Foundation pit pipeline settlement deformation prediction method based on heterogeneous spatio-temporal data source

Abstract

The invention discloses a method for predicting settlement deformation of a foundation pit pipeline based on a heterogeneous spatio-temporal data source. The method comprises the following steps: S1, preprocessing data; s2, spatial feature extraction: introducing a graph attention network, and extracting mutual influence weights among the monitoring points by using a distribution graph of the monitoring points around the foundation pit and a pipeline distribution graph to obtain the mutual influence weights among the monitoring points; s3, spatio-temporal feature fusion; s4, heterogeneous data processing; s5, time sequence prediction; by fusing the spatial features and the time sequence features of the foundation pit monitoring points, the prediction accuracy is improved in predicting the settlement deformation problem of the foundation pit pipeline; missing data is repaired by adopting a diffusion model, so that the utilization rate of the whole data set is improved; the blank of data missing is made up, the original information of the foundation pit monitoring data is reserved to the maximum extent, and the robustness and reliability of the deformation prediction model are further enhanced; the limitation of pipeline data can be effectively supplemented by utilizing multi-source heterogeneous data, and the overall prediction effect is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of foundation pit engineering, and in particular relates to a method for predicting settlement deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources. Background Art

[0002] Early methods for predicting the settlement deformation of foundation pit pipelines mainly relied on theoretical formula method, numerical simulation method and empirical coefficient method. The theoretical formula method is usually based on elastic-plastic theory and predicts deformation through approximate solutions; the numerical simulation method uses mathematical models to analyze the deformation law during foundation pit construction; the empirical coefficient law relies on historical data and engineering experience and uses empirical formulas to predict deformation. However, these methods are often too idealistic, greatly influenced by subjective experience, and ignore the complexity and actual deformation of foundation pit projects, so their prediction results are not ideal.

[0003] With the development of foundation pit engineering, traditional prediction methods have gradually shifted to statistical methods based on time series. Common prediction methods for foundation pit pipeline settlement deformation include the autoregressive moving average (ARMA) model and the GM (1,1) prediction model in the grey system theory. These methods analyze the linear relationship between the time series and its historical values ​​and make predictions in combination with random error terms. However, such methods have obvious limitations: on the one hand, they usually only focus on a single variable, fail to consider the combined effects of multiple factors in foundation pit deformation, and lack a comprehensive description of the entire foundation pit characteristics; on the other hand, traditional methods are mainly suitable for short-term predictions, and their prediction accuracy will gradually decrease as the prediction period increases.

[0004] In recent years, with the development of machine learning and deep learning technology, foundation pit deformation prediction has gradually adopted time series prediction models such as support vector machine (SVM), random forest (RF), back propagation neural network (BP) and long short-term memory network (LSTM). These methods are significantly higher than traditional methods in deformation prediction accuracy, and the prediction results have a good match with the actual monitoring data. However, for the huge foundation pit monitoring data, the above methods have failed to take into account the global nature of foundation pit prediction. They only use a few points as features to predict one point in the local part of the project, while ignoring the mutual connection between other points and the differences between different types of monitoring points; and the model's ability in long-term dependency learning still needs to be improved. For complex long-term time series data, many models have limited capabilities and cannot effectively capture the long-term dependency relationship between data; a large amount of data will also distract the model's attention and fail to focus on key data, resulting in poor prediction results.

[0005] To this end, the present invention proposes a method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for predicting the settlement deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources, the specific steps of which include: S1: data preprocessing:

[0008] For the foundation pit data whose missing values ​​of monitoring data of a monitoring point are less than 60% of the total data length, the data will be entered into the diffusion model to fill the missing values; then, the input time series data will be normalized;

[0009] S2: Extraction of spatial features:

[0010] The graph attention network is introduced to extract the mutual influence weights between the monitoring points using the distribution map of the monitoring points around the foundation pit and the distribution map of the pipelines, and the mutual influence weights between each monitoring point are obtained.

[0011] S3: Spatiotemporal feature fusion:

[0012] The spatiotemporal fusion coding GTE is used. The spatiotemporal fusion coding GTE uses the extracted spatial weights to fuse the time series data of each monitoring point with the time series data information of other monitoring points in the space.

