Ground surface settlement classification prediction method based on BP neural network model

Abstract

The invention discloses a ground surface settlement classification prediction method based on a BP neural network model, and relates to the technical field of foundation pit engineering construction, a foundation pit monitoring technology is a technology used for monitoring parameters such as soil deformation, underground water level and ground surface settlement in foundation pit engineering, and the foundation pit monitoring technology obtains and analyzes data of related parameters in real time. And accurate evaluation on engineering deformation and safety conditions is provided. According to the scheme, ground surface settlement monitoring data including ground surface settlement data of monitoring points and data of related environmental factors are collected and arranged, a data set is divided into a training set and a test set, the training set is used for training and optimizing a BP neural network model, and an optimal network structure and parameter configuration are obtained; and verifying and evaluating the trained model by using a test set, calculating accuracy and performance indexes of classification prediction, performing classification and risk evaluation of ground surface settlement according to a prediction result, and taking corresponding measures to perform engineering management and decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit construction, and in particular to a classification prediction method for surface settlement based on a BP neural network model. Background Art

[0002] The construction of foundation pit projects involves processes such as excavation, support, and backfilling of underground soil layers. These processes often have a certain impact on the surrounding environment and structures. The construction of urban rail transit projects is usually carried out in the central area of the city, with a complex surrounding environment and variable geological conditions. Therefore, problems such as ground settlement, groundwater influx, and soil collapse may be faced during the construction process. In order to detect and solve these problems in a timely manner, the application of foundation pit monitoring technology is particularly important.

[0003] In the prior art, foundation pit monitoring technology is a technology used to monitor parameters such as soil deformation, groundwater level, and surface settlement in foundation pit projects. A foundation pit project refers to a temporary or permanent excavation structure formed during the construction of a building or engineering structure for the excavation, support, and backfilling of underground space. Foundation pit monitoring technology provides an accurate assessment of the engineering deformation and safety status, as well as early warning and control of potential risks during the construction process by obtaining and analyzing relevant parameter data in real time. Therefore, a classification prediction method for surface settlement based on a BP neural network model is needed to meet people's needs. Summary of the Invention

[0004] The purpose of the present invention is to provide a classification prediction method for surface settlement based on a BP neural network model to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A classification prediction method for surface settlement based on a BP neural network model, comprising the following steps:

[0006] Step 1: Establish a sample database;

[0007] During the construction of urban rail transit foundation pits, monitor the surface settlement, check the deformation of the soil around the retaining structure. Conduct the monitoring every two days during the construction of the retaining structure and the foundation pit excavation, and twice a week during the construction of the main structure. The warning value is 20 mm, and the project control value is 30 mm. The monitoring points are located at the midpoint of the short side and at intervals of 30 m along the length direction of the foundation pit;

[0008] Surface settlement monitoring:

[0009] Equipment installation process: Arrange surface settlement monitoring points around the foundation pit to cover the entire construction area, and transmit the data to the monitoring center through a wireless transmission system;

[0010] Step 2: Extract samples;

[0011] For 950 groups of settlement data samples, a group of settlement data samples was selected as the training set, and the remaining settlement data was used to select 50 regions of interest as the test set, and there was no overlap between the training set and the test set;

[0012] Step 3: Establish an algorithm model;

[0013] Establish a three-layer neural network based on the BP algorithm, including an input layer, a hidden layer, and an output layer; among them, the input layer: receives external signals and data, and uses the settlement data as the nodes of the input layer; the output layer: converts qualitative into quantitative output through the BP network neural model, and comprehensively evaluates the set output result; the hidden layer: constructs the network according to the neuron number algorithm to establish a simple and efficient model;

[0014] Step 4: Train and validate the algorithm model;

[0015] Set the parameters on the basis of the neural network tool, and train and validate the neural network. During the training and validation process, cooperate with the confusion matrix to show the accuracy of the classification result;

[0016] Step 5: Detection;

[0017] Introduce the settlement data to be recognized into the algorithm model for processing;

[0018] Step 6: Output the result;

[0019] Use historical monitoring data as input to learn the laws and characteristics of ground settlement, and realize the prediction of future ground settlement data and whether it exceeds the warning value of 2 cm.

