A rapid quality assessment method for large batch concrete mixing piles

By combining distributed KPCA and Gustafson-Kessel clustering algorithms in Spark distributed architecture and RBF neural network model, the existing pile foundation detection efficiency is solved and the BP model is easily trapped in local optimality, and the pile foundation quality is achieved The rapid, stable and real-time evaluation improves detection efficiency and the authenticity of the model.

CN114595955BActive Publication Date: 2024-11-26SOUTHEAST UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202210197154.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-01
Publication Date
2024-11-26
Estimated Expiration
2042-03-01

AI Technical Summary

Technical Problem

The existing pile foundation quality detection methods are inefficient, costly and cannot effectively evaluate the overall pile quality. The BP neural network-based model is prone to local optimality and has a slow convergence speed.

Method used

The system is built based on Spark distributed architecture to realize distributed KPCA dimensionality reduction, and the mixing pile quality is evaluated using Gustafson-Kessel clustering combined with RBF neural network model. Through data preprocessing, distributed core principal component analysis and radial basis function Neural network models perform dimensionality reduction and quality evaluation.

Benefits of technology

It realizes fast, stable and real-time quality evaluation of large-scale pile foundations, improves detection efficiency and model authenticity, avoids local optimal problems, and has high computing speed and low latency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114595955B_ABST
    Figure CN114595955B_ABST
Patent Text Reader

Abstract

The present invention proposes a method for rapid quality assessment of large batches of concrete mixing piles, which solves the problems of low efficiency and poor real-time performance of quality assessment of concrete mixing piles. The specific method is as follows: a distributed data set is constructed based on existing engineering data. A distributed computing model is constructed based on the Spark platform, and distributed kernel principal component analysis is trained on the distributed data set to obtain a low-dimensional data set, and multiple groups of DKPCA models are obtained; then, on the Spark platform, the distributed low-dimensional data set is used to train the RBF neural network. During the evaluation, Kafka is used to transmit real-time data, and the samples to be evaluated are processed by the distributed KPCA model, and then the evaluation results are obtained by RBF neural network conversion. The distributed system can realize low-latency quality assessment of large-scale high-flow rate data, and give real-time suggestions for adjusting the operating parameters of the pile driver. Considering that the construction process has a great influence on the quality of the pile body, the pile driving process is parameterized and incorporated into the model to improve the authenticity and reliability of the model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of concrete pile formation, and in particular to a method for quickly evaluating the quality of large batches of concrete mixing piles. Background Art

[0002] As the foundation of the construction industry, the stability and firmness of pile foundation engineering is extremely important. A weak foundation will lead to extremely dangerous construction accidents. Therefore, the quality analysis and evaluation of piles is particularly important. For the quality inspection of pile foundations, the more commonly used technologies at home and abroad are: low-strain reflection wave method, low-strain dynamic pile test method, static load test method, sound wave transmission method and pile foundation inspection based on BP neural network. The inherent defect of the previous method is that it requires a lot of test costs and even requires sampling and testing of the pile body, resulting in low efficiency of large-scale inspection, relatively one-sided inspection, and unable to reflect the overall quality of piles; and the model based on BP neural network is prone to fall into local optimality and has a slow convergence speed.

[0003] In response to the above problems, this paper proposes to build a system based on Spark distributed architecture, realize distributed KPCA dimensionality reduction, and use Gustafson-Kessel clustering combined with RBF neural network model to realize the evaluation of mixing pile quality. This method can evaluate the quality of cement mixing piles more stably and quickly, and has real-time performance, which is of great significance to pile foundation engineering. Summary of the invention

[0004] The technical problem to be solved by the present invention is the detection problem of the quality of the mixing pile foundation. A system is provided based on the Spark distributed architecture to realize distributed KPCA dimensionality reduction, and Gustafson-Kessel clustering combined with the RBF neural network model is used to realize the evaluation of the mixing pile quality. The safety status of a large number of pile foundations can be intelligently evaluated, and the high real-time performance of the evaluation can be achieved, and guiding opinions are put forward for concrete piling projects.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for quickly evaluating the quality of large batches of concrete mixing piles comprises the following steps:

[0007] Based on the actual engineering measurement data, a distributed concrete mixing pile dataset is constructed and data preprocessing is performed. Distributed kernel principal component analysis (D-KPCA) is then used to reduce the dimensionality of high-dimensional data. Considering that the design length of the pile is closely related to the softness of the geology and customer needs, and the pile driving process has a great influence on the quality of the pile body, the design length of the pile and the pile driving process are parameterized and incorporated into the model as features. A distributed RBF neural network model is constructed, and the Gustafson-Kessel clustering algorithm is used to obtain the hidden layer nodes of the network model. The network is trained with the constructed distributed dataset to obtain the weights from the hidden layer to the output layer. Finally, the dataset to be evaluated is used as input, and the output obtained is the evaluation result.

