Foundation pit deformation intelligent prediction method and system considering earth excavation sequence

Abstract

The invention relates to an intelligent foundation pit deformation prediction method and system considering an earth excavation sequence. The method comprises the following steps: under the condition of considering an excavation sequence, processing foundation pit data in batches to obtain foundation pit response data, establishing an excavation database by utilizing the foundation pit response data, generating a sample data set, dividing the sample data set into a training set and a test set, and then performing joint training by utilizing the training set and multiple neural networks to obtain multiple prediction networks; and finally, testing the plurality of prediction networks by using the test set, comparing test results to output an optimal prediction network, and predicting the to-be-processed foundation pit data by using the optimal prediction network to obtain a foundation pit deformation prediction result. Compared with the prior art, the method has the advantages of high accuracy, good prediction performance, wide application range and high practicability.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit deformation prediction, and in particular to a foundation pit deformation intelligent prediction method and system taking earth excavation sequence into consideration. Background Art

[0002] With the deepening of my country's urbanization process, the continuous increase in building density, and the improvement of urban construction, the difficulty of constructing foundation pit projects in urban center areas has become increasingly greater. On the one hand, the surrounding environment of urban center areas is complex, with overpasses, trunk roads, new and old buildings, subway pipelines, etc. crisscrossing. These structures maintain people's daily lives. Once uneven settlement or cracking damage occurs due to foundation pit excavation, it will affect the basic functions, causing great economic losses and adverse social impacts. On the other hand, when the foundation pit is designed and constructed, deformation control is not paid attention to, resulting in frequent accidents of road cracking, surface subsidence, building cracking, tunnel cracking and other environmental damage accidents. It can be seen that when constructing foundation pit projects in urban center areas, it is particularly important to take appropriate engineering measures to control foundation pit deformation.

[0003] The deformation of foundation pit has a typical "time-space effect", which has been widely confirmed by engineering practice. The commonly used method to determine the excavation sequence is to select several reasonable excavation sequences based on engineering experience for comparison and selection. However, this method is very dependent on engineering experience and is difficult to promote on a large scale. On the other hand, this method can only obtain a local optimal solution and can hardly obtain a global optimal solution. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and system for intelligent prediction of foundation pit deformation taking into account the sequence of earth excavation.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to one aspect of the present invention, there is provided a method for intelligently predicting foundation pit deformation taking into account the excavation sequence of earthwork, comprising the following steps:

[0007] S1. Obtain foundation pit data;

[0008] S2. Considering the excavation sequence, batch process the foundation pit data to obtain the foundation pit response data and use it to establish an excavation database;

[0009] S3, processing the excavation database to obtain a sample data set, and dividing the sample data set into a training set and a test set;

[0010] S4, using the training set to jointly train with multiple neural networks to obtain multiple prediction networks;

[0011] S5. Use the test set to test multiple prediction networks, compare the test results to output the optimal prediction network, and use it to predict the foundation pit data to be processed to obtain the foundation pit deformation prediction result.

[0012] As a preferred technical solution, in S2, the specific process of batch processing foundation pit data is as follows, taking into account the excavation sequence:

[0013] S21, dividing the foundation pit data into blocks, and establishing a finite element model based on the blocks;

[0014] S22, calculating the foundation pit response data in the finite element model;

[0015] S23, changing the excavation sequence in the foundation pit data and returning to step S21 until all excavation sequences are traversed.

[0016] As an optimal technical solution, the block division process is specifically as follows: using the excavation sequence to obtain the excavation status of the soil on the left and right sides of each block, when the excavation status is not excavated, consider slope reduction; when the excavation status is excavated, consider completely excavating the remaining soil to determine the shape of the block; then combine the foundation pit size to obtain the point data of each block, and finally divide the excavation surface into different block forms according to the point data.

