CN120822099A - A deep foundation pit deformation prediction method and system based on neural network and rough set classification - Google Patents
A deep foundation pit deformation prediction method and system based on neural network and rough set classificationInfo
- Publication number
- CN120822099A CN120822099A CN202510940255.4A CN202510940255A CN120822099A CN 120822099 A CN120822099 A CN 120822099A CN 202510940255 A CN202510940255 A CN 202510940255A CN 120822099 A CN120822099 A CN 120822099A
- Authority
- CN
- China
- Prior art keywords
- data
- attribute
- foundation pit
- deep foundation
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明涉及基坑检测技术领域，且公开了一种基于神经网络与粗糙集分类的深基坑变形预测方法及系统，包括以下步骤：数据采集与预处理；基于粗糙集理论的属性约简；基于改进的注意力机制循环神经网络(A‑RNN)的模型构建与训练；可信度验证与动态反馈调整；输出结果与预警响应。该发明通过集成位移、土压力、地下水位、支护应力等多参数监测数据，全面捕捉深基坑变形的驱动因素，减少单一传感器数据偏差导致的误判，双向LSTM与注意力机制的结合，有效捕捉时序数据中的长期依赖和关键时间点的局部特征，显著提升变形预测的准确性。
The present invention relates to the field of foundation pit detection technology and discloses a deep foundation pit deformation prediction method and system based on neural networks and rough set classification. The method comprises the following steps: data acquisition and preprocessing; attribute reduction based on rough set theory; model construction and training based on an improved attention mechanism recurrent neural network (A-RNN); credibility verification and dynamic feedback adjustment; and output results and early warning response. By integrating multi-parameter monitoring data such as displacement, soil pressure, groundwater level, and support stress, the invention comprehensively captures the driving factors of deep foundation pit deformation, reduces misjudgments caused by single sensor data deviations, and effectively captures long-term dependencies in time series data and local features at key time points. The combination of a bidirectional LSTM and an attention mechanism significantly improves the accuracy of deformation prediction.
Description
Technical Field
The invention relates to the technical field of foundation pit detection, in particular to a deep foundation pit deformation prediction method and a deep foundation pit deformation prediction system based on neural network and rough set classification,
Background
Along with the acceleration of the urban process, the deep foundation pit engineering is widely applied to the fields of high-rise buildings, subway construction and the like. However, deformation problems caused by complicated geological conditions, dynamic environment changes and other factors frequently occur in the foundation pit construction process, and structural instability and even collapse accidents can be caused. The current deep foundation pit deformation prediction technology still has the following limitations.
1. Insufficient data dimension and reliability
The traditional monitoring method relies on single parameter monitoring such as displacement, stress and the like, omits multi-factor coupling effects such as soil pressure, groundwater level, supporting stress and the like, and easily causes omission of key driving factors. The single sensor data deviation is easy to cause misjudgment, and influences the comprehensiveness and accuracy of the prediction result.
2. Time sequence modeling capability is weak
The existing prediction model (such as the traditional RNN) is difficult to capture the long-term time sequence dependency relationship in the deep foundation pit deformation process, and the local feature recognition capability of the key time node is insufficient. The high computational complexity brought by the complex network structure makes it difficult to adapt to the real-time requirements of the on-site edge computing devices.
3. Model generalization and dynamic adaptation defects
The static model is easy to generate false alarm due to overfitting or abnormal data under the long-term complex working condition, and lacks on-line adjustment capability for dynamic change of the construction environment. The fixed risk threshold setting mode is difficult to adapt to different geological conditions, so that the early warning sensitivity is insufficient.
4. Redundancy feature and resource consumption problem
The high-dimensional monitoring data contains a large amount of redundant information, and the traditional method does not effectively reduce the dimension, so that the training efficiency of the neural network is low, the memory occupation is excessive, and the deployment feasibility of the embedded system is restricted.
