Deep foundation pit deformation prediction method based on neural network model

Abstract

The invention belongs to the technical field of deep foundation pit deformation prediction, and particularly relates to a deep foundation pit deformation prediction method based on a neural network model, which comprises the following steps: firstly, establishing a neural network prediction model, secondly, constructing a data set, thirdly, carrying out model initialization, then, carrying out model training, and finally, outputting a prediction result which is based on a BiLSTM-Attention combined neural network algorithm. The BiLSTM module carries out bidirectional processing on input time sequence data and controls information flow through an input gate, a forgetting gate and an output gate, the Attention module distributes an attention weight for the feature of each time step, weighted summation is carried out on the output feature of the BiLSTM module through the attention weight to obtain a feature representation which is focused on, and the feature representation is used for representing the feature of each time step. Therefore, the prediction model can focus on part of time sequence characteristics which have the most influence on the prediction result, and the final prediction model can predict a more accurate deep foundation pit excavation deformation result according to the input data through continuous training.

Description

Technical field:

[0001] The present invention belongs to the technical field of deep foundation pit deformation prediction, and specifically relates to a deep foundation pit deformation prediction method based on a neural network model, which provides basic data support for engineering construction quality acceptance and safety early warning. Background technology:

[0002] Since deep foundation pit excavation will have a great impact on surrounding facilities, the problem of deformation prediction of deep foundation pit excavation is one of the common prediction problems in the engineering field. Its goal is to accurately predict the deformation of deep foundation pit excavation by learning and training historical deformation data, taking into account key influencing factors, and then take corresponding control measures to ensure the safety of the construction process. When predicting the deformation of deep foundation pit excavation, the monitoring data of the past N days is usually used as input, combined with conventional algorithms such as LSTM (Long Short-Term Memory) for processing time series problems. For example, Chinese Patent 201710780941.5 discloses a foundation pit excavation monitoring and early warning method based on a neural network, which includes the following steps:

[0003] ⑴Arrangement of foundation pit displacement monitoring points

[0004] According to the Technical Specification for Monitoring of Construction Foundation Pit Engineering (GB50497-2009) and the Code for Monitoring of Construction of Foundation Pit Engineering (DG / TJ08-2001-2006), M displacement monitoring benchmark points are arranged in the stable non-deformation area of ​​3 times the excavation depth (3H) outside the excavation surface of the monitoring foundation pit, and L displacement monitoring points are arranged at the top of the foundation pit slope or the top of the retaining wall along the perimeter of the foundation pit. The displacement monitoring benchmark points and displacement monitoring points together form a displacement deformation monitoring control network of the excavation surface. With the excavation of the foundation pit, the horizontal displacement and vertical displacement of the top of the foundation pit slope at different excavation depths are monitored;

[0005] ⑵ Determination of horizontal and vertical displacements at different excavation depths of foundation pit top

[0006] Take a section around the top of the slope as an example, the number of monitoring points on this section is N;

[0007] ⑶ Determination of the mean horizontal and vertical displacements at different excavation depths of the foundation pit top

[0008] The average value of horizontal and vertical displacements of N monitoring points is used as the deformation of the foundation pit during the excavation process;

[0009] ⑷Establishment of BP neural network model based on foundation pit displacement

[0010] ①Establishment of BP neural network model

[0011] According to the modeling principle of BP neural network prediction method, it is determined that the BP network model consists of three parts, including input layer, hidden layer and output layer. The number of units in the input layer is one, which is represented by the excavation depth vector group; the number of units in the output layer is one, which is represented by the foundation pit displacement vector group; the weight and threshold from the input layer to the hidden layer are wjk and θjk respectively, and the weight and threshold from the hidden layer to the output layer are vkj and γkt respectively;

[0012] ② Algorithm for establishing BP neural network model

[0013] 1) According to the BP neural network model diagram, the net input and output of each neuron in the middle hidden layer are expressed by equations (1) and (2) respectively;

[0014] 2) The net input and actual output of each neuron in the output layer are expressed by equations (3) and (4) respectively;