[0017] As a preferred technical solution, the specific process of processing the excavation database in S3 is: firstly, the blocks of the foundation pit are numbered accordingly to obtain a discrete sequence, and then the discrete sequence is encoded;

[0018] As a preferred technical solution, the encoding processing method includes natural encoding or one-hot encoding.

[0019] As a preferred technical solution, the various neural networks in S4 include BP neural network, recurrent neural network, and convolutional neural network.

[0020] As a preferred technical solution, the test results of S5 include five model evaluation indicators: SmoothL1 loss function, explained variance, root mean square error, mean absolute error and determination coefficient; among them, the closer the SmoothL1 loss function, root mean square error and mean absolute error are to 0, and the closer the explained variance and determination coefficient are to 1, the better the model performance is.

[0021] As a preferred technical solution, the foundation pit deformation prediction results in S5 include foundation pit deformation data, retaining wall bending moment data, maximum lateral displacement of the foundation pit retaining wall, maximum settlement of the surface soil, maximum uplift of the foundation pit bottom and maximum bending moment of the foundation pit retaining wall.

[0022] As a preferred technical solution, the foundation pit deformation prediction results in S5 need to be normalized, and the normalization process uses a linear normalization formula:

[0023]

[0024] Where y is the predicted value of foundation pit deformation, min(Y) represents the minimum value of the predicted value of foundation pit deformation, max(Y) represents the maximum value of the predicted value of foundation pit deformation, and y′ is the normalized predicted value of foundation pit deformation.

[0025] According to another aspect of the present invention, there is provided a foundation pit deformation intelligent prediction system considering the earthwork excavation sequence, the system applies the above-mentioned foundation pit deformation intelligent prediction method considering the earthwork excavation sequence, the system includes a data acquisition and processing module, a model joint training module, a model testing and optimization module and a prediction execution module;

[0026] Among them, the data acquisition and processing module is used to obtain foundation pit data, and on the basis of considering the excavation sequence, batch process the foundation pit data to obtain the foundation pit response data and establish an excavation database, and then process the excavation database to obtain the sample data set, and divide it into a training set and a test set; the model joint training module is used to use the training set to jointly train with multiple neural networks to obtain multiple prediction networks; the model testing and optimization module is used to test multiple prediction networks using the test set, and output the optimal prediction network by comparing the test results; the prediction execution module is used to use the output optimal prediction network to predict the foundation pit data to be processed, and finally obtain the foundation pit deformation prediction result.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. The present invention batch processes foundation pit data while considering the excavation sequence, obtains foundation pit response data and uses it to establish an excavation database, generate a sample data set, and divide the sample data set into a training set and a test set, and then uses the training set to jointly train with multiple neural networks to obtain multiple prediction networks; finally, the test set is used to test the multiple prediction networks, and the test results are compared to output the optimal prediction network, and the optimal prediction network is used to predict the foundation pit data to be processed to obtain the foundation pit deformation prediction result. It considers the excavation sequence, is more accurate in deformation prediction, and is more reasonable in predicting the characteristics of foundation pit deformation changing with foundation pit excavation, so that its prediction performance is good, and provides a scientific basis for the earth excavation sequence during foundation pit excavation.

[0029] 2. In the present invention, the excavation database is used for processing. First, the blocks of the foundation pit are numbered accordingly to obtain a discrete sequence, and then the discrete sequence is coded; the coding processing method includes natural coding or one-hot coding. The earth excavation sequence is quantified, and the three-dimensional space-time effect is quantified by natural coding and one-hot coding by processing the excavation surface as input, and the foundation pit deformation is used as output to form an intelligent decision-making model, making it more suitable for machine automation process, enhancing practicality, and providing qualitative support for the influence of earth excavation sequence on retaining wall deformation.

[0030] 3. The foundation pit deformation prediction results in the present invention include foundation pit deformation data, retaining wall bending moment data, maximum lateral displacement of the foundation pit retaining wall, maximum settlement of the surface soil, maximum uplift of the foundation pit bottom and maximum bending moment of the foundation pit retaining wall, which makes it general and broadens the scope of application of the method.

