Deep foundation pit construction intelligent monitoring and early warning method based on multi-source characteristic parameter fusion

Abstract

The invention relates to the field of building construction, in particular to a deep foundation pit construction intelligent monitoring and early warning method based on multi-source characteristic parameter fusion, and constructs a comprehensive foundation pit intelligent monitoring and early warning index system by fusing multi-source characteristic parameters including deformation characteristics, mechanical characteristics and technical characteristics. By means of the combination of a one-dimensional convolutional neural network (1DCNN) and a support vector machine (SVM), complex feature information in foundation pit deformation monitoring data can be deeply excavated and extracted, compared with a traditional prediction method depending on a single feature or a simple model, the foundation pit deformation prediction precision is remarkably improved, and the foundation pit deformation prediction method has the advantages of being high in practicability and the like. And comprehensive monitoring and accurate early warning of the construction process are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of building construction, and in particular to an intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion. Background Art

[0002] In deep foundation pit engineering, accurate prediction of foundation pit deformation is crucial to ensure construction safety. However, most existing deformation prediction methods rely on single characteristic parameters or simple mathematical models, which makes it difficult to fully capture the complexity and dynamics of foundation pit deformation. Foundation pit deformation is affected by many factors, including the complexity of geological conditions, disturbances during construction, and changes in the hydrological environment. These factors make foundation pit deformation highly uncertain and nonlinear. Traditional prediction methods are usually based on limited monitoring data and cannot make full use of modern data processing and intelligent algorithm technologies. Therefore, there are obvious deficiencies in prediction accuracy and reliability. Therefore, how to effectively integrate multi-source data features and use advanced intelligent algorithms for accurate prediction has become a key technical problem that needs to be solved in the current field of deep foundation pit engineering. Summary of the invention

[0003] In view of the above problems, an embodiment of the present invention provides an intelligent monitoring and early warning method for deep foundation pit construction based on multi-source feature parameter fusion.

[0004] An embodiment of the present invention provides an intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion, comprising:

[0005] S1. Analyze the multi-source characteristic parameters that affect the deformation of the deep foundation pit during the construction period, and obtain the original monitoring data of the deep foundation pit deformation corresponding to the multi-source characteristic parameters during the construction process according to the real-time monitoring of the deep foundation pit project site;

[0006] S2. Use the interpolation method to convert the original data of deep foundation pit deformation monitoring into an equal time interval series to obtain more complete new data, and normalize the obtained new data;

[0007] S3. Use the deep learning framework to build a one-dimensional convolutional neural network 1DCNN, and use the convolution layer of the convolutional neural network 1DCNN to perform a convolution operation on the normalized new data to obtain a feature map of the new data;

[0008] S4, using the pooling layer of the convolutional neural network 1DCNN, deep mining of feature maps is performed to extract deep feature information of deep foundation pit deformation data;

[0009] S5. The deep feature information is integrated through the fully connected layer of the convolutional neural network 1DCNN to obtain the feature vector for deep foundation pit deformation monitoring;

[0010] S6. Build a support vector machine (SVM) model, input the feature vector into the SVM model for training, use the Grey Wolf Optimization (GWO) algorithm to optimize the model's hyperparameters during the training process, and obtain the GWO-1DCNN-SVM combined prediction model. Use the obtained combined prediction model for intelligent monitoring and early warning.

[0011] Optionally, multi-source characteristic parameters include three categories: one is deformation characteristic parameters, covering the deformation of the top of the foundation pit retaining piles, the deep deformation of the soil, the deformation of the columns, the deformation of the pit bottom uplift, and the surrounding surface deformation; the second is mechanical characteristic parameters, covering the internal force of the foundation pit retaining wall, the internal force of the support, and the lateral earth pressure of the retaining wall; the third is technical characteristic parameters, covering the excavation depth of the foundation pit.