Aiming at the problems, a deep foundation pit deformation prediction method integrating multisource data perception, efficient time sequence modeling and dynamic self-adaption capability is urgently needed, and real-time requirements of an engineering site are met while prediction accuracy is improved.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a deep foundation pit deformation prediction method and a deep foundation pit deformation prediction system based on neural network and rough set classification,
The invention provides a deep foundation pit deformation prediction method based on neural network and rough set classification, which comprises the following steps:
Step S1, data acquisition and preprocessing
Acquiring displacement, soil pressure, underground water level and support structure stress data of a deep foundation pit in real time through a multisource sensor array, and constructing a dynamic monitoring data set D raw;
step S2, attribute reduction based on rough set theory
Constructing a decision table S= (U, A, V, f) from the preprocessed data, wherein U is a domain, namely a set of all samples, A=C U-D, C is a conditional attribute set and comprises monitoring data attributes such as displacement, soil pressure and the like, D is a decision attribute set such as deformation state (normal, early warning, danger and the like) of a deep foundation pit, V is a set of attribute values, and f is a U×A-V is an information function and is used for determining the attribute value of each sample;
Calculating attribute importance, namely adopting an attribute importance calculating method based on information entropy, setting H (D) as information entropy of decision attribute D, setting H (D|C- { a i) as conditional information entropy of decision attribute D after removing attribute a i, and setting importance Sig (a i) of attribute a i as Sig (a i)＝H(D|C-{ai) -H (D|C) wherein P (D j) is the probability that the decision attribute D takes on the value D j; p (C i) is the probability that the conditional attribute set C takes the value of C i, and p (d j|ci) is the conditional probability that the decision attribute takes the value of d j when the conditional attribute takes the value of C i;
step S3, building and training a model of a recurrent neural network (A-RNN) based on an improved attentiveness mechanism
An improved A-RNN network structure is composed of input layer for receiving data set D red with reduced attributes, bidirectional LSTM layer for extracting features of input data, attention mechanism layer and output layerAndThe hidden states of the forward and backward LSTM units at time t are respectively shown, and then the output h t of the bidirectional LSTM layer is: The attention mechanism layer is mainly used for weighting the output of the bidirectional LSTM layer and highlighting important characteristics, and the training model is used for dividing the reduced data D red into a training set, a verification set and a test set;
Step S4, credibility verification and dynamic feedback adjustment
The reliability verification comprises the steps of carrying out reliability verification on a neural network output result through a rough set decision rule base, setting a predicted result of the neural network as y pred, judging whether the predicted result meets corresponding rules according to the rough set decision rule base, if yes, considering that the reliability is high, and if not, calculating the deviation delta of the predicted result and the rules;
The dynamic feedback adjustment comprises triggering a dynamic feedback mechanism if the error exceeds a preset threshold tau, firstly analyzing the reason of the error, readjusting the attribute reduction parameters if the attribute reduction is unreasonable, and updating a minimum condition attribute set C min;
Step S5, outputting the result and early warning response
Outputting a deep foundation pit deformation prediction curve and a risk grade classification result, wherein the risk grade is divided according to the predicted deformation and a preset threshold value, such as low risk, medium risk and high risk;
The linkage site early warning device implements grading response, can perform regular monitoring and recording for low risk states, sends out early warning signals for medium risk states, strengthens monitoring frequency, and immediately starts emergency measures such as evacuating personnel and reinforcing supporting structures for high risk states.
Preferably, preprocessing the data collected in step S1, including removing noise, filling in missing values, etc., for noise removal, adopting a moving average filtering algorithm, setting the original data sequence as x 1,x2,…xn, and the window size as w, wherein the filtered data y i is In order to round down the function,To round up the function, linear interpolation is used to fill in the missing values, if x k is missing and x k-1 and x k+1 are known
Preferably, the attention mechanism layer, let s t be the context vector of time t, alpha t, i be the attention weight, then: Where v, W h、Ws are the learnable weight matrices, b is the bias vector, and T is the time step.
Preferably, the model is trained by using a mean square error loss functionTraining the model, where N is the number of samples, y i is the true value,For the predicted values, parameters of the model are updated using an optimization algorithm such as random gradient descent (SGD).
