Soft soil foundation settlement automatic monitoring system based on multi-source data fusion

Abstract

The invention discloses an automatic soft soil foundation settlement monitoring system based on multi-source data fusion, and relates to the technical field of soft soil foundation monitoring, the system comprises an information acquisition module, a fusion processing module, a settlement prediction module and an intelligent monitoring module; the information acquisition module is used for acquiring foundation settlement sensing data and inputting the acquired data into the fusion processing module; the fusion processing module is used for preprocessing and integrating the collected data; the settlement prediction module is used for soft soil foundation settlement prediction; the intelligent monitoring module comprises a self-adaptive processing module and an interaction alarm module, the self-adaptive processing module is used for generating an optimization strategy, and the interaction alarm module is used for carrying out user interaction and multi-mode abnormal alarm reminding. An early warning response window is provided for engineering personnel, and the occurrence rate of sudden settlement accidents is reduced.

Description

Technical Field

[0001] This application relates to the field of soft soil foundation monitoring technology, specifically to an automatic monitoring system for soft soil foundation settlement based on multi-source data fusion. Background Technology

[0002] Real-time settlement monitoring of soft soil foundations refers to the continuous, automatic, and real-time observation and analysis of settlement occurring in soft soil foundations during engineering construction and operation, utilizing sensor technology and data acquisition systems. By monitoring settlement data in real time, abnormal settlement can be detected promptly, foundation stability can be assessed, future settlement trends can be predicted, and a scientific basis for engineering decisions can be provided. Real-time settlement monitoring of soft soil foundations is of great significance in the construction of infrastructure such as highways, railways, bridges, ports, and docks. It can effectively prevent engineering accidents caused by foundation settlement, ensure project safety, improve project quality, and reduce project risks.

[0003] However, existing technologies mainly rely on single sensor data and static experience models, which have shortcomings such as limited monitoring dimensions, delayed risk response, and rigid strategy adjustment. They cannot accurately capture the multi-physical field coupling changes in soft soil foundations, and have a high degree of dependence on human intervention, resulting in a high rate of missed accident warnings and serious waste of resources. It is difficult to achieve the dual optimization of settlement suppression and cost control. Summary of the Invention

[0004] In view of this, this application provides an automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion, which solves the technical problems of existing technologies that rely on single sensor data and static experience models, resulting in limited monitoring dimensions, delayed risk response, rigid strategy adjustment, inability to accurately capture the multi-physical field coupling changes of soft soil foundation, high dependence on manual intervention, high rate of missed accident warnings, serious waste of resources, and difficulty in achieving dual optimization of settlement suppression and cost control.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention mainly consists of an automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion. The system includes an information acquisition module, a fusion processing module, a settlement prediction module, and an intelligent monitoring module. The information acquisition module collects foundation settlement sensor data and inputs it into the fusion processing module. The output of the information acquisition module is connected to the input of the preprocessing unit in the fusion processing module via a hybrid RS-485 and LoRa communication link. The output of the preprocessing unit is directly connected to the data integration interface of the multi-source data fusion unit in the fusion processing module via a standardized time-series data stream. The fusion processing module performs preprocessing and integration of the collected data. The output of the fusion processing module is connected to the spatiotemporal characteristic input port of the settlement prediction module via a physical constraint verification channel. The settlement prediction module is used to predict settlement of soft soil foundations. The intelligent monitoring module is connected to both the fusion processing module and the settlement prediction module. The intelligent monitoring module includes an adaptive processing module and an interactive alarm module. The adaptive processing module generates optimization strategies and sends them to the real-time control execution and feedback submodule via a control command port. The prediction result output of the settlement prediction module is connected to the decision engine entry point of the adaptive processing module. The interactive alarm module provides user interaction and multi-mode anomaly alarm alerts. The alarm signal output of the interactive alarm module is connected in reverse to the strategy correction interface of the adaptive processing module, forming a closed-loop link from data acquisition to decision execution.

[0006] Furthermore, the information acquisition module includes a distributed sensor network, which comprises multiple sensor nodes. Each node includes one or more of the following devices: a vibrating wire soil settlement gauge, a fiber optic pore water pressure sensor, a resistive earth pressure cell, and an inclinometer. The sensor network is deployed differently based on the geological risk level. In high-risk areas (risk index R>0.7, thick layers of extremely compressible soft soil, large and unevenly distributed loads, drastic groundwater level changes, and strong nearby disturbances), a dual-node redundancy design is adopted to ensure data continuity. In high-risk areas of soft soil foundation settlement, two independent and functionally identical sensor nodes are deployed at each key monitoring point. The two nodes use different power supplies and communication links. When the primary node fails, the backup node automatically takes over, ensuring uninterrupted data acquisition. Simultaneously, the data from the two nodes are compared and verified in real time to improve the reliability of the monitoring results.