[0015] 3) According to the given expected output Y(hj), calculate the correction error dt of each neuron in the output layer, expressed as formula (5);

[0016] 4) The correction error ek of each neuron in the hidden layer is expressed as formula (6);

[0017] 5) Modify the connection weight V from the hidden layer to the output layer and the threshold γ of the output layer neurons;

[0018] 6) Modify the connection weight W from the input layer to the hidden layer and the threshold θ of the hidden layer neurons;

[0019] Randomly select the next learning pattern pair and provide it to the network until all learning pattern pairs are trained. Then judge whether the global error of the network meets the accuracy requirement. If it does, end the training, otherwise continue.

[0020] ⑸Compilation of neural network program in deep foundation pit deformation prediction

[0021] According to the BP neural network model determined in step (4), the displacement prediction value of the foundation pit can be obtained by using the MATLAB neural network toolbox. The excavation depth vector group is used as the input sample, and the displacement vector group is used as the target vector to be transferred into the toolbox to obtain the network output, which is the displacement prediction value of the next excavation depth hn+1.

[0022] ⑹ Prediction of mean horizontal and vertical displacements of monitoring points during step-by-step excavation of foundation pit

[0023] Step (5) The BP neural network program compiled obtains the displacement prediction value of the excavation depth hn+1 according to the displacement measured values ​​at the excavation depths h1, h2, ..., hn, and then obtains the prediction value of the excavation depth hn+2 according to the displacement measured values ​​at the excavation depths h1, h2, ..., hn, hn+1. The modeling and prediction are carried out simultaneously until the law of foundation pit deformation can be reflected.

[0024] ⑺Analysis and evaluation of foundation pit stability

[0025] In step (6), the displacement prediction value of the monitoring point is adjusted to meet the error requirements, and then compared with the warning values ​​of different levels. If the displacement prediction value of the monitoring point is less than the blue warning level alarm value, the foundation pit is stable and excavation is carried out according to the original plan; if the displacement prediction value of the monitoring point exceeds the blue alarm value in the table below, it indicates that the foundation pit has entered the primary stage of instability. At this time, a blue warning should be issued to attract the attention of construction personnel and strengthen reinforcement measures in advance; if the displacement prediction value of the monitoring point exceeds the orange alarm value in the table below, it indicates that the foundation pit has entered the intermediate stage of instability. At this time, an orange warning should be issued, and the excavation plan should be adjusted at the same time, and the construction personnel should further optimize the reinforcement measures; if the displacement prediction value of the monitoring point exceeds the red alarm value in the table below, it indicates that the foundation pit is about to become unstable, and after the next step of excavation, excessive displacement will occur, resulting in instability of the side wall of the foundation pit. At this time, excavation should be stopped, reinforcement measures should be further strengthened, and preparations for timely evacuation of personnel should be made. However, it has the problems of low prediction accuracy and poor adaptability, and the prediction results have a strong guiding significance for the improvement direction of construction.

[0026] As a special recurrent neural network, BiLSTM (Bidirectional Long Short-Term Memory) neural network can capture the previous and next context information at the same time by inputting data forward and backward into two independent LSTM networks and merging the outputs of the two networks. It has higher performance than traditional LSTM networks when processing sequence data.

[0027] BiLSTM effectively captures long-term dependencies by introducing three gate structures (input gate, forget gate, and output gate) to control the flow of information. These three gate structures are responsible for controlling input information, retaining useful information, and outputting information respectively:

[0028] (1) The Forget Gate determines which information needs to be forgotten or retained. Through a series of calculations, the Forget Gate outputs a value between 0 and 1 to each neuron cell. If a value in the Forget Gate is close to 1, the corresponding information is retained; if it is close to 0, the corresponding information is forgotten.

[0029] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0030] Among them, σ represents the sigmoid function, W f and b f Represent the weight and bias of the forget gate respectively,

[0031] [h t-1 ,x t ] represents the concatenation of the previous hidden state and the current input, and f t is the output of the forget gate.