[0020] Preferably, in the step 3, assume that the input layer has d neurons, X is the input layer data set, which itself is included in x1, x2…, xd, assume that the hidden layer has L neurons, and H is the hidden layer data set, which itself is included in h1, h2, …, hL;

[0021] The relationship between these data sets can be determined by the following formula:

[0022] Z (1) =W (1) ·X + b (1) (1)

[0023] Among them, W (1) is the weight matrix from the input layer to the hidden layer, with a size of W 1 (d×L); b (1) is the bias vector of the hidden layer, with a size of L×1; the output of the hidden layer is:

[0024] H=σ(Z (1) ) (2)

[0025] Where σ is the activation function, and the Tanh activation function is used in this paper:

[0026]

[0027] Assume that the output layer has k neurons, and the output layer dataset Y is included in y1, y2, …, yk, which can be expressed by the formula:

[0028] Z (2) = W (2) ·H + b (2) (4)

[0029] Where the weight matrix from the hidden layer W (2) (k×L to the output layer, with a size of k×L; b (2) The bias vector of the output layer, with a size of k×1;

[0030] The output of the output layer is:

[0031]

[0032] In the backpropagation stage, calculate the difference between the predicted output and the true output, and backpropagate this difference from the output layer to the hidden layer to adjust the weights and bias values of each layer to minimize the loss function. Assume that the mean squared error (MSE) is used as the loss function:

[0033]

[0034] Where Y is the true output, is the predicted output;

[0035] The gradient of the output layer:

[0036]

[0037] The gradient of the hidden layer:

[0038]

[0039] Finally, use the gradient descent method to update the weights and bias to obtain the best prediction result;

[0040] The weight update of the output layer:

[0041] W (2) ← W (2) - η·δ (2) ·(H) T (9)

[0042] The bias update of the output layer:

[0043] b (2) ← b(2) -η·δ (2) (10)

[0044] Weight update of the hidden layer:

[0045] W (1) ←W (1) -η·δ (1) ·(X) T (11)

[0046] Bias update of the hidden layer:

[0047] b (1) ←b (1) -η·δ (1) (12)

[0048] The learning process of the BP neural network is divided into two key stages: forward propagation and backward propagation. In the forward propagation stage, the input signal propagates from the input layer through the hidden layer to the output layer to generate a predicted output. In the backward propagation stage, by calculating the difference between the predicted output and the true output, and propagating this difference from the output layer back to the hidden layer, the weights and bias values of each layer are adjusted to minimize the loss function, thereby optimizing the network performance.

[0049] Preferably, in the confusion matrix of step four, where TP represents true positive, FN represents false negative, FP represents false positive, and TN represents true negative, the corresponding data of the model can be calculated by the following formula;

[0050] Accuracy is the proportion of the number of samples correctly classified by the classification model to the total number of samples:

[0051]

[0052] Precision (PC) is the number of positive samples correctly classified divided by the actual number of positive classes:

[0053]

[0054] Recall rate (SN) refers to the proportion of samples that are actually positive classes and are predicted as positive classes:

[0055]

[0056] Specificity (SP) measures the accuracy of the model in identifying negative class samples, that is, the proportion of samples that are truly negative classes among the samples predicted as negative classes by the model:

[0057]

[0058] F1 is the harmonic mean of precision and recall, providing a comprehensive evaluation that balances the precision and integrity of the model:

[0059]

[0060] Preferably, when the number of hidden neurons is different and the hidden layer is 2 layers with 20 hidden neurons in each layer, the accuracy rates of the training set and the test set are the highest.