A deep foundation pit deformation prediction system based on neural network and rough set classification consists of the following modules:
The distributed sensor array module comprises a plurality of types of sensors, such as a displacement sensor, a soil pressure sensor, a ground water level sensor, a stress sensor and the like, which are distributed at different positions of the deep foundation pit, acquire relevant data of the deep foundation pit in real time, and transmit the data to an edge computing node;
An edge computing node integrates a data preprocessing unit and performs preprocessing operations such as removing noise, filling missing values and the like on data acquired by a sensor;
An integrated rough set reduction processor, which is used for carrying out attribute reduction on the data according to the method in the step S2;
the integrated lightweight neural network reasoning chip is used for running an improved A-RNN model and carrying out deformation prediction on the reduced data;
The cloud collaborative training platform is used for storing a large amount of historical monitoring data and model training results and providing data support for training and optimizing the model;
The rule base version management module is provided for managing and updating the rough set decision rule base, providing a model iteration optimization interface and supporting the optimization of the neural network model by means of incremental learning and the like;
The three-dimensional visual terminal supports spatial mapping of the prediction result and the BIM model, intuitively displays the deformation prediction result of the deep foundation pit on the BIM model, provides a multi-working condition comparison analysis function, enables a user to select different time points and different working conditions for comparison analysis, displays the deformation condition of the stratum in a three-dimensional graph mode, and enables the user to view the deformation prediction result of the deep foundation pit and related data at any time point in the past.
Compared with the prior art, the invention has the following beneficial effects:
(1) The method has the advantages that the driving factors of deep foundation pit deformation are comprehensively captured through multi-parameter monitoring data such as integrated displacement, soil pressure, ground water level and supporting stress, misjudgment caused by single sensor data deviation is reduced, long-term dependence and local characteristics of key time points in time sequence data are effectively captured through combination of the bidirectional LSTM and an attention mechanism, accuracy of deformation prediction is remarkably improved, neural network prediction results are checked through a rough set rule base, misinformation of a model under fitting or abnormal working conditions is avoided, and reliability of early warning decision is ensured.
(2) The redundant features are removed, the input dimension is reduced to the minimum necessary set, the training time and memory occupation of the neural network are reduced, the real-time reasoning requirement of the edge computing equipment is adapted, the improved A-RNN structure simplifies the network hierarchy while guaranteeing the prediction precision through the cooperative design of the bidirectional LSTM and the attention mechanism, the computing energy consumption is reduced, and the method is suitable for on-site embedded deployment.
(3) Through the dynamic feedback adjustment module, the system can automatically update model parameters according to the change of the actual engineering environment, does not need to be completely retrained, adapts to long-term complex working conditions of the deep foundation pit, dynamically adjusts the risk level threshold based on historical data and real-time monitoring results, avoids misjudgment caused by fixed threshold, and improves early warning sensitivity.
(4) And the prediction result is superimposed into the BIM model, so that deformation thermodynamic diagram and support structure safety coefficient distribution are generated, multidimensional comparison analysis is supported, visual basis is provided for engineering decision, and full-closed loop control from risk identification to emergency treatment is realized through audible and visual alarm, short message pushing and equipment linkage, so that manual intervention delay is reduced, and construction safety is ensured.
Drawings
FIG. 1 is a logic diagram of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, a technical solution of the embodiments of the present disclosure will be clearly and completely described in conjunction with the accompanying drawings of the embodiments of the present disclosure, detailed descriptions of known functions and known components are omitted from the present disclosure in order to avoid unnecessarily obscuring the concept of the present disclosure,
Referring to fig. 1, a deep foundation pit deformation prediction method based on neural network and rough set classification includes the following steps:
Step S1, data acquisition and preprocessing
Acquiring displacement, soil pressure, underground water level and support structure stress data of a deep foundation pit in real time through a multisource sensor array, and constructing a dynamic monitoring data set D raw;
Preprocessing the acquired data, including removing noise, filling missing values and the like, removing high-frequency random interference for noise, repairing data break points by adopting a linear interpolation method, ensuring data continuity and effectiveness, and avoiding model training from being influenced by abnormal values.
