Foundation pit deformation prediction method based on transformer model

Through the foundation pit deformation prediction method based on the Transformer model, the multi-scale feature extraction and early warning mechanism is used to solve the problem that the existing technology cannot predict foundation pit deformation in advance, and the prediction accuracy and construction safety are improved.

CN118940911APending Publication Date: 2024-11-12CHINA CONSTR EIGHT ENG DIV CORP LTD +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411142323.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing foundation pit monitoring data cannot predict the deformation trend of foundation pit in advance, cannot provide technical support for the optimization of construction plans, and poses potential safety risks.

Method used

The foundation pit deformation prediction method based on the Transformer model is adopted, and historical timing monitoring data is obtained, preprocessing and feature extraction is performed, a multi-scale Transformer model is built to predict the foundation pit deformation, and an early warning signal is generated based on the prediction data.

Benefits of technology

It significantly improves the accuracy of foundation pit deformation prediction, can be early warning, prevent and avoid the occurrence of foundation pit disasters, and improves construction safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940911A_ABST
    Figure CN118940911A_ABST
Patent Text Reader

Abstract

The invention discloses a foundation pit deformation prediction method based on a transform model, and the method comprises the steps: collecting foundation pit monitoring data, and carrying out the preprocessing of the data; a multi-scale Transform model is established to extract features in collected data, prediction of foundation pit deformation monitoring data is realized through training the model, and early warning judgment is performed according to predicted data. Compared with a previous time sequence model, the prediction precision is greatly improved, and occurrence of foundation pit disasters is prevented and avoided to a certain extent. The problem that the deformation trend of the foundation pit cannot be predicted in advance according to the actual monitoring data of the existing foundation pit is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of building construction, and in particular to a foundation pit deformation prediction method based on a transformer model. Background Art

[0002] The construction process of the foundation pit will be affected by a variety of external forces, coupled with the combined influence of other uncertain factors, often resulting in complex underground deformation and settlement linearity. There are many potential unsafe factors in the construction process, so it is necessary to monitor and promptly feedback the changes around the foundation pit to ensure the safety of the construction. However, the actual monitoring data of the existing foundation pit can only reflect the current deformation of the foundation pit, and cannot be predicted in advance, and cannot provide technical support for the optimization of the construction plan.

[0003] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0004] In order to overcome the defects of the existing technology, a foundation pit deformation prediction method based on the transformer model is provided to solve the problem that the actual monitoring data of the existing foundation pit cannot predict the deformation trend of the foundation pit in advance.

[0005] To achieve the above purpose, a foundation pit deformation prediction method based on a transformer model is provided, comprising the following steps:

[0006] Acquire historical time series foundation pit monitoring data to construct a monitoring data set, wherein the monitoring data set includes a training data set and a test data set;

[0007] A transformer model is constructed, wherein the transformer model performs multiple scale segmentation on the input training data set to obtain multiple block sets of different blocks, wherein the block set generates an intra-block feature set by transforming the intra-block attention feature sets of all blocks, wherein after the dimension transformation, the block set performs a channel rearrangement operation and merges different time spans of the same block to perform attention calculation to obtain an extra-block feature set, wherein the intra-block feature set and the extra-block feature set are aggregated to obtain the features of the monitoring data set, and the predicted value of the foundation pit is output based on the features of the monitoring data set;

[0008] Calculating the error between the predicted value of the foundation pit and the measured detection value of the foundation pit to adjust the training parameters of the transformer model, and ending the training of the transformer model when the number of iterations reaches a specified number or the error meets the design requirements;

[0009] Predicting future time series using the trained transformer model to obtain prediction data;

[0010] Based on the comparison between the preset foundation pit warning value and the predicted data, when the predicted data exceeds the warning value, a warning signal is generated and sent to the outside.

[0011] Furthermore, before implementing the transformer model to perform multiple scale segmentation on the input training data set to obtain multiple block sets of different blocks, the foundation pit monitoring data is preprocessed, and the preprocessing includes missing value processing, outlier processing, normalization processing, and de-unitization.

[0012] Furthermore, the foundation pit monitoring data includes horizontal displacement, vertical displacement, and deep horizontal displacement.

[0013] Furthermore, the foundation pit monitoring data is divided into the training data set and the test data set in a ratio of 8:2.

