A concrete vibrating quality prediction method based on an AWMI-LightGBM algorithm

Abstract

The present application relates to the technical field of data processing, and specifically relates to a concrete vibration quality prediction method based on an AWMI-LightGBM algorithm. The present application realizes accurate prediction of concrete vibration quality by combining a multi-granularity fusion AWMI algorithm and a LightGBM algorithm. First, by obtaining original vibration data, including vibration characteristics and vibration quality, a basis is provided for subsequent processing. Then, the original data is weighted using a multi-granularity fusion AWMI algorithm, effectively improving the influence of key features, so that the data is more in line with actual prediction requirements. Next, the weighted data is used to train the LightGBM algorithm, obtaining a high-precision vibration quality prediction model. Finally, by inputting real-time vibration data, the model can quickly predict the corresponding vibration quality, providing strong support for quality control during construction. This method not only improves the accuracy of prediction, but also improves construction efficiency, and has important practical value.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically a method for predicting the quality of concrete vibration based on the AWMI-LightGBM algorithm. Background Technology

[0002] Concrete is one of the most important materials in the construction industry. Currently, the quality testing indicators for concrete mainly focus on properties such as strength and density. Traditional strength and density testing usually relies on the experience and judgment of construction workers, rebound tests, and core sampling. However, these methods have significant limitations. Complex construction environments and differences in the individual experience of construction workers can easily lead to errors in judging the vibration quality. Therefore, it is necessary to establish a scientific and precise concrete vibration quality testing system to achieve accurate detection of vibration quality and dynamic adjustment of vibration parameters during the vibration operation.

[0003] The aforementioned concrete quality inspections are all conducted after concrete pouring and cannot provide a reference for quality control during concrete construction. Currently, domestic quality inspections during concrete vibration mainly rely on real-time monitoring of vibration parameters and visual recognition technology. Real-time monitoring of vibration parameters typically uses sensors such as ultrasonic waves, tilt sensors, cameras, and satellite positioning to collect parameters such as vibration trajectory, vibration depth, and vibration tilt angle, which are then input into a pre-established vibration quality evaluation mechanism for analysis. However, the influencing mechanisms of concrete vibration quality in engineering practice are very complex, and relying solely on vibration parameters cannot achieve accurate assessment of concrete quality. Furthermore, harsh working environments in actual engineering applications can easily affect the accuracy of sensors, leading to detection errors. Visual recognition technology can analyze the image features of the concrete surface and evaluate quality by comparing features such as the area and distribution of air bubbles on the exposed concrete surface, which has some feasibility. However, visual recognition technology has high requirements for the accuracy and stability of image sensors, resulting in poor adaptability. It is difficult to achieve accurate image capture in harsh concrete vibration working environments. In addition, visual recognition technology can only detect the surface condition of concrete and has limited ability to assess the overall quality of concrete.

[0004] To address the issues of low accuracy and poor adaptability in current concrete vibration quality assessment schemes, it is necessary to find a highly feasible method for predicting concrete vibration quality. This would enable accurate detection of concrete vibration quality and real-time adjustment of vibration parameters during the operation, thereby ensuring the construction quality of the project. Summary of the Invention

[0005] To avoid and overcome the technical problems existing in the prior art, this invention provides a method for predicting the vibration quality of concrete based on the AWMI-LightGBM algorithm. This invention can predict the vibration quality of concrete with relatively high accuracy.

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

[0007] A method for predicting the vibration quality of concrete based on the AWMI-LightGBM algorithm includes the following prediction steps:

[0008] S1. Obtain the original vibration data during the concrete vibration process. The original vibration data includes vibration characteristics and vibration quality.

[0009] S2. Use the multi-granularity fusion AWMI algorithm to weight the original vibration data to obtain the corresponding weighted vibration data;

[0010] S3. Use weighted vibration data to train the LightGBM algorithm to obtain a vibration quality prediction model;

[0011] S4. Obtain real-time vibration data during the concrete vibration process and input the real-time vibration data into the vibration quality prediction model to predict the vibration quality corresponding to the real-time vibration data.