Adopting a moving average filtering algorithm, setting the original data sequence as x 1,x2,…xn, and setting the window size as w, wherein the filtered data y i is In order to round down the function,To round up the function, linear interpolation is used to fill in the missing values, if x k is missing and x k-1 and x k+1 are known
Step S2, attribute reduction based on rough set theory
The method comprises the steps of constructing a decision table S= (U, A, V, f) from preprocessed data, wherein U is a domain, namely a set of all samples, A=C U-D, C is a conditional attribute set and comprises monitoring data attributes such as displacement, soil pressure and the like, D is a decision attribute set, such as deformation states (normal, early warning, danger and the like) of a deep foundation pit, V is a set of attribute values, f is a function of U multiplied by A-V and is used for determining attribute values of each sample, preprocessed multidimensional monitoring data is converted into a structured table form, and logic relation between conditional attributes (monitoring parameters) and decision attributes (deformation states) is defined.
Calculating attribute importance, namely adopting an attribute importance calculating method based on information entropy, setting H (D) as information entropy of decision attribute D, setting H (D|C- { a i) as conditional information entropy of decision attribute D after removing attribute a i, and setting importance Sig (a i) of attribute a i as Sig (a i)＝H(D|C-{ai) -H (D|C) wherein P (D j) is the probability that the decision attribute D takes on the value D j; p (C i) is the probability of the condition attribute set C being C i, p (d j|ci) is the condition probability of the decision attribute being d j when the condition attribute is C i, the contribution degree of each attribute to the deformation state is quantized based on the information entropy theory, redundant attributes (such as soil pressure parameters with low correlation) are removed, the minimum condition attribute set is generated, a simplified decision table is output as the input feature of the neural network, and the training efficiency and generalization capability of the model are remarkably improved.
Step S3, building and training a model of a recurrent neural network (A-RNN) based on an improved attentiveness mechanism
An improved A-RNN network structure is composed of input layer for receiving data set D red with reduced attributes, bidirectional LSTM layer for extracting features of input data, attention mechanism layer and output layerAndThe hidden states of the forward and backward LSTM units at time t are respectively shown, and then the output h t of the bidirectional LSTM layer is: Through time sequence information transmission in the positive direction and the negative direction, the long-term dependence (such as the influence of continuous rainfall on foundation pit settlement) in the deformation process is captured.
The attention mechanism layer mainly weights the output of the attention mechanism layer for the bidirectional LSTM layer and highlights important characteristics, and s t is the context vector of time t, alpha t and i are attention weights, and then: Wherein v and W h、Ws are weight matrixes capable of learning, b is a bias vector, T is a time step, time step weights are dynamically allocated, attention to monitoring data in a key period (such as a construction peak period) is enhanced, and the problem that the traditional RNN is not sensitive enough to local features is solved.
Training model by dividing the reduced data D red into training set, validation set and test set, and using mean square error loss functionTraining the model, where N is the number of samples, y i is the true value,For predicting values, parameters of the model are updated by using optimization algorithms such as random gradient descent (SGD), and the model is ensured to be converged to a global optimal solution by adopting staged optimization (training set, verification set and test set) and combining an adaptive learning rate algorithm (Adam).
Step S4, credibility verification and dynamic feedback adjustment
The reliability verification comprises the steps of carrying out reliability verification on a neural network output result through a rough set decision rule base, setting a predicted result of the neural network as y pred, judging whether the predicted result meets corresponding rules according to the rough set decision rule base, if yes, considering that the reliability is high, and if not, calculating the deviation delta of the predicted result and the rules;
The dynamic feedback adjustment comprises triggering a dynamic feedback mechanism if the error exceeds a preset threshold tau, firstly analyzing the reason of the error, readjusting the attribute reduction parameters if the attribute reduction is unreasonable, and updating a minimum condition attribute set C min;
Step S5, outputting the result and early warning response
Outputting a deep foundation pit deformation prediction curve and a risk grade classification result, wherein the risk grade is divided according to the predicted deformation and a preset threshold value, such as low risk, medium risk and high risk;
The linkage site early warning device implements grading response, can perform regular monitoring and recording for low risk states, sends out early warning signals for medium risk states, strengthens monitoring frequency, and immediately starts emergency measures such as evacuating personnel and reinforcing supporting structures for high risk states.
