A real-time monitoring and second-level early warning system and method for deep foundation pit confined water

Abstract

The application discloses a kind of for deep foundation pit confined water real-time monitoring and second-level early warning system and method, constructs " perception layer-edge computing layer-cloud analysis layer " three layers of internet of things architecture: perception layer deployment miniaturization low-power IoT sensor array, real-time acquisition TDS, pH, water level and other key parameters;Edge computing layer embeds TinyML lightweight model, realizes data local cleaning, abnormality preliminary screening and instant preliminary warning;Cloud deployment deep learning model, carry out the comprehensive risk research and judge of multi-source data fusion.System uses LoRa / NB-IoT low-power wide-area network to realize data transmission, establish " edge end instant warning + cloud complex research and judge " two-level early warning system, realize 7×24 hours uninterrupted monitoring and second-level early warning push.The application effectively solves the problem of traditional monitoring hysteresis, greatly reduces hardware and operation and maintenance cost, and can be widely applied to deep foundation pit engineering confined water safety monitoring.

Description

Technical Field

[0001] This invention relates to the field of underground engineering safety monitoring technology, specifically to a real-time monitoring and second-level early warning system and method for confined aquifers in deep foundation pits, applicable to 24 / 7 uninterrupted monitoring and intelligent early warning of water level and water quality parameters of confined aquifers during the construction of deep foundation pit projects. Background Technology

[0002] Deep foundation pit projects in Shanghai often involve multiple confined aquifers, specifically layers 7 and 9. These layers have high water heads and large water volumes, making cross-layer leakage or sudden surges extremely serious. A key prerequisite for the safe management of confined aquifers is achieving real-time, accurate monitoring and rapid early warning of critical parameters such as water level and quality.

[0003] Currently, monitoring of confined water in deep foundation pits mainly relies on periodic manual sampling and laboratory analysis, which presents the following prominent problems:

[0004] (i) The timeliness of monitoring is seriously insufficient. Traditional monitoring mainly relies on periodic manual collection, with a low sampling frequency (usually 1 to 2 times per day), which cannot achieve continuous and uninterrupted monitoring and makes it difficult to capture instantaneous abnormal signals at the initial stage of leakage, resulting in a serious delay in early warning.

[0005] (ii) The early warning methods are singular and lack hierarchical classification. Existing early warning systems are mostly based on simple threshold alarms of a single indicator, resulting in a high false alarm rate. They lack a hierarchical early warning mechanism that links multiple indicators and cannot distinguish between sensor noise and real danger.

[0006] (iii) Data processing relies on the cloud, and the system becomes unusable when the network is down. The existing system relies on a remote cloud platform for data processing, analysis and evaluation. Sensors and data acquisition instruments only perform simple data acquisition. Once the network is interrupted, the monitoring system will not be able to work properly.

[0007] (iv) High hardware costs and difficult deployment. Traditional monitoring equipment is large in size, consumes a lot of power, requires on-site power supply and wired transmission, resulting in high deployment costs and long cycles, making it difficult to achieve dense monitoring in a wide range of locations.

[0008] To address the aforementioned issues, several patents have proposed improvement solutions. For example, patent document CN121482970A proposes a method and system for monitoring and warning of disasters during foundation pit construction, which utilizes edge computing devices to verify monitoring data before uploading it to a cloud platform for multi-algorithm analysis; patent document CN121545331A discloses a method and system for intelligent monitoring of deep foundation pits based on multi-source data fusion, which collects multi-physics field data through a distributed sensor network and uses edge computing nodes for preprocessing; another patent proposes an automated monitoring and early warning system for deep foundation pits based on edge computing, including a data acquisition terminal, an edge gateway, and an early warning cloud platform.

