Foundation pit deformation prediction method based on GCN-LSTM coupling model

Abstract

The invention relates to a foundation pit deformation prediction method based on a GCN-LSTM coupling model, and relates to the technical field of foundation pit deformation prediction. According to the foundation pit deformation prediction method based on the GCN-LSTM coupling model, the temperature and humidity sensors are arranged to collect the temperature and humidity meteorological data of the foundation pit, richer feature input can be provided for a subsequent prediction model, and the prediction accuracy and reliability are improved. According to the method, the meteorological characteristic data and the foundation pit deformation data are combined, so that the model can be helped to better capture the complex nonlinear relationship between the displacement of the foundation pit and the environmental factors, and the adaptability of the model can be enhanced. By extracting the long-term displacement trend term and combining the long-term displacement trend term with the short-term dynamic change, the method can make up for the deficiency of the long-term trend recognition capability of the model, ensures that the goodness of fit between the prediction result and the actual deformation process is higher, and more accurately reflects the deformation state of the foundation pit.

Description

Technical Field

[0001] This application relates to the technical field of foundation pit deformation prediction, and particularly to a foundation pit deformation prediction method based on a GCN-LSTM coupling model. Background Art

[0002] During the construction of foundation pit excavation, the relative position changes of the foundation pit and its surrounding soil and structures may have an adverse impact on the foundations of surrounding buildings and construction safety. Therefore, analyzing and predicting the deformation of the foundation pit is an important link in ensuring construction safety.

[0003] Currently, the implementation solutions in the field of foundation pit deformation prediction in the prior art mainly focus on predicting foundation pit deformation based on numerical simulation.

[0004] However: For traditional numerical simulation methods, their computational complexity is very high. Especially when considering large-scale complex projects, the calculation time is long and a large amount of computing resources are required, which makes it difficult to apply in engineering sites with high real-time requirements. And traditional numerical simulation methods usually ignore the spatial dependence between sensors and cannot effectively model the spatial characteristics of foundation pit deformation, which leads to certain limitations in processing multi-point sensor data and is difficult to meet the requirements of modern foundation pit projects, unable to accurately predict the deformation of the foundation pit and effectively ensure the safety of foundation pit construction. Summary of the Invention

[0005] To solve or partially solve the problems existing in the related art, this application provides a foundation pit deformation prediction method based on a GCN-LSTM coupling model, which can accurately predict the deformation of the foundation pit and ensure the safety of foundation pit construction.

[0006] This application provides a foundation pit deformation prediction method based on a GCN-LSTM coupling model, and this foundation pit deformation prediction method includes:

[0007] S1: Perform data preprocessing operations on the deformation data and meteorological characteristic data of the foundation pit to be measured collected by the working condition acquisition device, where the working condition acquisition device includes: a plurality of displacement sensors and temperature and humidity sensors;

[0008] S2: Construct a training data set according to the preprocessed deformation data and meteorological characteristic data, and train a spatio-temporal feature prediction model based on the GCN-LSTM coupling model according to this training data set. Among them, for the spatio-temporal feature prediction model based on the GCN-LSTM coupling model, the spatial features between sensor data are extracted through the GCN model, and the dynamic features in the time series data are extracted through the LSTM model;

[0009] S3: Collect the real-time deformation data and real-time meteorological characteristic data of the foundation pit to be measured through the working condition acquisition device, and input them into the spatio-temporal feature prediction model based on the GCN-LSTM coupled model that has been trained to obtain the preliminary predicted deformation data of the foundation pit;

[0010] S4: Add a deformation trend term to the preliminary predicted deformation data and perform anti-normalization processing to obtain the final predicted deformation data of the foundation pit. When the final predicted deformation data is greater than the preset safety threshold, output a warning result that there is a safety risk in the foundation pit to be measured. Among them, the deformation trend term is obtained by processing the deformation data of the foundation pit through the polynomial regression method.

[0011] Optionally, in some embodiments of the present application:

[0012] The data preprocessing operation on the deformation data and meteorological characteristic data of the foundation pit to be measured collected by the working condition acquisition device specifically includes:

[0013] Perform linear interpolation processing on the deformation data and meteorological characteristic data to fill the deformation data and meteorological characteristic data;

[0014] Perform normalization processing on the filled meteorological characteristic data and displacement data.

[0015] Optionally, in some embodiments of the present application:

[0016] The construction of the spatio-temporal feature prediction model based on the GCN-LSTM coupled model specifically includes:

[0017] Construction of graph convolution: Take each sensor layout point as a node of the graph, adopt an edge-weighted version of the graph model, and use the correlation of displacement changes between two sensor layout points as the weight of the edge between them to construct a correlation graph;

[0018] Perform correlation test on the collected data to measure the correlation of displacements between two sensor layout points, and construct the adjacency matrix of the correlation graph;

[0019] Construct a convolutional layer based on the adjacency matrix of the correlation graph;

[0020] Take the graph convolutional layer as a computing unit to construct a GCN-LSTM network.