[0008] Further, the specific steps include:

[0009] Step 1: Based on the existing data, a distributed data set is constructed and the data set is preprocessed. First, the data set is traversed to remove data rows with more than two missing values ​​and feature columns with missing value ratios exceeding 10%. Then, low variance filtering and high correlation filtering algorithms are used to select features of the data to filter out features with small variance and little implicit information; for multiple feature columns with high correlation, one of them is retained. Finally, the missing values ​​of the data set are filled. Here, the mean filling method is used to calculate the mean of the existing data in the column under the original data set and fill the missing empty rows. Finally, the data set is randomly sampled, and 30% of the data is taken out to hide the labels as a test set.

[0010] Step 2: Build a distributed KPCA model to perform data dimensionality reduction;

[0011] Step 3: Construct a quality assessment model of a distributed RBF neural network and use the Gustafson-Kessel algorithm to cluster and obtain the hidden layer of the RBF neural network;

[0012] Step 4: Determine whether the clustering result is stable. If it is stable, the training ends, otherwise go to step 3;

[0013] Step 5: Use the test data set to train the RBF neural network model to obtain the weights from the hidden layer to the output layer. Once the weights are determined, the prediction results can be obtained from the input. At this point, the model training is complete.

[0014] Step 6: Use Kafka middleware to pass the real-time pile driver parameters to be evaluated. After passing through the distributed KPCA model, they are converted through the RBF model. Finally, the distributed results are averaged to obtain the evaluation result.

[0015] Furthermore, in step 2, the distributed KPCA is an improvement on the algorithm implementation of the kernel principal component analysis (KPCA), and can effectively perform dimensionality reduction of nonlinear data of large data sets.

[0016] In the first step, the total training data set is shuffled and evenly distributed to P nodes. The size of P depends on the number of computing nodes and the amount of data on the Spark computing platform.

[0017] The second step is to execute the KPCA algorithm in parallel. For each sub-training data set of each computing node, the KPCA algorithm is executed in parallel, that is, for each sub-training data set, the original low-dimensional data X = [x 1 ,x 2 ,…,x N ] T After mapping to high dimensions, the first three feature columns with the largest eigenvalues ​​are selected. The specific process is as follows:

[0018] Define kernel function K = Φ(X) T Φ(X)=k(x i ,x j ).

[0019] Define the mapping φ:R n →H, then for any x i ∈R n , through the nonlinear transformation x i →φ(x i ), and x i Mapped into a high-dimensional space. Then the covariance matrix is

[0020]

[0021] The eigenvalue of C is λ, and the eigenvector is V, then

[0022] λV=CV (2)

[0023] The eigenvector V is given by {Φ(x 1 ),Φ(x 2 ),…,Φ(x m )}, then there exists a={a 1 ,a 2 ,…,a m}, so that

[0024]

[0025] Combining (3), (4), (5), we can get

[0026]

[0027] Substituting into the kernel function, we get

[0028]

[0029] Multiply both sides by Φ(x k ) T Available

[0030]

[0031] It can be abbreviated as Ka=mλa.

[0032] Here we use the Gaussian kernel As a mapping function.

[0033] h(x i )=V T Φ(x i ) (7)

[0034] Then we can find the vector x in the original space i The vector h(x) mapped to the high-dimensional space i ).

[0035] Sort the eigenvectors by eigenvalue from large to small, and take the eigenvectors corresponding to the first D eigenvalues ​​for orthogonal processing to obtain V D , D is the dimension after dimensionality reduction. Then x i After dimensionality reduction, it is V D T Φ(x i ).