[0013] S4: Heterogeneous data processing:

[0014] By dividing the multi-source heterogeneous data into pipeline deformation monitoring data and non-pipeline monitoring data;

[0015] S5: Timing Prediction:

[0016] By taking pipeline deformation data as endogenous variables, the internally fused features are segmented into non-overlapping segments of the same length to form patches of the same size with temporal and spatial characteristics. The temporal and spatial characteristics of pipeline deformation are obtained in the model through self-attention mechanism, cross-attention mechanism and feedforward neural network. Finally, the fused pipeline deformation features are passed through the fully connected layer to obtain the predicted value of pipeline deformation.

[0017] Preferably, in S1, a Z-score standardization method is used to standardize the time series data into a form having a standard normal distribution.

[0018] Preferably, in S3, the fused features include both temporal features and spatial features of the foundation pit.

[0019] Preferably, in S4, pipeline deformation data is used as an endogenous variable in the prediction model to focus on the impact of self-generated changes in time series through the self-attention mechanism in the Transformer architecture; non-pipeline deformation data is used as an exogenous variable in the prediction model to interact the impact of exogenous variables on endogenous variables through the cross-attention mechanism, so as to achieve auxiliary prediction of exogenous variables on endogenous variables.

[0020] Preferably, the multi-source heterogeneous data includes, in addition to pipeline deformation monitoring data, surface settlement profile monitoring data, support axial force monitoring data, and enclosure structure lateral displacement monitoring data.

[0021] Preferably, in S1, the diffusion model can fill in missing data with complete data of the same type, thereby avoiding wasting valuable data and improving the accuracy of the final prediction.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The advantages of adopting the prediction method of the present invention include: the accuracy of pipeline deformation prediction is significantly improved: by integrating the spatial characteristics and temporal characteristics of the foundation pit monitoring points, the present invention can significantly improve the accuracy of prediction in predicting the settlement and deformation of foundation pit pipelines.

[0024] Introducing foundation pit spatial features to enhance model interpretability: This technology extracts the spatial relationship features between the monitoring points in the foundation pit, identifies and quantifies the mutual influence weights between the monitoring points. On this basis, the spatial weights are fused with the temporal features of the monitoring points to enhance the accuracy of the prediction model. In addition, the introduction of spatial features not only improves the prediction accuracy, but also greatly improves the interpretability of the model, so that the prediction results can more clearly reflect the interaction between the monitoring points in the foundation pit.

[0025] Improved data utilization: During the construction process, due to sensor failure or damage, monitoring data is often missing, affecting the effectiveness of foundation pit deformation prediction; to solve this problem, the present invention uses an advanced diffusion model to repair missing data, significantly improving the utilization of the overall data set; this innovative method not only fills the gap in data missing, but also retains the original information of the foundation pit monitoring data to the maximum extent, further enhancing the robustness and reliability of the deformation prediction model.

[0026] Utilization of multi-source heterogeneous data: Another innovation of this invention is to make full use of multi-source heterogeneous data; in addition to pipeline deformation monitoring data, other relevant data sources in foundation pit engineering are also taken into consideration, such as surface settlement profile monitoring data, support axial force monitoring data, and lateral displacement monitoring data of retaining structures; these heterogeneous data sources provide more abundant information for pipeline deformation prediction, which can effectively supplement the limitations of pipeline data and thus improve the overall prediction effect. By integrating multimodal data, the prediction and generalization capabilities of the model are further enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] See also Figure 1 The present invention provides a technical solution: a method for predicting the settlement deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources, and the specific steps include: S1: data preprocessing:

[0030] The missing values ​​of the foundation pit data are repaired; for the data of a monitoring point whose missing values ​​are less than 60% of the total data length, the data is entered into the diffusion model for missing value repair; the diffusion model can fill the missing data with complete data of the same type, thereby avoiding the waste of valuable data and improving the accuracy of the final prediction. Then, the input time series data is normalized, and the Z score standardization method is used to standardize the time series data into a form with a standard normal distribution;

[0031] S2: Extraction of spatial features:

[0032] By introducing the graph attention network and using the distribution map of monitoring points around the foundation pit and the distribution map of pipelines, the weights of the mutual influence between the monitoring points are extracted, and the weights of the mutual influence between each monitoring point are obtained;

[0033] S3: Spatiotemporal feature fusion:

[0034] The spatiotemporal fusion coding GTE is used. The spatiotemporal fusion coding GTE uses the extracted spatial weight to fuse the time series data of each monitoring point with the time series data information of other monitoring points in the space, so that the fused features include both the time series features and the spatial features of the foundation pit.

