Deep foundation pit island area deformation prediction method based on BP neural network

Abstract

The invention belongs to the technical field of ocean engineering new energy, and particularly relates to a deep foundation pit island area deformation prediction method based on a BP neural network, finite element model forward prediction is reversely carried out on soil mass mechanical parameters obtained through soil mass displacement inversion analysis based on the BP neural network, and a forward prediction result and an actual measurement result are shown in the figure 2. It can be seen that the change trend of the measured data and the BP neural network prediction data is roughly the same as the change trend of the measured data, the fitting performance is good, and the prediction result of the back analysis parameter is superior to that of the original parameter. Therefore, the data of the soil displacement back analysis based on the BP neural network algorithm accords with the field actual situation, the inverted soil parameters are scientific and reliable, the prediction precision can be greatly improved when the method is used for predictive analysis of the deformation of the foundation pit of the project, and powerful support is provided for foundation pit excavation.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy in ocean engineering, and particularly relates to a deformation prediction method for deep foundation pit island areas based on BP neural network. Background Technique

[0002] Ocean engineering refers to the application of engineering technology to solve various engineering problems related to the marine environment, resources, energy, transportation, etc. It covers multiple fields, including but not limited to offshore oil and gas exploitation, offshore wind power, port construction, ocean platforms, submarine pipelines, marine environmental protection. Ocean engineering is a national key project. It is not only an important field to promote economic development, but also an important part to enhance the country's comprehensive competitiveness and strategic security. Ocean engineering is one of the key fields of national development, and the projects involved cover multiple aspects such as energy development, marine resource utilization, maritime transportation, environmental protection, and national defense security.

[0003] During the construction process of ocean engineering, for example, during the construction of deep foundation pits in island areas, one of the difficulties in analyzing the deformation and stress state of rock and soil is how to appropriately select the mechanical parameters of rock and soil. The methods based on indoor tests and on-site engineering tests effectively solve the above problems. However, there are also many problems with these two methods. For indoor tests, the selected test samples or the mechanical parameters of rock and soil obtained from local and limited on-site tests have great randomness, which in turn leads to the calculation results of the models using these soil parameters for finite element calculation often being unsatisfactory. To make up for the deficiencies of the above projects, the soil parameter back-analysis method based on on-site monitoring data has emerged, which effectively solves the important problem of the disconnection between rock and soil mechanics theory and actual rock and soil engineering practice. The present invention proposes a deformation prediction method for deep foundation pit island areas based on BP neural network. Summary of the Invention

[0004] The purpose of the present invention is to provide a deformation prediction method for deep foundation pit island areas based on BP neural network, which can predict in real time whether the deep foundation pit island area deforms during construction.

[0005] The technical solution adopted by the present invention is specifically as follows:

[0006] A deformation prediction method for deep foundation pit island areas based on BP neural network, the prediction method includes the following steps:

[0007] Step 1: Establish a finite element analysis model to complete the forward analysis process from soil mechanics parameters to the horizontal displacement value of the foundation pit support structure;

[0008] Step 2: Use the orthogonal test calculation method to determine the sample of soil mechanics parameters to be back-analyzed;

[0009] Step 3: Using the model established in Step 1, take the parameter samples generated in Step 2 as the input, and calculate the corresponding horizontal displacement value of the support structure;

[0010] Step 4: Establish a BP neural network, normalize the soil mechanics parameters and the horizontal displacement value of the support structure output by the finite element calculation to obtain the neural network training sample data;

[0011] Step 5: Train the BP neural network and gradually correct the connection weights of the network;

[0012] Step 6: Input the soil mechanics parameters into the trained neural network, and the network output is the horizontal displacement value of the support structure;

[0013] Step 7: Check and evaluate the back analysis results.

[0014] Preferably, Step 1 includes the following specific steps:

[0015] Step 101: Establish a geological model

[0016] First, according to the on-site soil investigation report, determine the thickness of the soil layer and the physical and mechanical properties of the soil

[0017] Define soil layers: Set different material properties according to the depth and type of different soil layers.

[0018] Step 102: Establish a foundation pit support structure model

[0019] It is necessary to establish a foundation pit support structure model, which includes the foundation pit support system and its physical properties; and define the geometric shape and size of the support structure; determine the material properties of the support structure: such as reinforced concrete, steel structure; in Midas GTS, draw the model of the support structure through 3D modeling tools;

[0020] Step 103: Mesh generation

[0021] According to the analysis requirements and calculation accuracy, divide the finite element mesh.