[0012] As a further aspect of the present invention, the specific content of step S2 is as follows:

[0013] S21. Select n sets of original samples (vibration features, vibration quality), and construct the original dataset after noise reduction.

[0014] S22. Based on the feature values ​​and vibration quality values, calculate the multi-granularity fusion mutual information and weighting function of vibration features for all categories;

[0015] S23. Based on multi-granularity fusion mutual information and weighting function, the original dataset is weighted to obtain weighted data, and this data is defined as weighted data.

[0016] As a further aspect of the present invention: the original dataset D is represented as:

[0017] D={(x ji ,y i )|j=1,2,…,J; i=1,2,…,n};

[0018] In the formula, x ji y represents the feature value of the j-th vibration feature in the i-th original sample group; i denoted by , where represents the value of the vibration mass in the i-th original sample group; J represents the total number of vibration feature categories.

[0019] As a further aspect of the present invention, the formula for calculating mutual information is expressed as follows:

[0020]

[0021] In the formula, I(x) j p(x,y) represents the mutual information between the j-th vibration characteristic and the vibration quality y; ji ,y i ) represents x ji and y i Joint probability distribution between; p(x) ji ) represents x ji Marginal probability distribution; p(y i ) represents y i The marginal probability distribution.

[0022] As a further aspect of the present invention, the calculation formula for the weighting function is expressed as follows:

[0023]

[0024] In the formula, w(x) j Var(x) represents the weighting function for the j-th vibration characteristic; j ) represents the variance of the j-th vibration feature in the original dataset; Var avg This represents the average variance of the vibration characteristics for all categories.

[0025] As a further aspect of the present invention, the calculation formula for multi-granularity fusion mutual information is expressed as follows:

[0026] When η1(I1)>max(η2(I2),η3(I3)):

[0027]

[0028] When η1(I1)≤max(η2(I2),η3(I3)):

[0029]

[0030] In the formula, m = 1, 2, 3, which represents normalized mutual information; It is a polynomial weighted summation of coarse-grained mutual information; α, β, γ, ρ are weighting coefficients that reflect the strength of the influence of each particle size level on the vibration quality.

[0031] As a further aspect of the present invention, the formula for calculating weighted data is expressed as follows:

[0032]

[0033] In the formula, x' ji x represents ji Weighted data formed after weighting processing.

[0034] As a further aspect of the present invention: the vibration characteristics include concrete water-cement ratio, concrete sand ratio, concrete temperature, vibration time, vibration depth and vibration frequency; and the vibration quality of concrete is characterized by compressive strength.

[0035] As a further aspect of this invention: weighted data replaces the feature values ​​of vibration features in the original samples to form weighted samples, and each group of weighted samples constitutes a weighted dataset; during the training of the LightGBM algorithm using the weighted dataset, the mean squared error is used as the loss function; the formula for calculating the mean squared error is as follows:

[0036]

[0037] In the formula, y i Indicates y i Predicted value; Indicates y i and y i The mean squared error between them.

[0038] As a further aspect of the present invention, the first derivative of the loss function is expressed as follows:

[0039]

[0040] In the formula, g i This represents the first derivative value of the i-th weighted sample group; This represents the multivariate differentiation symbol.

[0041] As a further aspect of the present invention: In the LightGBM algorithm, the calculation formula for the predicted vibration quality value is as follows:

[0042]

[0043] In the formula, This represents the predicted value during the (t+1)th iteration of the LightGBM algorithm. Let represent the predicted value in the t-th iteration of the LightGBM algorithm; η represents the learning rate. This indicates that in the LightGBM algorithm, the output value of the decision tree is obtained after t+1 rounds of iterative training.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] 1. This invention achieves accurate prediction of concrete vibration compaction quality by combining the AWMI and LightGBM algorithms. First, raw vibration data, including vibration characteristics and quality, is acquired, providing a foundation for subsequent processing. Then, the AWMI algorithm is used to weight the raw data, effectively enhancing the influence of key features and making the data more consistent with actual prediction needs. Next, the LightGBM algorithm is trained using the weighted data, resulting in a high-precision vibration quality prediction model. Finally, by inputting real-time vibration data, the model can quickly predict the corresponding vibration quality, providing strong support for quality control during construction. This method not only improves prediction accuracy but also increases construction efficiency, possessing significant practical value.