[0035] S4: Heterogeneous data processing:

[0036] By dividing multi-source heterogeneous data into pipeline deformation monitoring data and non-pipeline monitoring data; pipeline deformation data is used as an endogenous variable in the prediction model to focus on the impact of self-generated changes in time series through the self-attention mechanism in the Transformer architecture; non-pipeline deformation data is used as an exogenous variable in the prediction model to interact the impact of exogenous variables on endogenous variables through the cross-attention mechanism, so as to achieve auxiliary prediction of exogenous variables on endogenous variables;

[0037] S5: Timing Prediction:

[0038] By taking pipeline deformation data as endogenous variables, the internally fused features are divided into non-overlapping and equal-length segments to form patches of the same size with temporal and spatial characteristics; the spatiotemporal characteristics of pipeline deformation are obtained through self-attention mechanism, cross-attention mechanism and feedforward neural network in the model; the final fused pipeline deformation features are passed through the fully connected layer to obtain the predicted value of pipeline deformation; the model can obtain higher pipeline prediction accuracy under the condition of missing data and the characteristics of heterogeneous data sources, taking into account the spatial characteristic factors in foundation pit engineering, and guarantee better prediction effects under short-term and long-term conditions.

[0039] In this embodiment, preferably, the multi-source heterogeneous data includes not only pipeline deformation monitoring data, but also surface settlement profile monitoring data, support axial force monitoring data, and enclosure structure lateral displacement monitoring data.

[0040] Although the embodiments of the present invention have been shown and described, as detailed above, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources, characterized by: The specific steps include: S1: Data preprocessing: For the foundation pit data whose missing values ​​of monitoring data of a monitoring point are less than 60% of the total data length, the data will be entered into the diffusion model to fill the missing values; then, the input time series data will be normalized; S2: Extraction of spatial features: The graph attention network is introduced to extract the mutual influence weights between the monitoring points using the distribution map of the monitoring points around the foundation pit and the distribution map of the pipelines, and the mutual influence weights between each monitoring point are obtained. S3: Spatiotemporal feature fusion: The spatiotemporal fusion coding GTE is used. The spatiotemporal fusion coding GTE uses the extracted spatial weights to fuse the time series data of each monitoring point with the time series data information of other monitoring points in the space. S4: Heterogeneous data processing: By dividing the multi-source heterogeneous data into pipeline deformation monitoring data and non-pipeline monitoring data; S5: Timing Prediction: By taking pipeline deformation data as endogenous variables, the internally fused features are segmented into non-overlapping segments of the same length to form patches of the same size with temporal and spatial characteristics. The temporal and spatial characteristics of pipeline deformation are obtained in the model through self-attention mechanism, cross-attention mechanism and feedforward neural network. Finally, the fused pipeline deformation features are passed through the fully connected layer to obtain the predicted value of pipeline deformation.

2. The method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources according to claim 1 is characterized by: In S1, the Z-score standardization method is used to standardize the time series data into a form with a standard normal distribution.

3. The method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources according to claim 1 is characterized by: In S3, the fused features include both temporal features and spatial features of the foundation pit.

4. The method for predicting the settlement deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources according to claim 1 is characterized by: In S4, pipeline deformation data is used as an endogenous variable in the prediction model to focus on the impact of self-generated changes in time series through the self-attention mechanism in the Transformer architecture; non-pipeline deformation data is used as an exogenous variable in the prediction model to interact the impact of exogenous variables on endogenous variables through the cross-attention mechanism, so as to achieve auxiliary prediction of exogenous variables on endogenous variables.

5. The method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources according to claim 1 is characterized by: The multi-source heterogeneous data includes not only pipeline deformation monitoring data, but also surface settlement profile monitoring data, support axial force monitoring data, and enclosure structure lateral displacement monitoring data.

6. The method for predicting the settlement and deformation of foundation pit pipelines based on heterogeneous spatiotemporal data sources according to claim 1 is characterized by: In S1, the diffusion model can fill in missing data with complete data of the same type, thereby avoiding wasting valuable data and improving the accuracy of the final prediction.

Images