[0007] Geological risk assessment primarily quantifies the likelihood and potential severity of harmful settlement (especially uneven settlement) in soft soil foundations under load, requiring comprehensive consideration of geological conditions, load characteristics, and environmental factors. Geological risk assessment mainly includes the following three steps: 1. Data collection and basic parameter acquisition: Collect detailed engineering geological survey reports, obtain data on stratigraphic structure, physical and mechanical properties of soft soil, groundwater conditions, and historical settlement, and collect load, environmental, and site information; 2. Key risk indicator analysis and quantification: Analyze and quantify inherent vulnerability indicators (soft soil layer thickness, compressibility, water content, and void ratio) and external inducing factor indicators (load intensity and distribution, and groundwater level dynamics); 3. Risk level classification model: Based on engineering experience and sensitivity analysis, select core indicators, use the analytic hierarchy process (AHP) or expert scoring method to determine the weight of each indicator, and calculate the comprehensive risk index through indicator normalization. Finally, set thresholds to classify the risk index into 3-4 levels (e.g., high risk, medium risk, and low risk).

[0008] The principle for differentiated deployment of sensor networks based on risk level is as follows: high-density, high-precision, all-round, and real-time monitoring in high-risk areas; moderate deployment in medium-risk areas; and simplified deployment or reliance on routine inspections in low-risk areas. Regarding sensor types, high-precision GNSS receivers, hydrostatic levels, fixed inclinometers, and pore water pressure gauges are deployed in high-risk areas to monitor indicators such as absolute surface displacement, relative surface displacement, and excess pore pressure dissipation. Hydrostatic levels and mobile inclinometers are deployed in medium-risk areas to measure relative surface displacement. In low-risk areas, a small number of hydrostatic levels or periodically used electronic levels are used to measure relative surface displacement.

[0009] Furthermore, the information acquisition module captures key physical parameters of the foundation in real time through a distributed sensor network, including multi-dimensional information such as settlement, pore water pressure, soil stress, and tilt angle. Sensor nodes are deployed differently based on geological exploration results and risk levels. High-risk areas utilize dense coverage with high-precision vibrating wire soil settlement gauges and fiber optic pressure sensors, while low-risk areas are equipped with economical capacitive sensors. All devices are networked using industrial-grade communication protocols, supporting wired transmission and wireless LoRa dual-channel redundant backup to ensure no data loss even in extreme environments. The module has a built-in self-test function that periodically scans the sensor health status, automatically switching abnormal nodes to backup devices. It also supports manual data entry during inspections, providing comprehensive raw data support for subsequent analysis.

[0010] Furthermore, the fusion processing module includes a preprocessing unit and a multi-source data fusion unit; the preprocessing unit integrates wavelet transform algorithm and dynamic threshold correction mechanism to perform noise filtering, outlier processing and time sequence alignment on the acquired data.

[0011] The preprocessing unit standardizes the raw sensor data to eliminate environmental noise and equipment drift interference. A sliding window filtering algorithm is used to smooth high-frequency noise, and box plots are used to identify outliers. Data exceeding three standard deviations are removed or corrected using linear interpolation. Multi-source heterogeneous data are synchronized with timestamps using a high-precision clock synchronization protocol to eliminate timing misalignment caused by sampling frequency differences. Finally, characteristic parameters such as settling acceleration and pore water pressure change rate are extracted to form a standardized time-series dataset, providing high-quality input for higher-level analysis.

[0012] The multi-source data fusion unit is connected to the preprocessing unit. The multi-source data fusion unit is used to integrate and extract features from the preprocessed data through an improved Kalman filter framework, sensor dynamic weighting, and the Mohr-Coulomb physical constraint model.

[0013] The multi-source data fusion unit integrates multi-dimensional foundation state information based on an improved Kalman filter framework, addressing the limitations of single-sensor data. Leveraging the millimeter-level accuracy of settlement sensors and the spatial pointing characteristics of tilt sensors, information weights are dynamically allocated. When soil shear deformation occurs, the weight of tilt data is automatically increased to detect signs of instability. The fusion process incorporates soil mechanics constitutive equations as physical constraints, ensuring that the data fusion results conform to the laws of soil deformation mechanics. Outputs include derived indicators such as the comprehensive settlement index and regional stability score, providing accurate situational awareness for decision-makers.

[0014] Furthermore, the settlement prediction module includes an incremental learning unit, a confidence interval calculation engine, and an engineering case library. The incremental learning unit is used to predict future evolution trends by using a spatiotemporal convolutional neural network model and integrating historical settlement patterns with real-time monitoring data. It is also used to train the model based on the engineering case library and learn the nonlinear characteristics of settlement development under different geological conditions. During online operation, the parameters are updated incrementally according to a preset period to dynamically adapt to changes in soil creep and external loads. The confidence interval calculation engine is used to calculate the confidence interval.