[0046] 2. A detailed weighted processing procedure ensures the accuracy and validity of the data. First, the original dataset is constructed, providing a foundation for subsequent calculations. Then, multi-granularity fusion mutual information and weighting functions are calculated, fully considering the relationship between feature values ​​and vibration quality, as well as the differences between various features. Finally, based on this information, the original dataset is weighted to obtain weighted data that better meets prediction requirements. This optimization step makes the prediction model trained subsequently more accurate, improving overall prediction performance.

[0047] 3. The original dataset is clearly and concisely represented, visually displaying the values ​​of vibration features and vibration quality in each original sample through a matrix format. This representation not only facilitates data processing and analysis but also aids in subsequent weighted processing and model training. Furthermore, by specifying the total number of vibration feature categories, a foundation is laid for calculating mutual information and the weighting function, further ensuring the accuracy and validity of the data.

[0048] 4. The mutual information calculation formula fully considers the correlation between vibration characteristics and vibration quality. By calculating the joint probability distribution and marginal probability distribution, the mutual information value between the two is obtained. This formula not only reflects the degree of influence of features on vibration quality but also provides an important basis for subsequent weighted processing. By optimizing the calculation of mutual information, the importance of each feature can be assessed more accurately, thereby improving the accuracy of the prediction model.

[0049] 5. The weighting function calculation formula achieves weighted processing of features by considering the variance and mean variance of each vibration feature. This formula not only considers the differences between features but also fully takes into account the range of feature values, making the weighted data more consistent with actual prediction needs. By optimizing the weighting function, the performance of the prediction model can be further improved, enhancing the accuracy and stability of predictions.

[0050] 6. The weighted data calculation formula combines mutual information and a weighting function to achieve weighted processing of the original data. This formula not only considers the correlation between features but also fully considers the importance of feature values, making the weighted data more consistent with prediction needs. By optimizing the calculation of weighted data, the accuracy of the prediction model can be further improved, providing stronger support for quality control during the construction process.

[0051] 7. By selecting vibration characteristics such as concrete water-cement ratio, concrete sand ratio, concrete temperature, vibration time, vibration depth, and vibration frequency, and by characterizing the vibration quality of concrete through compressive strength, this method can comprehensively reflect the key factors and results in the concrete vibration process. This characterization method not only meets actual construction needs but also provides a foundation for training subsequent prediction models. By optimizing the characterization methods for vibration characteristics and vibration quality, the accuracy and practicality of the prediction model can be further improved.

[0052] 8. By creating a weighted dataset and using the mean squared error as the loss function, this method can more accurately evaluate the performance of the prediction model. The weighted dataset fully considers the importance of each feature, making the model training more in line with actual prediction needs. Meanwhile, the mean squared error as the loss function can intuitively reflect the difference between predicted and actual values, providing an important basis for model optimization. By optimizing the choice of weighted dataset and loss function, the accuracy and stability of the prediction model can be further improved.

[0053] 9. The representation of the first derivative of the loss function facilitates the implementation of the gradient descent algorithm during subsequent model training. By calculating the first derivative, the direction and step size of model parameter updates can be obtained, thereby optimizing model performance. This representation not only simplifies the model training process but also improves training efficiency. Optimizing the calculation of the first derivative of the loss function can further improve the accuracy and convergence speed of the prediction model.