[0014] The beneficial effect of the present invention is that the foundation pit deformation prediction method based on the transformer model of the present invention collects foundation pit monitoring data and pre-processes the data; builds a multi-scale Transformer model to extract features from the collected data, and through training the model, realizes the prediction of the foundation pit deformation monitoring data, and makes early warning judgments based on the predicted data. Compared with the previous time series model, the prediction accuracy is greatly improved, and the occurrence of foundation pit disasters is prevented and avoided to a certain extent. The foundation pit deformation prediction method based on the transformer model of the present invention improves the model's feature extraction capability for time series data by adopting the transformer model, and realizes the automatic prediction of foundation pit deformation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0016] Figure 1 It is a flow chart of a method for predicting foundation pit deformation based on a transformer model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0018] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] Reference Figure 1 As shown, the present invention provides a foundation pit deformation prediction method based on a transformer model, comprising the following steps:

[0020] S1. Obtain historical time series foundation pit monitoring data to construct a monitoring data set, which includes a training data set and a test data set.

[0021] The foundation pit monitoring data includes horizontal displacement, vertical displacement and deep horizontal displacement.

[0022] Specifically, the foundation pit monitoring data are divided into training data set and test data set in a ratio of 8:2.

[0023] S2. Build a transformer model. The transformer model performs multiple-scale segmentation on the input training data set to obtain multiple block sets of different blocks. The block set generates an intra-block feature set by transforming the intra-block attention feature set of all blocks through dimensionality transformation. After the block set is transformed through dimensionality transformation, the channel rearrangement operation is performed and different time spans of the same block are merged to perform attention calculation to obtain an extra-block feature set. The intra-block feature set is aggregated with the extra-block feature set to obtain the features of the monitoring data set. Based on the features of the monitoring data set, the predicted value of the foundation pit is output.

[0024] Before implementing the transformer model to perform multiple scale segmentation on the input training data set to obtain multiple block sets of different blocks, the foundation pit monitoring data is preprocessed, and the preprocessing includes missing value processing, outlier processing, normalization processing, and de-unitization.

[0025] Specifically, preprocess each data of the current monitoring time series T. The preprocessed monitoring data of the T time series are accumulated into the existing monitoring data set X.

[0026] The monitoring data set X covers all the time series data of foundation pit monitoring from time series 1 to time series T, as shown in Formula 1. T-1 These are all historical time series monitoring data.

[0027] X=*x1,x2…x T + (1).

[0028] Perform instance normalization on the monitoring dataset X:

[0029] Instance normalization is to normalize the data in the same channel to ensure the independence of each instance. The specific calculation is shown in Formula 2.i -x i Expectation, Var,x i -x i The variance of ∈ is a very small number, usually set to 1e-05. γ, β are training parameters.

[0030]

[0031] In this embodiment, see Figure 1 ,The transformer model includes: a multi-scale attention calculation module F, a bidirectional attention calculation module FD, and a predictor. The bidirectional attention calculation module FD includes an intra-block attention calculation module and an extra-block attention calculation module.

[0032] After normalizing the instance Input into the multi-scale attention calculation module F of the transformer model for feature learning. The calculation of the multi-scale attention calculation module F mainly includes the following three steps:

[0033] (1) Segment the input into multiple scales.

[0034] The block set P is obtained by splitting N blocks of different sizes, and the input The dimension is T*D, where D is the feature dimension. The size value set of N different blocks is S=*s1,s2…s N +, the calculation of the block set P is shown in formula 3, Contains S time spans.

[0035]

[0036] (2) The block set P is input into the bidirectional attention calculation module FD for calculation.

[0037] The bidirectional attention calculation module FD includes two sub-modules, namely the intra-block attention calculation module and the extra-block attention calculation module.

[0038] The calculation of the attention calculation module within the block is shown in Figure 4. First, The dimension is transformed through the embedding operation. The dimension is Through The transformation operation is performed to obtain L, where L represents the transformation operation. Represents the intra-block attention calculation result of the i-th block. The intra-block attention feature set of all blocks generates the intra-block feature set Feat intra , C is the feature aggregation operation.

[0039]

[0040]

[0041] The out-of-block attention calculation module mainly obtains the overall global relationship by calculating the relationship between different blocks. Specifically, as shown in Formula 5, the block set P first performs an embedding operation to transform the dimension, and then merges different time spans of the same block through the Re channel rearrangement operation to ensure that the subsequent attention calculation is for different blocks.

[0042] After transformation The dimension is Q inter , K inter 、V inter Through The transformation operation is performed, and L represents the transformation operation. The out-of-block feature set Feat is obtained by attention calculation. inter .

[0043]

[0044] (3) The feature Feat in the block intra and the out-of-block feature Feat inter Aggregation is performed to obtain the final overall feature Feat.