[0009] However, the above-mentioned existing technologies still have the following shortcomings: (1) The sensor selection and deployment scheme is not specifically optimized for pressurized water monitoring, and there is a lack of comprehensive perception capabilities for water quality parameters such as TDS, pH, and conductivity; (2) The edge end only performs data verification and simple processing, and does not embed a lightweight artificial intelligence model to realize local intelligent reasoning and real-time early warning; (3) A two-level early warning system of "real-time early warning at the edge end + complex analysis in the cloud" has not been established, and the second-level response capability for sudden anomalies is insufficient; (4) Low-power wide area network (LoRa / NB-IoT) and miniaturized sensor design are not adopted, and the hardware cost and operation and maintenance cost are still relatively high.

[0010] Therefore, there is an urgent need to develop a real-time monitoring and second-level early warning system for pressurized water that integrates IoT sensing, edge intelligent computing, and cloud-based deep analysis. Summary of the Invention

[0011] This invention addresses the technical challenges of insufficient timeliness, slow early warning response, and high hardware costs in existing deep foundation pit confined water monitoring systems. It provides a real-time monitoring and second-level early warning system and method for deep foundation pit confined water, aiming to achieve a fundamental shift from "passive data collection" to "active sensing and real-time early warning," and to build an all-weather, comprehensive monitoring network.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: A real-time monitoring and second-level early warning system for confined water in deep foundation pits includes: a sensing layer, an edge computing layer, a data transmission layer, a cloud analysis layer, and an early warning execution module. The sensing layer consists of a miniaturized low-power IoT sensor array deployed in the monitoring well of the foundation pit. The sensor array includes a water level sensor, a TDS sensor, a pH sensor, a temperature sensor, and a conductivity sensor, which are used to collect hydrological and hydrochemical parameters of the pressurized water in real time. The edge computing layer is communicatively connected to the perception layer, including an edge gateway and a TinyML lightweight inference model embedded therein, which is used to perform local cleaning, outlier removal and feature extraction on the raw data collected by the sensor, and to perform initial screening of anomalies and immediate preliminary warning judgment based on the lightweight model. The data transmission layer uses LoRa or NB-IoT low-power wide-area network protocols to upload the data processed by the edge computing layer to the cloud analysis layer; The cloud analysis layer is equipped with a deep learning comprehensive judgment model, which is used to receive data uploaded by the edge computing layer, perform multi-source heterogeneous data fusion analysis and complex trend prediction, and output comprehensive risk level and early warning instructions. The early warning execution module is communicatively connected to the edge computing layer and the cloud analysis layer, and is used to push early warnings in seconds through SMS, APP push, platform pop-up and sound and light alarms based on the results of the initial early warning at the edge or the comprehensive early warning in the cloud.

[0013] Furthermore, each sensor in the sensor array of the sensing layer adopts a low-power design, with an operating current ≤10mA, a standby current ≤50μA, a built-in battery life of ≥6 months, and supports solar-assisted power supply.

[0014] Furthermore, the TinyML lightweight model embedded in the edge computing layer adopts a quantized decision tree or lightweight neural network model with a model size ≤100KB and a single inference latency ≤50ms. The primary warning adopts an anomaly detection method based on dynamic thresholds, which triggers an immediate primary warning when any monitoring parameter exceeds the preset dynamic threshold range.

[0015] Furthermore, the data transmission layer uses LoRa or NB-IoT protocols for low-power wide-area data transmission, with a data reporting frequency of once every 5 to 15 minutes. When the edge triggers a primary warning, it automatically switches to encrypted reporting mode, increasing the reporting frequency to once every 10 to 30 seconds.

[0016] Furthermore, the deep learning comprehensive judgment model of the cloud analysis layer is a hybrid deep learning model that integrates LSTM time series prediction network and attention mechanism. It is used to comprehensively identify and judge the risk level of water level change trend, TDS and conductivity abnormal evolution, and multi-parameter coupling abnormal patterns, and output four-level warnings: green, yellow, orange and red.