[0021] Optionally, in some embodiments of the present application:

[0022] The correlation test on the collected data specifically includes:

[0023] Through the Pearson correlation test method, measure the strength of the correlation between the input parameter and the output parameter. The specific calculation expression is as follows:

[0024]

[0025] When there are pulse interferences or severe monotonic non - linear distortions in the collected data, through the Spearman correlation test method, the specific calculation expression is as follows:

[0026]

[0027] In the formula: Di is the difference in the rank values of the i - th data pair.

[0028] Optionally, in some embodiments of the present application:

[0029] The calculation expression of the GCN - LSTM network is specifically as follows:

[0030]

[0031] h ι = o ι ⊙tanh(c ι )

[0032]

[0033] Among them, ht - 1 and Xt are respectively the hidden state of the network at the (t - 1) - th step and the input value at the t - th step. [h t-1 , X t means concatenating ht - 1 and Xt. means inputting [h t-1 , X t into the graph convolutional layer for convolutional operation. ⊙ represents the Hadamard product, that is, element - wise product operation. ct - 1 and ct are respectively the cell states at the (t - 1) - th step and the t - th step; f t , i t , g t , o t are respectively the values of the forget gate, input gate, candidate memory state, and output gate at the t - th step; W i , W f , W g , W o and b i , b f , b g , b o are respectively the weights and biases of the forget gate, input gate, candidate memory state, and output gate; σ and tanh are activation functions.

[0034] Optionally, in some embodiments of the present application:

[0035] Constructing a training dataset based on the pre - processed deformed data and meteorological feature data specifically includes:

[0036] All sensor layout points share the same meteorological feature data. The normalized meteorological feature data and displacement data are vector spliced to obtain a feature vector, which is specifically expressed as:

[0037]

[0038] where x i，t represents the feature vector of layout point i at time t, is the numerical value of various meteorological features at time t, F w is the total number of meteorological feature categories, y i，t is the displacement data of site i at time t.

[0039] Optionally, in some embodiments of the present application:

[0040] The training of the spatio-temporal feature prediction model based on the GCN-LSTM coupled model specifically includes:

[0041] Set the time step of GCN-LSTM as the hyperparameter S, and then construct the feature matrix input into the GCN-LSTM network;

[0042] Train the GCN-LSTM network in a supervised learning manner: the label corresponding to the feature matrix Xt is the displacement data vector of each layout point at time t + 1.

[0043] Optionally, in some embodiments of the present application:

[0044] Adding a deformation trend term to the preliminary predicted deformation data and performing anti-normalization processing specifically includes: after the GCN-LSTM network is trained, input the feature matrix Xt at time t, and the predicted value yt+1 at time t + 1 can be output. At the same time, add a trend term to each component of yt+1 and perform anti-normalization to obtain the displacement prediction value. The specific formula is:

[0045] y i,t+1 ′=(y i，t+1 +tr t+1 )×(max(y i )-min(y i ))+min(y i )

[0046] where tr t+1 is the trend term value of layout point i on the day at time t + 1, and max(y i ) and min(y i ) are respectively the maximum and minimum values of the historical displacement data of layout point i.

[0047] Optionally, in some embodiments of the present application:

[0048] The working condition acquisition device further includes: a hub and a 4G signal wireless data transmission terminal module (DTU module). Among them, the hub is used to centrally manage the power supply and data connection of sensor devices, and the DTU module sends data to the database in real time through a 4G wireless network, providing the monitoring information required for subsequent data processing and prediction.

[0049] Optionally, in some embodiments of the present application:

[0050] The working condition acquisition device further includes: a strain gauge and an accelerometer. The data collected by the strain gauge and the accelerometer are transmitted through a 4G wireless network;

[0051] Based on various foundation pit monitoring data in the database, by constructing a virtual model of the foundation pit project based on digital twin technology, including the soil characteristics, formation parameters, construction stages, and displacement distribution information of the foundation pit, to simulate and monitor the deformation process of the foundation pit and its related influencing factors;

[0052] Use various foundation pit monitoring data in the database for visual display and decision support on the user interface, providing foundation pit deformation trend analysis, real-time warning, and construction safety risk assessment for on-site management personnel.

[0053] The technical solution provided by the present application may include the following beneficial effects:

[0054] By setting displacement sensors in the present application, the deformation trend of the foundation pit at different construction stages can be monitored through the displacement sensors; by setting temperature and humidity sensors to collect the temperature and humidity meteorological data of the foundation pit, it can provide richer feature inputs for the subsequent prediction model, improving the accuracy and reliability of the prediction.