[0015] The settlement prediction module employs a spatiotemporal convolutional neural network model, which predicts future evolution trends by integrating historical settlement patterns with real-time monitoring data. During model training, a ten-year engineering case library is loaded to learn the nonlinear characteristics of settlement development under different geological conditions. During online operation, parameters are updated incrementally every two hours to dynamically adapt to changes in soil creep and external loads. The prediction results include the settlement amount for the next six hours and its confidence interval. When the predicted value exceeds the warning threshold, a pre-intervention command is triggered, providing a scientific basis for proactive regulation.

[0016] Furthermore, the adaptive processing module includes a reinforcement learning decision submodule and a policy generation and policy library management submodule.

[0017] The reinforcement learning decision-making submodule continuously analyzes the foundation condition using deep reinforcement learning algorithms, including key indicators such as settlement, pore water pressure, and soil stress. Combined with future settlement trends provided by the prediction module, it constructs a multi-dimensional decision space. Its core lies in introducing a dynamic risk perception mechanism, which automatically adjusts the conservatism of the strategy based on real-time risk levels. When pore water pressure in a certain area approaches a critical value, the module rapidly increases the safety weight, prioritizing low-risk actions such as staged drainage over high-return but potentially more aggressive solutions that could exacerbate soil instability. Simultaneously, the module incorporates a long-term / short-term benefit balancing mechanism, achieving a balance between suppressing current settlement and preventing long-term risks, ensuring the treatment strategy is both effective and sustainable. The decision results are transmitted to downstream modules in real time, driving equipment execution and forming a closed-loop feedback loop for continuous strategy optimization.

[0018] The reinforcement learning decision-making submodule is used to continuously analyze the foundation state through deep reinforcement learning algorithms. The calculation formula for the dynamic safety weight reward function of the reinforcement learning decision-making submodule is as follows: , among which, S target S represents the target settlement threshold (preset safety value, such as 10mm). real C represents the real-time monitored settlement. energy P represents the normalized energy cost (0-1, 0 for no energy consumption). risk (t) represents the real-time risk probability, calculated from sensor data (such as the ratio of pore water pressure, soil stress to the critical value, ranging from 0 to 1); P safe λ(t) represents the risk safety threshold, λ(t) represents the dynamic risk weight coefficient, which is adaptively adjusted according to the foundation condition, γ represents the settlement control weight, with a default of 0.7, and η represents the energy consumption penalty weight, with a default of 0.2.

[0019] The strategy generation and strategy library management submodule is used for strategy parameter parsing and dynamic strategy adaptation. For example, by intelligently decomposing decision objectives, the "accelerate drainage" command is parsed into specific parameters such as the percentage increase in pump power and the valve switching frequency. Simultaneously, it combines historical successful cases and expert rule bases in the strategy library to quickly match the optimal strategy template under similar geological conditions. Its core lies in the dynamic strategy adaptation mechanism, which can comprehensively consider sensor data quality, equipment status, and environmental interference factors to adjust strategy parameters in real time. When a sudden decrease in soil permeability is detected in a certain area, the module automatically reduces the grouting pressure setpoint to prevent soil cracking and simultaneously calls the "gradual grouting in low-permeability areas" case from the strategy library for parameter correction. Furthermore, the module has a built-in strategy stability assessment function. By quantifying the success rate fluctuations of historical strategies in different scenarios, it prioritizes robust solutions and feeds the execution effect of new strategies back to the strategy library, forming a self-evolving closed loop.

[0020] The formula for calculating the dynamic policy adaptation index (USAI) of the policy generation and policy library management submodule is as follows: Among them, DataMatch represents the degree of matching between real-time monitoring data and the preset conditions of the strategy, ranging from [0,1]. It is calculated by dividing the real-time value by the preset threshold. If the real-time value is less than or equal to the threshold, the ratio is taken directly. If it exceeds the threshold, it is reset to zero. Stability represents the reliability and scene adaptability score of the historical strategy, ranging from [0,1]. EnvInterfere represents the influence strength of the external environment on the strategy execution, ranging from ≥0. SensorWeight represents the reliability score of sensor data, ranging from [0,1]. β represents the contribution weight of the strategy stability score, with a default value of 0.3. It forms a linear combination of the numerator with α. γ represents the suppression strength adjustment factor of the environmental interference coefficient, with a default value of 0.5. The larger the value, the more significant the interference effect. λ represents the amplification effect of the sensor weight on the final index, with a default value of 0.4. Nonlinear enhancement is achieved through the natural exponential function.

[0021] Furthermore, the adaptive processing module also includes a simulation verification and dynamic optimization submodule, which includes a digital twin model, a finite element simulation engine, a multi-objective optimization unit, and an adaptive correction mechanism unit.