[0017] Furthermore, the edge computing layer and the cloud analysis layer constitute a two-level early warning system: the first level is an immediate primary early warning at the edge, which is completed locally at the data acquisition node based on the TinyML model, with a response time of ≤1 second, and is used for immediate alarms of sudden anomalies; the second level is a comprehensive judgment by a complex model in the cloud, which performs trend analysis and risk prediction on multi-source data based on a deep learning model, with a response time of ≤30 seconds, and is used for accurate identification and graded early warning of complex leakage risks.

[0018] Furthermore, the edge gateway has the ability to operate autonomously when the network is interrupted. When communication with the cloud analysis layer is interrupted, the edge gateway can operate independently for no less than 72 hours, store monitoring data locally, and automatically retransmit it after the network is restored.

[0019] Furthermore, in the sensor array of the sensing layer, the water level sensor adopts a diffused silicon pressure water level gauge with a range of 0 to 100 m and an accuracy of ±0.05%FS; the TDS sensor adopts the conductivity method with a range of 0 to 5000 mg / L and an accuracy of ±1%FS; and the pH sensor adopts the glass electrode method with a range of 0 to 14 and an accuracy of ±0.1pH.

[0020] Furthermore, the early warning execution module also includes a GIS / BIM-based 3D visualization display submodule, which is used to mark the status and risk level of each monitoring point in the 3D model of the foundation pit in real time, automatically identify high-risk areas and generate electronic fences.

[0021] A method for real-time monitoring and second-level early warning of confined water in deep foundation pits using the aforementioned system includes the following steps: Step 1: Deploy a miniaturized low-power IoT sensor array inside the foundation pit monitoring well, connect the power supply and complete communication debugging, and start 24 / 7 uninterrupted data acquisition; Step 2: The sensor array collects water level, TDS, pH, temperature and conductivity parameters in real time, and transmits the raw data to the edge gateway via LoRa or NB-IoT protocol; Step 3: After receiving the raw data, the edge gateway uses the embedded TinyML lightweight model to perform local cleaning, outlier removal, and feature extraction on the data, and performs initial screening for anomalies based on dynamic thresholds. If an anomaly is detected, the first-level immediate primary warning is triggered immediately, and the warning execution module pushes the alarm information within seconds. Step 4: The edge gateway encapsulates the cleaned data and uploads it to the cloud analysis layer via LoRa or NB-IoT network; Step 5: The deep learning comprehensive judgment model of the cloud analysis layer performs fusion analysis and trend prediction on multi-source data, outputs a comprehensive risk level and early warning instruction, and triggers the second-level comprehensive early warning; Step Six: The early warning execution module pushes early warning information through multiple channels based on the results of the first or second level early warning, and updates the risk status and high-risk area identification in the GIS / BIM 3D model in real time.

[0022] Compared with the prior art, the present invention has the following significant advantages: 1. Achieve 24 / 7 uninterrupted monitoring. This invention utilizes a miniaturized low-power IoT sensor array, combined with LoRa / NB-IoT low-power wide-area networks, to achieve continuous and automatic acquisition of pressurized water level and water quality parameters, completely eliminating the time limitations of traditional manual sampling.

[0023] 2. Achieve second-level early warning response. This invention embeds a lightweight TinyML model into the edge gateway, enabling initial anomaly screening and immediate early warning locally at the data acquisition node, with a response time of ≤1 second. This solves the problem of delayed early warning caused by traditional monitoring relying on cloud analysis.

[0024] 3. Establish a two-tier early warning system that balances speed and accuracy. This invention features a unique two-tier early warning system that combines real-time edge warning with complex cloud-based analysis. The edge ensures a second-level response to sudden anomalies, while the cloud ensures accurate identification of complex leakage risks, achieving a balance between speed and accuracy.

[0025] 4. Significantly reduces hardware and maintenance costs. This invention uses commercially available low-cost sensors, a low-power design (standby current ≤50μA), and a low-power wide-area network (LoRa / NB-IoT), eliminating the need for on-site power supply lines and wired networks. The built-in battery provides a battery life of ≥6 months, significantly reducing hardware investment and subsequent maintenance costs.