[0031] 4. The present invention tests a variety of prediction networks to obtain test results, and the test results include five model evaluation indicators: SmoothL1 loss function, explained variance, root mean square error, mean absolute error and coefficient of determination; it considers multiple indicators and selects the global optimal network model through multi-faceted comparison, so that the final prediction model balances multiple capabilities including prediction ability and information capture ability to achieve optimal performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the steps of the intelligent prediction method of foundation pit deformation considering the earthwork excavation sequence in the present invention;

[0033] Figure 2 It is a schematic diagram of the excavation scheme in the embodiment;

[0034] Figure 3a It is a schematic diagram of block numbering of two-dimensional foundation pit excavation in the embodiment;

[0035] Figure 3b It is a schematic diagram of block numbering for three-dimensional foundation pit excavation in the embodiment;

[0036] Figure 4 It is the plan view of the foundation pit of the project in the embodiment;

[0037] Figure 5 It is a cross-sectional view of the foundation pit of the project in the embodiment;

[0038] Figure 6 It is a schematic diagram of a predetermined excavation plan for a project in the embodiment;

[0039] Figure 7a It is a schematic diagram of the north retaining wall data in the project foundation pit response in the embodiment;

[0040] Figure 7bIt is a schematic diagram of the north retaining wall data in the project foundation pit response in the embodiment;

[0041] Figure 7c It is a schematic diagram of the pit bottom uplift data in the project foundation pit response in the embodiment. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0043] With the deepening of my country's urbanization process, the continuous increase in building density, and the improvement of urban construction, the difficulty of constructing foundation pit projects in urban center areas has become increasingly greater. On the one hand, the surrounding environment of urban center areas is complex, with overpasses, trunk roads, new and old buildings, subway pipelines, etc. crisscrossing. These structures maintain people's daily lives. Once uneven settlement or cracking damage occurs due to foundation pit excavation, it will affect the basic functions, causing great economic losses and adverse social impacts. On the other hand, when the foundation pit is designed and constructed, deformation control is not paid attention to, resulting in frequent accidents of road cracking, surface subsidence, building cracking, tunnel cracking and other environmental damage accidents. It can be seen that when constructing foundation pit projects in urban center areas, it is particularly important to take appropriate engineering measures to control foundation pit deformation.

[0044] A commonly used measure to control foundation pit deformation is to optimize the excavation sequence of earthwork. A reasonable earthwork excavation sequence can give full play to the in-situ bearing capacity of the soil, optimize the stress of the retaining structure, and thus control the deformation of the foundation pit. Compared with other foundation pit control measures, it does not require changing the original foundation pit structure, and has the advantages of high efficiency, time saving and cost saving. Foundation pit deformation has a typical "time and space effect", which has been widely confirmed by engineering practice. At present, the commonly used method to determine the earthwork excavation sequence is to select several reasonable excavation sequences based on engineering experience for comparison and selection. However, on the one hand, this method is very dependent on engineering experience and is difficult to promote on a large scale. On the other hand, this method can often only obtain a local optimal solution and can hardly obtain a global optimal solution. Against the background of the increasingly prominent problem of foundation pit deformation control, exploring an intelligent method that can achieve global optimization of earthwork excavation sequence has become an urgent problem to be solved.

[0045] Example 1

[0046] In this embodiment, an intelligent prediction method for foundation pit deformation considering the excavation sequence is adopted. The method steps are as follows: Figure 1 As shown, the following steps are included:

[0047] S1. Obtain foundation pit data;

[0048] S2. Considering the excavation sequence, batch process the foundation pit data to obtain the foundation pit response data and use it to establish an excavation database;

[0049] S3, processing the excavation database to obtain a sample data set, and dividing the sample data set into a training set and a test set;

[0050] S4, using the training set to jointly train with multiple neural networks to obtain multiple prediction networks;

[0051] S5. Use the test set to test multiple prediction networks, compare the test results to output the optimal prediction network, and use it to predict the foundation pit data to be processed to obtain the foundation pit deformation prediction result.

