A method and system for predicting coarse-grained soil permeability based on PCA-BP neural network
Patent Information
- Application Number
- CN202610782126.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-02
- Publication Date
- 2026-10-23
AI Technical Summary
传统的渗透系数测试方法主要依赖于实验室渗透试验和现场试验，这些方法通常耗时长、成本高，且受试验条件和设备精度的影响较大
[0019]1.本发明基于设置PCA-BP神经网络，处理目标土壤，以输出预测粗粒土渗透率，能够提供高效准确的预测粗粒土渗透率的方法。
Smart Images
Figure CN122965259A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of soil data processing technology, and in particular to a method and system for predicting the permeability of coarse-grained soil based on PCA-BP neural network. Background Technology
[0002] Coarse-grained soils are widely used in water conservancy, transportation, and foundation engineering, and their permeability directly affects the safety and stability of these projects. Traditional methods for testing permeability coefficients mainly rely on laboratory permeability tests and field tests. These methods are typically time-consuming, costly, and highly susceptible to variations in test conditions and equipment accuracy. Therefore, accurately predicting the permeability coefficient of coarse-grained soils is crucial not only for improving the scientific and economical nature of engineering design but also for providing an important basis for developing efficient seepage prevention strategies. Summary of the Invention
[0003] This invention proposes a method and system for predicting the permeability of coarse-grained soil based on PCA-BP neural network, so as to provide a method for accurately predicting the permeability coefficient of coarse-grained soil.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0005] A method for predicting the permeability of coarse-grained soil based on PCA-BP neural network includes:
[0006] A PCA-BP neural network is set up to process the target soil and output the predicted permeability of coarse-grained soil.
[0007] The setup of the PCA-BP neural network includes:
[0008] Using the full gradation d10~d100 and void ratio as input variables, the output variable is the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
[0009] Collect publicly available information to obtain raw data;
[0010] The original data is linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
[0011] The setup of the PCA-BP neural network includes: performing principal component analysis on the original data using the Bartlett test for sphericity to change the input variables of the training samples.
[0012] The PCA-BP neural network configuration includes: a BP neural network comprising data input, 10 hidden layers, 1 output layer, and data output.
[0013] A system for predicting coarse-grained soil permeability based on PCA-BP neural network includes a prediction module for setting up the PCA-BP neural network, processing the target soil, and outputting the predicted coarse-grained soil permeability.
[0014] The PCA-BP neural network setup includes: using the full gradation d10~d100 and void ratio as input variables, and the output variable being the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
[0015] The raw data is obtained by collecting publicly available information; the raw data is then linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
[0016] The setup of the PCA-BP neural network includes: performing principal component analysis on the original data using the Bartlett test for sphericity to change the input variables of the training samples.
[0017] The PCA-BP neural network configuration includes: a BP neural network comprising data input, 10 hidden layers, 1 output layer, and data output.
[0018] By adopting the above technical solution, the present invention has the following beneficial effects:
[0019] 1. This invention is based on setting up a PCA-BP neural network to process the target soil and output a predicted coarse-grained soil permeability, which can provide an efficient and accurate method for predicting coarse-grained soil permeability. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of a method for predicting the permeability of coarse-grained soil based on a PCA-BP neural network proposed in this invention. Detailed Implementation
[0021] 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.
[0022] like Figure 1 The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network, as shown, includes:
[0023] S1. Set up a PCA-BP neural network to process the target soil and output the predicted permeability of coarse-grained soil.
[0024] The setup of the PCA-BP neural network includes:
[0025] Using the full gradation d10~d100 and void ratio as input variables, the output variable is the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
[0026] Collect publicly available information to obtain raw data;
[0027] The original data is linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
[0028] The setup of the PCA-BP neural network includes: performing principal component analysis on the original data using the Bartlett test for sphericity to change the input variables of the training samples.
[0029] The PCA-BP neural network configuration includes: a BP neural network comprising data input, 10 hidden layers, 1 output layer, and data output.
[0030] A system for predicting coarse-grained soil permeability based on PCA-BP neural network includes a prediction module for setting up the PCA-BP neural network, processing the target soil, and outputting the predicted coarse-grained soil permeability.
[0031] The PCA-BP neural network setup includes: using the full gradation d10~d100 and void ratio as input variables, and the output variable being the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
[0032] The raw data is obtained by collecting publicly available information; the raw data is then linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
[0033] The setup of the PCA-BP neural network includes: performing principal component analysis on the original data using the Bartlett test for sphericity to change the input variables of the training samples.
[0034] The PCA-BP neural network configuration includes: a BP neural network comprising data input, 10 hidden layers, 1 output layer, and data output.