[0055] By constructing a GCN-LSTM network in the present application, the spatial features between sensors can be extracted through GCN to capture the spatial correlation of foundation pit deformation, and the dynamic features in the time series data can be extracted through LSTM to capture the time series change law in the foundation pit deformation process. In addition, by combining bidirectional LSTM to enhance the understanding of long-term and short-term dependencies, and by using a multi-attention mechanism to weight the importance of different time steps and sensor data, the prediction accuracy of the model can be effectively improved.

[0056] By combining meteorological feature data with foundation pit deformation data in the present application, a more comprehensive and accurate prediction model can be established, which can help the model better capture the complex non-linear relationship between foundation pit displacement and environmental factors, enhance the adaptability of the model, and the model can automatically adjust the prediction of foundation pit displacement according to real-time meteorological information, making it maintain high reliability and adaptability under various meteorological conditions.

[0057] In this application, by extracting the long-term displacement trend term and combining it with the short-term dynamic changes, the model can more stably predict the future deformation trend, avoid prediction errors caused by random noise, enhance the prediction stability of the model. The displacement trend term can reflect the global characteristics of the foundation pit deformation, which can help construction personnel more intuitively understand the main driving factors of the foundation pit deformation process, and can help the model more effectively adapt to diverse data distribution characteristics, optimizing the generalization ability of the model. By adding the displacement trend term, this application can make up for the deficiency of the model in identifying long-term trends, ensure a higher degree of agreement between the prediction results and the actual deformation process, and more accurately reflect the deformation state of the foundation pit.

[0058] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Brief Description of the Drawings

[0059] By describing the exemplary embodiments of this application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of this application will become more apparent. Among them, in the exemplary embodiments of this application, the same reference numerals generally represent the same components.

[0060] Figure 1 It is a schematic structural diagram of a foundation pit deformation prediction method based on a GCN-LSTM coupled model in an embodiment of this application;

[0061] Figure 2 It is a schematic construction process diagram of a spatio-temporal feature prediction model based on a GCN-LSTM coupled model in an embodiment of this application;

[0062] Figure 3 It is a schematic prediction process diagram of a spatio-temporal feature prediction model based on a GCN-LSTM coupled model in an embodiment of this application. Detailed Embodiments

[0063] The embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0064] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, the meaning of "a plurality" is two or more unless otherwise specifically defined.

[0065] In the description of this application, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing this application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to this application.

[0066] Unless otherwise clearly specified and defined, the terms "installed", "connected", "connected to", "fixed", etc. should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0067] Currently, the implementation solutions in the field of foundation pit deformation prediction in the prior art mainly focus on predicting foundation pit deformation based on numerical simulation.

[0068] However: for traditional numerical simulation methods, their computational complexity is very high. Especially when considering large-scale complex projects, the calculation time is long and a large amount of computing resources are required, which makes it difficult to apply in engineering sites with high real-time requirements. And traditional numerical simulation methods usually ignore the spatial dependence between sensors and cannot effectively model the spatial characteristics of foundation pit deformation, which leads to certain limitations in processing multi-point sensor data, making it difficult to meet the requirements of modern foundation pit projects, unable to accurately predict foundation pit deformation, and unable to effectively ensure the safety of foundation pit construction.

[0069] In view of the above problems, the embodiments of this application provide a foundation pit deformation prediction method based on a GCN-LSTM coupling model, which can accurately predict foundation pit deformation and ensure the safety of foundation pit construction.

[0070] The technical solutions of the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0071] Figure 1 is a schematic structural diagram of a foundation pit deformation prediction method based on a GCN-LSTM coupled model in an embodiment of the present application;

[0072] Figure 2 is a schematic construction process diagram of a spatio-temporal feature prediction model based on a GCN-LSTM coupled model in an embodiment of the present application;

[0073] Figure 3 is a schematic prediction process diagram of a spatio-temporal feature prediction model based on a GCN-LSTM coupled model in an embodiment of the present application.

[0074] See Figures 1-3 , the present application provides a foundation pit deformation prediction method based on a GCN-LSTM coupled model, and the foundation pit deformation prediction method includes:

[0075] S1: Perform data preprocessing operations on the deformation data and meteorological feature data of the foundation pit to be measured collected by the working condition acquisition device, where the working condition acquisition device includes: a plurality of displacement sensors and temperature and humidity sensors.

[0076] In this embodiment, the displacement sensors are arranged according to the size of the foundation pit to ensure that the data collected by the displacement sensors can cover the entire foundation pit area, and the deformation trend of the foundation pit at different construction stages is monitored through the displacement sensors.