[0026] 5. Possesses autonomous operation capability during network outages. The edge gateway of this invention can operate independently for no less than 72 hours under network interruption conditions, ensuring uninterrupted monitoring and effective early warning, significantly improving the reliability and engineering practicality of the system.

[0027] 6. Upgrading from threshold alarms to intelligent trend prediction. This invention integrates edge-side TinyML fast inference with cloud-based deep learning model analysis, achieving an upgrade from simple threshold alarms to intelligent prediction based on trends, effectively reducing the false alarm rate. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall architecture of the real-time monitoring and second-level early warning system for confined water in deep foundation pits according to the present invention; Figure 2 This is a schematic diagram of the sensor array layout in the sensing layer of the present invention; Figure 3 This is a flowchart of the TinyML model inference for the edge computing layer of this invention; Figure 4 This is a schematic diagram of the workflow of the two-level early warning system of the present invention; Figure 5 This is an overall flowchart of the monitoring and early warning method of the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0030] like Figures 1 to 3 As shown in the figure, the present invention provides a real-time monitoring and second-level early warning system for confined water in deep foundation pits, including a sensing layer, an edge computing layer, a data transmission layer, a cloud analysis layer, and an early warning execution module.

[0031] (a) Perception layer The sensing layer consists of a miniaturized, low-power IoT sensor array deployed within the monitoring well of the foundation pit. The sensor array includes: Water level sensor: A diffused silicon pressure water level gauge is used, with a range of 0 to 100 m and an accuracy of ±0.05%FS, for real-time monitoring of changes in the water level of confined water. TDS sensor: Employs conductivity method, range 0~5000mg / L, accuracy ±1%FS, used to monitor total dissolved solids content; pH sensor: Employs glass electrode method, range 0-14, accuracy ±0.1pH, used to monitor the acidity and alkalinity of water. Temperature sensor: Used to monitor water temperature and assist in calibrating readings from other sensors; Conductivity sensor: Used to monitor the electrical conductivity of water and help determine changes in water quality.

[0032] Each sensor adopts a low-power design with an operating current of ≤10mA, a standby current of ≤50μA, a built-in battery life of ≥6 months, and supports solar-assisted power supply, which greatly reduces on-site power supply dependence and operation and maintenance costs.

[0033] (ii) Edge computing layer The edge computing layer includes edge gateways and an embedded TinyML lightweight inference model. The edge gateways are deployed on-site at the excavation site and connect wirelessly to the sensors in the perception layer. The TinyML lightweight model built into the edge gateways uses a quantized decision tree or lightweight neural network model, with a model size ≤100KB and a single inference latency ≤50ms.

[0034] The functions of the edge computing layer include: (1) Local data cleaning: outlier removal, missing value imputation and noise filtering are performed on the raw data collected by the sensor; (2) Feature extraction: Extract key features (rate of change, extreme values, spectral features, etc.) from the original time series data; (3) Initial screening of anomalies: An anomaly detection method based on dynamic thresholds is used to trigger an immediate primary warning when any monitored parameter exceeds the preset dynamic threshold range. The dynamic threshold is automatically updated based on a sliding window statistical analysis of historical data (mean ± 3 times the standard deviation) to adapt to water level changes at different construction stages; (4) Autonomous operation during network outage: The edge gateway has the ability to operate autonomously during network outage. When communication with the cloud analysis layer is interrupted, it can run independently for no less than 72 hours, store monitoring data locally and automatically retransmit it after the network is restored.

[0035] (III) Data Transmission Layer The data transmission layer uses LoRa or NB-IoT low-power wide-area network protocols to achieve wireless data transmission between the edge gateway and the cloud analytics layer. The data reporting frequency is once every 5 to 15 minutes; when a primary warning is triggered at the edge, it automatically switches to encrypted reporting mode, and the reporting frequency is increased to once every 10 to 30 seconds.