[0032] (2) The input gate is responsible for updating the cell state. First, the sigmoid function determines which values ​​need to be updated; second, the tanh function creates a new candidate value vector that can be added to the state; finally, the output of the sigmoid is multiplied by the output of the tanh to determine the actual state value to be updated.

[0033] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0034]

[0035] Among them, i t is the output of the input gate, W i and b i Represent the weight and bias of the input gate respectively, is the candidate cell state, tanh represents the hyperbolic tangent function, W C and b C Represent the weight and bias of the candidate cell state, respectively, and f t is the output of the forget gate, C t-1 is the previous cell state, and C t is the updated cell state.

[0036] (3) The output gate determines the next hidden state. First, the sigmoid function is used to determine which parts of the cell state will be output; then, the cell state is processed by the tanh function and the result is multiplied by the output of the sigmoid to obtain the final output.

[0037] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0038] h t =o t tanh(C t )

[0039] Among them, t is the output of the output gate. W oand b o Represent the weight and bias of the output gate, C t is the current cell state, and h t is the current hidden state.

[0040] Therefore, based on the neural network model, a deep foundation pit deformation prediction method was developed and designed to solve the above problems, so as to obtain accurate prediction results and provide a better reference for actual construction on specific issues. Summary of the invention:

[0041] The purpose of the present invention is to overcome the shortcomings of the prior art, develop and design a deep foundation pit deformation prediction method based on a neural network model, improve the model generalization ability and prediction accuracy by introducing an attention mechanism, and integrate the sparrow search algorithm to improve the model parameter optimization efficiency, so as to provide correct guidance for improving construction.

[0042] In order to achieve the above-mentioned purpose, the specific process of a deep foundation pit deformation prediction method based on a neural network model according to the present invention is as follows:

[0043] S1, establish a neural network prediction model

[0044] Establish a prediction model based on the neural network model, including an input layer, a combined network layer, a parameter optimization layer and an output layer;

[0045] Among them, the input layer is used to input feature vectors;

[0046] The combined network layer consists of a BiLSTM (Bidirectional Long Short-Term Memory) module and an Attention (attention mechanism) module. The BiLSTM module is used to process time series information, and the Attention module is used to enhance the prediction model's ability to capture key information.

[0047] The parameter optimization layer is the Sparrow Search Algorithm (SSA) layer, which is used to globally optimize the parameters of the entire neural network;

[0048] The output layer is used to output the prediction results;

[0049] S2, build dataset

[0050] First, the data required for the prediction model is normalized and divided into training set and test set in a ratio of 9:1;

[0051] Then, the input feature data are divided into three categories: the first category is environmental factors, the second category is construction factors, and the third category is monitoring data;

[0052] Finally, the monitoring data is normalized and split into an input vector and an output vector. The input vector includes all three types of feature data, and the output vector is the third type of monitoring data.

[0053] S3, model initialization

[0054] S31, setting the parameters of the BiLSTM module according to the characteristics of the input vector;

[0055] S32, set the number of heads of the Attention module and initialize the weight matrix;

[0056] S33, set the population size and number of iterations of the SSA algorithm, and use the mean square error (MSE) as the fitness function;

[0057] S4, model training

[0058] S41, forward propagation, inputs the training set data into the BiLSTM module in time steps through the input layer, and calculates the hidden state of each time step;

[0059] S42, calculate the attention score and weight, input the output of the BiLSTM module into the Attention module to calculate the attention weight of the hidden state at each time step;

[0060] S43, calculating the fitness of the current prediction model, and the SSA algorithm updates the prediction model parameters according to the fitness;

[0061] S44, performing back propagation according to the updated prediction model parameters;

[0062] S45, training is performed according to the operations of S41 to S44 until a termination condition is met;

[0063] S5, the output layer outputs the prediction results.