[0036] The third step is to calculate the data samples after the dimension reduction of the training sample set X based on the results obtained in the second step, that is, P Spark nodes respectively calculate the data samples after the dimension reduction of the training sample set X

[0037] V Dp T Φ(X),p=1,2,…,P (8)

[0038] Furthermore, in step 3, the RBF neural network model is constructed as follows: Assume that the training set vector is X=[x 1 ,x 2 ,…,x N ] T RBF neural network needs to determine the center c of each basis function m , width δ m and weight ω m The basis function is based on the point x in the input space. n With center c m The distance is taken as the independent variable of the function. The form of the basis function is

[0039] φ m (||x n -c m ||),m=1,2,3,…,M (9)

[0040] φ m represents the basis function corresponding to the m-th hidden layer. The output based on the radial basis function is defined as a linear combination of basis functions, and the expression is as follows

[0041]

[0042] where F(x) is the network output value, and ω m is the weight corresponding to the m-th basis function.

[0043] Let the number of training samples be N, the number of hidden layer nodes be M, and d n be the corresponding output component generated by the input x n .

[0044] The output linear equations of the training samples are

[0045]

[0046] Furthermore, in step 3, the Gustafson-Kessel algorithm is used to cluster to find the data center of each hidden layer.

[0047] The data set is denoted as X, X = {x 1 , x 2 , …, x N} is a finite set. Suppose there is a membership matrix U = [u ik c×N representing all subsets of u ik , then the fuzzy partition space set of X is denoted as

[0048]

[0049] The objective function of the Gustafson-Kessel algorithm is

[0050]

[0051] where is the inner product distance, where A i is the norm-induced matrix, and the expression is where ρ i is defaulted to 1, F i is the fuzzy covariance matrix of the i-th cluster, and F i The expression is

[0052]

[0053] The implementation steps of the Gustafson-Kessel algorithm are as follows. Select the number of clusters 1 < c < N, and the weighting exponent m = 2:

[0054] ​Repeat l=1,2,…

[0055] Calculate cluster centers

[0056]

[0057] Calculate the fuzzy covariance matrix of clustering

[0058]

[0059] Calculate the inner product distance

[0060]

[0061] Update the partition matrix. but

[0062]

[0063] otherwise

[0064] Until ||U (l) -U (l-1) || stops iteration when it is less than a given threshold. At this time, the membership matrix U and the center matrix V are the final results.

[0065] Furthermore, in step 5, let Φ represent φ nm N×M matrix, φ nm Represents the nth training data set x n With the mth center point c m The value of the radial basis function, Φω=d,

[0066]

[0067] ω and d represent the coefficient vector of weights [ω 1 ,ω 2 ,…,ω M ] T and the output vector [d 1 ,d 2 ,…,d N ] T The coefficient vector is then expressed as

[0068] ω=φ + d (18)

[0069] where φ + is the generalized inverse.

[0070] Furthermore, in step 7, the real-time pile driver parameter to be evaluated transmitted by Kafka is X=[x 1 ,x 2 ,…,xK ] T After the distributed KPCA model dimensionality reduction processing, we get P group data X' p =[x' 1 ,x' 2 ,…,x' K ] T , where p = 1, 2, ..., P. Then, the distributed RBF model is used for transformation to obtain P groups of prediction results {Y 1 ,Y 2 ,…,Y P}, finally take the average of the distributed results to get the evaluation result

[0071] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0072] In view of the characteristics of large data volume, nonlinearity and strong feature coupling of pile data, the data preprocessing of the present invention adopts distributed KPCA, which improves the operation efficiency of KPCA while realizing nonlinear dimensionality reduction to better retain the original data characteristics;

[0073] The present invention takes into account the diversity of factors affecting the quality of cement mixing piles, and incorporates the designed length of the piles and the pile driving process into the model as parameters, thereby improving the authenticity and reliability of the model.

[0074] After preprocessing, the pile foundation data is distributed in a partial elliptical shape in three-dimensional space. In view of this feature, the present invention adopts the Gustafson-Kessel algorithm to obtain the hidden layer nodes of the network model. Compared with the general clustering algorithm, the algorithm has better robustness.

[0075] The evaluation model is established using the memory-based Spark distributed framework. Compared with the ordinary BP model, the Spark-based RBF neural network model can not only solve the disadvantage of being easily trapped in the local optimum, but also has faster calculation speed and better real-time performance. It can achieve low-latency quality evaluation of large-scale high-velocity data, and has real-time performance, which can realize real-time adjustment of pile driver operating parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a model training flow chart of the present invention.

[0077] Figure 2 It is an evaluation flow chart of the present invention.