[0052] In this embodiment, the deformation of foundation pits is obtained in batches based on the PLAXIS script program, wherein the process of calling PLAXIS for batch processing using the Python script language mainly includes the following steps: first, the required finite element model is established through PLAXIS, and the materials, soil, structure, mesh style, and construction steps before the layer to be analyzed are created, and the model is saved; then, a script program is written in Python to realize the automatic blocking function of the program and complete the setting of the remaining analysis steps of the model to complete the calculation; then, a post-processing script program is written in Python to automatically obtain the calculation results of the model, including the deformation of the retaining wall, the deformation of the pit bottom uplift, the surface settlement outside the pit, etc.; the excavation order is changed, and the calculation is repeated until all calculations are completed.

[0053] In this embodiment, because the strength of the foundation pit soil in the soft soil area is relatively low and the vertical height is relatively small, it is necessary to consider slope when simulating the excavation process with finite element method. When dividing the blocks, in order to ensure that the size of each block is uniform, the centerline length of all excavated blocks is kept consistent. The logic of automatic block division of foundation pit excavation is to first determine the excavation status of the soil on the left and right sides of a block by the excavation sequence, consider slope when no excavation has been carried out, and consider excavating the remaining soil completely when excavation has been carried out, so as to determine the shape of the block, and then combine the foundation pit size to obtain the point data of each block, and finally divide the excavation surface into different block forms through the Polygon command in PLAXIS;

[0054] In this embodiment, the excavation width of the foundation pit is increased to 80m, and the excavation surface is divided into 8 blocks (a total of 8! = 40320 excavation schemes). Other conditions remain unchanged. On this basis, the batch processing method proposed by S1 is used to automatically obtain the foundation pit deformation database considering the excavation sequence. 1729 excavation sequence schemes are randomly selected to construct the excavation database. Figure 2 It is a schematic diagram of several excavation schemes;

[0055] In this embodiment, in order to illustrate the generality of the research method, in addition to recording the foundation pit deformation, the retaining wall bending moment data is also saved, and the maximum lateral displacement of the foundation pit retaining wall, the maximum settlement of the surface soil, the maximum uplift of the foundation pit bottom, and the maximum bending moment of the foundation pit retaining wall are recorded as output quantities (the maximum uplift has only one data, and the others contain two data on the left and right sides). The 1729 randomly generated excavation sequences (input quantities) are imported into the batch processing program, and the program will automatically run and save the foundation pit response data. An example of the sample database is shown in Table 1.

[0056] Table 1 Example of foundation pit deformation database considering excavation sequence

[0057]

[0058] In this embodiment, the excavation sequence information is a set of logical sequences that need to be processed before being processed by computer language. The processing flow is as follows: First, the blocks of the foundation pit are numbered accordingly, such as Figure 3a and Figure 3b As shown; then the excavation sequence is arranged in sequence, for example, a set of excavation sequence is "5-8-1-2-4-7-6-3", which means that the foundation pit starts with the block numbered 5, followed by the block numbered 8... until the block numbered 3 is excavated. At this time, a set of 8 discrete data sequences {5,8,1,2,4,7,6,3} can be obtained; then this set of discrete sequences is encoded. There are two main encoding methods:

[0059] One is natural coding, which directly takes the number of each excavated block, that is, the value between 1 and n as the input value. For example, if the number of the first excavated block is 5, then the first input value is also 5, and so on... The input value of the algorithm after natural coding is as follows: [5 8 1 2 4 7 6 3]