[0077] During the construction of the foundation pit, environmental parameters such as temperature and humidity will affect the physical and mechanical properties of the soil around the foundation pit. For example, changes in soil moisture content may cause a decrease in soil strength or swelling and shrinkage, and precipitation may lead to instability of the foundation pit slope, etc. By setting temperature and humidity sensors to collect the temperature and humidity meteorological data of the foundation pit, further understanding of the external influencing factors of foundation pit deformation can provide richer feature inputs for the subsequent prediction model and improve the accuracy and reliability of the prediction.

[0078] Specifically, the data preprocessing operations performed on the deformation data and meteorological feature data of the foundation pit to be measured collected by the working condition acquisition device specifically include:

[0079] Perform linear interpolation processing on the deformation data and meteorological feature data to fill the deformation data and meteorological feature data. Specifically: The missing data is processed by combining the monitoring value at the previous moment and the monitoring value at the next moment, and the specific calculation method is as follows:

[0080] x t = a 1 x t-1 + a 2 xt+1

[0081] Wherein, \(x_t\) is the monitored value at a certain moment; \(a_1\) and \(a_2\) are the weights of the corresponding data, both taking 0.5.

[0082] Normalize the filled meteorological feature data and displacement data. Specifically: perform Min - Max normalization on the filled meteorological feature data and displacement data, and the specific calculation method is as follows:

[0083]

[0084] Where \(x\) is the original data, \(x'\) is the value after normalization, \(x_{min}\) and \(x_{max}\) are the minimum and maximum values in the original data, and the normalized data is mapped to the interval [0, 1].

[0085] In this embodiment, in the monitoring process of the working condition acquisition device, there may be a situation of data loss. By using the linear interpolation method for filling and complementing, the missing values in the data can be effectively filled, ensuring the integrity and consistency of the data. At the same time, by normalizing the filled deformation data and meteorological feature data respectively, the data of different sensors can be within the same range, thereby improving the training efficiency of the model.

[0086] S2: Construct a training dataset according to the pre - processed deformation data and meteorological feature data, and train a spatio - temporal feature prediction model based on the GCN - LSTM coupled model according to this training dataset. Among them, for the spatio - temporal feature prediction model based on the GCN - LSTM coupled model, the spatial features between sensor data are extracted through the GCN model, and the dynamic features in the time - series data are extracted through the LSTM model.

[0087] Specifically, the construction of the spatio - temporal feature prediction model based on the GCN - LSTM coupled model specifically includes:

[0088] Construction of graph convolution: Take each sensor layout point as a node of the graph, adopt an edge - weighted version of the graph model, and use the correlation of displacement changes between two sensor layout points as the weight of the edge between them to construct a correlation graph.

[0089] In this embodiment, by taking the sensor layout points as the nodes of the graph and constructing an edge - weighted correlation graph, the spatial relationship between sensors can be effectively modeled, effectively improving the model's understanding of spatial relationships to improve the prediction accuracy.

[0090] Conduct a correlation test on the collected data, measure the correlation of displacements between two sensor layout points, and construct the adjacency matrix of the correlation graph; specifically:

[0091] By using the Pearson correlation test method, the strength of the correlation between the input parameters and the output parameters is measured. The specific calculation expression is as follows:

[0092]

[0093] Among them, the value range of the Pearson correlation coefficient is [-1, 1]. The closer its absolute value is to 1, the stronger the correlation between the two sets of data.

[0094] When there are pulse interferences or severe monotonic non-linear distortions in the collected data, the Spearman correlation test method is used. The specific calculation expression is as follows:

[0095]

[0096] In the formula: Di is the difference in the rank values of the i-th data pair.

[0097] The Pearson correlation coefficient is used to measure the correlation of displacements between two sensor layout points, and the adjacency matrix of the correlation graph is constructed. The specific calculation formula is as follows:

[0098]

[0099] Among them, N represents the number of sensor layout points.

[0100] In this embodiment, through the Pearson correlation test method, the correlation of displacements between two sensor layout points can be effectively evaluated, and the obtained correlation value is used to construct the adjacency matrix, thereby forming a relationship model between sensors, which can provide support for subsequent graph convolution analysis.

[0101] Based on the adjacency matrix of the correlation graph, a convolutional layer is constructed; specifically, based on the adjacency matrix of the correlation graph, a convolutional layer is constructed. The specific calculation formula is as follows:

[0102]

[0103] Among them, D is a diagonal matrix, and the diagonal elements represent the matrix obtained by taking the negative half power of each element in D; X (l) and X (l+1) are the matrices of the input and output of the l-th graph convolutional layer respectively. The matrix output by the l-th graph convolutional layer is the input of the (l + 1)-th graph convolutional layer; is a learnable parameter matrix, where F and F' are selected according to the input dimension requirements of the l-th and (l + 1)-th layers.

