CN111582632B - Multi-factor safety stage prediction method for whole process of underground large space construction - Google Patents

Abstract

The invention discloses a multi-factor safety prediction method and system in the whole process of underground large space construction, including: constructing and training a pre-construction safety prediction neural network model to form a nonlinear mapping relationship from an input layer to an output layer; constructing and training Training the safety prediction neural network model during construction so that it forms a nonlinear mapping relationship from the input layer to the output layer; connecting the pre-construction safety prediction neural network model and the construction safety prediction neural network model in series to form a construction prediction series model , using the construction prediction series model to carry out stage-by-stage safety prediction of the whole construction process. The invention establishes a construction safety prediction model based on a neural network with self-adaptability, nonlinearity and fault tolerance, and can predict the safety of underground large space construction before construction without relying on the internal working mechanism of the geotechnical system, and during construction Real-time prediction, and a series model is formed to predict the whole process of construction.

Description

A multi-factor safety phased prediction method for the entire process of underground large space construction

Technical Field

The present invention relates to the technical field of civil engineering construction, and in particular to a multi-factor safety prediction method and system for the entire process of underground large space construction.

Background Art

The practice of underground engineering shows that the geological environment, hydrological environment, surrounding building environment and the selection of construction methods during the construction process will have an impact on the deformation, stress, strain, settlement, displacement, etc. during construction. In most cases, engineering and technical personnel can only grasp the changes in deformation, stress, strain, settlement, displacement, etc. by collecting monitoring data during construction, and then improve the construction methods. They cannot have a rough prediction before construction, which often leads to various disasters and construction safety accidents due to untimely improvements.

In addition, theoretical calculation methods are currently mainly used to predict multiple safety factors such as deformation, stress, strain, settlement, displacement, etc. during construction. However, due to the lack of understanding of the internal working mechanism of the geotechnical system, there are bound to be great difficulties in establishing the corresponding theoretical calculation expressions, and the prediction calculation of the safety interference factors of the geotechnical engineering system shows very complex high-order nonlinear characteristics, and nonlinear calculation itself has certain difficulties.

Neural networks are particularly suitable for dealing with various nonlinear problems due to their adaptability, nonlinearity and strong fault tolerance. It can extract the causal relationship implicit in the samples through the learning of a large number of samples. Therefore, neural networks provide a research idea that is completely different from mathematical modeling in the field of underground engineering. It avoids complex constitutive models and becomes an effective way to solve underground engineering problems. At the same time, neural networks also have many applications in other fields, such as a method for assessing the risk of rainstorm disasters on the slopes of the foundation of a section of a transmission line tower. Based on disaster statistics and artificial rainfall slope erosion test results, an improved BP network is used to establish a mapping relationship between control factors and rainstorm landslide accident rates, and combined with an improved hierarchical analysis calculation program, the rainstorm disaster assessment results of the slopes of each section of the line are obtained. The difference between this patent and it is that this patent intends to establish a multi-factor safety stage prediction method for the entire process of underground large space construction, including two stages before construction and during construction, and consider the entire construction process by establishing a two-stage series model, making safety predictions before construction and real-time predictions during construction, and making safety judgments through relevant specifications or classification systems to ensure the safety of the entire process of construction.

Summary of the invention

At least one of the purposes of the present invention is to provide a multi-factor safety prediction method and system for large underground space construction in order to overcome the problems existing in the above-mentioned prior art. A construction safety prediction model is established based on a neural network with adaptability, nonlinearity and strong fault tolerance. The method and system can make safety predictions for large underground space construction before construction without relying on the internal working mechanism of the geotechnical system, and can make real-time predictions during construction.

In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention includes the following aspects.

A multi-factor safety prediction method for the entire process of underground large space construction, including:

Step 1, constructing a pre-construction safety prediction neural network model, and using the first training sample to train the pre-construction safety prediction neural network model so that the performance of the pre-construction safety prediction neural network model tends to be stable and forms a nonlinear mapping relationship from its input layer to the output layer;

The input parameters of the input layer of the pre-construction safety prediction neural network are: engineering geological conditions, hydrological conditions, surrounding building environment, construction methods, management level, and construction level; the output parameters of the pre-construction prediction neural network are the corresponding predicted values of stress, strain, displacement, and settlement;