[0064] The BiLSTM module involved in the present invention performs bidirectional processing on input time series data (such as daily monitoring data in foundation pit engineering, including soil pressure, groundwater level changes, enclosure structure displacement, etc.), and controls the flow of information through input gates, forget gates and output gates;

[0065] The Attention module assigns an attention weight to the features of each time step, and uses the attention weight to perform weighted summation on the output features of the BiLSTM module to obtain a focused feature representation, so that the prediction model can focus on the time series features that have the greatest impact on the prediction results.

[0066] The parameter optimization layer simulates the population strategy of sparrows foraging for the initial weights and biases of the neurons in the BiLSTM module and the Attention module to perform global optimization to prevent falling into the local optimal situation and achieve a better training effect for the combined network.

[0067] The steps for training and optimizing the combined network layer are as follows:

[0068] S11, randomly initialize the individual positions of the population;

[0069] S12, calculate the fitness of each individual, and update the role of each individual according to the fitness;

[0070] S13, updating individual positions, i.e. updating the parameters of the combined network layer;

[0071] S14, repeat the operations of S12-S13 until the termination condition is met.

[0072] Compared with the prior art, the present invention is based on the BiLSTM-Attention combined neural network algorithm. The BiLSTM module performs bidirectional processing on the input time series data, and controls the flow of information through an input gate, a forget gate and an output gate. The Attention module assigns an attention weight to the feature of each time step, and weighted sums the output features of the BiLSTM module through the attention weight to obtain a focused feature representation, so that the prediction model can focus on the part of the time series features that have the greatest impact on the prediction result. After continuous training, the prediction model will eventually predict a more accurate deep foundation pit excavation deformation result according to the input data; the BiLSTM-Attention combined neural network algorithm is used to train the deep foundation pit excavation deformation data, and the deep foundation pit excavation deformation under different influencing factors is predicted. The parameter optimization layer of the SSA algorithm simulates the population strategy of sparrows foraging to globally optimize the weight parameters of the combined neural network to prevent falling into a local optimal situation, thereby achieving a better training effect of the combined network and improving the accuracy of the prediction model. Description of the drawings:

[0073] Figure 1 It is a schematic diagram of the structure of the prediction model involved in the present invention.

[0074] Figure 2 The figure is a flow chart of the prediction model involved in the present invention. Specific implementation method:

[0075] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0076] Embodiment 1:

[0077] The specific process of a deep foundation pit deformation prediction method based on a neural network model involved in this embodiment is as follows:

[0078] S1, establish a neural network prediction model

[0079] Establish a prediction model based on the neural network model, including an input layer, a combined network layer, a parameter optimization layer and an output layer;

[0080] Among them, the input layer is used to input feature vectors;

[0081] The combined network layer consists of a BiLSTM (Bidirectional Long Short-Term Memory) module and an Attention (attention mechanism) module. The BiLSTM module is used to process time series information, and the Attention module is used to enhance the prediction model's ability to capture key information.

[0082] The parameter optimization layer is the Sparrow Search Algorithm (SSA) layer, which is used to globally optimize the parameters of the entire neural network;

[0083] The output layer is used to output the prediction results;

[0084] S2, build dataset

[0085] First, the data required for the prediction model is normalized and divided into training set and test set in a ratio of 9:1;

[0086] Then, the input characteristic data are divided into three categories: the first category is environmental factors, including average temperature, humidity and rainfall; the second category is construction factors, including excavation volume, excavation depth and number of cross braces; the third category is monitoring data, including average groundwater level, axial force, horizontal displacement of ground-connected wall, surrounding surface settlement and horizontal displacement of pile top;

[0087] Finally, the monitoring data is normalized and split into input vectors and output vectors. The input vector includes all feature data, and the output vector is the third type of monitoring data.