[0036] The selection of LoRa / NB-IoT technology has the following advantages: low power consumption (microampere level in terminal standby power consumption), wide coverage (coverage radius of 1-5km in urban environment), and deep penetration (can penetrate multiple underground structures), making it particularly suitable for monitoring and data transmission in deep foundation pit underground spaces.

[0037] (iv) Cloud-based analytics layer The cloud-based analytics layer deploys a deep learning-based comprehensive judgment model to receive data uploaded from the edge computing layer and perform multi-source heterogeneous data fusion analysis and complex trend prediction. This deep learning-based comprehensive judgment model is a hybrid deep learning model that integrates an LSTM time series prediction network and an attention mechanism, and is used for: (1) Time-series prediction and anomaly identification of water level change trends; (2) Pattern recognition of the abnormal evolution of TDS and conductivity; (3) Comprehensive identification of multi-parameter coupling anomaly patterns; (4) Output four warning levels: green (normal), yellow (attention), orange (warning), and red (danger).

[0038] (v) Early warning execution module The early warning execution module communicates with the edge computing layer and cloud analytics layer to send push notifications within seconds based on early warning results. Push methods include SMS, app push notifications, platform pop-ups, and on-site audible and visual alarms. The early warning execution module also includes a GIS / BIM-based 3D visualization submodule, used to annotate the status and risk level of each monitoring point in the 3D model of the foundation pit in real time, automatically identify high-risk areas, and generate electronic fences.

[0039] like Figure 4 As shown, this embodiment of the invention also proposes a two-tier early warning system that combines "real-time primary early warning at the edge + comprehensive analysis using complex models in the cloud": Level 1 (Edge-end Real-time Early Warning): Based on the TinyML model, reasoning and judgment are completed locally at the data acquisition node with a response time of ≤1 second. It is used for real-time alarms of sudden anomalies (such as sudden changes in water level or sudden changes in TDS), ensuring that early warnings can still be issued even under network interruption conditions. Level 2 (Cloud-based Comprehensive Analysis): Based on deep learning models, multi-source data is fused and analyzed for trend prediction, with a response time of ≤30 seconds, for accurate identification and graded early warning of complex leakage risks.

[0040] like Figure 5 As shown, this embodiment of the invention also provides a method for real-time monitoring and second-level early warning of confined water in deep foundation pits using the above-described system, comprising the following steps: Step 1: Deployment of the Perception Layer A sensor array was deployed in the confined water observation wells around and inside the foundation pit, with a total of 12 monitoring points (8 around the pit and 4 inside). Each monitoring point was equipped with an integrated sensor probe, including a water level sensor (diffuse silicon pressure type, range 0–50m), a TDS sensor (conductivity method, range 0–3000mg / L), a pH sensor (glass electrode method, range 0–14), a temperature sensor, and a conductivity sensor. Each sensor was connected to the data acquisition terminal via a waterproof connector and was powered by an internal battery with a designed battery life of 8 months.

[0041] Step 2: Deployment of the edge computing layer Three edge gateways (as backups) are deployed at the foundation pit construction site, with each edge gateway covering data access for 4-6 monitoring points. The edge gateways embed a TinyML lightweight model and pre-configure dynamic threshold calculation logic and basic early warning rules for each monitoring parameter.

[0042] After the edge gateway is deployed, communication debugging is performed to confirm that the wireless communication between each sensor and the edge gateway is normal, and that the LoRa / NB-IoT network between the edge gateway and the cloud platform is connected.

[0043] Step 3: System startup and 24 / 7 monitoring After the system starts up, each sensor collects water level, TDS, pH, temperature, and conductivity data every 5 minutes, and transmits them to the edge gateway in real time via wireless communication. After receiving the data, the edge gateway performs local cleaning (removing obvious outliers and smoothing noise), extracts key features (water level change rate, TDS change rate, etc.), and performs initial screening for anomalies based on dynamic thresholds.

