solution;

[0118] S323. Calculate the remaining individuals ω and The distance is used to update the positions of the gray wolf α3, β3, δ3 and the prey;

[0119] S324. Update the values ​​of convergence factor a3 and coefficient factors A3 and C3;

[0120] S325. Determine if the Grey Wolf Algorithm has reached the maximum number of iterations k3. If it has reached the maximum number of iterations, execute S526; otherwise, return to S522.

[0121] S326. Output the optimal solution X α3 That is, the position of the gray wolf α3, and the learning rate p, initial weight q and threshold s optimized by the gray wolf algorithm are obtained.

[0122] Specifically, the error index between the predicted values ​​and actual values ​​of the three individual prediction models is calculated, and the error index includes root mean square error and mean square error.

[0123] In this embodiment, three single-item prediction models were selected to predict the mass loss rate and relative dynamic elastic modulus of fiber-reinforced concrete in different devices. These three models are an improved grey prediction model, an improved correlation vector machine model, and an improved BP neural network model. These models are suitable for predicting small sample data. The grey wolf algorithm is used to optimize the parameters in the grey prediction model, the correlation vector machine model, and the BP neural network, which improves the predictive performance of the models.

[0124] S40. Integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset.

[0125] Furthermore, an accumulation method is used to integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset. The predicted value of the j-th individual prediction method in the i-th column of the first dataset can be expressed as:

[0126]

[0127] in This represents the predicted value of the k-th subsequence of the i-th column in the first dataset after improved EMD decomposition, obtained by using the j-th single-item prediction method.

[0128] S50. Construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset.

[0129] Figure 2 The flowchart for variable weight combination prediction provided in this embodiment of the invention is shown below. Figure 2As shown, specifically, the variable weight combination prediction model is constructed based on an induced operator and a similarity index; the information integration operator is an induced ordered weighted average operator; and the similarity index is the DICE coefficient.

[0130] Specifically, the expression for the induced ordered weighted average operator is:

[0131]

[0132] Among them, F l This is called the induced ordered weighted average operator, μ i i = 1, 2, ..., m is called a i The induced values, μ-index(i) are μ1, μ2, ..., μ m The index of the i-th largest number in the array, arranged in ascending order, is L = (l1, l2, ..., ln). m ) T It is a weighted vector that satisfies

[0133] Specifically, the DICE coefficient is an indicator of how close two sequences are, and its expression is as follows:

[0134]

[0135] Wherein, the x t The sequence X = (x1, x2, ..., x...) n The t-th number in ) x t ≥0; the y t It is a sequence Y = (y1, y2, ..., y n The t-th number in y) t ≥0, DICE(X,Y) represents the DICE coefficient of sequence X and sequence Y. As can be seen from the formula for the DICE coefficient, the DICE coefficient takes the value [0,1]. The less similar the two sequences are, the closer the DICE coefficient is to 0; the more similar the two sequences are, the closer the DICE coefficient is to 1; in particular, when the two sequences are completely identical, the DICE coefficient takes the value of 1.

[0136] Specifically, since a larger DICE coefficient indicates a higher degree of similarity between two sequences, a variable weighted combination prediction model is constructed by combining the induced ordered weighted average operator with the objective of maximizing the DICE coefficient. The expression of the variable weighted combination prediction model is as follows:

[0137]

[0138] Among them, the Let be the combined predicted value for the t-th period. The x itIt is the predicted value of the i-th single-item prediction method in the t-th period.

[0139] Specifically, the prediction accuracy of the individual predicted values ​​is extracted, substituted into the combined prediction model, and the weighting coefficients are output to obtain the combined predicted value. The prediction accuracy extracted from the individual predicted values ​​refers to the prediction accuracy between the predicted values ​​of each individual prediction model for periods 1 to 10 and the actual values. The expression for the prediction accuracy is:

[0140]

[0141] Where q represents the degradation degree characteristic data type, q=1, 2, where q=1 represents the mass loss rate and q=2 represents the relative dynamic elastic modulus; i represents the single prediction model, i=1, 2, 3, where i=1 represents the grey prediction model, i=2 represents the improved correlation vector machine model, and i=3 represents the BP neural network model; t represents the period, p=1, 2, ..., 40. This represents the actual value of the degradation characteristic data of type q in period t; This represents the predicted value of the i-th single-item prediction model for the degradation degree characteristic data of type q in period t; This represents the prediction accuracy between the predicted value and the actual value of the i-th single prediction model for the degradation degree characteristic data of class q in period t.