[0054] 10. The calculation formula for the vibration compaction quality prediction value in the LightGBM algorithm is used to progressively optimize model performance through iterative training. Each iteration adjusts the model based on the prediction results of the previous iteration, making the predicted value closer to the actual value. Simultaneously, by introducing parameters such as the learning rate and the output value of the decision tree, the model's complexity and generalization ability can be further controlled. This calculation formula not only improves the accuracy of predictions but also enhances the model's stability and robustness. By optimizing the calculation method for the vibration compaction quality prediction value in the LightGBM algorithm, the overall prediction performance can be further improved, providing more reliable support for quality control during construction. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the overall prediction process of the present invention.

[0056] Figure 2 This is a detailed prediction flowchart of the present invention.

[0057] Figure 3 This is a graph showing the predicted results of concrete vibration quality based on the AWMI-LightGBM model. Detailed Implementation

[0058] 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.

[0059] Please see Figure 1 and Figure 2 In this embodiment of the invention, a method for predicting the quality of concrete vibration based on the AWMI-LightGBM algorithm includes the following steps:

[0060] I. Obtaining the original dataset

[0061] Obtain raw vibration data during the concrete vibration process. The raw vibration data includes vibration characteristics and vibration quality.

[0062] In selecting concrete quality testing parameters, this invention combines engineering practice and testing accuracy requirements, and selects existing or easily obtainable vibration parameters and concrete characteristic parameters to form the original dataset. Concrete compressive strength is used as the testing index for concrete vibration quality. The vibration parameters are vibration time and vibration depth, and the concrete characteristic parameters are concrete water-cement ratio, sand ratio and temperature.

[0063] In practical engineering, due to various reasons such as fluctuations in raw material quality, environmental factors, and inconsistent construction operations, the characteristic parameters of concrete exhibit significant uncertainty, which in turn affects the accuracy of the trained model. Furthermore, the complex nonlinear relationship between features and target variables, especially when the number of input features is limited, severely impacts the accuracy of the prediction model. Therefore, this invention proposes an Adaptive Weighted Mutual Information (AWMI) algorithm based on multi-granularity fusion for data preprocessing. Compared to the traditional Weighted Mutual Information (WMI) algorithm, AWMI based on multi-granularity fusion can consider the complex influence of global statistics and local feature relationships on vibration quality. By adaptively adjusting feature weights, it effectively captures the complex relationship between features and the target vector, demonstrating advantages in addressing various uncertainties and complex dependencies in the concrete vibration process.

[0064] The water-cement ratio and sand ratio of concrete can be tested by taking samples on-site and using specialized equipment. For example, a concrete moisture content meter can be used to measure the water-cement ratio. The sand ratio is usually determined by laboratory analysis of the aggregate proportions in the concrete sample.

[0065] The temperature of concrete can be measured using a concrete thermometer or a thermal sensor. These devices can be inserted directly into the concrete to read temperature data in real time.

[0066] The vibration time can be recorded using a timer to ensure that the predetermined time is reached each time.

[0067] The vibration depth can be determined by measuring the insertion depth of the vibrator, which usually requires manual or automatic measurement during the vibration process.

[0068] The vibration frequency can be read from the frequency display of the vibration equipment, or calculated by measuring the number of vibrations of the vibrator per unit time.

[0069] II. Calculation of multi-particle-size weighted vibration data

[0070] 1. Construct the original dataset

[0071] The AWMI algorithm is used to weight the denoised vibration data to obtain the corresponding weighted vibration data. The AWMI algorithm can obtain complex dependencies between features and target variables by assigning different weights to different features.

[0072] Concrete water-cement ratio, concrete sand ratio, concrete temperature, vibration time, vibration depth, and vibration frequency are used as vibration features (input features), and concrete compressive strength is used as vibration quality (output feature). n sets of original samples (vibration features, vibration quality) are selected to construct the original dataset; the original dataset D is represented as:

[0073] D={(x ji ,y i )|j=1,2,…,J; i=1,2,…,n};

[0074] In the formula, x ji y represents the feature value of the j-th vibration feature in the i-th original sample group; i D represents the value of the vibration mass in the i-th original sample group; J represents the total number of vibration feature categories, which is 6 in this embodiment, i.e., D = {(x ji ,y i )|j=1,2,…,6;i=1,2,…,n}.