Step 2, constructing a neural network model for safety prediction during construction, and using the second training sample to train the neural network model for safety prediction during construction, so that the performance of the neural network for safety prediction during construction and the neural network for prediction before construction tend to be stable and form a nonlinear mapping relationship from its input layer to its output layer;

The input parameters of the input layer of the safety prediction neural network model during construction are: the input values of stress, strain, displacement and settlement at a certain time node during construction; the output parameters of the safety prediction neural network model during construction are the predicted values of stress, strain, displacement and settlement at the next time node during construction;

Step 3, using the output parameters of the pre-construction safety prediction neural network model as the input parameters of the in-construction safety prediction neural network model, connecting the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model in series to form a construction prediction series model, and using the construction prediction series model to perform safety predictions in stages throughout the entire construction process.

Preferably, the above method also includes: inputting the stress, strain, displacement and settlement monitored in real time during construction into the in-construction safety prediction neural network model, so as to perform real-time prediction of the stress, strain, displacement and settlement of the next time node during construction through the in-construction safety prediction neural network model.

Preferably, the current output value of the node weights of each layer of the construction prediction series model is calculated by the following formula:

为第k层第i元素的输入和；

为第k层第i元素的输出；

为第k-1层第i元素向第k层第j元素的连接权值；f为激励函数；

为与权矢量W和输入矢量X有关的第m层的第j元素的实际输出，其中，第m层即为输出层。in,

is the input sum of the i-th element of the k-th layer;

is the output of the i-th element of the k-th layer;

is the connection weight from the i-th element in the k-1th layer to the j-th element in the kth layer; f is the activation function;

is the actual output of the jth element of the mth layer related to the weight vector W and the input vector X, where the mth layer is the output layer.

Preferably, the weights of the nodes at each layer of the construction prediction series model are adjusted by the following formula:

When k = m,

When k<m,

为第k层第i元素的输入和；

为第k层第i元素的输出；

为第k-1层第i元素向第k层第j元素的连接权值；f为激励函数，可以是S_c(x)；

为与权矢量W和输入矢量X有关的第m层的第j元素的实际输出；y_i为与权矢量W和输入矢量X有关的第m层的第j元素的期望输出，其中，第m 层即为输出层。in,

is the input sum of the i-th element of the k-th layer;

is the output of the i-th element of the k-th layer;

is the connection weight from the i-th element in the k-1th layer to the j-th element in the kth layer; f is the activation function, which can be S _c (x);

is the actual output of the j-th element of the m-th layer related to the weight vector W and the input vector X; _yi is the expected output of the j-th element of the m-th layer related to the weight vector W and the input vector X, where the m-th layer is the output layer.

Preferably, when constructing the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model, their maximum number of iterations is set to 5000 and the learning rate η is set to 0.5.

Preferably, when the error rates of the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model are less than preset values, it is judged that the performance of the pre-construction prediction neural network tends to be stable.

In a further embodiment of the present invention, a multi-factor safety prediction system for the entire process of underground large space construction is also provided, including at least one processor, and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above method.

In summary, due to the adoption of the above technical solution, the present invention has at least the following beneficial effects:

Based on a neural network with strong adaptability, nonlinearity and fault tolerance, a pre-construction safety prediction neural network model and a construction safety prediction neural network model are established. The engineering geological conditions, hydrological conditions, surrounding building environment, construction methods, management level and construction level are used as input variables of the pre-construction safety prediction neural network model. The values of stress, strain, displacement and settlement corresponding to the input variables are used as output vectors of the pre-construction safety prediction neural network model. It is possible to make a safety prediction for the construction of a large underground space before construction, and connect the construction safety prediction neural network model in series to form a construction prediction series model. The construction prediction series model is used to make a stage-by-stage safety prediction for the entire construction process. The prediction method adopted by the present invention comprehensively considers the two stages before construction and during construction, and considers the entire construction process by establishing a two-stage series model, making a safety prediction before construction and a real-time prediction during construction, and making a safety judgment through construction specifications and/or a grading system to ensure the safety of the entire construction process.

The stress, strain, displacement, and settlement monitored in real time during construction are input into the in-construction safety prediction neural network model, so that the in-construction safety prediction neural network model can be used to make a real-time prediction of the stress, strain, displacement, and settlement at the next time node during construction.

BRIEF DESCRIPTION OF THE DRAWINGS

1 is a flowchart of a neural network model training for a pre-construction/mid-construction safety prediction neural network model according to an exemplary embodiment of the present invention.