A deep foundation pit deformation prediction system based on neural network and rough set classification consists of the following modules:
The distributed sensor array module comprises a plurality of types of sensors, such as a displacement sensor, a soil pressure sensor, a ground water level sensor, a stress sensor and the like, and is distributed at different positions of the deep foundation pit, so that relevant data of the deep foundation pit are collected in real time, and the data are transmitted to the edge computing nodes.
And the edge computing node integrates a data preprocessing unit and performs preprocessing operations such as removing noise, filling missing values and the like on the data acquired by the sensor.
And (3) integrating a rough set reduction processor to reduce the data according to the attribute of the method in the step S2.
And integrating a lightweight neural network reasoning chip, namely running an improved A-RNN model, and carrying out deformation prediction on the reduced data.
And the cloud collaborative training platform is used for storing a large amount of historical monitoring data and model training results and providing data support for training and optimizing the model.
And the rule base version management module is provided for managing and updating the rough set decision rule base, providing a model iteration optimization interface and supporting the optimization of the neural network model by means of incremental learning and the like.
And the three-dimensional visual terminal supports the spatial mapping of the prediction result and the BIM model, intuitively displays the deformation prediction result of the deep foundation pit on the BIM model, provides a multi-working condition comparison analysis function, and enables a user to select different time points and different working conditions for comparison analysis so as to better know the deformation condition of the deep foundation pit.
Stratum deformation drawing based on an improved MarchingCubes algorithm is realized, and the deformation condition of the stratum is displayed in a three-dimensional graph form.
And displaying space superposition information of the risk thermodynamic diagram and the safety coefficient of the supporting structure, so that a user can intuitively know the risk distribution condition of the deep foundation pit and the safety state of the supporting structure.
The deep foundation pit deformation prediction method has a history backtracking function, and a user can check the deep foundation pit deformation prediction result and related data at any time point in the past.
The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention, the scope of the present invention is defined by the claims, and various modifications or equivalent substitutions can be made by those skilled in the art within the spirit and scope of the present invention, and such modifications or equivalent substitutions should be considered to fall within the scope of the present invention.
Claims (5)
1. A deep foundation pit deformation prediction method based on neural network and rough set classification is characterized by comprising the following steps:
Step S1, data acquisition and preprocessing
Acquiring displacement, soil pressure, underground water level and support structure stress data of a deep foundation pit in real time through a multisource sensor array, and constructing a dynamic monitoring data set D raw;
step S2, attribute reduction based on rough set theory
Constructing a decision table S= (U, A, V, f) from the preprocessed data, wherein U is a domain, namely a set of all samples, A=C U-D, C is a conditional attribute set and comprises monitoring data attributes such as displacement, soil pressure and the like, D is a decision attribute set such as deformation state (normal, early warning, danger and the like) of a deep foundation pit, V is a set of attribute values, and f is a U×A-V is an information function and is used for determining the attribute value of each sample;
Calculating attribute importance, namely adopting an attribute importance calculating method based on information entropy, setting H (D) as information entropy of decision attribute D, setting H (D|C- { a i) as conditional information entropy of decision attribute D after removing attribute a i, and setting importance Sig (a i) of attribute a i as Sig (a i)＝H(D|C-{ai) -H (D|C) wherein P (D j) is the probability that the decision attribute D takes on the value D j; p (C i) is the probability that the conditional attribute set C takes the value of C i, and p (d j|ci) is the conditional probability that the decision attribute takes the value of d j when the conditional attribute takes the value of C i;
step S3, building and training a model of a recurrent neural network (A-RNN) based on an improved attentiveness mechanism