[0104] In this embodiment, by constructing a convolutional layer based on the adjacency matrix of the correlation graph, the feature extraction ability and expression ability of the model on graph data can be improved, the robustness can be enhanced, and the computational complexity can be reduced.

[0105] Construct a GCN-LSTM network with the graph convolutional layer as the computational unit. Specifically, the computational expression of this GCN-LSTM network is as follows:

[0106]

[0107] where \(h_{t - 1}\) and \(X_t\) are the hidden state of the network at the \((t - 1)\)-th step and the input value at the \(t\)-th step respectively, \([h_{t - 1},X_t]\) represents concatenating \(h_{t - 1}\) and \(X_t\), t―1 ,X t means inputting \([h_{t - 1},X_t]\) into the graph convolutional layer for convolution operation, ⊙ represents the Hadamard product, that is, element-wise product operation. \(c_{t - 1}\) and \(c_t\) are the cell states at the \((t - 1)\)-th step and \(t\)-th step respectively; \(f_t\), represents inputting \([h_{t - 1},X_t]\) into the graph convolutional layer for convolution operation, ⊙ represents the Hadamard product, that is, element-wise product operation. \(c_{t - 1}\) and \(c_t\) are the cell states at the \((t - 1)\)-th step and \(t\)-th step respectively; \(f_t\), t―1 ,X t into the graph convolutional layer for convolution operation, ⊙ represents the Hadamard product, that is, element-wise product operation. \(c_{t - 1}\) and \(c_t\) are the cell states at the \((t - 1)\)-th step and \(t\)-th step respectively; \(f_t\), \(i_t\), t ,i t ,g t ,o t are the values of the forget gate, input gate, candidate memory state, and output gate at the \(t\)-th step respectively; \(W_f\), i ,W f ,W g ,W o and \(b_f\), i ,b f ,b g ,b o are the weights and biases of the forget gate, input gate, candidate memory state, and output gate respectively; σ and tanh are activation functions.

[0108] In this embodiment, the constructed GCN-LSTM network can extract the spatial features between sensors through GCN, capture the spatial correlation of foundation pit deformation, and can extract the dynamic features in the time series data through LSTM, capture the time series change law in the process of foundation pit deformation. In addition, by combining bidirectional LSTM to enhance the understanding of long-term and short-term dependencies, and by using a multi-attention mechanism to weight the importance of different time steps and sensor data, the prediction accuracy of the model can be effectively improved.

[0109] Specifically, a training data set is constructed according to the preprocessed deformation data and meteorological feature data, which specifically includes:

[0110] All sensor layout points share the same meteorological feature data, and the normalized meteorological feature data and displacement data are subjected to vector concatenation processing to obtain a feature vector, which is specifically expressed as:

[0111]

[0112] Among them, x i，t represents the feature vector of the layout point i at time t, is the numerical value of various meteorological features at time t, and F w is the total number of meteorological feature categories, and y i，t is the displacement data of site i at time t.

[0113] In this embodiment, for the prediction of foundation pit deformation, it is necessary to prepare feature data for each sensor layout point for model training. However, for the situation where meteorological data cannot be collected separately for each sensor layout point, by sharing the same meteorological feature data for the sensor layout points, the complexity and cost of data collection can be reduced, and a feature dataset for model training can be effectively constructed. At the same time, by combining the meteorological feature data with the foundation pit deformation data, a more comprehensive and accurate prediction model can be established. As an additional input variable, the meteorological factor can help the model better capture the complex non-linear relationship between the foundation pit displacement and environmental factors, thereby significantly improving the prediction accuracy of the foundation pit displacement; at the same time, different meteorological conditions will cause changes in the mechanical environment where the foundation pit is located, and adding meteorological feature data can enhance the model adaptability. After adding the meteorological feature data, the model can automatically adjust the prediction of the foundation pit displacement according to the real-time meteorological information, so that it can maintain high reliability and adaptability under various meteorological conditions.

[0114] Specifically, the training of the spatio-temporal feature prediction model based on the GCN-LSTM coupling model specifically includes:

[0115] Set the time step of GCN-LSTM as the hyperparameter S, and then construct the feature matrix input into the GCN-LSTM network; where the feature matrix of the GCN-LSTM network is specifically expressed as:

[0116] X t =(X 1 , X 2 , …, X S )

[0117] Among them, X i is the sub-matrix composed of the feature vectors of all N layout points at time t+1-i, and X i =(x 1，t+1-i T , x 2，t+1-i T , …, x N,t+1--i T ) T When X t is input into the GCN-LSTM network, it is actually based on X 1 , X 2…, X s Input step by step in the order of

[0118] Train the GCN-LSTM network in a supervised learning manner: The label corresponding to the feature matrix Xt is the displacement data vector of each layout point at time t+1, specifically expressed as:

[0119] y t+1 =(y 1，t+1 , y 2，t+1 , …, y N，t+1 ) T

[0120] where yi,t+1 is the displacement data of layout point i at time t+1.