[0061] Another is One-Hot Encoding, which uses N-bit state registers to represent N states. The above sequence has 8 blocks, so 8 registers are set. The first excavation block is numbered 5, so register '5' is activated and other registers are suppressed. The encoding is [0 0 0 0 1 0 0 0], and so on... The input value of the algorithm after one-hot encoding is shown as follows:

[0062]

[0063] This conversion process can also be achieved in the form of a transposition matrix (a special square matrix in which each row and column has only one element that is 1 and the other elements are all 0);

[0064] In this embodiment, due to the different numerical ranges corresponding to the pit deformation, for example, the maximum deformation value of the retaining wall is between 40-60, while the maximum bending moment of the retaining wall is between 500-800, the difference between the two sets of data is large, which easily leads to slow convergence of the algorithm and the problem of gradient explosion or gradient disappearance. Therefore, the algorithm output value needs to be normalized, and the commonly used linear normalization formula is used:

[0065]

[0066] Where y represents the output value, min(Y) represents the minimum value in the output value, max(Y) represents the maximum value in the output value, and y′ represents the normalized output value;

[0067] In this embodiment, when the BP neural network algorithm is selected, the BP neural network is used to learn the foundation pit deformation database considering the excavation sequence obtained in S3, and the number of input layers is 8. If the excavation sequence is "5-8-1-2-4-7-6-3", the value of each neuron in the input layer is 5, 8, 1, 2, 4, 7, 6, 3. A total of 4 hidden layers are set. After parameter optimization, the number of neurons in each layer is 12, 15, 19, and 12 respectively. Each hidden layer is composed of a full-link layer and an activation layer. The ReLU function is used as the activation function. This function can effectively deal with the gradient vanishing problem and is widely used in neural network models. The calculation formula is shown in Formula 3.5. The number of output layers is set to 7, corresponding to 7 foundation pit response values;

[0068] f(x)=max(0,x)

[0069] The corresponding program was compiled using the PyTorch framework. During model training, 80% (1400) of the samples in the database were used as the training set, and the remaining 20% ​​(329) were used as the validation set. During training, the batch data input method was used, and only 128 data in the training set were used as input each time to participate in learning and updating the model structure until all the data in the training set were taken, that is, one iteration was completed. After 1000 iterations, the training stopped, and the prediction neural network of the maximum deformation of the foundation pit was obtained.

[0070] In this embodiment, when the recurrent neural network prediction method is selected, the input value in the database is first converted into a transposition matrix form by unique hot encoding. The number of neural network input layers is 8, each representing a row in the transposition matrix. The hidden layer adopts the LSTM layer. After parameter optimization, the number of hidden layers is set to 4, and the number of neurons in each layer is set to 24. In order to improve the ability of the neural network to learn data features, a bidirectional LSTM is designed so that the neural network can learn features through both positive and negative sequences. The last hidden layer and the output layer are connected by a fully connected layer (FC layer) to keep the number of neurons in the output layer at 7;

[0071] In this embodiment, when the convolutional neural network prediction method is selected. The input layer operates in the same way as the recurrent network, converting the sequence data into a transposed matrix form (1×8×8). The convolution layer extracts features by performing multiple convolution calculations on the input tensor. The number of convolution kernels is set to 128 each time. The first selected convolution kernel size is 1×8, 8 features are extracted, and a tensor of 128×8×1 is obtained. The second selected convolution kernel size is 2×8, 7 features are extracted, and a tensor of 128×7×1 is obtained... until the eighth selected convolution kernel size is 8×8, 1 feature is extracted, and a tensor of 128×1×1 is obtained, completing the convolution operation. The pooling layer uses the maximum pooling method to reduce the dimension of the tensor obtained after convolution. Corresponding to the convolution operation, the convolution kernel size of the first pooling is 8×1, and the tensor is 128×1. The convolution kernel size of the second pooling is 7×1, and the tensor is 128×1... This continues until the eighth pooling, when the selected convolution kernel size is 1×1, completing the pooling operation. The 8 tensors obtained after pooling are then merged into a tensor of 128×8, and the tensor is flattened to obtain a hidden layer with 1024 (128×8=1024) neurons, and then a fully connected layer is connected to keep the number of neurons in the output layer at 7, thus completing the construction of the convolutional neural network;