[0022] Step 104: Define boundary conditions and loads

[0023] Boundary conditions: Include the bottom of the foundation pit and the contact interface between the support structure and the soil;

[0024] Loads: Calculate the loads on the support structure;

[0025] Step 105: Contact relationship between the soil and the support structure

[0026] The friction contact model is adopted to simulate the interaction between the soil mass and the supporting structure; the contact surface and the friction coefficient are set through the contact surface definition tool in Midas GTS to simulate the reaction force of the soil mass on the supporting structure;

[0027] Step 106, solution analysis

[0028] After all the parameters, loads and boundary conditions are defined, finite element analysis is performed; the corresponding solver is selected for analysis according to the scale and complexity of the model.

[0029] Step 107: Result analysis and extraction

[0030] The analysis results include the deformation, displacement, stress and contact force of the foundation pit supporting structure.

[0031] Preferably, in the said step 2, the notation of the orthogonal test table: L v (t c );where L is the symbol of the orthogonal table; t is the number of levels of the factor; v is the number of tests; c is the number of columns of the factor; the steps of the orthogonal test are respectively: determining the evaluation index in combination with the actual problem and selecting the types of influencing factors and the number of levels of each influencing factor; determining the test table to be used according to the number of tests. Assuming that the number of levels of each factor is the same, the calculation formula of the orthogonal test is as follows:

[0032] v = c*(t - 1) + 1 (1);

[0033] A total of 3 categories and 9 parameters of the soil layer parameters above the bottom of the foundation pit selected by the orthogonal test, and each parameter takes 3 levels. It is calculated by the formula (1) of the orthogonal test that at least 19 tests are required. According to the principle of the orthogonal test that can accommodate all parameter variables and minimize the number of tests, therefore, L 27 (3 9) orthogonal test table is used for the orthogonal test.

[0034] Preferably, in the said step 3, the obtained parameter samples are input into the finite element calculation model to obtain the calculated values of the characteristic points of the horizontal displacement of the island main foundation pit supporting structure.

[0035] Preferably, in the said step 4, when constructing the BP neural network structure:

[0036] The elastic modulus, internal friction angle and cohesion of the three soil layers above the bottom surface of the foundation pit excavation are taken as 9 inversion parameters, and the number of nodes in the output layer is 9; five observation points of the horizontal displacement of the island foundation pit supporting structure are selected as the input, that is, the number of nodes in the input layer is 5;

[0037] The hidden layer includes the number of hidden layers of the neural network, the number of nodes in the hidden layer and the form of the transfer function. The value of the number of nodes in the hidden layer of the BP neural network is obtained by the following empirical formula: l≥log2n, where: n is the number of nodes in the input layer; l is the number of nodes in the hidden layer; m is the number of nodes in the output layer; α is a constant between [0, 10];

[0038] The transfer function of the BP neural network selects the hyperbolic tangent sigmoid function tansig, and its expression:

[0039] f(x) = 2 / [1 + exp(-2x)] - 1 (2).

[0040] Preferably, in step 4, after establishing the BP neural network model, during the data normalization process, the maximum-minimum method is used for data normalization:

[0041] x k =(x k -x min ) / (x max -x min ) (3)

[0042] where: x min is the minimum value in the sample sequence; x max is the maximum value in the sample sequence.

[0043] Preferably, in step 5, when training the BP neural network model, the following steps are included:

[0044] Step 501: Determination of the network: Determine

[0045] the number of nodes n in the input layer, the number of nodes l in the hidden layer, and the number of nodes m in the output layer according to the input sample X, output sample T, and prediction sample Y of the input system; determine the learning rate function and the neuron activation function; and determine the connection weights wij between the input layer and the hidden layer, the connection weights wjk between the hidden layer and the output layer, the hidden layer threshold a, the output layer threshold b, and the error value e of the neural network;

[0046] Step 502: Calculate the output of the hidden layer: Calculate the output H of the hidden layer according to the input sample Xn, the connection weights wij between the input layer and the hidden layer, and the hidden layer threshold a:

[0047] where: l is the number of nodes in the hidden layer; f is the hidden layer activation function;

[0048]