[0022] The simulation verification and dynamic optimization submodule constructs a high-precision virtual foundation model using digital twin technology to simulate soil response and potential risks under different treatment strategies. This module dynamically updates model parameters based on real-time monitoring data, employs a finite element method to predict stress redistribution, settlement trends, and pore water pressure changes in the foundation after strategy implementation, and uses a multi-objective optimization engine to find the optimal solution by balancing settlement suppression, resource consumption, and construction safety. If simulation results indicate that a certain strategy may trigger a chain reaction, such as local soil liquefaction or instability in adjacent areas, the module automatically triggers an adaptive correction mechanism. It generates alternative solutions by combining historical case studies and expert experience. These new solutions are recalculated by the multi-objective optimization engine and returned to the simulation environment for secondary verification, forming an iterative process of "real-time early warning - dynamic optimization - closed-loop verification" until an optimal solution without derivative risks is output.

[0023] Furthermore, the adaptive processing module also includes a real-time control execution and feedback submodule, which includes a device instruction parsing unit, an industrial IoT protocol interface unit, a dynamic compensation controller, and a feedback learning loop, supporting multi-device collaborative control and closed-loop data feedback.

[0024] Furthermore, the interactive alarm module includes a 3D visualization unit, a multi-level alarm unit, and an electronic work order closed-loop management unit. The 3D visualization unit is used to visualize and simulate monitoring data in 3D, the multi-level alarm unit is used to issue risk alarms, and the electronic work order closed-loop management unit is used to record monitoring-related parameter data.

[0025] As can be seen from the above technical solution, the advantages of the present invention are: 1. This application significantly improves the safety management level of soft soil foundation engineering through the synergy of a multi-dimensional sensor network and an intelligent decision-making model. The information acquisition module, relying on high-precision sensing equipment and data fusion algorithms, can accurately capture subtle changes in foundation settlement, soil stress, and pore water pressure. Combined with an advanced prediction model, it identifies instability risks in advance, providing engineers with a minute-level early warning response window. The adaptive processing module dynamically generates optimal control strategies based on reinforcement learning. When pore water pressure rises abnormally, it automatically switches to a graded drainage scheme. This avoids the lag of traditional experience-based decision-making and mitigates the risks of blind construction through virtual simulation pre-verification, thereby reducing the incidence of sudden settlement accidents and significantly extending the service life of infrastructure.

[0026] 2. Through intelligent resource scheduling and a closed-loop self-optimization mechanism, a precise balance between engineering costs and safety benefits is achieved. The adaptive control strategy, while suppressing settlement, dynamically adjusts the operating parameters of grouting and drainage equipment, reducing redundant energy consumption and material waste. The multi-source data fusion mechanism effectively integrates geological survey data and real-time monitoring information, providing customized solutions for different geological conditions, avoiding resource waste caused by traditional reinforcement methods. The user interface intuitively presents the overall safety situation, supports rapid location of risk points and automatic push of disposal work orders, improving incident handling and management efficiency. Attached Figure Description

[0027] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0028] Figure 1 This is a schematic diagram of the automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion proposed in this application.

[0029] Figure 2 This is a schematic diagram of the adaptive processing module in this embodiment.

[0030] Figure 3 This is a schematic diagram of the interactive alarm module in this embodiment. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0032] refer to Figures 1 to 3This embodiment provides an automatic monitoring system for settlement of soft soil foundations based on multi-source data fusion. This system can identify instability risks in advance, improve the safety management level of soft soil foundation projects, reduce the incidence of sudden settlement accidents, reduce redundant energy consumption and material loss, and improve event handling and management efficiency. Figure 1 As shown, the system includes: an information acquisition module, a fusion processing module, a settlement prediction module, and an intelligent monitoring module. The information acquisition module collects foundation settlement sensor data and inputs it into the fusion processing module. The output of the information acquisition module is connected to the input of the preprocessing unit in the fusion processing module via an RS-485 and LoRa hybrid communication link. The output of the preprocessing unit is directly connected to the data integration interface of the multi-source data fusion unit in the fusion processing module via a standardized time-series data stream. The fusion processing module performs preprocessing and integration processing on the collected data. The output of the fusion processing module is connected to the spatiotemporal feature input port of the settlement prediction module via a physical constraint verification channel. The settlement prediction module is used to predict settlement of soft soil foundations. The intelligent monitoring module is connected to the fusion processing module and the settlement prediction module. The intelligent monitoring module includes an adaptive processing module and an interactive alarm module. The adaptive processing module generates optimization strategies and sends them to the real-time control execution and feedback submodule via a control command port. The prediction result output of the settlement prediction module is connected to the decision engine entry of the adaptive processing module. The interactive alarm module is used for user interaction and multi-mode abnormal alarm reminders. The alarm signal output of the interactive alarm module is connected in reverse to the strategy correction interface of the adaptive processing module, forming a closed-loop link from data acquisition to decision execution.

[0033] Specifically, the information acquisition module includes a distributed sensor network, which comprises multiple sensor nodes. Each node includes one or more of the following devices: a vibrating wire soil settlement gauge, a fiber optic pore water pressure sensor, a resistive earth pressure cell, and an inclinometer. The sensor network is deployed according to the geological risk level, with a dual-node redundancy design used in high-risk areas to ensure data continuity.