[0142] Specifically, after substituting the predicted values ​​and prediction accuracy of the individual prediction models, the combined prediction model can be expressed as:

[0143]

[0144] in, The weight coefficients of the combined prediction model representing the degradation characteristics of class q are used to indicate the degree of degradation. This represents the predicted value corresponding to the i-th largest single prediction model in terms of prediction accuracy for the q-type degradation feature data over period t. This represents the combined predicted value of the degradation characteristic data of type q under period t. The actual values ​​of the mass loss rate and relative dynamic elastic modulus of the four devices, along with the predicted values ​​of the three individual prediction models, are substituted into the combined prediction model to output weighting coefficients. Based on these weighting coefficients, the combined predicted value of the mass loss rate and relative dynamic elastic modulus of the four devices in the test set is calculated.

[0145] In this embodiment, a variable-weighted combination prediction model is constructed by selecting the induced ordered weighted average operator and the DICE coefficient. This model is simple to calculate and has better prediction performance than the fixed-weighted combination prediction model. The predicted values ​​of the q-type degradation characteristic data of the single-item prediction model are reordered at different periods. The prediction accuracy is used as the induced value. The single-item prediction value with the highest prediction accuracy for the q-type degradation characteristic data at period t is placed in the first column, the second highest in the second column, and the lowest in the third column. The reordered single-item prediction values ​​are then substituted into the variable-weighted combination prediction model, and the optimal weight coefficient is output based on the condition of maximizing the DICE coefficient. This step can optimize the situation where different single-item prediction models have different prediction effects at different time points, and improve the prediction accuracy of the variable weight combination prediction model.

[0146] S60. Verify the effectiveness of the variable weight combination prediction model based on the error index.

[0147] Specifically, verifying the effectiveness of the combined prediction model involves calculating the error indices between the predicted and actual values ​​for periods 1 to 10 of the combined prediction model, including the root mean square error (RMSE) and the root mean square percentage error (RMSE%). The RMSE between the predicted and actual values ​​can be expressed as:

[0148]

[0149] Where X represents the actual value of the first dataset; This represents the combined predicted value of the first dataset. The mean squared percentage error between the combined predicted value and the actual value can be expressed as:

[0150]

[0151] Specifically, the root mean square error (RMSE) and mean square percentage error (MSP) of the three individual prediction models and the combined prediction model are compared. If the MSE and MSP of the combined prediction model are less than those of the three individual prediction models, the combined prediction model is considered a superior combination prediction. If the MSE and MSP of the combined prediction model are between those of the three individual prediction models, the combined prediction model is considered a non-inferior combination prediction. If the MSE and MSP of the combined prediction model are greater than those of the three individual prediction models, the combined prediction model is considered an inferior combination prediction.

[0152] In summary, this embodiment first collects characteristic data on the degradation degree of fiber-reinforced concrete in a high-altitude environment to construct a first dataset. An improved EMD decomposition method is then used to decompose the data in the first dataset. The improved EMD decomposition is a fully data-driven adaptive method that can extract nonlinear features from the original data. The improved EMD decomposition divides the mass loss rate and relative dynamic elastic modulus into seven subsequences, each representing a vibration mode at a different scale or frequency. Compared to the original sequence, the complexity of the decomposed subsequences is reduced. This nonlinear feature extraction makes the analysis and prediction of non-stationary and nonlinear degradation degree characteristic data more accurate. An improved grey prediction model, an improved correlation vector machine model, and an improved BP neural network model are used to predict the mass loss rate and relative dynamic elastic modulus of concrete. The Grey Wolf algorithm is used to optimize the parameters of the grey prediction model, correlation vector machine model, and BP neural network model. The Grey Wolf algorithm has advantages in parameter optimization, such as strong global search capability, fast convergence speed, ease of implementation, and strong robustness, which can improve the prediction effect of individual prediction models. Finally, the individual prediction results are integrated to obtain the individual prediction values ​​of the first dataset. Subsequently, a variable-weighted combined prediction model based on the induced ordered weighted average operator and DICE coefficients was constructed. Combining the predicted values ​​of three individual prediction models, the model outputs the optimal weight coefficients using the prediction accuracy of the individual predictions within the same period as the induced value. The combined prediction value is then obtained based on these output weight coefficients. Finally, the error indices of the variable-weighted combined prediction model and the individual prediction models were calculated and compared. The results show that the variable-weighted combined prediction model effectively avoids the drawback of assigning the same weights to individual prediction models at different time points in fixed-weighted combined prediction, thus improving the prediction performance.