[0035] Example
[0036] 1. Backpropagation (BP) neural network and its working principle
[0037] A backpropagation (BP) neural network is a unidirectional propagation multilayer feedforward network, consisting of an input layer, an output layer, and hidden layers. Its computational flow is divided into two stages: forward propagation and backward propagation. The basic operating logic is as follows: First, the pre-processed learning sample input signal is passed to the input layer; then, this information is passed to the output layer through a nonlinear transformation of the hidden layer; in the output layer, the network calculates the corresponding predicted value. If the error between the predicted value and the true value does not reach a preset accuracy standard, the BP neural network will initiate its backward propagation mechanism. In this backward process, the error is propagated from the output layer back to the input layer, and the weights and thresholds of each layer are adjusted according to the error gradient. Through iterative iterations, the output error is gradually reduced until the required number of iterations or accuracy is reached. The optimized weights and thresholds are then used to predict new unknown input samples. When the input information of these unknown samples passes through the trained network, based on the learned weights and thresholds, the network can output the corresponding prediction result, thus achieving a mapping from input to output.
[0038] Principal Component Analysis (PCA) and its Working Principle
[0039] Principal Component Analysis (PCA), as an efficient data dimensionality reduction technique, focuses on transforming multi-dimensional and interconnected variables into several independent principal components through orthogonal transformation strategies in mathematics. These principal components not only significantly reduce the dimensionality of the dataset but also retain most of the key information in the original data, effectively simplifying complex data structures and making problem-solving more intuitive and efficient.
[0040] Predicting the permeability of coarse-grained soil involves numerous influencing factors. Although some representative indicators have been identified through preliminary screening, these indicators often have direct or indirect quantitative relationships, leading to information overlap. Each feature indicator theoretically has the potential to affect the result, but in practical applications, considering all indicators increases the complexity and time cost of data processing and model training. While backpropagation (BP) neural networks, as a powerful nonlinear modeling tool, possess excellent adaptive learning capabilities, their performance is limited by the quality and dimensionality of the input data. Too many input features may introduce unnecessary noise and may also lead to model overfitting, reducing generalization ability.
[0041] Therefore, applying PCA to the preprocessing stage of coarse-grained soil permeability prediction models is crucial. Using the principal components extracted by PCA as input vectors for neural networks not only optimizes the model's input structure and improves training efficiency, but also effectively eliminates redundant information between features, highlights key influencing factors, and further enhances the accuracy and generalization ability of coarse-grained soil permeability prediction models while retaining sufficient information to support accurate predictions.
[0042] 2.1 Working principle of principal component analysis
[0043] PCA can transform data from high-dimensional to low-dimensional while preserving its main features. This method maps the original high-dimensional data to a new low-dimensional coordinate system through a linear transformation, maximizing the variance of the data in the new coordinate system, thus achieving dimensionality reduction. When using PCA for dimensionality reduction, weights are assigned to the indicators based on the contribution rate of the principal components' variance, and the coefficients within the linear combination of principal components are processed.
[0044] 3. Permeability of coarse-grained soil
[0045] Coarse-grained soils are widely used in water conservancy, transportation, and foundation engineering, and their permeability directly affects the safety and stability of these projects. Traditional methods for testing permeability coefficients mainly rely on laboratory permeability tests and field tests. These methods are typically time-consuming, costly, and highly susceptible to variations in test conditions and equipment accuracy. Therefore, accurately predicting the permeability coefficient of coarse-grained soils is crucial not only for improving the scientific and economical nature of engineering design but also for providing an important basis for developing efficient seepage prevention strategies.
[0046] 3.1 Data Selection and Processing
[0047] To construct the neural network sample set, 80 sets of data were collected and summarized from published literature. Cross-validation was used to compare and analyze the prediction results of three models. The study used the full gradation d10~d100 and void ratio (e) as input variables, and the output variable was the permeability coefficient k of coarse-grained soil at standard temperature (20℃). 20 Before model training, to reduce the similarity between data and ensure that the neural network gives equal importance to all input variables during training, the mapminmax function in Matlab is used. This function is used in MATLAB for data preprocessing. It reduces data scale differences by linearly mapping the data to a specified interval [-1,1], thereby ensuring model stability and training efficiency.
[0048] 3.2 Data PCA Processing
[0049] The original dataset contains 80 * 11 = 880 data points. To eliminate multicollinearity and improve the prediction accuracy of the neural network model, principal component analysis (PCA) is required. A prerequisite for using PCA is that there should be a strong linear correlation between the variables in the original dataset. If the linear correlation between the original variables is very weak, and there is no simplified data structure, then performing PCA is meaningless. Therefore, when applying PCA, its applicability must first be statistically tested. Testing methods include Bartlett's test of sphericity and KMO analysis; this study uses KMO analysis.