[0121] In this embodiment, through the training of the spatio-temporal feature prediction model based on the GCN-LSTM coupling model, the model can not only learn the change trend of the time series, but also understand the influence of the spatial structure, so as to achieve more efficient spatio-temporal feature prediction. By inputting the real-time data monitored in the foundation pit, the deformation prediction value of the foundation pit deformation can be output.

[0122] S3: Collect the real-time deformation data and real-time meteorological feature data of the foundation pit to be measured through the working condition acquisition device, and input them into the trained spatio-temporal feature prediction model based on the GCN-LSTM coupling model to obtain the preliminary predicted deformation data of the foundation pit.

[0123] S4: Add a deformation trend item to the preliminary predicted deformation data and perform anti-normalization processing to obtain the final predicted deformation data of the foundation pit. When the final predicted deformation data is greater than the preset safety threshold, output a warning result that there is a safety risk in the foundation pit to be measured, where the deformation trend item is obtained by polynomial regression method based on the deformation data of the foundation pit.

[0124] Specifically, adding a deformation trend item to the preliminary predicted deformation data and performing anti-normalization processing specifically includes:

[0125] After the GCN-LSTM network is trained, input the feature matrix Xt at time t, and the predicted value yt+1 at time t+1 can be output. At the same time, add a trend item to each component of yt+1 and perform anti-normalization to obtain the displacement predicted value. The specific formula is:

[0126] y i,t+1 '=(y i,t+1 +tr t+1 )×(max(y i )-min(y i ))+min(y i )

[0127] where tr t+1 is the trend item value of the layout point i on the day at time t + 1, and max(y i ) and min(y i ) are the maximum and minimum values of the historical displacement data of the layout point i, respectively.

[0128] In this embodiment, the foundation pit deformation data not only includes instantaneous changes, but is also affected by long-term change trends. Directly using the monitoring data cannot accurately describe the global characteristics of the displacement field. Therefore, by extracting the long-term displacement trend item and combining it with short-term dynamic changes, the model can more stably predict future deformation trends, avoid prediction errors caused by random noise, enhance the prediction stability of the model. At the same time, the displacement trend item can reflect the global characteristics of the foundation pit deformation, reflect the evolution law of the long-term displacement field distribution, and can help construction personnel more intuitively understand the main driving factors of the foundation pit deformation process. In addition, in different working conditions and scenarios, the addition of the displacement trend item can help the model more effectively adapt to diverse data distribution characteristics and optimize the generalization ability of the model. In summary, by adding the displacement trend item, the deficiency of the model in identifying long-term trends can be compensated, the consistency between the prediction result and the actual deformation process can be ensured to be higher, and the deformation state of the foundation pit can be reflected more accurately.

[0129] In this embodiment, the stability of the foundation pit is evaluated by combining the predicted foundation pit displacement field data with the set safety threshold. If the prediction result shows that the foundation pit deformation exceeds the safety threshold, the system will automatically trigger an early warning mechanism and send a safety warning message to the construction site personnel to prompt the possible deformation risk area. The construction personnel can adjust the on-site operation according to the warning message to avoid potential accidents.

[0130] Specifically, the working condition acquisition device further includes: a hub and a 4G signal wireless data transmission terminal module (DTU module). Among them, the hub is mainly used to centrally manage the power supply and data connection of the sensor devices, and the DTU module sends the data to the database in real time through the 4G wireless network to provide the monitoring information required for subsequent data processing and prediction. Among them:

[0131] The working condition acquisition device transmits the collected monitoring data through the DTU module, sends the monitoring data packet to the receiving port for preprocessing. The DTU module is based on the TCP protocol to provide a reliable transmission service to ensure that the transmitted data is error-free, not lost, not repeated, and arrives in order;

[0132] After the foundation pit displacement monitoring data is transmitted to the server via TCP, the developed ServerListener program is used to implement functions such as server port listening, message data parsing, and data storage in the database. This program is developed based on.NET Framework 4.5.2, uses the C# programming language, and has a console application as the output type. Specifically:

[0133] Create a socket through the Socket() function, bind the IP address and port number of the server with the Bind() function, and receive data through the EndReceiveFrom() function;

[0134] Based on the basic structure of the message data, parse the message data through functions such as ToString() and ToArray(), and temporarily store the parsed data in a custom function;

[0135] Create a connection with the MySQL database through the MySqlConnection() function, and upload data to the MySQL database with the MySqlCommand() function.