[0044] Step 4: Real-time early warning at the edge During a monitoring session, the water level in the observation well on the north side of the foundation pit changed at a rate far exceeding the dynamic threshold (normal rate of change <0.05 m / min). The TinyML model of the edge gateway identified the anomaly within 50 ms, determining it to be an "abnormal sudden drop in water level," and immediately triggered the first-level immediate primary warning. Within 1 second, the warning execution module sent an alarm message to the project management personnel via SMS and APP push: "The water level in a certain observation well in the foundation pit has dropped abnormally. Please investigate immediately."

[0045] Step 5: Comprehensive Analysis in the Cloud The edge gateway uploads the cleaned data (including anomaly markers) to the cloud analysis layer via the NB-IoT network. A cloud-based deep learning model (LSTM + attention mechanism) fuses and analyzes the multi-source data, comprehensively assessing the water level change trend and data from surrounding monitoring points. The model outputs an orange alert, identifying it as "suspected leakage from layer 7 to layer 9, risk level: high," and automatically generates a response suggestion: "Immediately check the integrity of the north-side water-stop curtain and prepare a grouting sealing plan."

[0046] Step Six: Early Warning Push and 3D Visualization The early warning execution module sends orange alert results to the project manager, technical manager, and safety director through multiple channels, including platform pop-ups, app push notifications, and SMS. Simultaneously, the BIM 3D visualization module marks observation wells as orange high-risk areas in the 3D model of the foundation pit, automatically generates electronic fences, and highlights abnormal data curves and trend prediction results on the monitoring screen.

[0047] Based on the early warning information, the engineering unit immediately organized an investigation and discovered localized seepage points in the cutoff wall. Grouting and sealing measures were promptly implemented, successfully preventing a sudden surge in the foundation pit. From the occurrence of the anomaly to the early warning notification, the entire process took approximately 1.5 seconds (edge-end early warning) to 35 seconds (cloud-based comprehensive analysis), representing a significant leap forward compared to traditional manual sampling and monitoring (which takes several hours).

[0048] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the concept and scope of the present invention. Therefore, all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A real-time monitoring and second-level early warning system for confined water in deep foundation pits, characterized in that, include: The system consists of a perception layer, an edge computing layer, a data transmission layer, a cloud analytics layer, and an early warning execution module. The sensing layer consists of a miniaturized low-power IoT sensor array deployed in the monitoring well of the foundation pit. The sensor array includes a water level sensor, a TDS sensor, a pH sensor, a temperature sensor, and a conductivity sensor, which are used to collect hydrological and hydrochemical parameters of the pressurized water in real time. The edge computing layer is communicatively connected to the perception layer, including an edge gateway and a TinyML lightweight inference model embedded therein, which is used to perform local cleaning, outlier removal and feature extraction on the raw data collected by the sensor, and to perform initial screening of anomalies and immediate preliminary warning judgment based on the lightweight model. The data transmission layer uses LoRa or NB-IoT low-power wide-area network protocols to upload the data processed by the edge computing layer to the cloud analysis layer; The cloud analysis layer is equipped with a deep learning comprehensive judgment model, which is used to receive data uploaded by the edge computing layer, perform multi-source heterogeneous data fusion analysis and complex trend prediction, and output comprehensive risk level and early warning instructions. The early warning execution module is communicatively connected to the edge computing layer and the cloud analysis layer, and is used to push early warnings in seconds through SMS, APP push, platform pop-up and sound and light alarms based on the results of the initial early warning at the edge or the comprehensive early warning in the cloud.

2. The system according to claim 1, characterized in that, Each sensor in the sensor array of the sensing layer adopts a low-power design, with an operating current of ≤10mA, a standby current of ≤50μA, a built-in battery life of ≥6 months, and supports solar-assisted power supply.