[0050] (1)
[0051] In the formula: —Simple correlation coefficient; —Partial correlation coefficient. Generally, a KMO measure > 0.5 is sufficient for factor analysis; the higher the KMO measure, the more suitable the sample is for factor analysis. In this example, the KMO measure value is 0.7488, indicating that the sample is suitable for PCA analysis.
[0052] After normalizing the 11 input variables of 80 training samples, an 80x11 matrix was constructed. Principal component analysis was performed using the PCA function in MATLAB. The eigenvalues, variance contribution rates, and cumulative contribution rates of each principal component were calculated. Specific results are as follows:
[0053] Among the 11 principal components, principal component 1 contributed 96.35%, indicating that principal component 1 retained most of the information in the original data. Therefore, principal component 1 was chosen to represent the original data and was labeled PC1. The data volume was reduced from the original 80x11 to 80x1.
[0054] 4. Constructing and training a BP neural network
[0055] This method uses MATLAB to construct a backpropagation (BP) neural network, which includes data input, 10 hidden layers, 1 output layer, and data output. 60 of the 80 data points were used to train the neural network, and the remaining 20 were used to validate its performance. Of the 60 data points, 42 were used as the training set, 9 as the validation set, and 9 as the test set. The neural network converged after 10 iterations, achieving the optimal validation mean squared error of 0.0051438 on the fourth iteration. Regression analysis showed that the correlation coefficients of all four data sets were greater than 0.85. The majority of the neural network's error values were near the zero error line, indicating good performance.
[0056] 5. Verify the performance of the BP neural network.
[0057] The remaining 20 data points were used to verify the performance of the neural network. In MATLAB, the following command was used: output=sim(net,input), where output represents the predicted output of the neural network, sim is the simulation command, net is the previously trained BP neural network, and input represents the 20 input data points used for verification. The verification / analysis results show that 75% of the 20 data points have an error rate within 10%, indicating that the BP neural network performs well.
[0058] 6. Comparison with data without principal component analysis
[0059] After normalizing the 11 input variables of 80 training samples, an 80x11 matrix was constructed. This matrix was used as the input data for a backpropagation (BP) neural network, thus constructing and training the network. The network iterated more than 35 times, and the optimal validation mean squared error was 0.08406 at the 33rd iteration. The neural network trained on data without principal component analysis showed a significant disadvantage in training time and prediction accuracy compared to the aforementioned neural networks. This demonstrates that principal component analysis can significantly improve the performance of neural networks, and this method can be applied in engineering practice.
[0060] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.
Claims
1. A method for predicting the permeability of coarse-grained soil based on a PCA-BP neural network, characterized in that, include: A PCA-BP neural network is set up to process the target soil test data and output the predicted permeability of coarse-grained soil.
2. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 1, characterized in that, The setup of the PCA-BP neural network includes: Using the full gradation d10~d100 and void ratio as input variables, the output variable is the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
3. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 2, characterized in that, Collect publicly available information to obtain raw data; The original data is linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
4. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 3, characterized in that, The setup of the PCA-BP neural network includes: Principal component analysis was performed on the original data using the Bartlett test of sphericity to modify the input variables of the training samples.
5. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 4, characterized in that, The setup of the PCA-BP neural network includes: A backpropagation (BP) neural network consists of data input, 10 hidden layers, 1 output layer, and data output.
6. A system for predicting the permeability of coarse-grained soil based on a PCA-BP neural network, characterized in that, It includes a prediction module for setting up a PCA-BP neural network to process target soil detection data and output predicted coarse-grained soil permeability.
7. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 6, characterized in that, The setup of the PCA-BP neural network includes: Using the full gradation d10~d100 and void ratio as input variables, the output variable is the permeability coefficient k of coarse-grained soil at a standard temperature of 20°C. 20 .
8. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 7, characterized in that, Collect publicly available information to obtain raw data; The original data is linearly mapped to the specified interval [-1, 1] using the mapminmax function in Matlab.
9. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 8, characterized in that, The setup of the PCA-BP neural network includes: Principal component analysis was performed on the original data using the Bartlett test of sphericity to modify the input variables of the training samples.
10. The method for predicting the permeability of coarse-grained soil based on PCA-BP neural network according to claim 8, characterized in that, The setup of the PCA-BP neural network includes: A backpropagation (BP) neural network consists of data input, 10 hidden layers, 1 output layer, and data output.