[0049] Step 3: Output layer calculation: From the output H of the hidden layer

[0050] , the connection weights w j between the hidden layer and the output layer, and the threshold b to construct jk

[0051] ​It becomes the predicted output sample Y of the BP neural network k :

[0052]

[0053] Where: m is the number of nodes in the output layer;

[0054] Step 4: Error calculation: The network prediction error e is calculated from the expected output, i.e., the output sample T and the predicted output sample Y k :

[0055] e k = T k - Y k , k = 1, 2,..., m (6)

[0056] Step 5: Weight update: According to the calculated network prediction error, in the error backpropagation stage, the connection weights w between the output layer and the hidden layer are adjusted jk , the connection weights w between the input layer and the hidden layer ij :

[0057] w jk = w jk + ηH j e k , j = 1, 2,..., l, k = 1, 2,..., m (7)

[0058]

[0059] Where: η is the neural network learning rate function

[0060] Step 7: Determine the termination condition of the neural network; if the prediction error e of the neural network k is less than the set error e, the neural network terminates iteration and outputs the prediction sample Y; if the prediction error e of the neural network k is greater than the set error e, the neural network enters Step 2 to continue iteration.

[0061] The technical effects achieved by the present invention are as follows:

[0062] In the present invention, based on the soil displacement back analysis of the BP neural network, the mechanical parameters of the soil are used for forward prediction of the finite element model in reverse. It can be seen that the variation trend of the measured data is roughly the same as that of the BP neural network prediction data and has a good fit, and the prediction result of the back analysis parameters is better than the original parameters. It can be seen that the data of the soil displacement back analysis based on the BP neural network algorithm conforms to the actual situation on site, and the inverted soil parameters are scientific and reliable. When used for the prediction analysis of the foundation pit deformation of this project, the prediction accuracy can be greatly improved, providing strong support for the foundation pit excavation. Description of the Drawings

[0063] Figure 1 is the flow chart of the present invention;

[0064] Figure 2 is the comparison chart of the prediction results of the present invention. Detailed implementation manners

[0065] In order to make the objectives and advantages of the present invention more clear, the present invention will be specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific implementation manners of the present invention, and does not strictly limit the scope of protection of the specific requests of the present invention.

[0066] A deformation prediction method for the isolated island area of deep foundation pits based on BP neural network, the prediction method includes the following steps:

[0067] Step 1: Establish a finite element analysis model to complete the forward analysis process from soil mechanics parameters to the horizontal displacement value of the foundation pit support structure;

[0068] Step 2: Use the orthogonal test calculation method to determine the soil mechanics parameter samples to be inverted;

[0069] Step 3: Use the model established in Step 1, take the parameter samples generated in Step 2 as inputs, and calculate the corresponding horizontal displacement values of the support structure;

[0070] Step 4: Establish a BP neural network, normalize the soil mechanics parameters and the horizontal displacement values of the support structure output by the finite element calculation to obtain the neural network training sample data;

[0071] Step 5: Train the BP neural network and gradually correct the connection weights of the network;

[0072] Step 6: Input the soil mechanics parameters into the trained neural network, and the network output is the horizontal displacement value of the support structure;

[0073] Step 7: Check and evaluate the back-analysis results.

[0074] Preferably, the Step 1 includes the following specific steps:

[0075] Step 101: Establish a geological model

[0076] First, it is necessary to determine the thickness of the soil layer and the physical and mechanical properties of the soil, such as elastic modulus, cohesion, friction angle, void ratio, etc. according to the on-site soil exploration report; in Midas GTS, the following method can be used to establish a geological model:

[0077] Define soil layers: Set different material properties according to the depths and types of different soil layers.

[0078] Select a suitable soil model: for example, the Mohr-Coulomb model, the Hardening Soil model, etc., to more realistically simulate the mechanical behavior of the soil;

[0079] Step 102: Establish a foundation pit support structure model

[0080] It is necessary to establish a foundation pit support structure model, which includes the foundation pit support system, such as steel supports, anchor rods, pile foundations, etc., and their physical properties; and define the geometry and dimensions of the support structure; determine the material properties of the support structure: for example, reinforced concrete, steel structure; in Midas GTS, draw the model of the support structure through 3D modeling tools;

[0081] Step 103: Mesh generation

[0082] According to the analysis requirements and calculation accuracy, divide the finite element mesh; the mesh should be fine enough to ensure the accuracy of the calculation results, but should not be too refined to avoid excessive calculation volume. According to the complexity of the model, select a suitable mesh type, such as tetrahedral or hexahedral mesh.