[0012] Optionally, in S2, the process of using the interpolation method to convert the original data of deep foundation pit deformation monitoring into an equidistant sequence to obtain more complete new data includes:

[0013] 1) Determine the data points of the original data of deep foundation pit deformation monitoring in the form of (x i ,y i ), where i = 1, 2, ..., n, and x i represents the time of the i-th data point, y i Represents the deformation value at the corresponding time;

[0014] 2) Calculate the slope m between adjacent data points i , quantifies the deformation rate between each pair of adjacent data points, and the formula is as follows:

[0015]

[0016] Among them, x i+1 -x i is the interval between two adjacent data points, y i+1 -y i Interpolate the deformation of two adjacent data points;

[0017] 3) For data points i=2,3,...,n-1, calculate the weighted slope t respectively i :

[0018]

[0019] Among them, |m i+1 -m i | and |m i-1 -m i-2 | is the absolute value of the slope change. For boundary points t1 and t n Directly use the slopes of adjacent points;

[0020] 4) For each interval [x i ,xi+1 ], based on the values ​​of the data points and the weighted slopes, construct a cubic interpolation polynomial P i (x):

[0021] P i (x) = P0 + P1 (x i )+P2(xx i ) 2 +P3(xx i ) 3

[0022] Where p0 = y i , p1=t i , p2 and p3 are coefficients obtained by considering the weighted slopes and actual values ​​of adjacent points;

[0023] 5) The cubic interpolation polynomials of all intervals are spliced ​​to obtain a complete interpolation curve as the new data.

[0024] Optionally, the formula for normalization processing in S2 is:

[0025]

[0026] Among them, Y norm is the normalized data, Y new is the new data, Y min and Y max are the minimum and maximum values ​​in the new data respectively.

[0027] Optionally, the process of performing a convolution operation on the normalized new data in S3 includes:

[0028] 1) Set the convolutional layer architecture of the convolutional neural network 1DCNN, including the convolution kernel size q and number r;

[0029] 2) Initialize multiple one-dimensional convolution kernels and randomly initialize the weight parameter K of each one-dimensional convolution kernel j and bias b j ;

[0030] 3) For the normalized new data Y norm ={y1,y2,...,y n} Perform convolution operation. Specifically, each one-dimensional convolution kernel slides each position of the normalized new data in turn, and performs convolution calculation to extract local features V j , the formula for convolution calculation is:

[0031]

[0032] Among them, V i,j represents the output of the j-th one-dimensional convolution kernel at position i, Kj,t is the weight of the j-th one-dimensional convolution kernel, Y norm,i+t is the value of the new data at position i+t, σ(·) is the activation function;

[0033] 4) After all one-dimensional convolution kernels are convolved, the local features V are obtained. j Get the feature map V:

[0034] V={V1,V2,...,V r}

[0035] Among them, V j = {V 1,j ,V 2,j ,...,V n-q+1,j} is the local feature generated by the jth convolution kernel.

[0036] Optionally, S4 specifically includes:

[0037] 1) Based on the maximum pooling method, set the size S and step size T of the pooling window;

[0038] 2) Map each local feature V in the feature map V j Input to the pooling layer and perform the maximum pooling operation:

[0039] P i,j =max(V i,j ,V i+1,j ,...,V i+s-1,j )

[0040] Among them, P i,j is the feature value after pooling;

[0041] 3) All are processed by maximum pooling to obtain deep feature information.

[0042] Optionally, S5 specifically includes:

[0043] 1) Flatten the deep feature information obtained after the maximum pooling process to obtain the flattened feature vector P flat ;

[0044] 2) The flattened eigenvector P flat Input into the fully connected layer, and obtain the feature vector F of deep foundation pit deformation monitoring through weighted summation and activation function processing fc :

[0045] F fc =σ(W fc ·P flat +b fc )

[0046] Among them, σ(·) is the activation function, W fc is the weight matrix of the fully connected layer, b fc is the bias of the fully connected layer.

[0047] Optionally, the kernel function of the support vector machine SVM model adopts the radial basis function K(x i ,x):

[0048] K(x i ,x)=exp(-γx i -x 2 )

[0049] Among them, γ is the kernel parameter, x i and x is the input vector.