FIG. 2 is a topological diagram of a neural network model for safety prediction before construction according to an exemplary embodiment of the present invention.

FIG. 3 is a topological diagram of a neural network model for safety prediction during construction according to an exemplary embodiment of the present invention.

FIG4 is a block diagram of construction phase prediction of a construction prediction series model according to an exemplary embodiment of the present invention.

5 is a schematic diagram of the structure of a multi-factor safety prediction system for the entire process of large underground space construction according to an exemplary embodiment of the present invention.

DETAILED DESCRIPTION

The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments to make the purpose, technical solutions and advantages of the present invention more clearly understood. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

Example 1

This embodiment provides a multi-factor safety prediction method for the entire process of underground large space construction, including the following steps:

Step 1, construct a pre-construction safety prediction neural network model, and use the first training sample to train the pre-construction safety prediction neural network model to make the performance of the pre-construction safety prediction neural network model tend to be stable and form a nonlinear mapping relationship from its input layer to the output layer; wherein, data on stress, strain, displacement, and settlement corresponding to different engineering geological conditions, hydrological conditions, surrounding building environment, construction methods, management levels, and construction levels are normalized into the first training sample.

Determination of input parameters. First, consider all items that affect the safety of the project itself and the safety of the environment during construction, such as geological conditions, hydrological conditions, surrounding building environment and construction methods, and then refer to specifications, standards, and expert evaluation scores to determine the value of the input parameters through normalization.

The output parameters are determined based on the measured data and normalized as the output parameters; a database is established, and the engineering conditions, i.e., the input parameters, are established in the database, and the measured data, i.e., the output parameters, are also established in the database.

The operation of the neural network. The learning algorithm is a multi-layer neural network algorithm guided by a teacher signal. It is a supervised learning process. It learns based on the given (input, output) sample pairs and reflects the learning effect by adjusting the network connection weights.

Neural networks have two states. In the learning phase, the input of the learning sample pair is first added to the input end of the network, and the output is generated in each layer of neurons along the forward direction (i.e., input layer---output layer) according to the input and excitation function (Sigmoid function). Then the difference between the actual output value and the expected output value of the output layer neurons is propagated to each layer of neurons in the reverse direction (i.e., output layer---input layer), and the connection weights are adjusted accordingly according to the size and sign of the error. This process continues until the weighted connection mode of the neural network can produce a given output result with a certain accuracy under the given input sample conditions, and the learning phase is considered to be over. In the working phase, when the sample to be tested is input to the input end of the learned neural network, according to the principle of similar output, the neural network generates the required solution at the output end by interpolation or extension.

Specifically, the implementation steps as shown in Figure 1 are: reading in learning samples---data normalization (normalization in this patent)---initializing neural network weights---calculating hidden layer node output values---calculating output node output values---calculating output layer errors---calculating hidden layer errors---adjusting weights - if it is within the allowable error range, then end; if not, then renormalize and conduct training.

The process of constructing the pre-construction safety prediction neural network model shown in Figure 2 includes: determining the initial parameters. Assigning small non-zero random real initial values to each network weight w _ij , and setting the learning rate η and the inertia coefficient α. Different random initial weights will result in different final weights. The number of hidden layers and hidden units of the network depends on different specific problems. According to the proven mapping theorem, as long as there is a three-layer neural network with one hidden layer, it can approximate any mapping function to complete a given mapping task.

There are many calculation formulas for the number of hidden layer units. Here we use the following formula:

n ₁ =log ₂ n (1)

Where: n ₁ ——number of hidden layer units,

n——the number of input layer units.

The number of samples, when learning, the number of sample pairs is closely related to the number of hidden layers and the number of units in the hidden layer. The more hidden layers there are, the more accurate the connection weights learned, but the worse the generalization ability of the network. According to research, in order to make the multi-layer network have generalization ability, the relationship between the total number of adjustable connection weights W of the network and the necessary number of training sample pairs N can be approximately expressed as

τ—coefficient, about 10.

When learning, the conditions for learning termination must be given. Generally, two methods are used to terminate network learning: one is to give a minimum error value, and terminate when the actual output error is less than the given error; the other is to specify the number of iterations (for example, 5000 times). The former is used in this system. In this embodiment, the maximum number of iterations is set to 5000 and the learning rate η is set to 0.5.

Neural network model training: Input the first sample pair among P (input, output) sample pairs.

