A method for intelligent evaluation of static and dynamic risks of foundation pit engineering based on multi-source data fusion

Abstract

The present application relates to the technical field of civil engineering safety monitoring, in particular to a method for foundation pit engineering static and dynamic risk intelligent evaluation system based on multi-source data fusion, which firstly sets personalized initial threshold value based on finite element model back analysis; collects sensor data, field images and texts, and realizes semantic vectorization of unstructured data by using computer vision and natural language processing technology; carries out data cleaning and fault diagnosis by using variational mode decomposition and spatial topology checking logic; applies D-S evidence theory to fuse multi-source data and calculate real-time risk probability; finally establishes a double-control early warning mechanism based on deformation total amount and acceleration, and corrects model parameters through measured data inversion. The present application effectively solves the problems of multi-source heterogeneous data fusion difficulty, low sensor reliability and threshold rigidity, and significantly improves the accuracy and forward-looking of foundation pit risk evaluation.

Description

Technical Field

[0001] This invention relates to the field of civil engineering safety monitoring technology, specifically to a method for intelligent evaluation of static and dynamic risks in foundation pit engineering based on multi-source data fusion. Background Technology

[0002] Foundation pit engineering, as a fundamental aspect of underground space development, is characterized by high risk, long duration, and complex environmental effects. The safety of foundation pits depends not only on prior geological investigation and support design, but also on real-time monitoring of soil deformation, groundwater levels, and the surrounding environment during construction. Therefore, establishing a scientific static and dynamic risk assessment system is crucial to preventing foundation pit collapse accidents.

[0003] Currently, existing foundation pit risk assessment technologies mainly rely on automated monitoring instruments to acquire structured data and achieve early warning by comparing monitoring values ​​with fixed alarm values ​​specified in national or industry standards. Before construction, static risk classification is usually carried out using expert scoring methods or the analytic hierarchy process; during construction, dynamic tracking is solely based on monitoring data.

[0004] The aforementioned existing technologies face the following pressing technical problems in practical engineering applications: Construction sites generate not only numerical data from sensors but also a large amount of unstructured data. Existing technologies struggle to convert this qualitative information into quantitative data for calculation, leading to the loss of crucial physical indicators in the evaluation model. Furthermore, the harsh environment of construction sites, with electromagnetic interference and mechanical collisions easily causing sensor readings to fluctuate or drift, makes it difficult to distinguish between equipment malfunctions and genuine emergencies, resulting in frequent false alarms and reduced vigilance among on-site personnel. Finally, existing technologies generally adopt fixed thresholds from standards, ignoring the differences in geological conditions and support systems, and often focusing only on cumulative deformation while neglecting changes in deformation acceleration. This leads to situations where, under certain special conditions, although the total deformation may not exceed the standard, the soil has already entered a stage of accelerated failure, yet the system fails to provide timely warnings. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for intelligent assessment of static and dynamic risks in foundation pit engineering based on multi-source data fusion. This method can achieve quantitative fusion of unstructured data, intelligently identify and eliminate false sensor data, and set personalized dynamic thresholds based on physical and mechanical mechanisms, thereby improving the accuracy, robustness, and foresight of foundation pit risk assessment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent assessment of static and dynamic risks in foundation pit engineering based on multi-source data fusion, comprising the following steps: Step S1: Static baseline modeling and personalized threshold initialization Based on the geological survey parameters (such as soil physical and mechanical properties) and support design scheme of the foundation pit project, a three-dimensional finite element numerical model is constructed. Using the strength reduction method or surcharge method, the entire process from foundation pit excavation to instability failure is simulated in the model to obtain the ultimate bearing state displacement of this specific foundation pit. U limit Select a certain ratio between the allowable value specified in the standard and the calculated critical displacement value ( k U limit The smaller value among the values ​​is used as the specific red alarm threshold for this foundation pit project. T red To implement a tailored approach for each pit.

[0007] Step S2: Multi-source heterogeneous data acquisition and semantic vectorization The IoT monitoring system collects structured time-series data of the foundation pit (such as deep horizontal displacement and support axial force); at the same time, it collects unstructured data such as on-site inspection images and construction log text.