[0072] In this embodiment, the remaining 20% ​​(329) samples in the database are used as validation sets to verify different algorithms. In order to ensure that the model can effectively and reliably predict foundation pit deformation, five model evaluation indicators, including SmoothL1 loss function, explained variance (VAF), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2), are selected to measure the model prediction effect. Generally speaking, the closer the SmoothL1 loss function, root mean square error, and mean absolute error are to 0, and the closer the explained variance and coefficient of determination are to 1, the better the model performance;

[0073] Table 2 Comparison of the performance of some intelligent algorithm models in predicting foundation pit response

[0074]

[0075] In this embodiment, three intelligent prediction models are used to predict the validation set, and the prediction results are shown in Table 2. From the comparison of the performance of the four models, it can be seen that the prediction effect of the recurrent neural network (RNN) is the best, followed by the convolutional neural network (CNN), which is slightly worse than RNN, and then the BP neural network (Back-Propagation Neural Network, BPNN). Compared with CNN, RNN and BPNN use a unique hot encoding data processing method, which expands the characteristics of the data to a certain extent, and makes the distance calculation between the input features more reasonable during the algorithm optimization calculation, which is conducive to the improvement of model performance; in addition, RNN and CNN have the ability to learn sequence order features, so that the excavation order features of the input data can be captured by the model to a certain extent, which is also conducive to the improvement of model performance. Compared with CNN, the RNN model uses a long short-term memory neural network model, which can not only capture the short-term memory features of the sequence, but also capture the long-term memory features. In addition, after using the form of a bidirectional long short-term memory neural network model, it can also improve its ability to capture data features by obtaining the form of reverse sequence features. Compared with the CNN model, its ability to capture information is stronger and the model performance is better.

[0076] In this embodiment, the output recurrent neural network is the optimal prediction network, and it is used to predict the foundation pit data to be processed to obtain the foundation pit deformation prediction result.

[0077] Example 2

[0078] This embodiment adopts an intelligent prediction system for foundation pit deformation that takes into account the order of earth excavation. The system includes a data acquisition and processing module, a model joint training module, a model testing and optimization module, and a prediction execution module;

[0079] Among them, the data acquisition and processing module is used to obtain foundation pit data, and on the basis of considering the excavation sequence, batch process the foundation pit data to obtain the foundation pit response data and establish an excavation database, and then process the excavation database to obtain the sample data set, and divide it into a training set and a test set; the model joint training module is used to use the training set to jointly train with multiple neural networks to obtain multiple prediction networks; the model testing and optimization module is used to test multiple prediction networks using the test set, and output the optimal prediction network by comparing the test results; the prediction execution module is used to use the output optimal prediction network to predict the foundation pit data to be processed, and finally obtain the foundation pit deformation prediction result.

[0080] In this system, the specific process of batch processing foundation pit data is as follows, taking into account the excavation sequence:

[0081] S21, dividing the foundation pit data into blocks, and establishing a finite element model based on the blocks;

[0082] S22, calculating the foundation pit response data in the finite element model;

[0083] S23, changing the excavation sequence in the foundation pit data and returning to step S21 until all excavation sequences are traversed.

[0084] In this system, the specific process of block division is as follows: using the excavation sequence, the excavation status of the soil on the left and right sides of each block is obtained. When the excavation status is not excavated, the slope is considered; when the excavation status is excavated, the remaining soil is considered to be completely excavated to determine the shape of the block; then the point data of each block is obtained in combination with the foundation pit size, and finally the excavation surface is divided into different block forms by the point data. When dividing the blocks, the centerline length of all excavation blocks is kept consistent.