[0153] Figure 3 The flowchart of the concrete deterioration prediction device based on the combined algorithm in the plateau environment provided in the embodiment of the present invention is shown below. Figure 3 As shown, the specific technical solution is as follows:

[0154] S210: First construction module, used to construct a first dataset based on the characteristic data of fiber-reinforced concrete deterioration in a plateau environment; S220: Decomposition module, used to decompose the first dataset according to the improved EMD decomposition model to obtain a second dataset; S230: Prediction module, used to perform single-item prediction on the second dataset to obtain the single-item prediction value of the second dataset; S240: Integration module, used to integrate the single-item prediction values ​​of the second dataset to obtain the single-item prediction value of the first dataset; S250: Second construction module, used to construct a variable-weight combination prediction model based on the single-item prediction value of the first dataset; S260: Validation module, used to verify the effectiveness of the variable-weight combination prediction model according to the error index.

[0155] In summary, this embodiment first collects characteristic data on the degradation degree of fiber-reinforced concrete in a high-altitude environment to construct a first dataset. An improved EMD decomposition method is then used to decompose the data in the first dataset. The improved EMD decomposition is a fully data-driven adaptive method that can extract nonlinear features from the original data. The improved EMD decomposition divides the mass loss rate and relative dynamic elastic modulus into seven subsequences, each representing a vibration mode at a different scale or frequency. Compared to the original sequence, the complexity of the decomposed subsequences is reduced. This nonlinear feature extraction makes the analysis and prediction of non-stationary and nonlinear degradation degree characteristic data more accurate. An improved grey prediction model, an improved correlation vector machine model, and an improved BP neural network model are used to predict the mass loss rate and relative dynamic elastic modulus of concrete. The Grey Wolf algorithm is used to optimize the parameters of the grey prediction model, correlation vector machine model, and BP neural network model. The Grey Wolf algorithm has advantages in parameter optimization, such as strong global search capability, fast convergence speed, ease of implementation, and strong robustness, which can improve the prediction effect of individual prediction models. Finally, the individual prediction results are integrated to obtain the individual prediction values ​​of the first dataset. Subsequently, a variable-weighted combined prediction model based on the induced ordered weighted average operator and DICE coefficients was constructed. Combining the predicted values ​​of three individual prediction models, the model outputs the optimal weight coefficients using the prediction accuracy of the individual predictions within the same period as the induced value. The combined prediction value is then obtained based on these output weight coefficients. Finally, the error indices of the variable-weighted combined prediction model and the individual prediction models were calculated and compared. The results show that the variable-weighted combined prediction model effectively avoids the drawback of assigning the same weights to individual prediction models at different time points in fixed-weighted combined prediction, thus improving the prediction performance.

[0156] Specifically, the working principle of this device is based on the above-mentioned method for predicting concrete deterioration in high-altitude environments based on combined algorithms.

[0157] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting concrete deterioration in high-altitude environments based on a combined algorithm, characterized in that, include: S10. Collect characteristic data on the deterioration degree of fiber-reinforced concrete under high-altitude environment and construct the first dataset; S20. Decompose the first dataset according to the improved EMD decomposition model to obtain the second dataset; The improved EMD decomposition model is based on the EMD model and uses a stepwise threshold function to calculate the threshold of the intrinsic mode function (IMF). S30. Perform single-item prediction on the second dataset to obtain the single-item prediction value of the second dataset; S40. Integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset; S50. Construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset; The variable weight combination prediction model is constructed based on the induced weighted average operator and the DICE similarity index; The expression for the induced ordered weighted average operator is: The expression for the DICE coefficient is as follows: Wherein, the x t It is a sequence X(x1,x2,…,x) n The t-th number in ) is y t It is a sequence Y(y1,y2,…,y n In the sequence X, the t-th number is given, and DICE(X, Y) represents the DICE coefficients of sequences X and Y. The combined prediction model expression is: Among them, the For combined predicted values, The x it It is the t-th number in the sequence of predicted values ​​for the i-th single-item prediction method; S60. Verify the effectiveness of the variable weight combination prediction model based on the error index.

2. The method according to claim 1, characterized in that, S10 includes: the fiber-reinforced concrete deterioration characteristic data includes mass loss rate and relative dynamic elastic modulus; the higher the mass loss rate, the higher the degree of concrete deterioration; the lower the relative dynamic elastic modulus, the higher the degree of concrete cracking.