[0075] 2. Calculate multi-granularity weighted mutual information

[0076] Let x be the set of feature values ​​for the j-th vibration feature. j ={x j1 ,x j2 ,…,x jn The formula for calculating mutual information can be expressed as:

[0077]

[0078] The formula for calculating the weighting function is as follows:

[0079]

[0080] General feature-weighted methods only consider global features, limiting the capture of multi-dimensional influencing factors on concrete vibration quality. For example, temperature and vibration frequency may have a combined effect on the final compressive strength of concrete. Multi-grain size weighted mutual information can analyze the relationship between features and vibration quality at different grain sizes, thereby assessing the importance of features. Here, three levels of weighted mutual information calculation are used: coarse-grained, medium-grained, and fine-grained.

[0081] In coarse-grained analysis, the weighted mutual information between all vibration features and vibration quality is calculated to evaluate the impact of each feature on vibration quality at a global level. The calculation formula is as follows:

[0082]

[0083] Medium-grain size analysis can incorporate the direct or indirect relationships between undefined vibration characteristics in concrete vibration into the model training process. Based on the physical properties of the concrete vibration characteristics, the characteristics are divided into two subsets: S1 = {water-cement ratio, sand ratio, temperature}, and S2 = {vibration depth, vibration time, vibration frequency}. For each subset, the mutual information between it and the vibration quality is calculated, and the joint contribution of the feature subset to the vibration quality is analyzed. The calculation formula is as follows:

[0084]

[0085] Fine-grained analysis primarily examines the local relationships between features, calculating the mutual information between vibration features and their neighboring features to discover potential local dependencies between features. The calculation formula is as follows:

[0086]

[0087] In the formula, x j ,x j+1 Indicates the vibration characteristic x j and its neighboring features x j+1 Mutual information between them.

[0088] A weighted fusion method is used to sum the weighted mutual information of coarse-grained, medium-grained, and fine-grained data to obtain the final multi-grained comprehensive weighted mutual information.

[0089] When η1(I1)>max(η2(I2),η3(I3)):

[0090]

[0091] When η1(I1)≤max(η2(I2),η3(I3)):

[0092]

[0093] In the formula, m = 1, 2, 3, which represents normalized mutual information; It is a polynomial weighted summation of coarse-grained mutual information; α, β, γ, ρ are weighting coefficients that reflect the strength of the influence of each particle size level on the vibration quality.

[0094] Based on mutual information and a weighting function, the original dataset is weighted to obtain weighted data, which is then defined as weighted data.

[0095] The formula for calculating weighted data is as follows:

[0096]

[0097] In the formula, x' ji x represents ji Weighted data formed after weighting processing.

[0098] III. Obtaining a Prediction Model for Vibration Quality

[0099] The vibration feature values ​​in the original samples are replaced with weighted data to form weighted samples, and each group of weighted samples constitutes a weighted dataset. The LightGBM algorithm is trained using the weighted dataset to obtain the vibration quality prediction model.

[0100] The LightGBM algorithm is a gradient boosting ensemble method based on decision trees. When dealing with regression tasks, it can gradually reduce prediction errors by training multiple decision trees. The final output of the model is the weighted prediction result of all decision trees. LightGBM uses a histogram algorithm, which consumes less memory and significantly reduces the computational cost. Compared with other algorithms, it can effectively improve computational efficiency in practical engineering applications.

[0101] The modeling approach of the concrete vibration quality prediction model based on the LightGBM algorithm is as follows: Input the preprocessed concrete characteristic parameters and vibration parameter dataset, and use the histogram algorithm to find the optimal split point of the features; use a leaf-wise leaf growth strategy with depth constraints to generate a decision tree; calculate the residual of the first round of decision trees, and use the residual as the training sample for the next decision tree, continuously fitting the residual for iterative training; and sum the decision trees generated in each round with weights to obtain the final prediction model.

[0102] The LightGBM concrete vibration quality prediction model is a decision tree-based regression model. The model hyperparameters need to be set, and the relevant parameters are as follows.