[0008] Computer vision techniques (such as convolutional neural networks, CNNs) are used to process images, automatically identifying and extracting features such as crack length, width, and seepage area, which are then mapped to a normalized image risk index. V img .

[0009] Natural language processing techniques (such as the BERT model) are used to perform semantic analysis on the text, extract risk keywords (such as sand inrush and abnormal noise), and combine them with sentiment analysis to calculate the text risk index. V text .

[0010] Finally, based on the BIM model, the aforementioned risk indices are mapped to corresponding three-dimensional spatial coordinates to achieve spatiotemporal alignment of multi-source data.

[0011] Step S3: Data cleaning and fault diagnosis based on spatiotemporal correlation First, the original sensor data is decomposed using a variational mode decomposition algorithm to remove high-frequency noise components.

[0012] Secondly, a sensor spatial topology network is constructed, and the principle of neighborhood consistency is used for fault diagnosis: if the reading of a sensor node changes abruptly, but the readings of its spatially adjacent nodes and the image and text risk index of the corresponding area do not show any abnormalities, then the change is determined to be a device failure or local interference and is eliminated.

[0013] Finally, a long short-term memory network is used to perform high-precision interpolation to fill in the removed or missing data.

[0014] Step S4: Multi-source evidence fusion and dynamic risk calculation A fusion model is established using DS evidence theory. This model integrates cleaned sensor data and image risk indices. V img Text risk index V text These are defined as three independent pieces of evidence within the identification framework. The weights are dynamically adjusted based on the real-time conflict coefficient between the evidence, and the confidence level of the fused foundation pit safety status is calculated, outputting a comprehensive dynamic risk probability value. P risk .

[0015] Step S5: Dual-control dynamic early warning and feedback correction Establish a dual-control early warning mechanism based on both total amount and trend. An alarm will be triggered when any of the following conditions are met: ① The cumulative value monitored in real-time exceeds... T red ② The second derivative (acceleration) of the monitoring data is greater than zero and shows a continuous increasing trend, and P risk It is in a high-risk zone.

[0016] Meanwhile, the measured data and the numerical model prediction data are compared regularly. When the deviation exceeds a preset threshold, the soil parameters in the numerical model are corrected using a back-analysis algorithm, and the data is recalculated. U limit And update the alarm threshold for the next stage.

[0017] This invention provides a method for intelligent assessment of static and dynamic risks in foundation pit engineering based on multi-source data fusion. It has the following beneficial effects: 1. This invention introduces convolutional neural network image recognition and natural language processing technology to convert on-site photos and log text into numerical vectors, and uses DS evidence theory for mathematical fusion to incorporate physical signs (qualitative information) discovered during inspections into the evaluation system, which greatly improves the ability to perceive early micro-risks.

[0018] 2. This invention employs signal decomposition and denoising, combined with neighborhood consistency verification logic based on spatial topology. This effectively distinguishes between real soil deformation and mechanical vibration / equipment malfunction, significantly reducing the false alarm rate caused by harsh on-site environments.

[0019] 3. This invention sets a threshold by calculating the physical ultimate bearing capacity of a specific foundation pit based on finite element inverse analysis, and introduces an acceleration trend term as a dual control indicator. This avoids both collapse accidents caused by excessively loose thresholds (where the value is not exceeded) and blind work stoppages caused by excessively strict thresholds; furthermore, acceleration monitoring can detect signs of sudden instability in advance.

[0020] 4. This invention compares measured and simulated values ​​in real time and corrects model parameters through inversion, enabling the evaluation system to have adaptive evolution capabilities. As construction progresses, the model's understanding of soil properties becomes more accurate, and the prediction and evaluation results become more reliable. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: This invention provides a method for intelligent assessment of static and dynamic risks in foundation pit engineering based on multi-source data fusion. Taking a deep foundation pit project of a subway station in a certain city as an example, the foundation pit is 18m deep, the retaining structure is a diaphragm wall with internal support, and the stratum is mainly silty soft soil. The specific implementation steps are as follows: S1. Static baseline modeling and threshold setting A finite element model of the foundation pit was created using Plaxis 3D software. Soil parameters from the investigation report were input (unit weight γ = 17 kN / m³). 3 (Cohesion c = 15 kPa, internal friction angle ϕ = 12°).