[0078] Figure 3 It is a flow chart of determining the partition matrix U and the center matrix V of the present invention. DETAILED DESCRIPTION

[0079] The technical solution provided by the present invention will be described in detail below in conjunction with specific embodiments. It should be understood that the following specific implementation methods are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0080] This example first constructs a distributed data set based on existing engineering data. Relevant algorithms are used to perform a series of preprocessing on the data, such as feature selection and missing value filling. Distributed kernel principal component analysis (D-KPCA) is then used to reduce the dimension of the constructed data set to obtain a three-dimensional distributed data set. A distributed RBF neural network algorithm is then used to construct a quality prediction model. The hidden layer of the RBF neural network is obtained using the fuzzy algorithm Gustafson-Kessel, and the model is trained using a distributed data set. Finally, the quality of cement mixing piles is evaluated using the trained RBF neural network model. Taking into account the diversity of factors affecting the quality of cement mixing piles, the present invention trains soft soil characteristics, curing agents, and construction processes as features to improve the authenticity and reliability of the model, specifically including the following steps:

[0081] Step 1: Based on the existing data, a distributed dataset is constructed. The original dataset contains the following multiple valid feature columns: longitude, latitude, running time, expected length, actual length, spraying time, total slurry, total concrete, upspeed, downspeed, internal current, external current, verticality, state, score, and label. The dataset is preprocessed here. First, the dataset is traversed to remove data rows with more than two missing values ​​and feature columns with a missing value ratio of more than 10%. Then, low variance filtering and high correlation filtering algorithms are used to select features of the data to filter out features with small variance and less implicit information; for multiple feature columns with high correlation, one of them is retained. Finally, the missing values ​​of the dataset are filled. Here, the mean filling method is used to calculate the mean of the existing data in the column under the original dataset and fill the missing empty rows. Finally, the dataset is randomly sampled, and 30% of the data is taken out to hide the labels as the test set.

[0082] Step 2: Construct a distributed KPCA model to perform data dimensionality reduction. Distributed KPCA is an improvement on the algorithm implementation of kernel principal component analysis (KPCA), which can effectively reduce the dimensionality of nonlinear data of large data sets.

[0083] In the first step, the total training data set is shuffled and evenly distributed to P nodes. The size of P depends on the number of computing nodes and the amount of data on the Spark computing platform.

[0084] The second step is to execute the KPCA algorithm in parallel. For each sub-training data set of each computing node, the KPCA algorithm is executed in parallel, that is, for each sub-data set, the original data set X = [x 1 ,x 2 ,…,x N ] T After mapping to high dimensions, the first three feature columns with the largest eigenvalues ​​are selected. The specific process is as follows:

[0085] Define kernel function K = Φ(X) T Φ(X)=k(x i ,x j ).

[0086] Define the mapping φ:R n →H, then for any x i ∈R n , through the nonlinear transformation x i →φ(x i ), and x i Mapped into a high-dimensional space. Then the covariance matrix is

[0087]

[0088] The eigenvalue of C is λ, and the eigenvector is V, then

[0089] λV=CV (2)

[0090] The eigenvector V is given by {Φ(x 1 ),Φ(x 2 ),…,Φ(x m )}, then there exists a={a 1 ,a 2 ,…,a m}, so that

[0091]

[0092] Combining (3), (4), (5), we can get

[0093]

[0094] Substituting into the kernel function, we get

[0095]

[0096] Multiply both sides by Φ(x k )T Available

[0097]

[0098] It can be abbreviated as Ka=mλa.

[0099] Here we use the Gaussian kernel As a mapping function.

[0100] h(x i )=V T Φ(x i ) (7)

[0101] Then we can find the vector x in the original space i The vector h(x) mapped to the high-dimensional space i ).

[0102] Sort the eigenvectors by eigenvalue from large to small, and take the eigenvectors corresponding to the first D eigenvalues ​​for orthogonal processing to obtain V D , D is the dimension after dimensionality reduction. Then x i After dimensionality reduction, it is V D T Φ(x i ).