[0061] The beneficial effects of the present invention are as follows:

[0062] In the present invention, this solution collects and collates surface settlement monitoring data, including surface settlement data of monitoring points and data of related environmental factors. The data set is divided into a training set and a test set. The BP neural network model is trained and optimized with the training set to obtain the best network structure and parameter configuration. Next, the trained model is verified and evaluated with the test set, the accuracy and performance indicators of classification prediction are calculated, surface settlement classification and risk assessment are carried out according to the prediction results, and corresponding measures are taken for project management and decision-making. Description of the Drawings

[0063] Figure 1 It is a schematic diagram of the BP neural network structure proposed by the present invention;

[0064] Figure 2 It is a schematic diagram of the surface settlement values obtained from 5 monitoring points proposed by the present invention;

[0065] Figure 3 It is a schematic diagram of the flow of a surface settlement classification prediction method based on a BP neural network model proposed by the present invention. Detailed Embodiments

[0066] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0067] In Embodiment 1, the following steps are included:

[0068] Step 1: Establish a sample database

[0069] During the construction of the foundation pit of urban rail transit, surface settlement is mainly monitored importantly to ensure the safety and stability of the project construction. The deformation of the soil around the retaining structure is mainly checked. It is carried out once every two days during the construction of the retaining structure and the excavation of the foundation pit, and twice a week during the construction of the main structure. The warning value is 20 mm, and the project control value is 30 mm. The midpoint of the short side and the spacing along the length direction of the foundation pit is 30 m (appropriately densified according to the surrounding building conditions);

[0070] Surface settlement monitoring:

[0071] Monitoring Instruments: Commonly used surface settlement monitoring instruments include level gauges, total stations, inclinometers, etc. These instruments can measure parameters such as relative surface elevation and tilt angle, and can transmit data to the monitoring center through a wireless transmission system;

[0072] Equipment Installation Process: The layout of surface settlement monitoring points usually selects a certain number of monitoring points around the foundation pit to cover the entire construction area;

[0073] The specific installation process is as follows:

[0074] Determine the Location of Monitoring Points: According to the engineering design and geological exploration data, select appropriate locations for the monitoring points to cover the areas where settlement is expected to occur;

[0075] Install Instruments: According to the monitoring requirements, use instruments such as level gauges and total stations to set up measurement points and install instruments;

[0076] Connect the Transmission System: Connect the monitoring instruments to the wireless transmission system to ensure that real-time data can be transmitted to the monitoring center;

[0077] Data Collection and Monitoring: Start the monitoring instruments and conduct data collection to monitor the surface settlement situation in real time;

[0078] Step 2: Extract Samples

[0079] Using image visualization environment software, 1000 groups of settlement data were manually extracted during the construction process. The settlement data includes geological conditions, soil layer properties, construction methods, and surface settlement data of the monitoring points. The specific operations are as follows: For 950 groups of settlement data samples, a group of settlement data samples was selected as the training set, and for the remaining settlement data, 50 regions of interest were selected as the test set. By processing the settlement data samples, a total of 100 settlement data were selected, among which 950 settlement data were set as the training set and 50 settlement data were set as the test set, and there is no overlap between the training set and the test set;

[0080] Step 3: Establish an Algorithm Model

[0081] Establish a three-layer neural network based on the BP algorithm, including an input layer, a hidden layer, and an output layer; among them, the input layer: receives external signals and data, and uses the settlement data as the nodes of the input layer; the output layer: converts qualitative data into quantitative output through the BP network neural model, and comprehensively evaluates the set output results; the hidden layer: constructs the network according to the neuron number algorithm to establish a simple and efficient model;

[0082] Step 4: Train and Validate the Algorithm Model

[0083] Set the parameters of the neural network tool and train and validate the neural network. During the training and validation process, a confusion matrix is used to show the accuracy of the classification results;

[0084] Step Five: Detection

[0085] Introduce the settlement data to be recognized into the algorithm model for processing;

[0086] Step Six: Output Results

[0087] Use historical monitoring data as input to learn the laws and characteristics of surface settlement, and realize the prediction of future surface settlement data and whether it exceeds the warning value of 2 cm;

[0088] Furthermore, in Step Three,

[0089] Assume that the input layer has d neurons, X is the input layer data set, which itself is included in x1, x2…, xd. Assume that the hidden layer has L neurons, and H is the hidden layer data set, which itself is included in h1, h2,…, hL.