[0083] Step 104: Define boundary conditions and loads

[0084] Boundary conditions: include the bottom of the foundation pit and the contact interface between the support structure and the soil; common boundary conditions include fixed boundaries; for example, the horizontal and vertical displacements at the bottom of the foundation pit, and the contact conditions on the foundation pit wall surface, such as frictional contact;

[0085] Loads: Calculate the loads on the support structure, considering the self-weight of the soil, hydraulic loads, wind loads, traffic loads, etc. that may affect the support structure during the foundation pit excavation process;

[0086] Step 105: Contact relationship between the soil and the support structure

[0087] Defining the contact relationship between the foundation pit support structure and the surrounding soil is crucial; usually, a frictional contact model is used to simulate the interaction between the soil and the support structure; set the contact surface and friction coefficient through the contact surface definition tool in Midas GTS to simulate the reaction force of the soil on the support structure;

[0088] Step 106, Solve and analyze

[0089] After all parameters, loads, and boundary conditions are defined, perform a finite element analysis; Midas GTS provides different types of solvers, and the corresponding solver can be selected for analysis according to the scale and complexity of the model;

[0090] Step 107: Result analysis and extraction

[0091] The analysis results include the deformation, displacement, stress, and contact force of the foundation pit support structure. To obtain the horizontal displacement value of the foundation pit support structure, the results can be extracted in the following ways. In Midas GTS, the displacement data of the support structure can be extracted by specifying the positions of the monitoring points. The output results can be displayed as displacement nephograms or the horizontal displacement values of specific points can be extracted.

[0092] Preferably, in step 2, the notation of the orthogonal experiment table: L v (t c ); where L is the symbol of the orthogonal table; t is the number of levels of the factor; v is the number of experiments; c is the number of columns of the factor. The steps of the orthogonal experiment are as follows: Determine the evaluation index in combination with the actual problem and select the types of influencing factors and the number of levels of each influencing factor. According to the number of experiments, determine the selected experiment table. Assuming that the number of levels of each factor is the same, the calculation formula of the orthogonal experiment is as follows:

[0093] v = c*(t - 1) + 1 (1);

[0094] A total of 3 categories and 9 parameters of the soil layer parameters above the bottom of the foundation pit selected by the orthogonal experiment, and each parameter takes 3 levels. It can be calculated by the formula (1) of the orthogonal experiment that at least 19 experiments are required. According to the principle of the orthogonal experiment that can accommodate all parameter variables and minimize the number of experiments, the orthogonal experiment table of L 27 (3 9 ) is selected for the orthogonal experiment, and the orthogonal experiment is as follows

[0095] as shown in Table 1.

[0096]

[0097] Table 1 Orthogonal scheme experiment table of soil parameters

[0098] Preferably, in step 3, the parameter samples in the above table are input into the finite element calculation model to obtain the calculated values of the characteristic points of the horizontal displacement of the foundation pit support structure of the island main body, as shown in Table 2 below.

[0099]

[0100] Table 2: Finite element simulation values of the characteristic points of the horizontal displacement of the foundation pit support structure of the island main body

[0101] Preferably, in step 4, when constructing the BP neural network structure:

[0102] Take the elastic modulus, internal friction angle, and cohesion of the three layers of soil above the bottom surface of the foundation pit excavation as 9 inversion parameters, and then the number of output layer nodes is 9; Select five observation points of the horizontal displacement of the foundation pit support structure of the island as the input, that is, the number of input layer nodes is 5;

[0103] The hidden layer includes the number of hidden layers in the neural network, the number of nodes in the hidden layer, and the form of the transfer function.