[0136] Specifically, the working condition acquisition device also includes: a strain gauge and an accelerometer, and the data collected by the strain gauge and the accelerometer are transmitted via a 4G wireless network;

[0137] Based on various types of foundation pit monitoring data in the database, construct a virtual model of the foundation pit project based on digital twin technology, including the soil characteristics, formation parameters, construction stage, and displacement distribution information of the foundation pit, to simulate and monitor the foundation pit deformation process and its related influencing factors;

[0138] Use various types of foundation pit monitoring data in the database for visual display and decision support on the user interface, so as to provide foundation pit deformation trend analysis, real-time warning, and construction safety risk assessment for on-site management personnel.

[0139] Among them, the system of the virtual model of the foundation pit project based on digital twin technology adopts a front-end and back-end separated development mode. The front end uses the ElementUI open-source framework, and the back end uses the ThinkPHP open-source framework, and is developed in combination with programming languages such as Vue, PHP, and Python. The system uses MySQL as the database to store structured, semi-structured, and unstructured data. The server is based on the Linux operating system, and the Web server is configured through Apache.

[0140] The technical solution provided by the embodiment of this application has the following beneficial effects:

[0141] By arranging displacement sensors in this application, the deformation trend of the foundation pit at different construction stages can be monitored through the displacement sensors; by arranging temperature and humidity sensors to collect the temperature and humidity meteorological data of the foundation pit, richer characteristic inputs can be provided for the subsequent prediction model, improving the accuracy and reliability of the prediction.

[0142] By constructing a GCN-LSTM network in this application, spatial features between sensors can be extracted through GCN to capture the spatial correlation of foundation pit deformation, and dynamic features in time series data can be extracted through LSTM to capture the time series variation law during the foundation pit deformation process. In addition, by combining bidirectional LSTM to enhance the understanding of long-term and short-term dependencies and using a multi-attention mechanism to weight the importance of different time steps and sensor data, the prediction accuracy of the model can be effectively improved.

[0143] By combining meteorological characteristic data with foundation pit deformation data in this application, a more comprehensive and accurate prediction model can be established, which can help the model better capture the complex non-linear relationship between foundation pit displacement and environmental factors, enhance the adaptability of the model, and the model can automatically adjust the prediction of foundation pit displacement according to real-time meteorological information, making it maintain high reliability and adaptability under various meteorological conditions.

[0144] By extracting the long-term displacement trend term and combining it with short-term dynamic changes in this application, the model can more stably predict the future deformation trend, avoid prediction errors caused by random noise, enhance the prediction stability of the model. The displacement trend term can reflect the global characteristics of foundation pit deformation, help construction personnel more intuitively understand the main driving factors of the foundation pit deformation process, and help the model more effectively adapt to diverse data distribution characteristics, optimizing the generalization ability of the model. By adding the displacement trend term in this application, the deficiency of the model in identifying long-term trends can be made up, ensuring a higher degree of coincidence between the prediction result and the actual deformation process, and more accurately reflecting the deformation state of the foundation pit.

[0145] Finally, it should also be noted that in this article, relationships such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "including", "comprising" or any other variant are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0146] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0147] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0148] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application.

[0149] The above has described the embodiments of the present application. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, the actual application, or the improvement of the technology in the market, or to enable other ordinary technical personnel in the technical field to understand the embodiments disclosed herein.

Claims

1. A foundation pit deformation prediction method based on the GCN-LSTM coupling model, characterized by: The foundation pit deformation prediction method comprises: S1: performing data preprocessing operations on deformation data and meteorological characteristic data of the foundation pit to be tested collected by a working condition collection device, wherein the working condition collection device includes: a plurality of displacement sensors and temperature and humidity sensors; S2: constructing a training data set according to the preprocessed deformation data and meteorological feature data, and training a spatiotemporal feature prediction model based on the GCN-LSTM coupling model according to the training data set, wherein the spatiotemporal feature prediction model based on the GCN-LSTM coupling model extracts spatial features between sensor data through the GCN model and extracts dynamic features in time series data through the LSTM model; S3: The real-time deformation data and real-time meteorological characteristic data of the foundation pit to be tested are collected by the working condition collection device, and input into the trained spatiotemporal characteristic prediction model based on the GCN-LSTM coupling model to obtain preliminary predicted deformation data of the foundation pit; S4: Add a deformation trend item to the preliminary predicted deformation data and perform a denormalization process to obtain the final predicted deformation data of the foundation pit. When the final predicted deformation data is greater than a preset safety threshold, output a warning result indicating that the foundation pit to be tested has a safety risk, wherein the deformation trend item is obtained by performing a polynomial regression method on the deformation data of the foundation pit.

2. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 1 is characterized in that: The deformation data and meteorological characteristic data of the foundation pit to be measured collected by the working condition collection device are subjected to data preprocessing operations, specifically including: Performing linear interpolation processing on the deformation data and the meteorological characteristic data to fill the deformation data and the meteorological characteristic data; The filled meteorological characteristic data and displacement data are normalized.

3. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 2 is characterized in that: The construction of the spatiotemporal feature prediction model based on the GCN-LSTM coupling model specifically includes: Construction of graph convolution: Each sensor deployment point is used as a node of the graph, and the edge-weighted version of the graph model is adopted. The correlation of the displacement changes of two sensor deployment points is used as the weight of the edge between them to construct a correlation graph; the collected data is tested for correlation, the correlation of the displacement between the two sensor deployment points is measured, and the adjacency matrix of the correlation graph is constructed; Construct convolutional layers based on the adjacency matrix of the correlation graph; The graph convolution layer is used as the computing unit to construct the GCN-LSTM network.

4. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 3 is characterized in that: The correlation test of the collected data specifically includes: The Pearson correlation test method is used to measure the correlation between the input parameters and the output parameters. The specific calculation expression is as follows: When the collected data has pulse interference or severe monotonic nonlinear distortion, the Spearman correlation test method is used. The specific calculation expression is as follows: Where: Di is the difference in the rank values ​​of the i-th data pair.

5. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 4 is characterized in that: The calculation expression of the GCN-LSTM network is as follows: c ι =f ι ⊙c ι―1 +i ι ⊙g ι h ι =o ι ⊙tanh(c ι ) Among them, h t-1 and X t are the hidden state of the network at step t-1 and the input value at step t, respectively. t―1 ,X t ] means to set h t-1 and X t Splicing, Indicates that [h t―1 ,X t ] Input graph convolution layer for convolution operation, ⊙ represents Hadamard product, that is, element-wise product operation, c t-1 and ct are the cell states at step t-1 and step t, respectively; f t ,i t ,g t ,o t are the values ​​of the forget gate, input gate, candidate memory state, and output gate at step t respectively; i ,W f ,W g ,W o and b i ,b f ,b g ,b o are the weights and biases of the forget gate, input gate, candidate memory state, and output gate respectively; σ and tanh are activation functions.

6. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 5 is characterized in that: The construction of a training data set based on the pre-processed deformation data and meteorological characteristic data specifically includes: All sensor deployment points share the same meteorological characteristic data, and the normalized meteorological characteristic data and displacement data are vector-joined to obtain the characteristic vector, which is specifically expressed as: Among them, x i，t represents the characteristic vector of the layout point i at time t, is the value of various meteorological characteristics at time t, F w is the total number of meteorological characteristic categories, y i，t is the displacement data of site i at time t.

7. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 6 is characterized in that: The training of the spatiotemporal feature prediction model based on the GCN-LSTM coupling model specifically includes: Set the time step of GCN-LSTM to the hyperparameter S, and then construct the feature matrix of the input GCN-LSTM network; The GCN-LSTM network is trained in a supervised learning manner: the label (1abel) corresponding to the feature matrix Xt is the displacement data vector of each layout point at time t+1.

8. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to any one of claims 1 to 7, characterized in that: The adding of deformation trend items to the preliminary predicted deformation data and performing denormalization processing specifically includes: After the GCN-LSTM network is trained, the feature matrix Xt at time t is input, and the predicted value yt+1 at time t+1 can be output. At the same time, trend terms are added to each component of yt+1, and denormalized to be used as the displacement prediction value. The specific formula is: and i,t+1 ′=(and i,t+1 +tr t+1 )×(max(and i )―min(and i ))+min(y i ) Among them, tr t+1 is the trend item value of the layout point i at time t+1, max(y i ) and min(y i ) are the maximum and minimum values ​​of the historical displacement data of the layout point i, respectively.

9. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 8 is characterized in that: The working condition collection device also includes: a hub and a 4G signal wireless data transmission terminal module (DTU module), wherein the hub is used to centrally manage the power supply and data connection of the sensor equipment, and the DTU module sends data to the database in real time via the 4G wireless network, providing the required monitoring information for subsequent data processing and prediction.

10. The foundation pit deformation prediction method based on the GCN-LSTM coupling model according to claim 9 is characterized in that: The working condition acquisition device also includes: a strain gauge and an accelerometer, and the data collected by the strain gauge and the accelerometer are transmitted via a 4G wireless network; Based on various foundation pit monitoring data in the database, a virtual model of the foundation pit engineering based on digital twin technology is constructed, including the soil characteristics, formation parameters, construction stage, and displacement distribution information of the foundation pit, so as to simulate and monitor the deformation process of the foundation pit and its related influencing factors; Various types of foundation pit monitoring data in the database are used for visual display and decision support in the user interface to provide on-site managers with foundation pit deformation trend analysis, real-time warning, and construction safety risk assessment.

Images