3. The system according to claim 1, characterized in that, The TinyML lightweight model embedded in the edge computing layer adopts a quantized decision tree or lightweight neural network model with a model size ≤100KB and a single inference latency ≤50ms. The primary warning adopts an anomaly detection method based on dynamic thresholds, which triggers an immediate primary warning when any monitored parameter exceeds the preset dynamic threshold range.

4. The system according to claim 1, characterized in that, The data transmission layer uses LoRa or NB-IoT protocols for low-power wide-area data transmission, with a data reporting frequency of once every 5 to 15 minutes. When the edge triggers a primary warning, it automatically switches to encrypted reporting mode, increasing the reporting frequency to once every 10 to 30 seconds.

5. The system according to claim 1, characterized in that, The deep learning comprehensive judgment model of the cloud analysis layer is a hybrid deep learning model that integrates LSTM time series prediction network and attention mechanism. It is used to comprehensively identify and judge the risk level of water level change trend, TDS and conductivity abnormal evolution, and multi-parameter coupling abnormal patterns, and output four-level warnings: green, yellow, orange and red.

6. The system according to claim 1, characterized in that, The edge computing layer and the cloud analysis layer constitute a two-level early warning system: the first level is an immediate primary early warning at the edge, which is completed locally at the data acquisition node based on the TinyML model, with a response time of ≤1 second, and is used for immediate alarms of sudden anomalies; The second level is a comprehensive analysis of complex cloud models, which uses deep learning models to perform trend analysis and risk prediction on multi-source data with a response time of ≤30 seconds. This is used for the accurate identification and graded early warning of complex leakage risks.

7. The system according to claim 1, characterized in that, The edge gateway has the ability to operate autonomously when the network is interrupted. When communication with the cloud analysis layer is interrupted, the edge gateway can run independently for no less than 72 hours, store monitoring data locally and automatically retransmit it after the network is restored.

8. The system according to claim 1, characterized in that, In the sensor array of the sensing layer, the water level sensor adopts a diffused silicon pressure water level gauge with a range of 0-100m and an accuracy of ±0.05%FS; the TDS sensor adopts the conductivity method with a range of 0-5000mg / L and an accuracy of ±1%FS; and the pH sensor adopts the glass electrode method with a range of 0-14 and an accuracy of ±0.1pH.

9. The system according to claim 1, characterized in that, The early warning execution module also includes a GIS / BIM-based 3D visualization display submodule, which is used to mark the status and risk level of each monitoring point in the 3D model of the foundation pit in real time, automatically identify high-risk areas and generate electronic fences.

10. A method for real-time monitoring and second-level early warning of confined water in deep foundation pits using the system described in any one of claims 1 to 9, characterized in that, Includes the following steps: Step 1: Deploy a miniaturized low-power IoT sensor array inside the foundation pit monitoring well, connect the power supply and complete communication debugging, and start 24 / 7 uninterrupted data acquisition; Step 2: The sensor array collects water level, TDS, pH, temperature and conductivity parameters in real time, and transmits the raw data to the edge gateway via LoRa or NB-IoT protocol; Step 3: After the edge gateway receives the raw data, the embedded TinyML lightweight model performs local cleaning, outlier removal, and feature extraction on the data, and performs initial screening of anomalies based on dynamic thresholds. If an anomaly is detected, the first-level immediate primary warning is immediately triggered, and the warning execution module pushes the alarm information within seconds; Step 4: The edge gateway encapsulates the cleaned data and uploads it to the cloud analysis layer via LoRa or NB-IoT network; Step 5: The deep learning comprehensive judgment model of the cloud analysis layer performs fusion analysis and trend prediction on multi-source data, outputs a comprehensive risk level and early warning instruction, and triggers the second-level comprehensive early warning; Step Six: The early warning execution module pushes early warning information through multiple channels based on the results of the first or second level early warning, and updates the risk status and high-risk area identification in the GIS / BIM 3D model in real time.

Images