[0050] Optionally, the process of optimizing the hyperparameters of the model using the Grey Wolf Optimization Algorithm GWO in S6 includes:

[0051] 1) Set the number of gray wolf populations and randomly initialize the position of each gray wolf, where each position represents a hyperparameter combination;

[0052] 2) Evaluate the position of each gray wolf and obtain the fitness value;

[0053] 3) Sort the gray wolves in the population from best to worst according to their fitness values, select the three gray wolves with the best fitness values ​​as α wolf, β wolf, and δ wolf, and the rest of the wolves as ω wolves. α wolf represents the optimal solution, and β wolf and δ wolf are suboptimal solutions;

[0054] 4) Use the positions of α wolf, β wolf, and δ wolf to guide the position of ω wolf, and update the position of ω wolf. The update formula is as follows:

[0055]

[0056] in, A represents the value of the position in the dth dimension after t+1 iterations. j and C j They are the absorption that affects the location update, is the position of the corresponding dimension of α wolf, β wolf, and δ wolf at the tth iteration;

[0057] 5) Return to the process of evaluating the position of each gray wolf and obtaining the fitness value. After multiple iterations and position updates, when the fitness value meets the convergence condition or reaches the maximum number of iterations, the current optimal hyperparameter combination is output.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] (1) The present invention constructs a comprehensive foundation pit intelligent monitoring and early warning indicator system by integrating multi-source characteristic parameters, including deformation characteristics, mechanical characteristics and technical characteristics. With the combination of one-dimensional convolutional neural network (1DCNN) and support vector machine (SVM), the present invention can deeply mine and extract complex characteristic information in foundation pit deformation monitoring data. Compared with the traditional prediction method that relies on a single feature or a simple model, it significantly improves the prediction accuracy of foundation pit deformation and ensures comprehensive monitoring and accurate early warning of the construction process.

[0060] (2) The present invention innovatively introduces the Grey Wolf Optimization (GWO) algorithm to optimize and adjust the hyperparameters of the model. By continuously iteratively optimizing the hyperparameters, the model can better adapt to complex and changeable geological conditions and construction environments, avoiding the defects of traditional methods that perform inconsistently under different conditions, thereby improving the applicability and reliability of the prediction model and enabling it to be widely used in actual engineering.

[0061] (3) The present invention utilizes on-site real-time monitoring data and combines it with the Akima interpolation method to perform equidistant sequence conversion to generate more complete and high-quality data input. With the help of feature extraction and optimization algorithms under the deep learning framework, the model can dynamically update and provide foundation pit deformation prediction and early warning information in real time. This real-time monitoring and early warning mechanism effectively reduces the errors caused by data discreteness and outliers, making the prediction results more reliable, providing a solid guarantee for construction safety, and helping to take timely measures to prevent potential risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0063] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0064] Figure 2 It is the flow chart of the algorithm principle of the present invention;

[0065] Figure 3 The loss iteration change curve of the combined model training loss function of the present invention;

[0066] Figure 4 It is the iterative change curve of the mean square error RMSE of the combined model training loss of the present invention;

[0067] Figure 5 It is the iterative change curve of the fitness value during the optimization process of the gray wolf optimization algorithm of the present invention;

[0068] Figure 6 This is a comparison chart of prediction results of the combined prediction model training set of the present invention;

[0069] Figure 7 This is a comparison chart of prediction results of the combined prediction model test set of the present invention;

[0070] Figure 8 It is a prediction error curve diagram of the test set of the combined prediction model of the present invention;

[0071] Fig. 9 A bar graph showing relative error of the combined prediction model of the present invention;

[0072] Fig.10 This is a training set effect diagram of the combined prediction model of the present invention;

[0073] Fig.11 This is a test set effect diagram of the combined prediction model of the present invention;

[0074] Fig.12 This is a sample fitting prediction effect diagram of the combined prediction model of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0076] See also Figure 1 and Figure 2 The embodiment of the present invention provides an intelligent monitoring and early warning method for deep foundation pit construction based on multi-source feature parameter fusion, comprising:

[0077] S1. Analyze the multi-source characteristic parameters that affect the deformation of the deep foundation pit during the construction period, and obtain the original monitoring data of the deep foundation pit deformation corresponding to the multi-source characteristic parameters during the construction process according to the real-time monitoring of the deep foundation pit project site;

[0078] S2. Use Akima interpolation method to convert the original data of deep foundation pit deformation monitoring into equal time interval series to obtain more complete new data, and normalize the obtained new data;

[0079] S3. Use the deep learning framework to build a one-dimensional convolutional neural network 1DCNN, and use the convolution layer of the convolutional neural network 1DCNN to perform a convolution operation on the normalized new data to obtain a feature map of the new data;

[0080] S4, using the pooling layer of the convolutional neural network 1DCNN, deep mining of feature maps is performed to extract deep feature information of deep foundation pit deformation data;

[0081] S5. The deep feature information is integrated through the fully connected layer of the convolutional neural network 1DCNN to obtain the feature vector for deep foundation pit deformation monitoring;

[0082] S6. Build a support vector machine (SVM) model, input the feature vector into the SVM model for training, use the Grey Wolf Optimization (GWO) algorithm to optimize the model's hyperparameters during the training process, and obtain the GWO-1DCNN-SVM combined prediction model. Use the obtained combined prediction model for intelligent monitoring and early warning.

[0083] In implementation, the feature vectors can be divided into training set and test set according to a certain ratio. The training set is used for model training, and the test set is used to verify the model effect. Figure 3-12 shown.

[0084] During implementation, multi-source characteristic parameters include three categories: one is deformation characteristic parameters, covering the deformation of the top of the foundation pit retaining piles, the deep deformation of the soil, the deformation of the columns, the deformation of the pit bottom uplift, and the surrounding surface deformation; the second is mechanical characteristic parameters, covering the internal force of the foundation pit retaining wall, the internal force of the support, and the lateral earth pressure of the retaining wall; the third is technical characteristic parameters, covering the excavation depth of the foundation pit.

[0085] In the implementation, the Akima interpolation method is used in S2 to convert the original data of deep foundation pit deformation monitoring into equidistant series to obtain more complete new data. The process includes:

[0086] 1) Determine the data points of the original data of deep foundation pit deformation monitoring in the form of (x i ,y i ), where i = 1, 2, ..., n, and x i represents the time of the i-th data point, y i Represents the deformation value at the corresponding time;

[0087] 2) Calculate the slope m between adjacent data points i , quantifies the deformation rate between each pair of adjacent data points, and the formula is as follows:

[0088]

[0089] Among them, x i+1 -x i is the interval between two adjacent data points, y i+1 -y i Interpolate the deformation of two adjacent data points;

[0090] 3) For data points i=2,3,...,n-1, calculate the weighted slope t respectively i :

[0091]

[0092] Among them, |m i+1 -m i | and |m i-1-m i-2 | is the absolute value of the slope change. For boundary points t1 and t n Directly use the slopes of adjacent points;

[0093] 4) For each interval [x i ,x i+1 ], based on the values ​​of the data points and the weighted slopes, construct a cubic interpolation polynomial P i (x):

[0094] P i (x) = P0 + P1 (x i )+P2(xx i ) 2 +P3(xx i ) 3

[0095] Where p0 = y i , p1=t i , p2 and p3 are coefficients obtained by considering the weighted slopes and actual values ​​of adjacent points to ensure that the interpolation curve is continuous and smooth at each data point;

[0096] 5) The cubic interpolation polynomials of all intervals are spliced ​​to obtain a complete interpolation curve as the new data.

[0097] Optionally, the formula for normalization processing in S2 is:

[0098]

[0099] Among them, Y norm is the normalized data, Y new is the new data, Y min and Y max are the minimum and maximum values ​​in the new data respectively.