[0045] As shown in Formula 6, Feat inter First, rearrange the channels and transform them into dimensions Then perform dimension transformation to obtain Finally with Feat intra Add them together to get the final features.

[0046]

[0047] The feature Feat is input again into the adaptive multi-scale attention calculation module F for calculation. This process can be repeated n times. After the cycle calculation is completed, the feature feat is obtained.

[0048] The feature feat is input into the predictor H to obtain the model output result. The predictor can be a simple combination of a fully connected layer or a convolutional layer. The model output result is the observation value result XF of the predicted future time series.

[0049] XF=*x T+1 ,x T+2 …x T+l + = H (feat)

[0050] (4) Model training steps include:

[0051] Divide the accumulated monitoring data into training and testing sets;

[0052] Input data of different time series lengths into the multi-scale Transformer model M to calculate the model's predicted value;

[0053] Calculate the error between the predicted value and the actual monitored value, adjust the model training and update the model parameters;

[0054] When the number of iterations reaches the specified number or the error meets the conditions, the training ends;

[0055] Verify the accuracy of the trained model on the test set.

[0056] S3. Calculate the error between the predicted value of the foundation pit and the measured detection value of the foundation pit to adjust the training parameters of the transformer model. When the number of iterations reaches the specified number or the error meets the design requirements, the training of the transformer model is terminated.

[0057] S4. Use the trained transformer model to predict future time series and obtain prediction data.

[0058] S5. Based on the comparison between the preset foundation pit warning value and the predicted data, when the predicted data exceeds the warning value, a warning signal is generated and sent to the outside.

[0059] The transformer-based foundation pit deformation prediction method of the present invention collects foundation pit monitoring data and pre-processes the data; builds a multi-scale transformer model to extract features from the collected data, and predicts the foundation pit deformation monitoring data through training the model, and makes early warning judgments based on the predicted data. Compared with the previous time series model, the prediction accuracy is greatly improved, and the occurrence of foundation pit disasters can be prevented and avoided to a certain extent.

[0060] The transformer model-based foundation pit deformation prediction method of the present invention improves the model's feature extraction capability for time series data by adopting the transformer model, thereby realizing automatic prediction of foundation pit deformation.

[0061] The foundation pit deformation prediction method based on the transformer model of the present invention proposes a multi-scale mechanism based on the transformer model, and an intra-block attention and an extra-block attention calculation module, which can enhance the model's ability to extract feature relationships in the latent space, better obtain the relationship between input and output, and improve the accuracy of model prediction to a certain extent.

[0062] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A foundation pit deformation prediction method based on transformer model, characterized in that: The following steps are involved: Acquire historical time series foundation pit monitoring data to construct a monitoring data set, wherein the monitoring data set includes a training data set and a test data set; A transformer model is constructed, wherein the transformer model performs multiple scale segmentation on the input training data set to obtain multiple block sets of different blocks, wherein the block set generates an intra-block feature set by transforming the intra-block attention feature sets of all blocks, wherein after the dimension transformation, the block set performs a channel rearrangement operation and merges different time spans of the same block to perform attention calculation to obtain an extra-block feature set, wherein the intra-block feature set and the extra-block feature set are aggregated to obtain the features of the monitoring data set, and the predicted value of the foundation pit is output based on the features of the monitoring data set; Calculating the error between the predicted value of the foundation pit and the measured detection value of the foundation pit to adjust the training parameters of the transformer model, and ending the training of the transformer model when the number of iterations reaches a specified number or the error meets the design requirements; Predicting future time series using the trained transformer model to obtain prediction data; Based on the comparison between the preset foundation pit warning value and the predicted data, when the predicted data exceeds the warning value, a warning signal is generated and sent to the outside.

2. The foundation pit deformation prediction method based on the transformer model according to claim 1 is characterized in that: Before implementing the transformer model to perform multiple scale segmentation on the input training data set to obtain multiple block sets of different blocks, the foundation pit monitoring data is preprocessed, and the preprocessing includes missing value processing, outlier processing, normalization processing, and de-unitization.

3. The foundation pit deformation prediction method based on transformer model according to claim 2 is characterized in that: The foundation pit monitoring data includes horizontal displacement, vertical displacement, and deep horizontal displacement.

4. The foundation pit deformation prediction method based on transformer model according to claim 1 is characterized in that: The foundation pit monitoring data is divided into the training data set and the test data set in a ratio of 8:2.

Citation Information

Cited By

  • Dynamic risk prediction and optimization decision-making method and system for deep foundation pit construction process

    CN120974895A