[0103] 1. Generate a decision tree

[0104] Predicting the quality of concrete vibration is a regression problem. We use the mean squared error as the loss function, which is:

[0105]

[0106] In the formula, y i Indicates y i Predicted value; Indicates y i and y i The mean squared error between them.

[0107] The gradient is the first derivative of the loss function with respect to the model output, that is:

[0108]

[0109] In the formula, g i This represents the first derivative value of the i-th weighted sample group; This represents the multivariate differentiation symbol.

[0110] LightGBM iterates through each feature data point, calculates the gain of each possible split point, and selects the optimal split point. The gain is the information gain G after the split.

[0111]

[0112] Where, n k It is the left subset of the split point; n s It is the right subset of the split point.

[0113] By calculating the split gain, we can obtain the degree of improvement in model performance before and after a feature split. The split gain of each feature split point is compared to select the optimal split point. Then, among all candidate split points, the feature with the largest gain is selected for splitting. This process is repeated until preset stopping conditions such as the maximum tree depth or maximum number of leaves are met, thus completing the construction of the decision tree.

[0114] 2. Model Update

[0115] Based on the predicted values ​​of the already generated decision tree, the decision tree generation process is repeated, the sample gradient after the prediction value is updated is calculated, a new split point is selected to generate a new decision tree, and the newly generated decision tree is added to the previous round of predictions to update the model. That is:

[0116]

[0117] In the formula, This represents the predicted value during the (t+1)th iteration of the LightGBM algorithm. Let represent the predicted value in the t-th iteration of the LightGBM algorithm; η represents the learning rate. This indicates that in the LightGBM algorithm, the output value of the decision tree is obtained after t+1 rounds of iterative training.

[0118] This process is repeated continuously. Each new decision tree is built based on the predictions of the previous model, thus continuously updating the model to reduce errors and improve its predictive ability. If the model's error does not improve significantly within 10 iterations, training is stopped prematurely.

[0119] 3. Hyperparameter optimization

[0120] Hyperparameters are parameters that need to be manually set before model training and cannot be learned by the model itself. In the LightGBM model, the selection of hyperparameters directly affects the model's prediction performance, so they need to be set through experience or experimental tuning.

[0121] Random search, by randomly sampling within a predefined parameter space and then using a loss function to find the optimal parameter combination, can effectively solve the problem of low computational efficiency caused by the large parameter space of the LightGBM model, and is more practical than other hyperparameter optimization algorithms.

[0122] Let the set of hyperparameters to be optimized be:

[0123] θ = [θ1, θ2, θ3, θ4, θ5];

[0124] The hyperparameters θ1, θ2, θ3, θ4, and θ5 represent the maximum number of leaves, learning rate, tree depth, number of trees, and minimum number of samples per leaf node, respectively.

[0125] For each combination of hyperparameters selected by random search, 5-fold cross-validation is used to evaluate performance, and the optimal combination of parameters is finally selected.

[0126]

[0127] Where, θ best The optimal combination of hyperparameters; θ is the mean square error; m Θ represents the combination of model hyperparameters; Θ represents the hyperparameter space.

[0128] The optimal LightGBM model is obtained through hyperparameter tuning, and the trained model is evaluated by root mean square error (RMSE) on the test set.

[0129]

[0130] This invention achieves accurate prediction of concrete vibration compaction quality by combining the AWMI and LightGBM algorithms. First, raw vibration data, including vibration characteristics and quality, is acquired, providing a foundation for subsequent processing. Then, the AWMI algorithm is used to weight the raw data, effectively enhancing the influence of key features and making the data more consistent with actual prediction needs. Next, the LightGBM algorithm is trained using the weighted data, resulting in a high-precision vibration quality prediction model. Finally, by inputting real-time vibration data, the model can quickly predict the corresponding vibration quality, providing strong support for quality control during construction. This method not only improves prediction accuracy but also increases construction efficiency, demonstrating significant practical value.