[0100] In practice, the process of convolution operation on the normalized new data in S3 includes:

[0101] 1) Set the convolutional layer architecture of the convolutional neural network 1DCNN, including the convolution kernel size q and number r;

[0102] 2) Initialize multiple one-dimensional convolution kernels and randomly initialize the weight parameter K of each one-dimensional convolution kernel j and bias b j ;

[0103] 3) For the normalized new data Y norm ={y1,y2,…,y n} Perform convolution operation. Specifically, each one-dimensional convolution kernel slides each position of the normalized new data in turn, and performs convolution calculation to extract local features V j , the formula for convolution calculation is:

[0104]

[0105] Among them, V i,j represents the output of the j-th one-dimensional convolution kernel at position i, K j,t is the weight of the j-th one-dimensional convolution kernel, Y norm,i+t is the value of the new data at position i+t, σ(·) is the activation function;

[0106] 4) After all one-dimensional convolution kernels are convolved, the local features V are obtained. j Get the feature map V:

[0107] V={V1,V2,...,V r}

[0108] Among them, V j = {V 1,j ,V 2,j ,...,V n-q+1,j} is the local feature generated by the jth convolution kernel.

[0109] In implementation, S4 specifically includes:

[0110] 1) Based on the maximum pooling method, set the size S and step size T of the pooling window;

[0111] 2) Map each local feature V in the feature map V j Input to the pooling layer and perform the maximum pooling operation:

[0112] P i,j =max(V i,j ,V i+1,j ,...,V i+s-1,j )

[0113] Among them, P i,j is the feature value after pooling;

[0114] 3) After all the max pooling processes, deep feature information is obtained. These features have lower tails than the original feature maps, but retain important spatial structure information in the data.

[0115] In implementation, S5 specifically includes:

[0116] 1) Flatten the deep feature information obtained after the maximum pooling process to obtain the flattened feature vector P flat ;

[0117] 2) The flattened eigenvector P flat Input into the fully connected layer, and obtain the feature vector F of deep foundation pit deformation monitoring through weighted summation and activation function processing fc :

[0118] F fc =σ(W fc ·P flat +b fc )

[0119] Among them, σ(·) is the activation function, W fc is the weight matrix of the fully connected layer, b fc is the bias of the fully connected layer.

[0120] In the implementation, the kernel function of the support vector machine SVM model adopts the radial basis function K(x i ,x):

[0121] K(x i ,x)=exp(-γ x i -x 2 )

[0122] Among them, γ is the kernel parameter, x i and x is the input vector.

[0123] In implementation, the process of optimizing the hyperparameters of the model using the Grey Wolf Optimization Algorithm GWO in S6 includes:

[0124] 1) Set the number of gray wolf populations and randomly initialize the position of each gray wolf, where each position represents a hyperparameter combination;

[0125] 2) Evaluate the position of each gray wolf and obtain the fitness value;

[0126] 3) Sort the gray wolves in the population from best to worst according to their fitness values, select the three gray wolves with the best fitness values ​​as α wolf, β wolf, and δ wolf, and the rest of the wolves as ω wolves. α wolf represents the optimal solution, and β wolf and δ wolf are suboptimal solutions;

[0127] 4) Use the positions of α wolf, β wolf, and δ wolf to guide the position of ω wolf, and update the position of ω wolf. The update formula is as follows:

[0128]

[0129] in, A represents the value of the position in the dth dimension after t+1 iterations. j and C j They are the absorption that affects the location update, is the position of the corresponding dimension of α wolf, β wolf, and δ wolf at the tth iteration;

[0130] 5) Return to the process of evaluating the position of each gray wolf and obtaining the fitness value. After multiple iterations and position updates, when the fitness value meets the convergence condition or reaches the maximum number of iterations, the current optimal hyperparameter combination is output.

[0131] Compared with the prior art, the present invention has the following beneficial effects:

[0132] (1) The present invention constructs a comprehensive foundation pit intelligent monitoring and early warning indicator system by integrating multi-source characteristic parameters, including deformation characteristics, mechanical characteristics and technical characteristics. With the combination of one-dimensional convolutional neural network (1DCNN) and support vector machine (SVM), the present invention can deeply mine and extract complex characteristic information in foundation pit deformation monitoring data. Compared with the traditional prediction method that relies on a single feature or a simple model, it significantly improves the prediction accuracy of foundation pit deformation and ensures comprehensive monitoring and accurate early warning of the construction process.