[0034] Specifically, high-risk areas refer to monitoring points with a risk index R > 0.7, thick layers of extremely highly compressible soft soil, large and unevenly distributed loads, drastic groundwater level fluctuations, and strong adjacent disturbances. In high-risk areas for soft soil foundation settlement, two independent and functionally identical sensor nodes are deployed at each key monitoring point. The two nodes use different power supplies and communication links; when the primary node fails, the backup node automatically takes over to ensure uninterrupted data acquisition. Simultaneously, the data from the two nodes are compared and verified in real time to improve the reliability of the monitoring results.

[0035] Geological risk assessment primarily quantifies the likelihood and potential severity of harmful settlement (especially uneven settlement) in soft soil foundations under load, requiring comprehensive consideration of geological conditions, load characteristics, and environmental factors. Geological risk assessment mainly includes the following three steps: 1. Data collection and basic parameter acquisition: Collect detailed engineering geological survey reports, obtain data on stratigraphic structure, physical and mechanical properties of soft soil, groundwater conditions, and historical settlement, and collect load, environmental, and site information; 2. Key risk indicator analysis and quantification: Analyze and quantify inherent vulnerability indicators (soft soil layer thickness, compressibility, water content, and void ratio) and external inducing factor indicators (load intensity and distribution, and groundwater level dynamics); 3. Risk level classification model: Based on engineering experience and sensitivity analysis, select core indicators, use the Analytic Hierarchy Process (AHP) or expert scoring method to determine the weight of each indicator, and calculate the comprehensive risk index through indicator normalization. Finally, set thresholds to classify the risk index into 3-4 levels (e.g., high risk, medium risk, and low risk).

[0036] The principle for differentiated deployment of sensor networks based on risk level is as follows: high-density, high-precision, all-round, and real-time monitoring in high-risk areas; moderate deployment in medium-risk areas; and simplified deployment or reliance on routine inspections in low-risk areas. Regarding sensor types, high-precision GNSS receivers, hydrostatic levels, fixed inclinometers, and pore water pressure gauges are deployed in high-risk areas to monitor indicators such as absolute surface displacement, relative surface displacement, and excess pore pressure dissipation. Hydrostatic levels and mobile inclinometers are deployed in medium-risk areas to measure relative surface displacement. In low-risk areas, a small number of hydrostatic levels or periodically used electronic levels are used to measure relative surface displacement.

[0037] In this embodiment, the information acquisition module captures key physical parameters of the foundation in real time through a distributed sensor network, including settlement, pore water pressure, soil stress, and tilt angle. Sensor nodes are deployed differently based on geological exploration results and risk levels. For example, high-risk areas utilize dense coverage with high-precision vibrating wire settlement gauges and fiber optic pressure sensors, while low-risk areas are equipped with economical capacitive sensors. All devices are networked using industrial-grade communication protocols, supporting wired transmission and wireless LoRa dual-channel redundant backup to ensure no data loss even in extreme environments. The module has a built-in self-test function that periodically scans the sensor health status, automatically switching abnormal nodes to backup devices. It also supports manual data entry during inspections, providing comprehensive raw data support for subsequent analysis.

[0038] The fusion processing module includes a preprocessing unit and a multi-source data fusion unit. The preprocessing unit integrates a wavelet transform algorithm and a dynamic threshold correction mechanism to perform noise filtering, outlier handling, and time-series alignment on the acquired data. The multi-source data fusion unit is connected to the preprocessing unit and is used to integrate and extract features from the preprocessed data using an improved Kalman filter framework, sensor dynamic weighting, and the Mohr-Coulomb physical constraint model.

[0039] In this embodiment, the preprocessing unit standardizes the raw sensor data to eliminate environmental noise and equipment drift interference. A sliding window filtering algorithm is used to smooth high-frequency noise, and box plots are used to identify outliers. Data exceeding three standard deviations are removed or corrected using linear interpolation. Multi-source heterogeneous data are synchronized with timestamps using a high-precision clock synchronization protocol to eliminate timing misalignment caused by sampling frequency differences. Finally, feature parameters such as settling acceleration and pore water pressure change rate are extracted to form a standardized time-series dataset, providing high-quality input for upper-level analysis.

[0040] The multi-source data fusion unit integrates multi-dimensional foundation state information based on an improved Kalman filter framework, addressing the limitations of single-sensor data. Leveraging the millimeter-level accuracy of settlement sensors and the spatial pointing characteristics of tilt sensors, information weights are dynamically allocated. For example, when soil shear deformation occurs, the weight of tilt data is automatically increased to detect signs of instability. The fusion process incorporates soil mechanics constitutive equations as physical constraints, ensuring that the data fusion results conform to the laws of soil deformation mechanics. Outputs include a comprehensive settlement index and derived indicators corresponding to the regional stability score, providing precise situational awareness for decision-makers.