[0088] S3, model initialization

[0089] S31, setting the parameters of the BiLSTM module according to the characteristics of the input vector, including the input dimension, the number of hidden units and the activation function;

[0090] S32, set the number of heads of the Attention module and initialize the weight matrix;

[0091] S33, set the population size and number of iterations of the SSA algorithm, and use the mean square error (MSE) as the fitness function;

[0092] S4, model training

[0093] S41, forward propagation, inputs the training set data into the BiLSTM module in time steps through the input layer, and calculates the hidden state of each time step;

[0094] S42, calculate the attention score and weight, input the output of the BiLSTM module into the Attention module to calculate the attention weight of the hidden state at each time step;

[0095] S43, calculating the fitness of the current prediction model, and the SSA algorithm updates the prediction model parameters according to the fitness;

[0096] S44, performing back propagation according to the updated prediction model parameters;

[0097] S45, training is performed according to the operations of S41 to S44 until a termination condition is met;

[0098] S5, the output layer outputs the prediction results.

[0099] Among them, the BiLSTM module is based on bidirectional processing and gating mechanism, which can extract rich time series features. These features include the changing rules and mutual relationships of data in the time dimension;

[0100] The parameter optimization layer uses the SSA algorithm to quickly locate a possible better parameter area through the random exploration of the scout and the guidance of the producer. As the iteration proceeds, the search is further refined in this better area and the prediction model parameters are adjusted so that the prediction model can better fit the time series data of foundation pit deformation and improve the accuracy of the prediction.

[0101] Embodiment 2:

[0102] This embodiment involves a deep foundation pit deformation prediction method based on a neural network model to predict a foundation pit adjacent to the Yangtze River;

[0103] The general situation of the foundation pit is as follows: the plane is approximately trapezoidal, the excavation depth is about 25.15m, and the area is about 17,000m 2 , on the north side of the excavation, 140m away from the edge of the excavation is the Yangtze River;

[0104] The enclosure structure adopts underground continuous wall and adds lateral internal support to restrain the ground deformation;

[0105] The construction method adopts the excavation-support-precipitation mode;

[0106] The underlying soil layers in the foundation pit area within the scope of the project are miscellaneous fill, plain fill, silty clay, fine sand, fine medium sand, strongly weathered rock, and moderately weathered rock. The strata within the excavation area of ​​the foundation pit are mainly clay layers.

[0107] The monitoring object is mainly underground continuous walls. The horizontal and vertical displacement monitoring points of the underground continuous walls are arranged at intervals of about 20-25m along the perimeter of the foundation pit. The lateral supports are arranged at the corners of the foundation pit, and the lateral displacement of the pile tops is evenly arranged inside the foundation pit.

[0108] Environmental monitoring mainly focuses on the surface settlement of the pit. The surface settlement monitoring points outside the foundation pit are arranged at the center of the four sides of the foundation pit, with five on each side. The groundwater level is arranged at intervals of 30-50 meters around the foundation pit and evenly arranged inside the pit.

[0109] After construction begins, continuously analyze the trend of monitoring data changes, dynamically adjust monitoring strategies, and provide timely feedback on monitoring results;

[0110] The specific prediction process is:

[0111] The data set was collated based on the 227-day construction logs. The first 200 days of data were used as the training set to train the parameters of the SSA-BiLSTM-Attention neural network and dynamically adjust the parameters. The last 27 days of data were used as the validation set and the neural network parameters were not adjusted.

[0112] Based on the above parameters, the groundwater level change, axial force of the horizontal support, surface settlement, horizontal deformation of the ground-connected wall and horizontal displacement of the pile top during the foundation pit excavation process are predicted. The environmental factors, construction factors and monitoring values ​​of the previous day are used as input features, and the predicted values ​​at the monitoring points are used as output features. The input features and output features are normalized and the loss function value is calculated. MSE and R2 are used as the loss function and evaluation index of the prediction model.