[0133] (2) The present invention innovatively introduces the Grey Wolf Optimization (GWO) algorithm to optimize and adjust the hyperparameters of the model. By continuously iteratively optimizing the hyperparameters, the model can better adapt to complex and changeable geological conditions and construction environments, avoiding the defects of traditional methods that perform inconsistently under different conditions, thereby improving the applicability and reliability of the prediction model and enabling it to be widely used in actual engineering.

[0134] (3) The present invention utilizes on-site real-time monitoring data and combines it with the Akima interpolation method to perform equidistant sequence conversion to generate more complete and high-quality data input. With the help of feature extraction and optimization algorithms under the deep learning framework, the model can dynamically update and provide foundation pit deformation prediction and early warning information in real time. This real-time monitoring and early warning mechanism effectively reduces the errors caused by data discreteness and outliers, making the prediction results more reliable, providing a solid guarantee for construction safety, and helping to take timely measures to prevent potential risks.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. An intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion, characterized in that: include: S1. Analyze the multi-source characteristic parameters that affect the deformation of the deep foundation pit during the construction period, and obtain the original monitoring data of the deep foundation pit deformation corresponding to the multi-source characteristic parameters during the construction process according to the real-time monitoring of the deep foundation pit project site; S2. Use the interpolation method to convert the original data of deep foundation pit deformation monitoring into an equal time interval series to obtain more complete new data, and normalize the obtained new data; S3. Use the deep learning framework to build a one-dimensional convolutional neural network 1DCNN, and use the convolution layer of the convolutional neural network 1DCNN to perform a convolution operation on the normalized new data to obtain a feature map of the new data; S4, using the pooling layer of the convolutional neural network 1DCNN, deep mining of feature maps is performed to extract deep feature information of deep foundation pit deformation data; S5. The deep feature information is integrated through the fully connected layer of the convolutional neural network 1DCNN to obtain the feature vector for deep foundation pit deformation monitoring; S6. Build a support vector machine (SVM) model, input the feature vector into the SVM model for training, use the Grey Wolf Optimization (GWO) algorithm to optimize the model's hyperparameters during the training process, and obtain the GWO-1DCNN-SVM combined prediction model. Use the obtained combined prediction model for intelligent monitoring and early warning.

2. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 1 is characterized in that: Multi-source characteristic parameters include three categories: one is deformation characteristic parameters, covering the deformation of the top of the foundation pit retaining piles, the deep deformation of the soil, the deformation of the columns, the deformation of the pit bottom uplift, and the surrounding surface deformation; the second is mechanical characteristic parameters, covering the internal force of the foundation pit retaining wall, the internal force of the support, and the lateral earth pressure of the retaining wall; the third is technical characteristic parameters, covering the excavation depth of the foundation pit.

3. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 1 is characterized in that: In S2, the process of using the interpolation method to convert the original data of deep foundation pit deformation monitoring into an equidistant sequence to obtain more complete new data includes: 1) Determine the data points of the original data of deep foundation pit deformation monitoring in the form of (x i ,y i ), where i = 1, 2, ..., n, and x i represents the time of the i-th data point, y i Represents the deformation value at the corresponding time; 2) Calculate the slope m between adjacent data points i , quantifies the deformation rate between each pair of adjacent data points, and the formula is as follows: Among them, x i+1 -x i is the interval between two adjacent data points, y i+1 -y i Interpolate the deformation of two adjacent data points; 3) For data points i=2,3,...,n-1, calculate the weighted slope t respectively i : Among them, |m i+1 -m i | and |m i-1 -m i-2 | is the absolute value of the slope change. For boundary points t1 and t n Directly use the slopes of adjacent points; 4) For each interval [x i ,x i+1 ], based on the values ​​of the data points and the weighted slopes, construct a cubic interpolation polynomial P i (x): P i (x)=P0+P1(x-x i )+P2(x-x i ) 2 +P3(x-x i ) 3 Where p0 = y i , p1=t i , p2 and p3 are coefficients obtained by considering the weighted slopes and actual values ​​of adjacent points; 5) The cubic interpolation polynomials of all intervals are spliced ​​to obtain a complete interpolation curve as the new data.

4. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 3 is characterized in that: The formula for normalization in S2 is: Among them, Y norm is the normalized data, Y new is the new data, Y min and Y max are the minimum and maximum values ​​in the new data respectively.

5. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 4 is characterized in that: The process of performing convolution operations on normalized new data in S3 includes: 1) Set the convolutional layer architecture of the convolutional neural network 1DCNN, including the convolution kernel size q and number r; 2) Initialize multiple one-dimensional convolution kernels and randomly initialize the weight parameter K of each one-dimensional convolution kernel j and bias b j ; 3) For the normalized new data Y norm ={y1,y2,...,y n } Perform convolution operation. Specifically, each one-dimensional convolution kernel slides each position of the normalized new data in turn, and performs convolution calculation to extract local features V j , the formula for convolution calculation is: Among them, V i,j represents the output of the j-th one-dimensional convolution kernel at position i, K j,t is the weight of the j-th one-dimensional convolution kernel, Y norm,i+t is the value of the new data at position i+t, σ(·) is the activation function; 4) After all one-dimensional convolution kernels are convolved, the local features V are obtained. j Get the feature map V: V={V1,V2,...,V r } Among them, V j = {V 1,j ,V 2,j ,...,V n-q+1,j } is the local feature generated by the jth convolution kernel.

6. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 5, characterized in that: S4 specifically includes: 1) Based on the maximum pooling method, set the size S and step size T of the pooling window; 2) Map each local feature V in the feature map V j Input to the pooling layer and perform the maximum pooling operation: P i,j =max(V i,j ,V i+1,j ,...,V i+s-1,j ) Among them, P i,j is the feature value after pooling; 3) All are processed by maximum pooling to obtain deep feature information.

7. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 6 is characterized in that: S5 specifically includes: 1) Flatten the deep feature information obtained after the maximum pooling process to obtain the flattened feature vector P flat ; 2) The flattened eigenvector P flat Input into the fully connected layer, and obtain the feature vector F of deep foundation pit deformation monitoring through weighted summation and activation function processing fc : F fc =σ(W fc ·P flat +b fc ) Among them, σ(·) is the activation function, W fc is the weight matrix of the fully connected layer, b fc is the bias of the fully connected layer.

8. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 1, characterized in that: The kernel function of the support vector machine SVM model adopts the radial basis function K(x i ,x): K(x i ,x)=exp(-γx i -x 2 ) Among them, γ is the kernel parameter, x i and x is the input vector.

9. The intelligent monitoring and early warning method for deep foundation pit construction based on multi-source characteristic parameter fusion according to claim 1, characterized in that: The process of optimizing the model's hyperparameters using the Grey Wolf Optimization Algorithm (GWO) in S6 includes: 1) Set the number of gray wolf populations and randomly initialize the position of each gray wolf, where each position represents a hyperparameter combination; 2) Evaluate the position of each gray wolf and obtain the fitness value; 3) Sort the gray wolves in the population from best to worst according to their fitness values, select the three gray wolves with the best fitness values ​​as α wolf, β wolf, and δ wolf, and the rest of the wolves as ω wolves. α wolf represents the optimal solution, and β wolf and δ wolf are suboptimal solutions; 4) Use the positions of α wolf, β wolf, and δ wolf to guide the position of ω wolf, and update the position of ω wolf. The update formula is as follows: in, A represents the value of the position in the dth dimension after t+1 iterations. j and C j They are the absorption that affects the location update, is the position of the corresponding dimension of α wolf, β wolf, and δ wolf at the tth iteration; 5) Return to the process of evaluating the position of each gray wolf and obtaining the fitness value. After multiple iterations and position updates, when the fitness value meets the convergence condition or reaches the maximum number of iterations, the current optimal hyperparameter combination is output.

Images