[0085] In this system, the specific process of processing the excavation database is: first, the blocks of the foundation pit are numbered accordingly to obtain a discrete sequence, and then the discrete sequence is encoded; the encoding processing method includes natural encoding or one-hot encoding.

[0086] In this system, the training set accounts for 80% of the sample data set and is used to train various neural networks to obtain their corresponding prediction networks; the test set accounts for 20% of the sample data set and is used to test various prediction networks to obtain test results in order to output the optimal prediction network.

[0087] In this system, multiple neural networks include BP neural network, recurrent neural network, and convolutional neural network.

[0088] In this system, the test results include five model evaluation indicators: SmoothL1 loss function, explained variance, root mean square error, mean absolute error and determination coefficient. Among them, the closer the SmoothL1 loss function, root mean square error and mean absolute error are to 0, and the closer the explained variance and determination coefficient are to 1, the better the model performance is.

[0089] In this system, the foundation pit deformation prediction results include foundation pit deformation data, retaining wall bending moment data, maximum lateral displacement of the foundation pit retaining wall, maximum settlement of the surface soil, maximum uplift of the foundation pit bottom and maximum bending moment of the foundation pit retaining wall.

[0090] In this embodiment, the project consists of a podium building and a main building, and both the main building and the podium building have three-story underground garages. Figure 4 As shown, the foundation pit is approximately square in plan, with an excavation area of ​​7452m 2The excavation depth of the main building is 17.15m, and the excavation depth of the podium is 14.95m. The underground continuous wall is used as the enclosure structure. The enclosure structure is 0.8m wide and inserted to a depth of 28.6m. Three concrete supports are set in the pit, and the excavation is carried out to the bottom of the pit in four times. The simulated section is as follows Figure 5 As shown, the parameters of soil layer and surrounding structure are shown in Table 3 and Table 4. The deformation prediction and optimization are carried out for the third layer of soil (5m thick);

[0091] Table 3 Project foundation pit retaining structure parameters

[0092]

[0093] Table 4 Project soil layer HSS model parameters

[0094]

[0095] In this embodiment, considering the effect of the size of the blocks on the deformation of the foundation pit, the first two layers of soil are excavated in only two layers (the difference between excavating in 2 blocks and excavating in 8 blocks is less than 0.5 mm), and the excavation of the first two layers of soil induces less deformation of the foundation pit than the third layer of soil. Therefore, the first two layers of soil are excavated in 2 blocks, and the third layer of soil is excavated in 8 blocks. Figure 6 As shown, this can reduce the amount of finite element calculation while ensuring the calculation accuracy.

[0096] In this embodiment, the simulation is shown in Figure 7. The displacement pattern of the retaining wall is similar to the monitoring result, which is a "bow shape" with a large center and small sides. Figure 7a As shown in the figure, the simulated curve is similar to the measured curve. The measured maximum retaining wall displacement is 37.0 mm, and the simulated maximum retaining wall displacement is 35.5 mm, with a difference of 1.5 mm. The simulation results are consistent with the measured results. The comparison between the simulated value and the measured value of the south retaining wall is as follows: Figure 7b As shown in the figure, the simulated curve of the south retaining wall is also very similar to the measured curve. The measured maximum retaining wall displacement is 30.8 mm, and the simulated maximum retaining wall displacement is 31.0 mm, with a difference of only 0.2 mm. The simulation results are in good agreement with the measured results. If the excavation sequence is not considered, the maximum retaining wall displacement calculated for both retaining walls is 35.0 mm, which is 2 mm different from the measured displacement of the north retaining wall and 4.2 mm different from the displacement of the south retaining wall. It can be seen that when the excavation sequence is not considered, the finite element model can only obtain the average value of the deformation of the left and right retaining walls, while the finite element model considering the excavation sequence can simultaneously obtain the deformation values ​​of the left and right retaining walls in deformation prediction. Compared with the finite element model that does not consider the excavation sequence, it is more accurate in deformation prediction and more reasonable in predicting the characteristics of foundation pit deformation as the foundation pit excavation changes.