[0104] The value of the number of nodes in the hidden layer of the BP neural network. The selection of the node parameters in the hidden layer of the BP neural network is crucial for the prediction accuracy of the neural network. If the number of nodes in the hidden layer is too small, the BP neural network cannot express the complex functional relationship between the input samples and the output samples, so its prediction error is large. If the number of nodes in the hidden layer is too large, the learning time of the neural network will be too long, and there is likely to be an overfitting phenomenon, that is, the data of the training samples is predicted well, but the data is predicted poorly when predicting the samples. The number of nodes in the hidden layer should be selected according to the actual situation of the sample data or calculated through an empirical formula to determine the approximate range of the number of nodes in the hidden layer. Compare the number of nodes in the hidden layer within this range by trial calculation respectively. However, when there is a large amount of data, the method is more troublesome. Therefore, the radial neural network mentioned below is a neural network algorithm that improves the shortcomings of the BP neural network. The selection of the number of nodes in the hidden layer of the BP neural network should meet the requirements of both few iteration times and high fault tolerance rate. The current empirical formula can only determine its approximate range and does not necessarily perfectly express the mapping relationship between the input samples and the output samples. The empirical formula is as follows: l≥log2n, where: n is the number of nodes in the input layer; l is the number of nodes in the hidden layer; m is the number of nodes in the output layer; α is a constant between [0,10];

[0105] The transfer function of the BP neural network selects the hyperbolic tangent s-type function tansig, and its expression:

[0106] f(x) = 2 / [1 + exp(-2x)] - 1 (2);

[0107] Under the condition that the connection weights, thresholds, and network structures of the BP neural network are the same, its prediction error and mean square error are shown in Table 3.

[0108]

[0109] Table 3: Prediction errors corresponding to transfer functions

[0110] Preferably, in step 4, after establishing the BP neural network model, during the data normalization process, data normalization is to transform the data of the input sample X and output sample T of the neural network to between [0, 1] to eliminate the difference in the magnitude of each data. More importantly, the transfer function of the neuron varies greatly between [0, 1]. If it is greater than 1, the transfer function changes little, that is, the derivative or slope of the transfer function is small, which is not conducive to the execution of the error backpropagation algorithm. When performing error backpropagation, the gradient information of each neuron's transfer function is required. When the input of the neuron is too large, the gradient value of the corresponding independent variable is too small to smoothly adjust the weights and thresholds. Therefore, the sample data should be normalized first. The maximum-minimum method is used for data normalization:

[0111] x k =(x k -x min ) / (x max -x min )(3)

[0112] In the formula: x min is the minimum value in the sample sequence; x max is the maximum value in the sample sequence; Table 4 shows the values after normalizing each parameter, and Table 5 shows the values after normalizing each feature point.

[0113]

[0114]

[0115] Table 4 Parameter sample normalization

[0116]

[0117]

[0118] Table 5 Feature point horizontal displacement normalization

[0119] Preferably, in step 5, when training the BP neural network model, it includes the following steps:

[0120] Step 501: Determination of the network: Determine

[0121] the number of input layer nodes n, the number of hidden layer nodes l, and the number of output layer nodes m according to the input sample X, output sample T, and prediction sample Y of the input system; determine the learning rate function and neuron activation function; and determine the connection weights wij between the input layer and the hidden layer, the connection weights wjk between the hidden layer and the output layer, the hidden layer threshold a, the output layer threshold b, and the error value e of the neural network.

[0122] Step 502: Calculate the output of the hidden layer: Based on the input sample Xn, the connection weights wij between the input layer and the hidden layer, and the threshold a of the hidden layer, calculate the output H of the hidden layer:

[0123] Where: l is the number of nodes in the hidden layer; f is the activation function of the hidden layer;

[0124]

[0125] In the formula: l is the number of nodes in the hidden layer; f is the activation function of the hidden layer;

[0126] Step 3: Output layer calculation: From the output H of the hidden layer j , the connection weights w jk between the hidden layer and the output layer, and the threshold b constitute

[0127] the predicted output sample Y of the BP neural network k :

[0128]

[0129] In the formula: m is the number of nodes in the output layer;

[0130] Step 4: Error calculation: Calculate the network prediction error E from the expected output, that is, the output sample T and the predicted output sample Y:

[0131] e k = T k - Y k , k = 1, 2,..., m (6)

[0132] Step 5: Weight update: According to the calculated network prediction error, perform the error backpropagation stage to adjust the connection weights w jk between the output layer and the hidden layer, and the connection weights w ij between the input layer and the hidden layer:

[0133] w jk = w jk + ηH j e k , j = 1, 2,..., l, k = 1, 2,..., m (7)

[0134]

[0135] In the formula: η is the neural network learning rate function

[0136] Step 7: Determine the termination condition of the neural network; if the prediction error e k of the neural network is less than the set error e, then the neural network terminates the iteration and outputs the predicted sample Y; if the prediction error e k of the neural network is greater than the set error e, then the neural network enters Step 2 to continue the iteration.