[0090] The relationship between these data sets can be determined by the following formula:

[0091] Z (1) =W (1) ·X + b (1) (1)

[0092] Among them, W (1) is the weight matrix from the input layer to the hidden layer, with a size of W 1 (d×L); b (1) is the bias vector of the hidden layer, with a size of L×1. The output of the hidden layer (the value after the activation function) is:

[0093] H=σ(Z (1) ) (2)

[0094] In the formula, σ is the activation function. The Tanh activation function used in this article is:

[0095]

[0096] Assume that the output layer has k neurons, and the output layer data set Y is included in y1, y2,…, yk. It can be expressed by the formula:

[0097] Z (2) =W (2) ·H + b (2) (4)

[0098] In the formula, the weight matrix from the hidden layer W (2) (k×L to the output layer, with a size of k×L; b(2) The bias vector of the output layer, with a size of k×1.

[0099] The output of the output layer (the value after the activation function):

[0100]

[0101] In the backpropagation stage, calculate the difference between the predicted output and the true output (i.e., the loss function), and backpropagate this difference from the output layer to the hidden layer to adjust the weights and bias values of each layer to minimize the loss function. Assume the mean squared error (MSE) is used as the loss function:

[0102]

[0103] In the formula, Y is the true output, is the predicted output.

[0104] The gradient (error term) of the output layer:

[0105]

[0106] The gradient (error term) of the hidden layer:

[0107]

[0108] Finally, use the gradient descent method to update the weights and biases to obtain the best prediction results.

[0109] Weight update of the output layer:

[0110] W (2) ←W (2) -η·δ (2) ·(H) T (9)

[0111] Bias update of the output layer:

[0112] b (2) ←b (2) -η·δ (2) (10)

[0113] Weight update of the hidden layer:

[0114] W (1) ←W (1) -η·δ (1) ·(X) T (11)

[0115] Bias update of the hidden layer:

[0116] b (1) ←b (1) -η·δ (1)(12)

[0117] The learning process of the BP neural network is divided into two key stages: forward propagation and backward propagation. In the forward propagation stage, the input signal propagates from the input layer through the hidden layer to the output layer to generate a predicted output. In the backward propagation stage, by calculating the difference between the predicted output and the true output (i.e., the loss function), and propagating this difference from the output layer back to the hidden layer, the weights and bias values of each layer are adjusted to minimize the loss function, thereby optimizing the network performance.

[0118] Furthermore, in the confusion matrix of step four, where TP represents true positive, FN represents false negative, FP represents false positive, and TN represents true negative, the corresponding data of the model can be calculated through the following formula;

[0119] Accuracy is the proportion of the number of samples correctly classified by the classification model to the total number of samples:

[0120]

[0121] Precision (PC) is the ratio of the number of positive samples correctly classified to the actual positive class:

[0122]

[0123] Recall rate (SN) refers to the proportion of samples that are truly positive and are predicted to be positive:

[0124]

[0125] Specificity (SP) measures the accuracy of the model in identifying negative class samples, that is, the proportion of samples that are truly negative among the samples predicted to be negative by the model:

[0126]

[0127] F1 is the harmonic mean of precision and recall rate, providing a comprehensive evaluation that balances the precision and integrity of the model:

[0128]

[0129] Furthermore, under different numbers of hidden neurons, when the hidden layer is 2 layers and the number of hidden neurons in each layer is 20, the accuracy of the training set and the test set is the highest.

[0130] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for classifying and predicting ground settlement based on a BP neural network model, comprising the following steps: Step 1: Establish a sample database; During the construction of the foundation pit of urban rail transit, monitor the ground settlement, check the deformation of the soil around the retaining structure. It is carried out once every two days during the construction of the retaining structure and the excavation of the foundation pit, and twice a week during the construction of the main structure. The warning value is 20mm, and the project control value is 30mm. The midpoint of the short side, with a spacing of 30m along the length direction of the foundation pit; Ground settlement monitoring: Equipment installation process: Arrange ground settlement monitoring points around the foundation pit to cover the entire construction area, and transmit the data to the monitoring center through a wireless transmission system; Step 2: Extract samples; For 950 groups of settlement data samples, a group of settlement data samples are selected as the training set, and the remaining settlement data selects 50 regions of interest as the test set, and there is no overlap between the training set and the test set; Step 3: Establish an algorithm model; Establish a three-layer neural network based on the BP algorithm, including an input layer, a hidden layer, and an output layer; among them, the input layer: receives external signals and data, and uses the settlement data as the nodes of the input layer; Output layer: Qualitatively convert to quantitative output through the BP network neural model, and comprehensively evaluate the set output results; Hidden layer: Build a network according to the neuron number algorithm to establish a concise and efficient model; Step 4: Train and verify the algorithm model; Set the parameters on the basis of the neural network tool, and train and verify the neural network. During the training and verification process, cooperate with the confusion matrix to show the accuracy of the classification results; Step 5: Detection; Introduce the settlement data to be recognized into the algorithm model for processing; Step 6: Output results; Use historical monitoring data as input to learn the laws and characteristics of ground settlement, and realize the prediction of future ground settlement data, whether it exceeds the warning value of 2cm.