[0103] The third step is to calculate the data samples after the dimension reduction of the training sample set X based on the results obtained in the second step, that is, P Spark nodes respectively calculate the data samples after the dimension reduction of the training sample set X

[0104] V Dp T Φ(X),p=1,2,…,P (8)

[0105] Furthermore, in step 3, the RBF neural network model is constructed as follows: Assume that the training set vector is X=[x 1 ,x 2 ,…,x N ] T RBF neural network needs to determine the center c of each basis function m , width δ m and weight ω m The basis function is based on the point x in the input space. n With center c m The distance is taken as the independent variable of the function. The form of the basis function is

[0106] φ m (||x n -c m ||),m=1,2,3,…,M (9)

[0107] φ mDenote the basis function corresponding to the m-th hidden layer. The output based on the radial basis function is defined as a linear combination of basis functions, and the expression is as follows

[0108]

[0109] In the formula, F(x) is the network output value, and ω m is the weight corresponding to the m-th basis function.

[0110] Let the number of training samples be N, the number of hidden layer nodes be M, and d n be the corresponding output component generated by the input x n .

[0111] The output linear equations of the training samples are

[0112]

[0113] Step 3: Construct a quality evaluation model for the distributed RBF neural network, and use the Gustafson-Kessel algorithm for clustering to obtain the hidden layer of the RBF neural network.

[0114] The data set is denoted as X, X = {x 1 , x 2 , …, x N} is a finite set. Assume there is a membership matrix U = [u ik c×N representing all subsets of u ik , then denote the fuzzy partition space set of X as

[0115]

[0116] The objective function of the Gustafson-Kessel algorithm is

[0117]

[0118] where is the inner product distance, where A i is the norm-induced matrix, and the expression is In the formula, ρ i is default taken as 1, F i is the fuzzy covariance matrix of the i-th cluster, and F i The expression is

[0119]

[0120] The implementation steps of the Gustafson-Kessel algorithm are as follows. Select the number of clusters 1 < c < N, and the weighting exponent m = 2:

[0121] Repeat l = 1, 2, …​

[0122] Calculate cluster centers

[0123]

[0124] Calculate the fuzzy covariance matrix of clustering

[0125]

[0126] Calculate inner product distance

[0127]

[0128] Update the partition matrix. but

[0129]

[0130] otherwise

[0131] Until ||U (l) -U (l-1) || stops iteration when it is less than a given threshold. At this time, the membership matrix U and the center matrix V are the final results.

[0132] Furthermore, in step 5, let Φ represent φ nm N×M matrix, φ nm Represents the nth training data set x n With the mth center point c m The value of the radial basis function, Φω=d,

[0133]

[0134] ω and d represent the coefficient vector of weights [ω 1 ,ω 2 ,…,ω M ] T and the output vector [d 1 ,d 2 ,…,d N ] T The coefficient vector is then expressed as

[0135] ω=φ + d (18)

[0136] where φ + is the generalized inverse.

[0137] Furthermore, in step 7, the real-time pile driver parameter to be evaluated transmitted by Kafka is X=[x 1 ,x 2 ,…,x K ]T After the distributed KPCA model dimensionality reduction processing, we get P group data X' p =[x' 1 ,x' 2 ,…,x' K ] T , where p = 1, 2, ..., P. Then, the distributed RBF model is used for transformation to obtain P groups of prediction results {Y 1 ,Y 2 ,…,Y P}, finally take the average of the distributed results to get the evaluation result

[0138] Step 4: Determine whether the clustering result is stable. If it is stable, the training ends, otherwise go to step 3;

[0139] Step 5: Use the test data set to train the RBF neural network model to obtain the weights from the hidden layer to the output layer. Once the weights are determined, the prediction results can be obtained from the input. At this point, the model training is complete.

[0140] Step 6: Use Kafka middleware to pass the real-time pile driver parameters to be evaluated, and after passing through the distributed KPCA model, they are converted through the RBF model. Finally, the distributed results are averaged to obtain the evaluation result. Let the real-time pile driver parameters to be evaluated be X = [x 1 ,x 2 ,…,x K ] T After the distributed KPCA model dimensionality reduction processing, we get P group data X' p =[x' 1 ,x' 2 ,…,x' K ] T , where p = 1, 2, ..., P. Then, the distributed RBF model is used to predict the P group evaluation results {Y 1 ,Y 2 ,…,Y P}, finally take the average of the distributed results to get the evaluation result