[0137] In the present invention, through the training of training samples and the prediction of test samples, the BP neural network can fully meet the requirements of back analysis of soil displacement parameters. Substitute the measured data into the neural network for back analysis of soil parameters. Table 6 shows the results of back inversion of soil parameters. The results of back inversion of soil parameters are excluded from human interference through an error avoidance method, and the average value is taken as the soil parameter value of the soil displacement back analysis based on the BP neural network this time.

[0138]

[0139] Table 6 Results of back inversion of soil parameters

[0140] Based on the mechanical parameters of the soil obtained from the back analysis of soil displacement by the BP neural network, the forward prediction of the finite element model is carried out in the reverse direction. The forward prediction results and the measured results are as Figure 2 shown. It can be seen that the change trend of the measured data and the predicted data of the BP neural network is roughly the same as that of the measured data and has good fitting. The prediction results of the back analysis parameters are better than the original parameters. It can be seen that the data of the soil displacement back analysis based on the BP neural network algorithm conforms to the actual situation on site, and the back-inverted soil parameters are scientific and reliable. Used for the prediction and analysis of the foundation pit deformation of this project, it can greatly improve the prediction accuracy and provide strong support for the foundation pit excavation.

[0141] The above is only the preferred implementation mode of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. The structures, devices and operation methods not specifically described and explained in the present invention, unless otherwise specified and limited, are implemented according to the conventional means in the art.

Claims

1. A deformation prediction method for isolated island area of ​​deep foundation pit based on BP neural network, characterized by: The prediction method comprises the following steps: Step 1: Establish a finite element analysis model and complete the forward analysis process from soil mechanics parameters to the horizontal displacement value of the foundation pit support structure; Step 2: Use the orthogonal test calculation method to determine the soil mechanical parameter sample to be inverted; Step 3: Using the model established in step 1, taking the parameter sample generated in step 2 as input, calculate the corresponding horizontal displacement value of the support structure; Step 4: Establish a BP neural network, normalize the soil mechanics parameters and the horizontal displacement values ​​of the support structure output by the finite element calculation, and obtain the neural network training sample data; Step 5: Train the BP neural network and gradually correct the connection weights of the network; Step 6: Input the soil mechanics parameters into the trained neural network, and the network output is the horizontal displacement value of the support structure; Step 7: Verify and evaluate the reverse analysis results.

2. The method for predicting deformation of isolated island areas in deep foundation pits based on BP neural network according to claim 1 is characterized in that: The step 1 comprises the following specific steps: Step 101: Build a geological model First, it is necessary to determine the thickness of the soil layer and the physical and mechanical properties of the soil according to the on-site soil survey report, and then define the soil layer: set different material properties according to the depth and type of different soil layers; Step 102: Establish foundation pit support structure model It is necessary to establish a foundation pit support structure model, which includes the foundation pit support system and its physical characteristics; define the geometric shape and size of the support structure; determine the material properties of the support structure; Step 103: Meshing: dividing the finite element mesh according to the analysis requirements and calculation accuracy; Step 104: Define boundary conditions and loads Boundary conditions: including the bottom of the foundation pit and the contact interface between the support structure and the soil; Loads: Calculate the loads on the supporting structure.

3. The method for predicting deformation of isolated island area in deep foundation pit based on BP neural network according to claim 2 is characterized in that: The step 1 also includes: Step 105: Contact relationship between soil and supporting structure; use friction contact model to simulate the interaction between soil and supporting structure; set contact surface and friction coefficient through contact surface definition tool in MidasGTS to simulate the reaction force of soil on supporting structure; Step 106, solving the analysis; after all parameters, loads and boundary conditions are defined, perform finite element analysis; select the corresponding solver for analysis according to the scale and complexity of the model; Step 107: Result analysis and extraction; the analysis results include deformation, displacement, stress, and contact force of the foundation pit support structure.