[0023] When the excavation reaches 15m, the strength reduction method was used for calculation. It was found that the calculation did not converge when the maximum horizontal displacement reached 50mm, which was determined to be the ultimate displacement. U limit =50mm.

[0024] According to the specifications, the alarm threshold for this level of foundation pit is 40mm. We compared 0.8 × 50mm = 40mm with the specified value and ultimately set the red alarm threshold for this working condition. T red =40mm.

[0025] S2. Data Acquisition and Processing At time T, the IoT system collected data showing that the inclinometer numbered CX-10 had a displacement of 32 mm at a depth of 12 m.

[0026] Meanwhile, the on-site inspector uploaded a photo of a crack in the crown beam near CX-10, along with a voice recording: "Large dump trucks frequently pass by here."

[0027] The system uses a pre-trained U-Net model to identify photos, calculates the crack width as 1.5mm, and converts it into an image risk index. V img =0.6 (medium risk).

[0028] The system uses an NLP model to extract keywords related to dump trucks from the speech and identifies them as sources of dynamic load interference.

[0029] S3. Data Cleaning and Diagnostics The system detected that the CX-10 reading jumped from 28 mm to 32 mm within one hour.

[0030] The system immediately checked the CX-09 on the left and the CX-11 on the right of the CX-10 and found that their readings were 27mm and 27.5mm respectively, with no sudden changes.

[0031] Based on the interference information extracted from the dump truck using natural language processing technology, the system determined that the sudden change in the CX-10 was mainly affected by vehicle vibration. A signal decomposition and denoising algorithm was used to remove high-frequency components, and the corrected CX-10 reading was 28.5 mm.

[0032] S4. Integration and Evaluation Although the corrected sensor reading of 28.5 mm is far less than the threshold of 40 mm (seemingly safe), image evidence shows a solid crack. V img =0.6).

[0033] The DS evidence fusion algorithm, when dealing with conflicts between low-risk data from sensors and high-risk data in images, considers the physical authenticity of the cracks, thus improving the trustworthiness of image evidence. The final output is a comprehensive dynamic risk probability. P risk =0.65 (belongs to the yellow warning zone).

[0034] S5. Early Warning and Feedback Calculating the deformation rate reveals that the deformation acceleration a > 0 over the past 24 hours (deformation is accelerating).

[0035] Although the cumulative value (28.5mm) did not reach the red line, due to abnormal acceleration, the presence of cracks, and a high overall risk probability, the system triggered a yellow warning, indicating to the project department that the deformation rate was diverging and accompanied by structural cracks, and recommended suspending the passage of dump trucks and increasing monitoring frequency.

[0036] After the project team adopted the suggestion, the monitoring data the following day showed that the acceleration turned negative (deformation tended to stabilize).

[0037] Meanwhile, the system found that the measured deformation was slightly greater than the predicted value of the three-dimensional finite element numerical model, and automatically inverted and corrected the soil resilient modulus parameter in the model. The updated model will make more accurate predictions for the next stage of excavation.

[0038] This embodiment verifies that the method of the present invention can effectively avoid false alarms and achieve accurate risk management by integrating multi-source data and physical mechanisms under complex working conditions.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent assessment of static and dynamic risks in foundation pit engineering based on multi-source data fusion, characterized in that, Includes the following steps: Step S1: Static baseline modeling and personalized threshold initialization Based on the geological survey parameters and support design scheme of the foundation pit project, a three-dimensional finite element numerical model is constructed. The ultimate bearing state displacement of the foundation pit is obtained through simulation calculation, and an initial dynamic early warning threshold is set accordingly. Step S2: Multi-source heterogeneous data acquisition and semantic vectorization Structured data from IoT sensors at the foundation pit site, as well as unstructured data including on-site inspection images and construction log text, are collected. Computer vision and natural language processing technologies are used to transform the unstructured data into risk numerical vectors, and spatiotemporal coordinates are aligned based on the BIM model. Step S3: Data cleaning and fault diagnosis based on spatiotemporal correlation The collected sensor data is decomposed and denoised, and spatial correlation verification logic is used to identify and remove false data anomalies caused by equipment failure, and missing or removed data is intelligently filled in. Step S4: Multi-source evidence fusion and dynamic risk calculation The cleaned sensor data, image risk vector, and text risk vector are used as independent evidence, and a multi-source information fusion algorithm is used to calculate the real-time comprehensive dynamic risk probability value of the foundation pit. Step S5: Dual-control dynamic early warning and feedback correction A dual-control early warning mechanism is established based on the total deformation and deformation acceleration, and graded early warning is carried out in combination with the comprehensive dynamic risk probability value. When the deviation between the measured data and the numerical model prediction data exceeds the preset range, the numerical model parameters in step S1 are inverted and corrected, and the early warning threshold for the next stage is updated.

2. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, In step S1, setting the initial dynamic early warning threshold specifically includes: The strength parameters of the soil in the three-dimensional finite element numerical model are gradually reduced using the strength reduction method until the model calculation fails to converge or abrupt instability occurs, and the critical displacement value at this point is obtained. U limit ; Select the smaller value of a certain proportion of the specified allowable value and the calculated critical displacement value ( k·U limit , where 0 < k < 1) as the specific red alarm threshold for this foundation pit project T red .

3. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, In step S2, the transformation of the unstructured data specifically includes: For on-site inspection images, a convolutional neural network model is used to extract features of crack length, width, and seepage area, which are then mapped to a normalized image risk index. V img ; For construction logs and inspection records, a pre-trained language model based on the BERT architecture is used for semantic analysis to extract pre-defined risk keywords and combine them with sentiment analysis algorithms to calculate the text risk index. V text .

4. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, In step S3, the data cleaning and fault diagnosis specifically include: The variational mode decomposition algorithm is used to decompose the original sensor data sequence into several modal components, and the trend signal is reconstructed after removing high-frequency noise components; Construct a sensor spatial topology network. If the reading change rate of a certain sensor node exceeds a set threshold, but the reading change rates of its neighboring sensor nodes and the image risk index of the corresponding area are also considered... V img Text risk index V text If none of the data exceeds the anomaly detection criteria, the sensor node data is determined to be due to equipment failure or local interference and is therefore discarded.

5. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, In step S4, the multi-source information fusion algorithm adopts the DS evidence theory, defining sensor data, image risk index, and text risk index as three evidence bodies under the recognition framework, and constructs a basic probability allocation function. During the fusion process, the weights are dynamically adjusted according to the real-time conflict coefficients of each evidence body, the confidence level of the foundation pit safety status after fusion is calculated, and a comprehensive dynamic risk probability value is output. P risk .

6. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, In step S5, the specific triggering condition for the dual-control early warning mechanism is to satisfy any of the following conditions: Condition 1: The real-time cumulative value monitored by the sensor after cleaning exceeds the red alarm threshold set in step S1. T red ; Condition 2: The second derivative of the monitoring data is greater than zero and shows a continuous increasing trend, while the comprehensive dynamic risk probability value calculated in step S4 is... P risk The risk exceeds the preset high-risk probability threshold.

7. The intelligent assessment system method for static and dynamic risks of foundation pit engineering based on multi-source data fusion as described in claim 1, characterized in that, The feedback correction in step S5 specifically includes: Periodically calculate the root mean square error between the measured monitoring curve and the numerical model predicted curve; When the root mean square error exceeds the preset tolerance limit, the intelligent back analysis algorithm is activated to infer the actual physical and mechanical parameters of the soil based on the measured displacement. The back-derived parameters are updated in the three-dimensional finite element numerical model, and step S1 is executed again to generate a new warning threshold applicable to the next construction stage.

8. A static and dynamic risk intelligent assessment system for foundation pit engineering implementing the method of any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to acquire IoT sensor data, as well as on-site images and text data; The data processing module is used to perform data vectorization, cleaning, noise reduction, and fault diagnosis. The computational analysis module is used to run finite element numerical simulations, evidence fusion algorithms, and risk probability calculations. The early warning decision module is used to output early warning signals and dynamically update model parameters based on the dual control mechanism.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method of any one of claims 1 to 8.