Claims

1. A method for rapid quality assessment of large batches of concrete mixing piles, characterized by: First, a distributed data set is constructed based on the existing data. Then, a distributed KPCA model is constructed based on the Spark platform to reduce the dimension of the data set, and a distributed RBF neural network prediction model is established. The hidden layer nodes of the network model are obtained using a clustering algorithm, and the network is trained with the existing data set to obtain the weights from the hidden layer to the output layer. Finally, the data stream to be evaluated is used as input, and the output obtained is the evaluation result. The specific steps include: Step 1: Based on the existing data, a distributed data set is constructed. The original data set contains the following multiple valid feature columns: longitude, latitude, running time, expected length, actual length, spraying time, total slurry, total concrete, upspeed, downspeed, internal current, internal current, external current, verticality, state, score, label. The data set is preprocessed here. First, the data set is traversed to remove data rows with more than two missing values ​​and feature columns with missing value ratios exceeding 10%. Then, low variance filtering and high correlation filtering algorithms are used to select features of the data to filter out features with small variance and less implicit information. For multiple feature columns with high correlation, one is retained. Finally, the missing values ​​of the data set are filled. Here, the mean filling method is used to calculate the mean of the existing data in the column under the original data set and fill the missing empty rows. Finally, the data set is randomly sampled, and 30% of the data is taken out to hide the labels as a test set. Step 2: Build a distributed KPCA model to perform data dimensionality reduction; Step 3: Construct a quality assessment model of a distributed RBF neural network and use the Gustafson-Kessel algorithm to cluster and obtain the hidden layer of the RBF neural network; Step 4: Determine whether the clustering result is stable. If it is stable, the training ends. Otherwise, go to step 3. Step 5: Use the test data set to train the RBF neural network model to obtain the weights from the hidden layer to the output layer. Once the weights are determined, the prediction results can be obtained from the input. At this point, the model training is complete. Step 6: Use Kafka middleware to transfer the real-time pile driver parameters to be evaluated. After passing through the distributed KPCA model, they are converted through the RBF model. Finally, the distributed results are averaged to obtain the evaluation result. In the step 2, a distributed kernel-based principal component analysis dimensionality reduction method D-KPCA is used. This is a nonlinear data processing method, and its core idea is to project the data in the original space into a high-dimensional feature space through a nonlinear mapping, and then use PCA to reduce the dimensionality of the data in the high-dimensional feature space. Compared with PCA, the KPCA method can achieve high-dimensional linear separability of any data as long as a suitable kernel function is used, so that more features can be retained after dimensionality reduction. The distributed computing method can greatly improve the running speed and alleviate the problem of slow KPCA calculation under high-dimensional and large data sets.

2. A method for rapid quality assessment of large batch concrete mixing piles as claimed in claim 1, characterized in that: In step 1, the data set is actual test pile data in the project; The sample labels will be labeled "0" or "1" based on the sample test results of the piles, combined with the engineering parameters of pile length and inclination, and combined with expert experience, where "0" represents unqualified and "1" represents qualified.

3. A method for rapid quality assessment of large batch concrete mixing piles as claimed in claim 1, characterized in that: In step 3, the RBF neural network is used because the model can approximate any nonlinear function with arbitrary precision and has global approximation capability, which solves the defect that the BP neural network is prone to fall into local optimality; on a distributed platform, the RBF neural network is easier to build than the BP neural network and occupies less memory; therefore, the distributed RBF neural network is used to construct the prediction model based on comprehensive consideration.

4. A method for rapid quality assessment of large batch concrete mixing piles as claimed in claim 1, characterized in that: In step 3, the Gustafson-Kessel algorithm clustering is used to calculate the hidden layer; in the case of a large amount of data, the algorithm has a faster calculation speed than the gradient descent method; compared with the KMeans clustering algorithm, which is also a clustering algorithm and the cluster types are limited to convex or circular shapes, the Gustafson-Kessel algorithm can handle a richer shape of clusters.

5. A method for rapid quality assessment of large batch concrete mixing piles as claimed in claim 1, characterized in that: In step 6, Kafka's high throughput and distributed architecture can greatly reduce data transmission delays; and Spark's memory-based computing model can quickly process data and achieve real-time quality assessment.

Citation Information

Patent Citations

  • High-confidence reconstruction quality and false reduction transient quantitative evaluation method based on cascaded neural network

    CN110837523A

  • Pile foundation horizontal bearing capacity prediction method based on BAS-BP

    CN112926117A

  • Network operation and maintenance situation assessment method based on machine learning

    CN113225346A