Calculate the actual output value. According to the formula

Calculate the actual output value of each layer of the network. Adjust the weight of each connection according to the formula

Adjust the weights of each connection, where:

When k = m,

When k<m,

After the first sample pair is completed, the subsequent sample pairs are input and the above steps are repeated until the end. P sample pairs are recycled until w _ij becomes stable and unchanged. When r ≤ ε is satisfied during network training, the learning process ends.

The symbols in the above calculation formula have the following meanings:

第k层第i元素的输入和；

The input sum of the i-th element of the k-th layer;

第k层第i元素的输出；

The output of the i-th element of the k-th layer;

第k-1层第i元素向第k层第j元素的连接权值；

The connection weight of the i-th element in the k-1th layer to the j-th element in the kth layer;

f: activation function, which can be S _c (x);

与权矢量W和输入矢量X有关的第m层(即输出层)的第j元素的实际输出；

The actual output of the jth element of the mth layer (i.e., the output layer) associated with the weight vector W and the input vector X;

_yi : the expected output of the jth element of the mth layer (ie, output layer) associated with the weight vector W and the input vector X.

Step 2, constructing a neural network model for safety prediction during construction as shown in Figure 3, and using the second training sample to train the neural network model for safety prediction during construction, so that the performance of the neural network for safety prediction during construction and the neural network for prediction before construction tend to be stable and form a nonlinear mapping relationship from its input layer to the output layer; wherein, the stress, strain, displacement, and settlement data at a certain time node during construction, as well as the stress, strain, displacement, and settlement data at the next time node, are normalized to form the second training sample.

Similar to step 1, input parameters are determined. Prediction during construction: The input parameters of the input layer of the safety prediction neural network model during construction are: the input values of stress, strain, displacement, and settlement at a certain time node during construction; the output parameters of the safety prediction neural network model during construction are the predicted values of stress, strain, displacement, and settlement at the next time node during construction; initial parameters are determined. Assign small non-zero random real number initial values to each network weight w _ij , and set the learning rate η and inertia coefficient α. Different random initial weights will result in different final weights. The number of hidden layers and hidden units of the network depends on different specific problems. The specific determination method is based on the above description and formulas (1) and (2). Input the first sample pair of P (stress, strain, displacement, and settlement data at a certain time node during construction, and stress, strain, displacement, and settlement data at the next time node) sample pairs, and calculate the actual output value according to the above formula (3). According to the above formulas (4) to (6), adjust the weights of each connection. After the first sample pair is completed, input the subsequent sample pairs, and repeat the above steps until the end. P sample pairs are reused repeatedly until w _ij becomes stable and unchanged. When r≤ε is satisfied during network training, the learning process ends. When the model training is completed, the working stage can be entered to carry out security classification, and the corresponding emergency and response methods can also be determined according to the classification.

Step 3, using the output parameters of the pre-construction safety prediction neural network model as the input parameters of the in-construction safety prediction neural network model, connecting the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model in series to form a construction prediction series model, and using the construction prediction series model to perform safety predictions in stages throughout the entire construction process.

As shown in Figure 2, the prediction process of the construction prediction series model. In practical application, the two safety prediction neural network models before construction and during construction are used in series. The first input of this series model is the engineering geology, hydrogeology, surrounding environmental conditions, construction methods, management level, and construction level before construction. Through the pre-construction prediction model, the stress, strain, displacement, and settlement before a certain construction process in construction are predicted. Then, through the in-construction prediction model, the stress, strain, displacement, and settlement of the monitoring data at the next construction time point in construction are predicted. At the same time, according to the stress, strain, displacement, and settlement of the monitoring data during construction, the stress, strain, displacement, and settlement of the monitoring data at the next construction time point in construction are predicted through the in-construction prediction model, and real-time prediction is performed.

Furthermore, in actual use, we input the data of stress, strain, displacement and settlement of the current time node monitored in real time during the construction process into the construction safety prediction neural network model, so as to make real-time prediction of the stress, strain, displacement and settlement of the next time node during construction through the construction safety prediction neural network model.

Example 2

Figure 5 shows a multi-factor safety prediction system for the entire process of underground large space construction according to an exemplary embodiment of the present invention, that is, an electronic device 310 (such as a computer server with a program execution function), which includes at least one processor 311, a power supply 314, and a memory 312 and an input-output interface 313 that are communicatively connected to the at least one processor 311; the memory 312 stores instructions that can be executed by the at least one processor 311, and the instructions are executed by the at least one processor 311 so that the at least one processor 311 can execute the method disclosed in any of the aforementioned embodiments; the input-output interface 313 may include a display, a keyboard, a mouse, and a USB interface for inputting and outputting data; the power supply 314 is used to provide power to the electronic device 310.

Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), disks or optical disks, etc. Various media that can store program codes.

When the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

The above is only a detailed description of the specific implementation of the present invention, rather than a limitation of the present invention. Various substitutions, modifications and improvements made by those skilled in the relevant art without departing from the principle and scope of the present invention should be included in the protection scope of the present invention.

Claims (7)

1. A multi-factor safety prediction method for the whole process of underground large space construction is characterized by comprising the following steps:

step 1, constructing a safety prediction neural network model before construction, and training the safety prediction neural network model before construction by using a first training sample so as to enable the performance of the safety prediction neural network model before construction to tend to be stable and form a nonlinear mapping relation from an input layer to an output layer;

wherein, the input parameters of the input layer of the safety prediction neural network before construction are as follows: engineering geological conditions, hydrological conditions, surrounding building environment, construction method, management level and construction level; the output parameters of the prediction neural network before construction are the predicted values of corresponding stress, strain, displacement and settlement; normalizing the data of stress, strain, displacement and settlement under different engineering geological conditions, hydrological conditions, surrounding building environments, construction methods, management levels and construction levels and corresponding conditions of the engineering geological conditions, the hydrological conditions and the surrounding building environments into a first training sample; the values of the input parameters are determined through normalization by reference to standards, standards and expert argumentation; normalizing the data as output parameters based on the measured stress, strain, displacement and settlement data;

step 2, constructing a safety prediction neural network model in construction, and training the safety prediction neural network model in construction by using a second training sample so as to enable the performance of the safety prediction neural network in construction before construction to tend to be stable and form a nonlinear mapping relation from an input layer to an output layer of the safety prediction neural network in construction;

wherein, the input parameters of the input layer of the safety prediction neural network model in construction are as follows: input values of stress, strain, displacement and settlement of a certain time node in construction; the output parameters of the safety prediction neural network model in construction are stress, strain, displacement and settlement prediction values of the next time node in construction;

and 3, connecting the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model in series by taking the output parameters of the pre-construction safety prediction neural network model as the input parameters of the in-construction safety prediction neural network model to form a construction prediction series model, and performing staged safety prediction in the whole construction process by using the construction prediction series model.

2. The method of claim 1, further comprising: and inputting the stress, strain, displacement and settlement monitored in real time in the construction into the in-construction safety prediction neural network model so as to predict the stress, strain, displacement and settlement of the next time node in real time in the construction through the in-construction safety prediction neural network model.

3. The method of claim 1, wherein the current output value of the node weight of each layer of the construction prediction series model is calculated by the following formula:

wherein,

is the input sum of the ith element of the kth layer;

Is the output of the ith element of the kth layer;

The connection weight value from the ith element of the kth-1 layer to the jth element of the kth layer is obtained; f is an excitation function;

Is the actual output of the jth element of the mth layer relative to the weight vector W and the input vector X, wherein the mthThe layer is the output layer.

4. The method of claim 3, wherein the weight of each layer node of the construction prediction series model is adjusted by the following formula:

when k = m, the number of the magnetic poles is zero,

when k is<When m is greater than the total number of the carbon atoms,

wherein,

is the input sum of the ith element of the kth layer;

Is the output of the ith element of the kth layer;

The connection weight from the ith element of the kth-1 layer to the jth element of the kth layer is set; f is an excitation function, which may be S _c (x)；

Is the actual output of the jth element of the mth layer relative to the weight vector W and the input vector X; y is _i Is the desired output of the jth element of the mth layer associated with the weight vector W and the input vector X, where the mth layer is the output layer.

5. The method according to claim 1, wherein the pre-construction safety prediction neural network model and the in-construction safety prediction neural network model are constructed with a maximum number of iterations set to 5000 and a learning rate η set to 0.5.

6. The method of claim 1, wherein the pre-construction safety prediction neural network performance is determined to be stable when the pre-construction safety prediction neural network model error rate and the in-construction safety prediction neural network model error rate are less than preset values.

7. The multi-factor safety prediction system for the whole process of underground large space construction is characterized by comprising at least one processor and a memory which is in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6.

Images