2. The method for classifying and predicting ground settlement based on the BP neural network model according to claim 1, wherein: In the said Step 3, assume that the input layer has d neurons, X is the input layer data set, which itself is included in x1, x2…, xd, assume that the hidden layer has L neurons, H is the hidden layer data set, which itself is included in h1, h2, …, hL; The relationship between these data sets can be determined by the following formula: Z (1) = W (1) ·X + b (1) (1) Among them, W (1) is the weight matrix from the input layer to the hidden layer, with a size of W 1 (d × L); b (1) is the bias vector of the hidden layer, with a size of L × 1; the output of the hidden layer is: H = σ(Z (1) ) (2) In the formula, σ is the activation function, and the Tanh activation function used in this article: Assume that the output layer has k neurons, and the output layer data set Y is included in y1, y2,…, yk, and can be expressed by the formula: Z (2) = W (2) ·H + b (2) (4) Where, the hidden layer W (2) (The weight matrix from the k×L to the output layer, with a size of k×L; b (2) The bias vector of the output layer, with a size of k×1; The output of the output layer is: In the backpropagation stage, calculate the difference between the predicted output and the true output, and backpropagate this difference from the output layer to the hidden layer to adjust the weights and bias values of each layer to minimize the loss function. Assume that the mean square error (MSE) is used as the loss function: where Y is the true output, is the predicted output; The gradient of the output layer: The gradient of the hidden layer: Finally, use the gradient descent method to update the weights and biases to obtain the best prediction results; Weight update of the output layer: W (2) ←W (2) -η·δ (2) ·(H) T (9) Bias update of the output layer: b (2) ←b (2) -η·δ (2) (10) Weight update of the hidden layer: W (1) ←W (1) -η·δ (1) ·(X) T (11) Bias update of the hidden layer: b (1) ←b (1) -η·δ (1) (12) The learning process of the BP neural network is divided into two key stages: forward propagation and backward propagation. In the forward propagation stage, the input signal propagates from the input layer through the hidden layer to the output layer to generate a predicted output. In the backward propagation stage, the difference between the predicted output and the true output is calculated, and this difference is propagated backward from the output layer to the hidden layer to adjust the weights and bias values of each layer to minimize the loss function, thereby optimizing the network performance.

3. A ground settlement classification prediction method based on the BP neural network model according to claim 1, characterized in that: For the confusion matrix in step 4, where TP represents true positive, FN represents false negative, FP represents false positive, and TN represents true negative, the corresponding data of the model can be calculated by the following formula; Accuracy is the proportion of the number of samples correctly classified by the classification model to the total number of samples: Precision (PC) is the ratio of the number of positive samples correctly classified to the actual positive class: Recall (SN) refers to the proportion of samples that are actually positive and are predicted to be positive: Specificity (SP) measures the accuracy of the model in identifying negative class samples, that is, the proportion of samples that are truly negative among the samples predicted to be negative by the model: F1 is the harmonic mean of precision and recall, providing a comprehensive evaluation that balances the precision and completeness of the model:

4. A method for classifying and predicting ground settlement based on a BP neural network model according to claim 1, characterized in that: Under different numbers of hidden neurons, when there are 2 hidden layers and 20 hidden neurons in each layer, the accuracy of the training set and the test set is the highest.

Images