[0041] The settlement prediction module includes an incremental learning unit, a confidence interval calculation engine, and an engineering case library. The incremental learning unit is used to predict future evolution trends by using a spatiotemporal convolutional neural network model and integrating historical settlement patterns with real-time monitoring data. It is also used to train the model based on the engineering case library and learn the nonlinear characteristics of settlement development under different geological conditions. During online operation, the parameters are updated incrementally according to a preset period to dynamically adapt to changes in soil creep and external loads. The confidence interval calculation engine is used to calculate the confidence interval.

[0042] In this embodiment, the settlement prediction module employs a spatiotemporal convolutional neural network model, which predicts future evolution trends by fusing historical settlement patterns with real-time monitoring data. During model training, a ten-year engineering case library is loaded to learn the nonlinear characteristics of settlement development under different geological conditions. During online operation, parameters are incrementally updated every two hours to dynamically adapt to changes in soil creep and external loads. The prediction results include the settlement amount for the next six hours and its confidence interval. When the predicted value exceeds the warning threshold, a pre-intervention command is triggered, providing a scientific basis for proactive adjustment.

[0043] Specifically, the adaptive processing module includes a reinforcement learning decision-making submodule and a strategy generation and strategy library management submodule. The reinforcement learning decision-making submodule continuously analyzes the foundation condition using deep reinforcement learning algorithms, including key indicators such as settlement, pore water pressure, and soil stress. It then combines this with future settlement trends provided by the prediction module to construct a multi-dimensional decision space. Its core lies in introducing a dynamic risk perception mechanism, which automatically adjusts the conservatism of the strategy based on the real-time risk level. For example, when the pore water pressure in a certain area is detected to be close to a critical value, the module quickly increases the safety weight, prioritizing low-risk actions such as staged drainage rather than high-return but potentially aggravating soil instability aggressive solutions. Simultaneously, it strikes a balance between suppressing current settlement and preventing long-term risks, ensuring that the processing strategy is both effective and sustainable. The decision results are transmitted to downstream modules in real time, driving equipment execution and forming a closed-loop feedback loop to achieve continuous strategy optimization. The reinforcement learning decision-making submodule is used to continuously analyze the foundation condition using deep reinforcement learning algorithms. The calculation formula for the dynamic safety weight reward function of the reinforcement learning decision-making submodule is as follows: , among which, S target S represents the target settlement threshold (preset safety value, such as 10mm). real C represents the real-time monitored settlement. energy P represents the normalized energy cost (0-1, 0 for no energy consumption). risk (t) represents the real-time risk probability, calculated from sensor data (such as the ratio of pore water pressure, soil stress to the critical value, ranging from 0 to 1); P safe λ(t) represents the risk safety threshold, λ(t) represents the dynamic risk weight coefficient, which is adaptively adjusted according to the foundation condition, γ represents the settlement control weight, with a default of 0.7, and η represents the energy consumption penalty weight, with a default of 0.2.

[0044] The strategy generation and strategy library management submodule is used for strategy parameter parsing and dynamic strategy adaptation. For example, by intelligently decomposing decision objectives, the "accelerate drainage" command is parsed into specific parameters such as the percentage increase in pump power and the frequency of valve switching. Simultaneously, it combines historical successful cases and expert rule bases in the strategy library to quickly match the optimal strategy template under similar geological conditions. Its core lies in the dynamic strategy adaptation mechanism, which comprehensively considers sensor data quality, equipment status, and environmental interference factors to adjust strategy parameters in real time. When a sudden decrease in soil permeability is detected in a certain area, the module automatically reduces the grouting pressure setpoint to prevent soil cracking and simultaneously calls the "gradual grouting in low-permeability areas" case from the strategy library for parameter correction. Furthermore, the module has a built-in strategy stability assessment function. By quantifying the success rate fluctuations of historical strategies in different scenarios, it prioritizes robust solutions and feeds the execution effect of new strategies back to the strategy library, forming a self-evolutionary closed loop. The dynamic strategy adaptation index (USAI) of the strategy generation and strategy library management submodule is calculated using the following formula: Among them, DataMatch represents the degree of matching between real-time monitoring data and the preset conditions of the strategy, ranging from [0,1]. It is calculated by dividing the real-time value by the preset threshold. If the real-time value is less than or equal to the threshold, the ratio is taken directly. If it exceeds the threshold, it is reset to zero. Stability represents the reliability and scene adaptability score of the historical strategy, ranging from [0,1]. EnvInterfere represents the influence strength of the external environment on the strategy execution, ranging from ≥0. SensorWeight represents the reliability score of sensor data, ranging from [0,1]. β represents the contribution weight of the strategy stability score, with a default value of 0.3. It forms a linear combination of the numerator with α. γ represents the suppression strength adjustment factor of the environmental interference coefficient, with a default value of 0.5. The larger the value, the more significant the interference effect. λ represents the amplification effect of the sensor weight on the final index, with a default value of 0.4. Nonlinear enhancement is achieved through the natural exponential function.