[0113] The results show that among the five prediction tasks on the test set, the numerical prediction trend is good, indicating that the SSA-BiLSTM-Atention neural network model can handle the deformation during the foundation pit excavation process well;

[0114] From the perspective of the MSE of the loss function, for the prediction tasks of groundwater level, axial force and surface settlement, the loss function is relatively low, which are 3.55, 3.40 and 3.12 respectively: for the prediction tasks of horizontal deformation of ground-anchored wall and horizontal displacement of pile top, the loss function after stabilization has a larger amplitude, which are 4.49 and 3.95 respectively;

[0115] From the determination coefficient R2, the R2 of the five prediction tasks are 0.95, 0.91, 0.97, 0.87, and 0.89 respectively. The first three have better fitting degree, and the latter two have the second best fitting degree;

[0116] From the overall prediction effect, the predicted value is closer to the true value and the effect is better.

Claims

1. A deep foundation pit deformation prediction method based on a neural network model, characterized in that: The specific process is as follows: S1, establish a neural network prediction model Establish a prediction model based on the neural network model, including an input layer, a combined network layer, a parameter optimization layer and an output layer; S2, build dataset First, the data required for the prediction model is normalized and divided into training set and test set in a ratio of 9:1; Then, the input feature data are divided into three categories: the first category is environmental factors, the second category is construction factors, and the third category is monitoring data; Finally, the monitoring data is normalized and split into an input vector and an output vector. The input vector includes all three types of feature data, and the output vector is the third type of monitoring data. S3, model initialization S31, setting the parameters of the BiLSTM module according to the characteristics of the input vector; S32, set the number of heads of the Attention module and initialize the weight matrix; S33, sets the population size and number of iterations of the SSA algorithm, taking the mean square error as the fitness function; S4, model training S41, forward propagation, inputs the training set data into the BiLSTM module in time steps through the input layer, and calculates the hidden state of each time step; S42, calculate the attention score and weight, input the output of the BiLSTM module into the Attention module to calculate the attention weight of the hidden state at each time step; S43, calculating the fitness of the current prediction model, and the SSA algorithm updates the prediction model parameters according to the fitness; S44, performing back propagation according to the updated prediction model parameters; S45, training is performed according to the operations of S41 to S44 until a termination condition is met; S5, the output layer outputs the prediction results.

2. A deep foundation pit deformation prediction method based on a neural network model according to claim 1, characterized in that: In step S1, the input layer is used to input the feature vector; The combined network layer consists of a BiLSTM module and an Attention module. The BiLSTM module is used to process time series information, and the Attention module is used to enhance the prediction model's ability to capture key information. The parameter optimization layer is a sparrow search algorithm layer, which is used to globally optimize the parameters of the entire neural network; The output layer is used to output the prediction results.

3. The method for predicting deformation of deep foundation pit based on neural network model according to claim 1 is characterized in that: In step S1, environmental factors include average temperature, humidity and rainfall; construction factors include excavation volume, excavation depth and number of cross braces; the third type of monitoring data includes average groundwater level, axial force, horizontal displacement of ground-anchored wall, surrounding surface settlement and horizontal displacement of pile top.

4. A deep foundation pit deformation prediction method based on a neural network model according to claim 1 or 2, characterized in that: The BiLSTM module processes the input time series data bidirectionally and controls the flow of information through the input gate, forget gate, and output gate.

5. A method for predicting deep foundation pit deformation based on a neural network model according to claim 1 or 2, characterized in that: The Attention module assigns an attention weight to the features of each time step, and uses the attention weight to perform weighted summation on the output features of the BiLSTM module to obtain a focused feature representation, allowing the prediction model to focus on the time series features that have the greatest impact on the prediction results.

6. A method for predicting deformation of a deep foundation pit based on a neural network model according to claim 1 or 2, characterized in that: The parameter optimization layer performs global optimization on the initial weights and biases of the neurons in the BiLSTM module and the Attention module, simulating the population strategy of sparrows foraging.

7. The method for predicting deformation of deep foundation pit based on neural network model according to claim 6 is characterized in that: The steps for training the parameter optimization layer and optimizing the combined network layer are as follows: S11, randomly initialize the individual positions of the population; S12, calculate the fitness of each individual, and update the role of each individual according to the fitness; S13, updating individual positions, i.e. updating the parameters of the combined network layer; S14, repeat the operations of S12-S13 until the termination condition is met.

Images