[0131] The concrete vibration quality prediction results based on the AWMI-LightGBM model are as follows: Figure 3 As shown in Table 1, it can be observed that using multi-granularity fusion mutual information to preprocess concrete vibration data can effectively reduce the impact of abnormal data on the results, thus improving the prediction results. Table 1 shows that the AWMI-LightGBM model used in this invention has higher prediction accuracy for concrete vibration quality than the SVM and LSBoost models.

[0132] Table 13 Prediction Accuracy of Concrete Vibration Quality for Different Models

[0133]

[0134] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the quality of concrete vibration based on the AWMI-LightGBM algorithm, characterized in that, The prediction steps include the following: S1. Obtain the original vibration data during the concrete vibration process. The original vibration data includes vibration characteristics and vibration quality. S2. Use the multi-granularity fusion AWMI algorithm to weight the original vibration data to obtain the corresponding weighted vibration data; S3. Use weighted vibration data to train the LightGBM algorithm to obtain a vibration quality prediction model; S4. Obtain real-time vibration data during the concrete vibration process and input the real-time vibration data into the vibration quality prediction model to predict the vibration quality corresponding to the real-time vibration data. The specific details of step S2 are as follows: S21, Select The original dataset is constructed by denoising the original samples (vibration features, vibration quality); S22. Based on the feature values ​​and vibration quality values, calculate the multi-granularity fusion mutual information and weighting function of vibration features for all categories; S23. Based on multi-granularity fusion mutual information and weighting function, the original dataset is weighted to obtain weighted data, and this data is defined as weighted data. Original dataset Represented as: ； In the formula, Indicates the first In the original sample group, the first The characteristic values ​​of the vibration compaction feature; Indicates the first The values ​​of vibration mass in the original samples of the group; This represents the total number of categories of vibration compaction characteristics; The formula for calculating mutual information is as follows: ； In the formula, Indicates the first Vibration characteristics and vibration quality Mutual information between them; express and The joint probability distribution between them; express The marginal probability distribution; express The marginal probability distribution; The formula for calculating the weighting function is as follows: ； In the formula, Indicates the first A weighted function for various vibration characteristics; Represents the first in the original dataset The variance of the vibration characteristics; This represents the average variance of the vibration characteristics for all categories; The formula for calculating multi-granularity fusion mutual information is as follows: when hour: ； when hour: ； In the formula, It is normalized mutual information; It is a polynomial weighted summation of coarse-grained mutual information; It is a weighting coefficient that reflects the strength of the influence of each particle size level on the vibration quality; The formula for calculating weighted data is as follows: ； In the formula, express The weighted data is generated after weighting processing.

2. The method for predicting concrete vibration quality based on the AWMI-LightGBM algorithm according to claim 1, characterized in that, Vibration characteristics include concrete water-cement ratio, concrete sand ratio, concrete temperature, vibration time, vibration depth, and vibration frequency; the vibration quality of concrete is characterized by compressive strength.

3. The method for predicting concrete vibration quality based on the AWMI-LightGBM algorithm according to claim 2, characterized in that, Weighted data replaces the vibration feature values ​​in the original samples to form weighted samples, and each group of weighted samples constitutes a weighted dataset. During the training of the LightGBM algorithm using the weighted dataset, the mean squared error is used as the loss function. The formula for calculating the mean squared error is as follows: ； In the formula, express Predicted value; express and The mean squared error between them.

4. The method for predicting concrete vibration quality based on the AWMI-LightGBM algorithm according to claim 3, characterized in that, The first derivative of the loss function is expressed as follows: ； In the formula, Indicates the first The first derivative value of the group-weighted sample; Represents the multivariate differentiation symbol; In the LightGBM algorithm, the formula for calculating the predicted compaction quality is as follows: ； In the formula, Represents the first step of the LightGBM algorithm. Predicted values ​​from multiple rounds of iterative training; Represents the first step of the LightGBM algorithm. Predicted values ​​from multiple rounds of iterative training; Indicates the learning rate; In the LightGBM algorithm, Rounds of iterative training change the output value of the decision tree.

Images