[0045] In this embodiment, the adaptive processing module further includes a simulation verification and dynamic optimization submodule and a real-time control execution and feedback submodule. The simulation verification and dynamic optimization submodule includes a digital twin model, a finite element simulation engine, a multi-objective optimization unit, and an adaptive correction mechanism unit. The simulation verification and dynamic optimization submodule constructs a high-precision virtual foundation model using digital twin technology to simulate soil response and potential risks under different processing strategies. This module dynamically updates model parameters based on real-time monitoring data, uses a finite element algorithm to predict stress redistribution, settlement trends, and pore water pressure changes in the foundation after strategy implementation, and uses a multi-objective optimization engine to find the optimal solution by balancing settlement suppression, resource consumption, and construction safety. If the simulation results show that a certain strategy may trigger a chain reaction, such as local soil liquefaction or instability in adjacent areas, the module automatically triggers the adaptive correction mechanism, generating alternative solutions by combining historical case libraries and expert experience. After recalculation by the multi-objective optimization engine, the new solution is returned to the simulation environment for secondary verification, forming an iterative process of "real-time early warning - dynamic optimization - closed-loop verification" until an optimal solution without derivative risks is output.

[0046] The real-time control execution and feedback submodule includes an equipment instruction parsing unit, an industrial IoT protocol interface unit, a dynamic compensation controller, and a feedback learning loop, supporting multi-device collaborative control and closed-loop data feedback. This submodule translates optimized strategies into executable equipment instructions and tracks execution effects in real-time. Through industrial IoT protocols, it precisely controls the start-up and shutdown timing and operational intensity of equipment such as drainage pumps and grouting machines, coordinating the actions of multiple devices synchronously to avoid conflicts, and temporarily closing drainage valves in adjacent areas before grouting to prevent grout loss. During execution, it collects equipment status data and foundation response indicators in real time, compares them with expected targets to calculate the deviation rate, and immediately triggers a dynamic compensation mechanism if the settlement suppression effect fails to reach the threshold or energy consumption exceeds limits, increasing pump power or extending operation time. All operation records and feedback data are encrypted and transmitted back to the decision-making center, driving model iteration and strategy library updates, forming a self-evolving closed loop of "strategy optimization – precise execution – effect verification."

[0047] The interactive alarm module includes a 3D visualization unit, a multi-level alarm unit, and an electronic work order closed-loop management unit. The 3D visualization unit presents the overall foundation safety status data through a visual cockpit, performs 3D geological model rendering and heat map risk labeling, and also visualizes settlement curves, equipment operating status, and predictive warning information. The multi-level alarm unit is used for risk alarms, and the electronic work order closed-loop management unit records historical monitoring data and generates comparative analysis reports. The user interface presents the overall foundation safety status through a visual cockpit, supports 3D geological model rendering and heat map risk labeling, and allows operators to view real-time settlement curves, equipment operating status, and predictive warning information in layers. Historical data is archived according to project stages, supporting comparative analysis and automatic report generation. The multi-level alarm unit adopts a tiered response mechanism: a yellow alert automatically pushes a message to the duty terminal; a red alert triggers an audible and visual alarm and directly connects to the emergency plan database, simultaneously generating an emergency work order containing location coordinates, risk level, and handling suggestions, ensuring the response process is initiated within five minutes. This application improves the safety management level of soft soil foundation engineering, identifies instability risks in advance by combining predictive models, provides an early warning and response window for engineers, and reduces the incidence of sudden settlement accidents.

[0048] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to the embodiments of this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion, characterized in that, include: Information acquisition module, fusion processing module, settlement prediction module, and intelligent monitoring module; The information acquisition module is used to collect foundation settlement sensor data and input the collected data into the fusion processing module; the fusion processing module is used to preprocess and integrate the collected data; the output of the fusion processing module is connected to the spatiotemporal feature input port of the settlement prediction module, which is used to predict the settlement of soft soil foundations; the intelligent monitoring module is connected to the fusion processing module and the settlement prediction module, and includes an adaptive processing module and an interactive alarm module. The adaptive processing module is used to generate optimization strategies and send them to the real-time control execution and feedback submodule through the control command port; the interactive alarm module is used for user interaction and multi-mode abnormal alarm reminders.

2. The automatic monitoring system for soft soil foundation settlement based on multi-source data fusion according to claim 1, characterized in that, The information acquisition module includes a distributed sensor network, which includes multiple sensor nodes. Each node includes one or more of the following devices: a vibrating wire soil settlement gauge, a fiber optic pore water pressure sensor, a resistive earth pressure cell, and an inclinometer.

3. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 1, characterized in that, The fusion processing module includes a preprocessing unit and a multi-source data fusion unit; the preprocessing unit is used to perform noise filtering, outlier processing and time-series alignment on the acquired data based on wavelet transform algorithm and dynamic threshold correction mechanism. The multi-source data fusion unit is connected to the preprocessing unit. The multi-source data fusion unit is used to integrate and extract features from the preprocessed data using an improved Kalman filter framework, sensor dynamic weighting, and the Mohr-Coulomb physical constraint model.

4. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 1, characterized in that, The settlement prediction module includes an incremental learning unit, a confidence interval calculation unit, and an engineering case library. The incremental learning unit is used to predict future evolution trends by using a spatiotemporal convolutional neural network model and integrating historical settlement patterns with real-time monitoring data. It is also used to train the model based on the engineering case library and learn the nonlinear characteristics of settlement development under different geological conditions. During online operation, the parameters are updated incrementally according to a preset period to dynamically adapt to changes in soil creep and external loads. The confidence interval calculation unit is used to calculate the confidence interval.

5. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 1, characterized in that, The adaptive processing module includes a reinforcement learning decision submodule and a policy generation and policy library management submodule; The reinforcement learning decision-making submodule is used to analyze the foundation state through deep reinforcement learning algorithms and automatically adjust the conservatism of the strategy according to the real-time risk level. The calculation formula of the dynamic safety weight reward function of the reinforcement learning decision-making submodule is as follows: , of which S target S represents the target settlement threshold. real C represents the real-time monitored settlement. energy P represents the normalized energy cost. risk (t) represents the real-time risk probability, calculated from sensor data; P safe λ(t) represents the risk safety threshold, λ(t) represents the dynamic risk weight coefficient, which is adaptively adjusted according to the changes in foundation condition, γ represents the settlement control weight, and η represents the energy consumption penalty weight. The strategy generation and strategy library management submodule is used for strategy parameter parsing and dynamic strategy adaptation. The formula for calculating the dynamic strategy adaptation index (USAI) of the strategy generation and strategy library management submodule is as follows: Among them, DataMatch represents the degree of matching between real-time monitoring data and the preset conditions of the strategy, ranging from [0,1]. It is calculated by dividing the real-time value by the preset threshold. If the real-time value is less than or equal to the threshold, the ratio is taken directly. If it exceeds the threshold, it is reset to zero. Stability represents the reliability and scene adaptability score of the historical strategy, ranging from [0,1]. EnvInterfere represents the influence strength of the external environment on the strategy execution, ranging from ≥0. SensorWeight represents the sensor data credibility score, ranging from [0,1]. β represents the contribution weight of the strategy stability score, which together with α constitutes a linear combination of the numerator. γ represents the suppression strength adjustment factor of the environmental interference coefficient. The larger the value, the more significant the interference effect. λ represents the amplification effect of the sensor weight on the final index, which is achieved through the natural exponential function to achieve nonlinear enhancement.

6. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 5, characterized in that, The adaptive processing module also includes a simulation verification and dynamic optimization submodule, which includes a digital twin model, a finite element simulation unit, a multi-objective optimization unit, and an adaptive correction mechanism unit. The digital twin model is used to construct a virtual foundation model to simulate the soil response and potential risks under different processing strategies. The finite element simulation unit is used to predict the stress redistribution, settlement trend, and pore water pressure changes of the foundation after the finite element algorithm is implemented; the multi-objective optimization unit is used to calculate the multi-objective optimization solution for suppressing settlement, resource consumption, and construction safety; the adaptive correction mechanism unit is used to generate alternative solutions by combining historical case database and expert experience data when the adaptive correction mechanism is triggered.

7. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 5, characterized in that, The adaptive processing module further includes a real-time control execution and feedback submodule, which includes a device instruction parsing unit, an industrial IoT protocol interface unit, a dynamic compensation controller, and a feedback learning loop. The device instruction parsing unit is used to convert the optimization strategy into device executable instructions; The industrial IoT protocol interface unit is used to control the start-up and stop sequence and operation intensity of drainage pumps, grouting machines and related industrial equipment through industrial IoT protocols, and to coordinate the actions of multiple devices synchronously. The dynamic compensation controller is used to acquire equipment status data and foundation response indicators in real time, and calculate the deviation rate by comparing them with the expected target. If the settlement inhibition data does not reach the threshold or the energy consumption exceeds the limit, the dynamic compensation mechanism is immediately triggered to increase the pump power or extend the operation time. The feedback learning loop is used to encrypt and transmit all operation records and feedback data back to the intelligent monitoring module, driving model iteration and strategy library updates.

8. The automatic monitoring system for settlement of soft soil foundation based on multi-source data fusion according to claim 1, characterized in that, The interactive alarm module includes a 3D visualization unit, a multi-level alarm unit, and an electronic work order closed-loop management unit. The 3D visualization unit is used to display the overall foundation safety status data through a visualization cockpit, and to perform 3D geological model rendering and heat map risk labeling. It is also used to visualize settlement curves, equipment operating status, and predictive early warning information. The multi-level alarm unit is used to perform risk alarms using a graded response mechanism. The electronic work order closed-loop management unit is used to record historical monitoring data, generate emergency work orders, and comparative analysis reports.

Images