3. The method according to claim 1, characterized in that, S20 includes: The EMD decomposition model is improved by using a stepwise threshold function. The specific steps include: S21. Set the data in the first dataset as the original signal; S22. Perform a single-layer EMD decomposition on the original signal to obtain the IMF; S23. Calculate the standard deviation of the IMF as the initial value for the threshold; S24. Calculate the threshold of the IMF according to the formula of the stepwise threshold function; S25. Using the calculated threshold, set the values ​​in the IMF that are less than the threshold to 0 to obtain the threshold component of the IMF; S25. Subtract the threshold component from the IMF to obtain the denoised component of the IMF; S26. Output the threshold component of the IMF, and use the denoised component as the new original signal. Repeat S22 to S25 to obtain a new IMF and denoised component. S27. After multiple iterations, until the predetermined number of decomposition layers or the stopping criterion is reached, the decomposed IMF series is obtained.

4. The method according to claim 1, characterized in that, S30 includes: The second dataset is predicted simultaneously using three individual prediction models; The three single-item prediction models are the improved grey prediction model, the improved correlation vector machine model, and the improved BP neural network model. The parameters in the three single-item prediction models were optimized using the Grey Wolf algorithm. The steps for optimizing parameters using the Grey Wolf algorithm include: S31. Initialize parameters and the gray wolf pack H = (H1, H2, ..., H...) N The position H of each gray wolf j =(H j1 H j2 ,…,H jN ) T j = 1, 2, ..., N; S32. Calculate the fitness value E for each individual gray wolf. j The positions of the top three gray wolf individuals in terms of fitness value are denoted as X. α X β X δ And select the X with the best fitness value α Let this be the optimal solution; S33. Calculate the remaining individuals ω and X. α X β X δ The distance is used to update the positions of the gray wolf α, β, δ and the prey; S34. Update the values ​​of convergence factor a and coefficient factors A and C; S35. Determine if the Grey Wolf Algorithm has reached the maximum number of iterations K. If it has reached the maximum number of iterations, execute S36; otherwise, return to S32. S36. Output the optimal solution X α That is, the position of gray wolf α, and the parameters optimized by the gray wolf algorithm; The parameters of the grey prediction model include the development coefficient f; The parameters of the related vector machine model include the kernel function width σ. 2 Regularization parameter λ; The parameters of the BP neural network model include the learning rate p, the initial weights q, and the threshold s.

5. The method according to claim 1, characterized in that, S40 includes: integrating the individual predicted values ​​of the second dataset using an accumulation method to obtain the individual predicted values ​​of the first dataset.

6. The method according to claim 1, characterized in that, S40 and S50 further include: Calculate the error indices for single-item prediction models and variable-weight combination prediction models; The error metrics include root mean square error and root mean square percentage error.

7. The method according to claim 1, characterized in that, The S60 includes: The effectiveness of the variable weight combination prediction model is verified by comparing the error index of the variable weight combination prediction model with that of the single-item prediction model. The error indices of the single-item prediction model refer to the root mean square error and the root mean square percentage error between the predicted and actual values ​​of a single item in the first dataset.

8. A device for predicting the deterioration of metal fiber reinforced concrete based on a combined prediction model, characterized in that, include: The first construction module is used to construct the first dataset based on the characteristic data of the deterioration degree of fiber-reinforced concrete in the plateau environment; The decomposition module is used to decompose the first dataset according to the improved EMD decomposition model to obtain the second dataset; The prediction module is used to perform single-item predictions on the second dataset and obtain single-item predicted values ​​for the second dataset. An integration module is used to integrate the individual predicted values ​​of the second dataset to obtain the individual predicted values ​​of the first dataset. The second construction module is used to construct a variable weight combination prediction model based on the individual predicted values ​​of the first dataset. The verification module is used to verify the effectiveness of the variable weight combination prediction model based on the error index.

9. The apparatus according to claim 8, characterized in that, The first building module includes: Based on the characteristic data of the deterioration degree of fiber-reinforced concrete in plateau environments, the first dataset was constructed. The deterioration characteristic data of the fiber-reinforced concrete include mass loss rate and relative dynamic elastic modulus. The higher the mass loss rate, the higher the degree of concrete deterioration; The lower the relative dynamic elastic modulus, the higher the degree of concrete cracking.

10. The apparatus according to claim 8, characterized in that, The prediction module includes: The second dataset is predicted simultaneously using three individual prediction models; The three single-item prediction models are the improved grey prediction model, the improved correlation vector machine model, and the improved BP neural network model. The parameters in the three single-item prediction models were optimized using the Grey Wolf algorithm.

Images