4. The method for predicting deformation of isolated island areas in deep foundation pits based on BP neural network according to claim 3 is characterized in that: In step 2, the notation of the orthogonal test table is: L v (t c ); where L is the symbol of the orthogonal table; t is the number of levels of the factor; v is the number of trials; c is the number of columns of the factor; the steps of the orthogonal test are: determine the evaluation index in combination with the actual problem and select the type of influencing factors and the number of levels of each influencing factor; determine the selected test table according to the number of trials, assuming that the number of levels of each factor is the same, the calculation formula of the orthogonal test is as follows: v = c*(t-1)+1 (1); The soil layer parameters above the bottom of the foundation pit selected in the orthogonal test are 9 parameters of 3 categories in total. Each parameter is taken at 3 levels and the orthogonal test formula (1) is used to calculate that at least 19 tests are carried out. According to the principle that the orthogonal test can accommodate all parameter variables and minimize the number of tests, L is selected this time. 27 (3 9 ) to conduct orthogonal experiment.

5. The method for predicting deformation of isolated island area in deep foundation pit based on BP neural network according to claim 4 is characterized by: In step 3, the obtained parameter samples are input into the finite element calculation model to obtain the calculated values ​​of the horizontal displacement characteristic points of the isolated island main body foundation pit support structure.

6. The method for predicting deformation of isolated island areas in deep foundation pits based on BP neural network according to claim 5 is characterized in that: In step 4, when constructing the BP neural network structure: The elastic modulus, internal friction angle and cohesion of the three layers of soil above the bottom of the foundation pit are taken as inversion parameters, and the number of output layer nodes is 9; five observation points of the horizontal displacement of the isolated foundation pit support structure are selected as input, that is, the number of input layer nodes is 5; The hidden layer includes the number of hidden layers, the number of nodes in the hidden layer and the form of the transfer function of the neural network. The empirical formula for the number of nodes in the hidden layer of the BP neural network is as follows: l≥log2n, where: n is the number of input layer nodes; l is the number of hidden layer nodes; m is the number of output layer nodes; α is a constant between [0,10]; The transfer function of the BP neural network uses the hyperbolic tangent s-type function tansig, and its expression is: f(x)=2 / [1+exp(-2x)]-1 (2).

7. The method for predicting deformation of isolated island area in deep foundation pit based on BP neural network according to claim 6 is characterized by: In step 4, after the BP neural network model is established, the maximum and minimum value method is used to normalize the data during the data normalization process: x k =(x k -x min ) / (x max -x min ) (3) Where: x min is the minimum value in the sample sequence; x max is the maximum value in the sample sequence.

8. The method for predicting deformation of isolated island areas in deep foundation pits based on BP neural network according to claim 7 is characterized by: In step 5, when training the BP neural network model, the following steps are included: Step 501: Determination of the network: Determine based on the input sample X, output sample T, and prediction sample Y of the input system The number of input layer nodes is n, the number of hidden layer nodes is l, and the number of output layer nodes is m; determine the learning rate function and neuron excitation function; and determine the connection weight w between the input layer and the hidden layer ij , the connection weight w between the hidden layer and the output layer jk , hidden layer threshold a, output layer threshold b, neural network error value e; Step 502: Calculate the hidden layer output: Based on the input sample X n , the connection weight w between the input layer and the hidden layer ij ,hidden Containing layer threshold a, calculate the hidden layer output H j : Where: l is the number of hidden layer nodes; f is the hidden layer activation function; Step 3: Output layer calculation: The hidden layer outputs H j , the connection weight w between the hidden layer and the output layer jk , threshold b structure Become the predicted output sample Y of the BP neural network k : Where: m is the number of nodes in the output layer; Step 4: Error calculation: The network prediction error e is calculated by predicting the output sample Y from the expected output, i.e. the output sample T. k : e k =T k -Y k ,k=1,2,...,m (6) Step 5: Weight update: According to the calculated network prediction error, the error back propagation stage is performed to adjust the connection weight w between the output layer and the hidden layer jk , the connection weight w between the input layer and the hidden layer ij : w jk =w jk +ηH j e k ,j=1,2,...,l;k=1,2,...,m (7) Where: η is the neural network learning rate function; Step 7: Determine the termination condition of the neural network; if the prediction error of the neural network e k If the prediction error of the neural network is less than the set error e, the neural network terminates the iteration and outputs the predicted sample Y; if the prediction error of the neural network is e k If it is greater than the set error e, the neural network enters step 2 to continue iterating.

Images