An improved A-RNN network structure is composed of input layer for receiving data set D red with reduced attributes, bidirectional LSTM layer for extracting features of input data, attention mechanism layer and output layerAndThe hidden states of the forward and backward LSTM units at time t are respectively shown, and then the output h t of the bidirectional LSTM layer is: The attention mechanism layer is mainly used for weighting the output of the bidirectional LSTM layer and highlighting important characteristics, and the training model is used for dividing the reduced data D red into a training set, a verification set and a test set;
Step S4, credibility verification and dynamic feedback adjustment
The reliability verification comprises the steps of carrying out reliability verification on a neural network output result through a rough set decision rule base, setting a predicted result of the neural network as y pred, judging whether the predicted result meets corresponding rules according to the rough set decision rule base, if yes, considering that the reliability is high, and if not, calculating the deviation delta of the predicted result and the rules;
The dynamic feedback adjustment comprises triggering a dynamic feedback mechanism if the error exceeds a preset threshold tau, firstly analyzing the reason of the error, readjusting the attribute reduction parameters if the attribute reduction is unreasonable, and updating a minimum condition attribute set C min;
Step S5, outputting the result and early warning response
Outputting a deep foundation pit deformation prediction curve and a risk grade classification result, wherein the risk grade is divided according to the predicted deformation and a preset threshold value, such as low risk, medium risk and high risk;
The linkage site early warning device implements grading response, can perform regular monitoring and recording for low risk states, sends out early warning signals for medium risk states, strengthens monitoring frequency, and immediately starts emergency measures such as evacuating personnel and reinforcing supporting structures for high risk states.
2. The deep foundation pit deformation prediction method based on neural network and rough set classification as claimed in claim 1, wherein the preprocessing of the data collected in step S1 comprises removing noise, filling missing values, etc., for noise removal, a moving average filtering algorithm is adopted, the original data sequence is set as x 1,x2,…xn, the window size is w, and the filtered data y i is In order to round down the function,To round up the function, linear interpolation is used to fill in the missing values, if x k is missing and x k-1 and x k+1 are known
3. The deep foundation pit deformation prediction method based on neural network and rough set classification of claim 1, wherein the attention mechanism layer is provided with s t as a context vector of time t, and a t, i as attention weight, and the following steps are: Where v, W h、Ws are the learnable weight matrices, b is the bias vector, and T is the time step.
4. The deep foundation pit deformation prediction method based on neural network and rough set classification as set forth in claim 1, wherein the training model is a mean square error loss function Training the model, where N is the number of samples, y i is the true value,For the predicted values, parameters of the model are updated using an optimization algorithm such as random gradient descent (SGD).
5. The deep foundation pit deformation prediction system based on the neural network and the rough set classification is characterized by adopting the deep foundation pit deformation prediction method based on the neural network and the rough set classification as claimed in any one of claims 1-4, and comprising the following modules:
The distributed sensor array module comprises a plurality of types of sensors, such as a displacement sensor, a soil pressure sensor, a ground water level sensor, a stress sensor and the like, which are distributed at different positions of the deep foundation pit, acquire relevant data of the deep foundation pit in real time, and transmit the data to an edge computing node;
An edge computing node integrates a data preprocessing unit and performs preprocessing operations such as removing noise, filling missing values and the like on data acquired by a sensor;
An integrated rough set reduction processor, which is used for carrying out attribute reduction on the data according to the method in the step S2;
the integrated lightweight neural network reasoning chip is used for running an improved A-RNN model and carrying out deformation prediction on the reduced data;
The cloud collaborative training platform is used for storing a large amount of historical monitoring data and model training results and providing data support for training and optimizing the model;
The rule base version management module is provided for managing and updating the rough set decision rule base, providing a model iteration optimization interface and supporting the optimization of the neural network model by means of incremental learning and the like;