[0097] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An intelligent prediction method for foundation pit deformation considering the excavation sequence, characterized in that: The following steps are involved: S1. Obtain foundation pit data; S2. Considering the excavation sequence, batch process the foundation pit data to obtain the foundation pit response data and use it to establish an excavation database; S3, processing the excavation database to obtain a sample data set, and dividing the sample data set into a training set and a test set; S4, using the training set to jointly train with multiple neural networks to obtain multiple prediction networks; S5. Use the test set to test multiple prediction networks, compare the test results to output the optimal prediction network, input the foundation pit data to be processed into the optimal prediction network, and obtain the foundation pit deformation prediction result.

2. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: In S2, the specific process of batch processing foundation pit data in consideration of the excavation sequence is as follows: S21, dividing the foundation pit data into blocks, and establishing a finite element model based on the blocks; S22, calculating the foundation pit response data in the finite element model; S23, changing the excavation sequence in the foundation pit data and returning to step S21 until all excavation sequences are traversed.

3. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: The specific process of block division is as follows: using the excavation sequence, the excavation status of the soil on the left and right sides of each block is obtained. When the excavation status is not excavated, slope is considered; when the excavation status is excavated, the remaining soil is considered to be completely excavated to determine the shape of the block; then the point data of each block is obtained in combination with the foundation pit size, and finally the excavation surface is divided into different block forms according to the point data.

4. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: The specific process of processing the excavation database in S3 is: firstly, the blocks of the foundation pit are numbered accordingly to obtain a discrete sequence, and then the discrete sequence is encoded.

5. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 4 is characterized in that: The encoding processing method includes natural encoding or one-hot encoding.

6. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: The various neural networks in S4 include BP neural network, recurrent neural network and convolutional neural network.

7. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: The test results of S5 include five model evaluation indicators: SmoothL1 loss function, explained variance, root mean square error, mean absolute error and determination coefficient; among them, the closer the SmoothL1 loss function, root mean square error and mean absolute error are to 0, and the closer the explained variance and determination coefficient are to 1, the better the model performance is.

8. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1 is characterized in that: The foundation pit deformation prediction result in S5 includes foundation pit deformation data, retaining wall bending moment data, maximum lateral displacement of the foundation pit retaining wall, maximum settlement of the surface soil, maximum uplift of the foundation pit bottom and maximum bending moment of the foundation pit retaining wall.

9. The method for intelligent prediction of foundation pit deformation considering the excavation sequence according to claim 1, characterized in that: The foundation pit deformation prediction result in S5 needs to be normalized, and the normalization process adopts a linear normalization formula: Where y is the predicted value of foundation pit deformation, min(Y) represents the minimum value of the predicted value of foundation pit deformation, max(Y) represents the maximum value of the predicted value of foundation pit deformation, and y′ is the normalized predicted value of foundation pit deformation.

10. An intelligent prediction system for foundation pit deformation considering the excavation sequence, characterized in that: The system works by applying a method for intelligent prediction of foundation pit deformation taking into account the excavation sequence as described in any one of claims 1 to 9, and the system includes a data acquisition and processing module, a model joint training module, a model testing and optimization module, and a prediction execution module; The data acquisition and processing module is used to obtain foundation pit data, and batch process the foundation pit data on the basis of considering the excavation sequence, obtain foundation pit response data and establish an excavation database, then process the excavation database to obtain a sample data set, and divide it into a training set and a test set; The model joint training module is used to use the training set to jointly train with multiple neural networks to obtain multiple prediction networks; The model testing and optimization module is used to test multiple prediction networks using a test set, and output the optimal prediction network by comparing the test results; The prediction execution module is used to use the output optimal prediction network to predict the foundation pit data to be processed, and finally obtain the foundation pit deformation prediction result.

Images