The three-dimensional visual terminal supports spatial mapping of the prediction result and the BIM model, intuitively displays the deformation prediction result of the deep foundation pit on the BIM model, provides a multi-working condition comparison analysis function, enables a user to select different time points and different working conditions for comparison analysis, displays the deformation condition of the stratum in a three-dimensional graph mode, and enables the user to view the deformation prediction result of the deep foundation pit and related data at any time point in the past.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510940255.4A CN120822099A (en) | 2025-07-08 | 2025-07-08 | A deep foundation pit deformation prediction method and system based on neural network and rough set classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510940255.4A CN120822099A (en) | 2025-07-08 | 2025-07-08 | A deep foundation pit deformation prediction method and system based on neural network and rough set classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN120822099A true CN120822099A (en) | 2025-10-21 |
Family
ID=97370638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510940255.4A Pending CN120822099A (en) | 2025-07-08 | 2025-07-08 | A deep foundation pit deformation prediction method and system based on neural network and rough set classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120822099A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN121385015A (en) * | 2025-10-30 | 2026-01-23 | 中国电建集团吉林省电力勘测设计院有限公司 | Composite green concrete salt-freezing-resistant durability evaluation system under soil environment of frozen soil area |
| CN121615722A (en) * | 2026-01-30 | 2026-03-06 | 华中科技大学 | Foundation pit prediction model training method, foundation pit monitoring method and system |
-
2025
- 2025-07-08 CN CN202510940255.4A patent/CN120822099A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN121385015A (en) * | 2025-10-30 | 2026-01-23 | 中国电建集团吉林省电力勘测设计院有限公司 | Composite green concrete salt-freezing-resistant durability evaluation system under soil environment of frozen soil area |
| CN121615722A (en) * | 2026-01-30 | 2026-03-06 | 华中科技大学 | Foundation pit prediction model training method, foundation pit monitoring method and system |
| CN121615722B (en) * | 2026-01-30 | 2026-04-10 | 华中科技大学 | Foundation pit prediction model training method, foundation pit monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN120822099A (en) | A deep foundation pit deformation prediction method and system based on neural network and rough set classification | |
| CN120340223B (en) | Intelligent early warning system and method for foundation pit deformation based on multimodal fusion | |
| CN113159431A (en) | Analysis early warning method and device based on ground disaster data | |
| CN118692225B (en) | Hydraulic blasting safety risk early warning system based on deep learning | |
| CN120726788B (en) | Intelligent Early Warning Methods and Systems for Geological Hazards in Geotechnical Engineering | |
| CN120911980B (en) | Construction environment abrupt change risk field prediction method fusing geological meteorological data | |
| CN120296571A (en) | Road slope stability prediction method, device, storage medium and electronic equipment | |
| CN120561455A (en) | Building correction settlement prediction method and system based on multi-source data and artificial intelligence fusion | |
| CN120318988A (en) | Geological disaster early warning method and geological disaster early warning system | |
| CN120493056A (en) | Multi-standard intelligent aggregation evaluation method and system for quality of underground engineering surrounding rock | |
| CN114673246A (en) | Anti-blocking measurement method and measurement system for sewage pipeline | |
| KR102579897B1 (en) | Disaster Prevention System According to Building Resilience Destruction and Method Using the Same | |
| CN119647935B (en) | A regional geological disaster risk intelligent assessment method and system | |
| CN120783491A (en) | Disaster monitoring and early warning method and system based on iron tower data | |
| CN120471456A (en) | Non-coal mine safety risk dynamic monitoring and early warning system based on point-surface fusion technology | |
| CN119961842A (en) | A hydrological sequence data anomaly detection method and system based on cloud computing | |
| CN119933109A (en) | A Slope Stability Monitoring System for Spoil Sites | |
| CN117689210A (en) | Intelligent Assessment Method for Road Natural Disaster Risks | |
| CN121071565B (en) | A method for predicting the state of an operational tunnel near a foundation pit based on a spatiotemporal decoupled network | |
| CN120340204B (en) | Deep learning-driven multi-level geological disaster precipitation early warning method | |
| CN121119932A (en) | Micropile construction control, graph neural network training method and device | |
| CN121615957B (en) | Traffic network toughness diagnosis method under flood disasters based on space-time diagram neural network | |
| CN121640666A (en) | High-fill reinforced slope monitoring and early warning method based on artificial intelligence | |
| CN120611871A (en) | Drainage flow prediction model construction method, equipment, medium and program product | |
| CN121677799A (en) | Rock-soil layer movement monitoring system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |