Deep foundation pit safety early warning escape system based on multi-source data deep learning

Abstract

The invention discloses a deep foundation pit safety early warning escape system based on multi-source data deep learning, and the system comprises a multi-source monitoring module which is composed of multi-source data collection equipment disposed in a deep foundation pit and is used for collecting multi-source data in the deep foundation pit; the early warning platform is in communication connection with the multi-source monitoring module through the data transmission module; the early warning platform is internally provided with a deep learning model, and the deep learning model is used for carrying out fusion analysis on multi-source data, predicting a foundation pit deformation trend and a safety risk level, and outputting a graded early warning signal; and the escape execution module is in communication connection with the early warning platform and is used for responding to the early warning signal and automatically starting an escape program. Through deep fusion of multi-source data, development of the early warning platform and design of the escape execution module, the safety and reliability of deep foundation pit construction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep foundation pit escape technology, and in particular to a deep foundation pit safety early warning escape system based on multi-source data deep learning. Background Art

[0002] With the continuous acceleration of urbanization, urban ground space is becoming increasingly scarce, and seeking new development space in underground space has become an inevitable trend. The scale of high-rise buildings and underground projects continues to expand, and the number of deep foundation pit projects has also increased accordingly. Especially in densely populated and busy urban center areas, deep foundation pit construction faces multiple challenges from complex geological conditions and the surrounding environment. As a typical example, the deep foundation pit with jacking is subjected to complex forces and has low stability during construction. Accidents such as foundation pit collapse, sudden water and mud gushing, etc., are very likely to occur due to improper design or construction. These accidents not only endanger the lives of construction workers, but may also cause serious damage to surrounding buildings and underground facilities. Especially in the Hangzhou area, due to the large variability of the strata and the frequent presence of harmful gases such as gas, the safety risks of deep foundation pit construction are further exacerbated.

[0003] Currently, deformation monitoring and early warning for deep foundation pit projects primarily rely on traditional methods such as theoretical analysis, numerical simulation, and on-site monitoring. While these methods can predict foundation pit deformation trends to a certain extent, due to the complex and changing construction environment, a single model struggles to accurately capture dynamic changes. Prediction accuracy is often insufficient, especially under the influence of external factors such as rainfall and groundwater level fluctuations. Furthermore, existing monitoring systems often rely on manual data collection, which is inefficient and lacks real-time transmission and analysis, resulting in delayed early warnings. Furthermore, the development of deep foundation pit escape systems is relatively lagging. Existing escape devices, mostly simple iron ladders or ropes, lack protective measures and are limited by the depth of the foundation pit, posing significant safety risks. This is particularly true in deep pipe jacking pits, where the small surface area and large depth make escape difficult for construction workers in the event of emergencies such as sudden water and mud surges. Summary of the Invention

[0004] The purpose of this invention is to provide a deep foundation pit safety warning and escape system based on deep learning of multi-source data. Through the deep integration of multi-source data, the development of a warning platform, and the design of an escape execution module, this invention improves the safety and reliability of deep foundation pit construction.

[0005] The technical solution of the present invention is a deep foundation pit safety early warning and escape system based on multi-source data deep learning, comprising:

[0006] The multi-source monitoring module is composed of multi-source data acquisition equipment deployed in the deep foundation pit and is used to collect multi-source data in the deep foundation pit;

[0007] An early warning platform is connected to the multi-source monitoring module through a data transmission module. The early warning platform has a built-in deep learning model, which is used to integrate and analyze multi-source data, predict foundation pit deformation trends and safety risk levels, and output graded early warning signals.

[0008] The escape execution module is connected to the early warning platform and is used to automatically start the escape program in response to the early warning signal.

[0009] The above-mentioned deep foundation pit safety early warning and escape system based on multi-source data deep learning, the multi-source data acquisition equipment includes a camera, a concrete strain gauge, an earth pressure box, a rebar meter, a displacement sensor, a water level monitor, an inclinometer, a point displacement meter and a gas detector; the multi-source data acquisition equipment transmits the collected data to the early warning platform via wireless communication.

[0010] The aforementioned deep foundation pit safety warning and escape system based on deep learning of multi-source data, the escape execution module includes a hanging basket type escape device, the hanging basket type escape device includes a lifting basket, a vertical channel and a buffer base located at the bottom of the vertical channel; the lifting basket is arranged in the vertical channel, using an electric hoist as power and a steel wire rope for traction, and the hanger where the electric hoist is located can be adjusted left and right to ensure that the hook position is aligned with the center of the vertical channel when in use; a lighting lamp and a harmful gas detector are provided on the top of the basket; the vertical channel uses cylindrical pipe piles, and mud wall protection is done when the pipe piles are excavated to ensure the necessary strength to prevent local collapse, and the diameter of the pipe piles is between 0.8 meters and 1.2 meters.

[0011] The aforementioned deep foundation pit safety warning and escape system based on deep learning of multi-source data, the escape execution module is a cylindrical escape channel, and a liftable car is provided in the cylindrical escape channel; the cylindrical escape channel is composed of multiple sections of steel pipes axially spliced ​​through flanges, and each section of steel pipe is provided with an escape exit; a traction machine for pulling the car is provided at the top of the cylindrical escape channel, and a fall arrester is provided on the side wall of the cylindrical escape channel, and when the falling speed is greater than 1.1m / s, the fall arrester triggers the rack brake; a buffer is provided at the bottom end of the cylindrical escape channel.

[0012] The aforementioned deep foundation pit safety warning and escape system based on multi-source data deep learning, the warning platform includes:

[0013] The data fusion unit processes multi-source data using pixel-level and feature-level fusion methods to obtain time series monitoring data;

[0014] The LSTM neural network prediction unit takes time series monitoring data as input and outputs the deformation prediction value and risk probability for the next 5-30 minutes;

[0015] The threshold comparison unit compares the predicted value with the preset safety threshold and triggers an early warning.

[0016] The aforementioned deep foundation safety early warning and escape system based on multi-source data deep learning, the early warning includes:

[0017] Primary warning: 10 minutes in advance, start the sound and light alarm;

[0018] Secondary warning: 5 minutes in advance, activate the ventilation system and escape path indication;

[0019] Tertiary warning: 1 minute in advance, forcibly start the escape mode of the escape execution module.

[0020] The aforementioned deep foundation safety early warning and escape system based on multi-source data deep learning, the LSTM neural network prediction unit is trained by the following steps:

[0021] The input layer receives time series monitoring data;

[0022] The hidden layer uses memory cells to store long-term dependencies, and the optimization algorithm selects Adam algorithm;

[0023] The output layer generates predicted values, with average prediction accuracy σp and mean relative error MAPE as evaluation indicators.

[0024] The aforementioned deep foundation safety early warning and escape system based on multi-source data deep learning, further comprising a linkage control module; the linkage control module connects the field devices of the deep foundation through the Internet of Things protocol, and performs:

[0025] After receiving the early warning signal, automatically turn on the exhaust fan and emergency lighting;

[0026] Push the evacuation path map to the construction personnel terminal;

[0027] Lock the overloaded escape execution module and start the sound and light alarm.

[0028] The aforementioned deep foundation safety early warning and escape system based on multi-source data deep learning, further comprising an evacuation simulation unit:

[0029] The evacuation simulation unit constructs a three-dimensional model of the deep foundation based on the Pathfinder software, and dynamically generates the optimal escape path based on real-time personnel positioning data and pushes it to the construction personnel terminal through the linkage control module.

[0030] The aforementioned deep foundation safety early warning and escape system based on multi-source data deep learning, the data fusion unit uses pixel-level and feature-level fusion methods to process multi-source data to obtain time series monitoring data, specifically including:

[0031] Pixel-level fusion: multi-camera field of view splicing is performed on the foundation pit image data to generate a panoramic image, and the crack expansion track is identified by comparing the pixel gray value difference, and the crack width change sequence and damage area change curve in units of pixels are output;

[0032] Feature-level fusion: the original signals of the sensors are synchronously sampled and spatio-temporally aligned, the coupling relationship between displacement and stress is analyzed, and the fused joint time sequence curve is output; the mean, variance, trend item and frequency domain component features of the time sequence data are extracted to generate a multi-dimensional feature vector; the correlation analysis and dimensionality reduction processing are performed on the harmful gas concentration and construction load data to generate the joint time sequence curve of the gas safety index and the construction load;

[0033] Collaborative application: pixel-level fusion provides original deformation data for feature-level fusion, and pixel-level image acquisition frequency is triggered to increase when feature-level fusion is abnormal, realizing data complementation and dynamic optimization.

[0034] Compared with the prior art, the present application proposes a deep learning technology based on multi-source data fusion, combines an automatic monitoring and intelligent early warning platform, and constructs a complete deep foundation pit safety guarantee system. By integrating underground water level, soil deformation, enclosure structure stress and other multi-source data, a deep learning model is used to dynamically predict the deformation trend of the foundation pit, and a hierarchical early warning mechanism is established to realize real-time perception and rapid response to potential risks. At the same time, the escape execution module of the present application can ensure that the construction personnel can quickly evacuate in an emergency, and the escape execution module has the characteristics of modular design and reusability, and can adapt to the needs of deep foundation pits. In addition, the present application can also connect the field equipment through the Internet of Things protocol through the linkage control module, automatically open the exhaust fan and emergency lighting when warning to ensure the safety of the escape environment; at the same time, the dynamic escape path map is pushed to the construction personnel terminal to ensure the shortest path and the least obstacles. Therefore, the present application can realize accurate perception and rapid response to construction risks, significantly improve the safety, reliability and economic benefits of deep foundation pit engineering, and has important popularization value for urban underground space development. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a model architecture diagram of the system of the present application;

[0036] Figure 2 is a plane layout schematic diagram of a deep foundation pit monitoring point;

[0037] Figure 3 is a schematic diagram of a gas detector;

[0038] Figure 4 is a flowchart of an LSTM model prediction;

[0039] Figure 5 is a schematic diagram of a basket-type escape device;

[0040] Figure 6 is a schematic view of a cylindrical escape passage.

[0041] Reference signs: DETAILED DESCRIPTION

[0042] The application will be further described below in conjunction with the accompanying drawings and examples, but not as the basis for limiting the application.

[0043] Example: A deep foundation safety warning escape system based on multi-source data deep learning, its specific implementation is as follows. As shown in Figure 1 The overall architecture of the system includes a multi-source monitoring module, a warning platform, an escape execution module, a linkage control module, and a evacuation simulation unit. Each module is closely coordinated through data flow and communication protocol, thereby building a complete safety guarantee system. In practical application, the system is deployed at a deep foundation construction site of a certain city subway station, which is used to monitor the deformation of the foundation pit, the change of gas concentration and the fluctuation of underground water level in real time, and generate graded warning signals according to the data analysis results, guiding the safe evacuation of construction personnel.

[0044] First, the multi-source monitoring module is the core data source of the entire system, which mainly includes cameras, concrete strain gauges, soil pressure boxes, reinforcement meters, displacement sensors, water level monitors, inclinometers, point displacement meters, and gas detectors, etc. Figure 2 The monitoring plan of the foundation pit is shown, these devices are arranged in different positions of the foundation pit, such as the side wall, the bottom and the surrounding soil layer of the foundation pit, to ensure comprehensive coverage of the key areas. The concrete body stress sensor adopts strain gauge design, which can collect real-time stress change data of the side wall of the foundation pit; the underground water level monitor measures the water level height through the pressure sensor, and at the same time combines the temperature compensation algorithm to improve the measurement accuracy; the gas detector can detect the concentration change of oxygen, carbon monoxide, methane and other gases. All the collected data is sent to the warning platform for processing through wireless transmission technology. In practical application, the sampling frequency of the multi-source monitoring module is set to one per second to ensure the real-time and continuity of the data. Figure 3The system architecture of a gas detector is demonstrated. It consists of a sensor module, a signal conditioning module, a power supply adjustment module, a keypad input module, an LCD display module, an alarm module, and a microcontroller module. The gas sensor is primarily used to accurately detect the concentration of toxic gases in foundation pits, including CO, CH4, H2S, and CO2. The signal conditioning circuit converts the weak current output by the sensor into a voltage signal with a specific dynamic range. This signal is then input to the ADC port of the LPC2131 for analog signal acquisition. After signal processing, the concentration of the corresponding gas is calculated and sent to the LCD display module. If the concentration exceeds the set warning threshold, the LPC2131 microcontroller automatically issues an alarm, prompting personnel to take appropriate protective measures. The keypad input and LCD display modules form the human-computer interface, facilitating user configuration of various detector parameters.

[0045] The early warning platform has a built-in deep learning model that is responsible for integrating and analyzing the data transmitted by the multi-source monitoring modules and predicting the foundation pit deformation trend and safety risk level. Figure 4 As shown, the deep learning model adopts an LSTM neural network structure, and its input layer receives time series data from a multi-source monitoring module, including parameters such as soil stress, groundwater level, and gas concentration. The hidden layer stores long-term dependencies through memory units, thereby capturing nonlinear changes in complex environments. The output layer generates a predicted value of foundation pit deformation and risk probability for the next 5-30 minutes. The specific prediction process is as follows: the collected multi-source data is preprocessed, including normalization and denoising operations, and abnormal values ​​collected by the sensor (such as sudden jumps in soil stress data) are eliminated. Then, spatiotemporal alignment is performed to unify the timestamps of different sensors (such as synchronization of acquisition time between groundwater level meters and inclinometers) to ensure that the data are comparable in the time dimension; then, unstructured data (such as foundation pit images, construction logs) are converted into a structured format for easy computer processing. Then, the preprocessed data is processed in the data fusion unit using pixel-level and feature-level fusion methods to process the multi-source data to obtain time series monitoring data. In this embodiment, the application scenarios and implementation methods of data fusion are as follows:

[0046] Scenario 1: Foundation pit image monitoring

[0047] Data source: Images captured by high-definition cameras placed around the foundation pit (pixel-level data).

[0048] Fusion method: The field of view of multiple cameras is stitched into a panoramic image, covering the entire surface of the foundation pit; pixel data at different times are compared to identify the crack expansion trajectory. For example, by calculating the difference in pixel grayscale values ​​of adjacent images, new crack areas are automatically marked.

[0049] Output time series data: crack width change sequence in pixels, damage area change curve.

[0050] Scenario 2: Multi-sensor raw signal fusion

[0051] Data source: original output of analog sensors such as displacement meter (voltage signal) and axial force meter (current signal).

[0052] Fusion method: A unified clock triggers synchronous sampling across multiple sensors, ensuring temporal alignment of the data. Displacement signals are then superimposed with axial force signals to analyze the displacement-stress coupling relationship of the foundation pit support structure. For example, when a displacement meter detects a sudden change in horizontal displacement, the axial force meter is simultaneously checked to see if a stress surge occurs, thus determining the risk of local instability.

[0053] Output time series data: The fused "displacement-stress" joint time series curve is used for LSTM model training.

[0054] Scenario 3: Time Series Data Feature Extraction

[0055] Data source: Time series monitoring data of groundwater level, soil stress, and retaining structure deformation.

[0056] Feature extraction method: Calculate the mean, variance, extreme values, etc. to reflect the data fluctuation range (such as the change in the daily average of the groundwater level); extract trend items through sliding average and polynomial fitting to identify long-term deformation trends (such as the linear growth trend of surface settlement around the foundation pit); use Fourier transform to convert time domain data into frequency domain and analyze vibration frequency components (such as the main frequency of soil vibration caused by pipe jacking construction).

[0057] Output feature vector: a multidimensional feature vector consisting of "timestamp + deformation rate + stress growth rate + vibration frequency".

[0058] Scenario 4: Multi-source data correlation analysis

[0059] Data sources: harmful gas concentration (CO, CH4), dust content, construction load (jacking thrust, excavation depth).

[0060] Feature correlation method: Calculate the Pearson correlation coefficient to identify strongly correlated features (such as the correlation between CH4 concentration and pipe jacking thrust); reduce high-dimensional features to a small number of principal components to eliminate data redundancy. For example, combine "CO concentration, CH4 concentration, and oxygen content" into the principal component of "gas safety index."

[0061] Output time series data: The fused "gas safety index-construction load" joint time series curve is used for risk level classification model training.

[0062] Scenario 5: Collaborative application of pixel-level and feature-level fusion

[0063] Pixel-level fusion provides the basis for feature-level analysis: for example, raw time series data of crack width (such as daily crack width measurements) is obtained through pixel-level image analysis, and then the crack expansion rate (Δwidth / Δtime) is calculated through feature-level processing as a key feature for risk assessment.

[0064] Feature-level fusion guides pixel-level data acquisition: For example, when feature-level analysis finds an abnormal soil stress-displacement correlation, it automatically triggers an increase in the frequency of pixel-level image acquisition (e.g., from once a day to once an hour) to capture subtle deformation features.

[0065] After obtaining the time series monitoring data, it is input into the LSTM neural network prediction unit, and the time series features in the data are gradually extracted through the memory unit; finally, the output layer outputs the deformation prediction value for the next 5-30 minutes and risk probability, and calculate the average prediction accuracy σ by the following formula p And the mean relative error MAPE:

[0066]

[0067] in, is the predicted value, y i is the actual value, and N is the number of samples.

[0068] In actual tests, the average prediction accuracy of the model reached over 95%, and the average relative error was less than 5%, indicating that it has high prediction reliability.

[0069] After obtaining the predicted value, the predicted value is compared with the preset safety threshold in the threshold comparison unit to trigger an early warning.

[0070] The hierarchical and classified forecast and warning system is set up as follows:

[0071] Level 1 warning: 10 minutes in advance, using a general alarm (sound and light alarm);

[0072] Level 2 warning: 5-minute advance warning, emergency alarm, ventilation system and escape route instructions;

[0073] Level 3 warning: 1 minute advance warning, urgent alarm and forced activation of the escape execution module’s escape mode; including water gushing alarm, mud burst alarm, gas alarm, foundation pit retaining structure deformation alarm, etc.

[0074] When the early warning platform generates a graded warning signal, the escape execution module automatically initiates the corresponding escape procedure based on the signal level. The escape execution module includes two main devices: a hanging basket escape device and a cylindrical escape channel.

[0075] like Figure 5As shown, the hanging basket type escape device consists of an electric hoist, a buffer base and a harmful gas detector. The electric hoist pulls the hanging basket up and down along the vertical channel through a steel cable to ensure that the construction workers can quickly evacuate to the ground. The buffer base is installed at the bottom of the vertical channel and adopts a spring shock-absorbing design. It can absorb the impact force when the hanging basket reaches the ground to protect the safety of personnel. The harmful gas detector is installed at the top of the channel to monitor the changes in gas concentration in the channel in real time and feed back the data to the early warning platform. If the concentration of harmful gases exceeds the standard, the system will immediately sound an alarm and suspend the operation of the hanging basket until the gas concentration returns to a safe range. Specifically, the vertical channel is carried out using cylindrical pipe piles. When the pipe piles are excavated, mud wall protection is made to ensure the necessary strength to prevent local collapse. The diameter of the pipe piles can be between 0.8 meters and 1.2 meters. The size of the pipe piles does not affect the use of the upper lifting device. The diameter of the pipe piles in this embodiment is 1.2 meters. The following are the specific indicators of the hanging basket type escape device:

[0076]

[0077] Table 1

[0078] The electrical parameters of the hanging basket type escape device are as follows:

[0079]

[0080]

[0081] Table 2

[0082] The following describes them separately by structure.

[0083] The above-ground section is used for the entry and exit of goods and personnel in emergencies. An electric hoist with a rated load of 1 ton is installed on the top. Since the vertical channel is a circular hole, to prevent the hanging basket from deviating from the center of the circle, the hanger on which the electric hoist is located can be adjusted left and right to ensure that the hook is positioned exactly at the center of the pile hole when in use.

[0084] The hanging basket is used for transporting cargo up and down. It has an effective area of ​​0.8 m² and a maximum load of 400 kg. When the load exceeds 400 kg, an audible alarm sounds and the hoisting mechanism stops operating. The basket is equipped with a top light and a hazardous gas monitoring device. If excessive hazardous gases are detected, an alarm is sent to the ground via Wi-Fi. Due to the unique conditions of the site, the basket does not have a track for up and down movement. To prevent swaying under eccentric loads, two sheaves are installed on the top and bottom of the basket. Two parallel steel ropes run from the surface to the bottom of the well through the four sheaves of the basket, and the ropes are always taut. The restraint of the ropes effectively prevents the basket from tilting. As the basket moves up and down within the pile hole, it may rub against the inner wall due to eccentricity. Rollers are installed on the top and bottom of the basket to convert sliding friction into rolling friction when rubbing against the inner wall of the pile hole, reducing frictional resistance and improving operational reliability. The hazardous gas monitoring device primarily detects CO and SO2.

[0085] Buffer base pile hole: A buffer spring is installed at the bottom to prevent the impact when the hanging basket falls to the ground.

[0086] like Figure 6 As shown, a liftable car is provided in the cylindrical escape passage; the cylindrical escape passage is formed by axially splicing multiple sections of steel pipes (DN1200) through flanges, and each section of the steel pipe is provided with an escape exit; the cylindrical steel pipe (DN1200) has good compressive resistance, and even if a local collapse occurs on the working surface, it can still maintain its overall integrity to a considerable extent. When an underground disaster occurs, a valuable escape passage is provided for underground construction workers. A traction machine for pulling the car is provided at the top of the cylindrical escape passage, and a fall arrester is provided on the side wall of the cylindrical escape passage. When the falling speed is greater than 1.1m / s, the fall arrester triggers the rack brake; a buffer is provided at the bottom end of the cylindrical escape passage. The following are the specific indicators of the cylindrical escape passage:

[0087]

[0088]

[0089] Table 3 The electrical parameters of the cylindrical escape route are as follows:

[0090] Name Power Unit Voltage Remark Machine of traction 6 KW 340V Power supply of frequency converter Brake of traction 1 KW DC 110V Brake of power-off Pump of pollution discharge 3 KW 3PH380V 4KW when lift 32m Fan of exhaust air 1.1 KW 3PH380V Lighting lamp 100 W DC 24V Power supply of safety voltage Power supply of car 100 W DC 24V Power supply of safety voltage

[0091] Table 4 below describes the structure separately.

[0092] Ground section: For ground personnel to enter and exit, and is also equipped with a lifting traction machine, exhaust fans and sewage outlets.

[0093] The middle section allows for the vertical movement of the car. It can be connected in series, with lengths of four meters, and any number of cascades can be created to accommodate different scenarios depending on the depth of the underground operation. It is equipped with guide rails, racks, and supporting accessories. Each section has a circular access door, which can be closed if there is no corresponding transverse passage connecting the section. The middle section has a smooth outer wall, allowing for easy removal from the pile hole and reuse after construction. Each section has eight symmetrical locating holes on each end: two for axial alignment and six for bolt connections. Tightening is performed by the installer inside the pipe. Some groundwater may seep into the end joints, which will accumulate at the bottom and be discharged upwards by a sewage pump. Two T75 guide rails and racks are arranged symmetrically within the pipe. The guide rails provide lateral limits for the vertical movement of the car, while the racks work in conjunction with the safety device to prevent the car from accidentally falling. The car itself is pulled by five steel ropes.

[0094] The bottom section provides structural support for the entire escape system, a final buffer to prevent the car from falling, and temporary storage and drainage of groundwater that infiltrates the escape system. Groundwater at construction sites often contains a high concentration of impurities, such as plastic bags, fibers, cloth strips, paper, and straw. Therefore, the sewage pump within the system features an alloy steel cutterhead that cuts these impurities and discharges them along with the liquid water through a hose to the ground level. The personnel lift car is designed with multiple safety measures to prevent accidental falls. If other measures fail, the buffer within the bottom section provides a final deceleration step, minimizing impact on passengers. The bottom section consists of a container with a bottom plate that can hold up to approximately two cubic meters of sewage. To prevent excessive sewage from rusting the buffer, a level switch is installed inside, one-quarter of the height from the bottom. When the sewage reaches the level switch, the sewage pump is activated to drain the water.

[0095] Car: The car is used for transporting people, with an effective area of ​​0.6m 2The maximum load is 400 kg. When the load exceeds 400 kg, an alarm sounds and the hoisting mechanism stops operating. The car is equipped with a light, ventilation fan, operating buttons, and a communication system. A fall arrester is installed on the upper left and upper right sides of the car. If the traction ropes break and the car free falls, the speed of the fall arrester will exceed the set speed. At this time, the clutch slider inside the fall arrester, pulled by centrifugal force, will apply a braking action, preventing further acceleration of the car. This ensures the safety of occupants. The escape window on the top of the car can be tilted upward. In the event of a sudden power outage during movement, occupants can climb out through the escape window. For enhanced safety, the power supply voltage in the car is 24V DC. Even if there is a leakage due to moisture underground, it will not cause harm to people. A communication system is installed in the car for voice communication with personnel on the ground. To prepare for emergencies, drinking water, food, and first aid medical equipment are placed in the car for immediate use in the event of an emergency. In this embodiment, the anti-fall device realizes the emergency braking function through the rack brake mechanism. When the car falling speed v>1.1m / sv>1.1m / s, the braking is triggered and the braking force F brake The calculation formula is as follows:

[0096] F brake =μ·m·g+k·v 2 ;

[0097] Where μ is the friction coefficient, ranging from 0.1 to 0.3; m is the car mass, in kilograms; g is the acceleration due to gravity, with a value of 9.8 m / s 2 k is the speed-related coefficient, ranging from 0.5 to 1.0. In actual tests, the braking system was able to completely stop the car within 0.5 seconds, effectively avoiding the safety hazards caused by high-speed falls.

[0098] The linkage control module connects to on-site equipment via the Internet of Things (IoT) protocol. Upon triggering a warning signal, it automatically activates exhaust fans and emergency lighting, and sends a dynamic escape route map to construction personnel's terminals. Exhaust fans are primarily used to remove harmful gases from the foundation pit, ensuring that air quality within the corridor meets safety standards. Emergency lighting is installed at key points in the escape route, providing clear visual guidance for construction personnel.

[0099] like Figure 5As shown in the figure, the evacuation simulation unit dynamically generates the optimal escape path based on the Pathfinder three-dimensional modeling software and pushes the path map to the construction personnel's terminal. The path generation process is as follows: First, the system calculates the distribution of obstacles in the current environment based on the three-dimensional structural model of the foundation pit and real-time monitoring data. Then, the path planning algorithm is used to determine the shortest escape route with the least obstacles. Finally, the generated path map is displayed in a visual form on the construction personnel's handheld terminal, while marking the location of key nodes and the evacuation direction. In actual application, this module significantly improved the evacuation efficiency of construction personnel, and the average evacuation time was shortened by more than 30%.

[0100] The following is a detailed description of the application of each system module in actual scenarios to highlight the implementation of technology and functions:

[0101] Application scenario: A deep foundation pit project for a city center subway station. The pit is 25 meters deep and covers an area of ​​800 square meters. It is adjacent to subway lines and commercial buildings and has complex geological conditions (including silty soil and gas layers).

[0102] 1. System deployment and core module configuration:

[0103] 1. Multi-source monitoring module: full-dimensional data collection;

[0104] Device deployment:

[0105] Stress and strain monitoring: 12 steel bar gauges are arranged on the main reinforcement of the retaining piles and 8 concrete strain gauges are installed on the support beams to monitor changes in the internal forces of the structure.

[0106] Deformation monitoring: 10 displacement sensors and 4 inclinometers (buried at a depth of 30 meters) are arranged around the foundation pit, and 6 point displacement meters are set at the bottom of the pit to capture horizontal displacement and settlement in real time.

[0107] Environmental monitoring: Groundwater level monitors (accuracy ±1cm) are arranged at the four corners of the foundation pit, and gas detectors (CO range 0-1000ppm, CH4 range 0-4%LEL) cover the construction surface and escape routes.

[0108] Data transmission: All devices access the early warning platform through the LoRa wireless gateway, with a sampling frequency of 1 time per second, and abnormal data triggers real-time encrypted transmission.

[0109] 2. Early Warning Platform: Intelligent Forecasting and Graded Response

[0110] Model training:

[0111] Input data: Integrate one month of historical monitoring data (including 10 types of parameters such as soil stress, water level, and gas concentration) to construct a time series dataset.

[0112] LSTM network structure: 10 nodes in the input layer, 2 hidden layers (128 units each), 2 nodes in the output layer (deformation prediction value, risk probability), and the optimization algorithm used is Adam (learning rate 0.001).

[0113] Training results: average prediction accuracy σp = 96.2%, average relative error MAPE = 4.1%, meeting engineering requirements.

[0114] Warning rules:

[0115] Level 1 warning (deformation reaches 80% of the threshold): an audible and visual alarm is triggered 10 minutes in advance to alert on-site personnel to the risk.

[0116] Level 2 warning (deformation reaches 90% of the threshold): start the ventilation system 5 minutes in advance (air volume 4900m 3 / h), the escape route LED indicator lights up and shows the direction of "Emergency Exit".

[0117] Level 3 warning (deformation exceeds the threshold or CH4 concentration is greater than 1% LEL): The escape execution module is forcibly activated one minute in advance, and a red alarm and escape route are pushed to 30 construction personnel terminals on site.

[0118] 3. Escape execution module: dual-mode emergency evacuation;

[0119] Hanging basket escape device (applicable to daily construction):

[0120] The vertical channel uses 1.2-meter diameter pipe piles, the mud wall depth is 30 meters, and the bottom buffer base spring stiffness is 20kN / m.

[0121] The electric hoist has a rated load of 400KG and can transport 4 people at a time (each person has a load of 100KG). It only takes 3.1 minutes (lifting speed 8m / min) to move from the bottom of the pit to the ground (25 meters).

[0122] Actual case: When the monitored CO concentration reaches 200ppm (exceeding the threshold of 100ppm), the hanging basket automatically suspends operation until the ventilation system reduces the concentration to a safe range (<50ppm).

[0123] Cylindrical escape passage (applicable to deep foundation pit):

[0124] The passage is made of 6 sections of 4-meter steel pipes (total length 24 meters), the effective area of ​​the car is 0.56㎡, and it is equipped with 5 Φ10mm traction steel ropes.

[0125] Arrester test: Simulating a wire rope breakage scenario, the car falls from a height of 10 meters. The arrester triggers the rack brake at a speed of 1.2m / s and stops the car within 0.4 seconds. The impact load is less than 5kN.

[0126] 4. Linkage control and evacuation simulation

[0127] Linked response:

[0128] When the third-level warning is triggered, the exhaust fan (power 1.1kW) automatically turns on, reducing the CH4 concentration in the foundation pit from 1.5% LEL to below 0.5% LEL within 10 minutes.

[0129] Emergency lighting (DC24V, 50W) is arranged every 5 meters along the escape route to ensure that the light intensity is greater than 100Lux.

[0130] Evacuation simulation:

[0131] A three-dimensional model of the foundation pit was constructed based on Pathfinder software, and real-time personnel positioning data (Bluetooth beacon positioning, accuracy of ±1.5 meters) was imported.

[0132] Dynamic path generation example: When a local landslide occurs on the east side of the foundation pit, the system automatically avoids the area and generates the shortest path (total length 28 meters) of "construction surface → central cylindrical passage → ground exit", which is 40% shorter than the traditional path.

[0133] II. Typical emergency scenario drills

[0134] Scenario: Risk of sudden water and mud inrush in foundation pit (Level 3 warning)

[0135] Monitoring trigger:

[0136] The water level monitor showed that the groundwater level rose sharply by 1.2 meters per hour, and the inclinometer detected that the deep horizontal displacement of the retaining wall reached 50 mm (exceeding the warning threshold of 40 mm).

[0137] After comprehensive analysis, the early warning platform determined that mudsurge might occur within the next minute and immediately issued a level 3 warning.

[0138] Escape execution:

[0139] The cylindrical escape passage car automatically descended to the bottom of the pit, and a voice announcement was made: "Please evacuate immediately." All 20 construction workers were evacuated within 5 minutes (the average boarding time per person was 15 seconds).

[0140] As the car rose to the ground, the anti-fall device was not triggered and the buffer absorbed the impact of landing, resulting in no casualties.

[0141] Linkage effect:

[0142] The exhaust fan and emergency lighting will start within 10 seconds after the warning, the escape route will be unobstructed, and the terminal will display the real-time escape progress (the remaining number of people will be dynamically updated).

[0143] Subsequent statistics: It took only 6 minutes from the early warning to the evacuation of all personnel, which is 70% shorter than traditional manual evacuation.

[0144] In summary, the present invention proposes a deep learning technology based on multi-source data fusion, combined with automated monitoring and intelligent early warning platforms, to build a complete deep foundation pit safety assurance system. By integrating multi-source data such as groundwater level, soil deformation, and retaining structure stress, a deep learning model is used to dynamically predict the deformation trend of the foundation pit, and a hierarchical early warning mechanism is established to achieve real-time perception and rapid response to potential risks. At the same time, the escape execution module of the present invention can ensure that construction personnel can evacuate quickly in an emergency. The escape execution module has the characteristics of modular design and reusability, and can adapt to the needs of foundation pits of different depths. In addition, the present invention can also connect to on-site equipment through the Internet of Things protocol through the linkage control module, automatically turn on exhaust fans and emergency lighting when an early warning is issued, and ensure the safety of the escape environment; at the same time, a dynamic escape path map is pushed to the construction personnel terminal to ensure the shortest path with the least obstacles. As a result, the present invention can achieve accurate perception and rapid response to construction risks, significantly improve the safety, reliability and economic benefits of deep foundation pit projects, and has important promotion value for urban underground space development.

Claims

1. A deep foundation pit safety warning and escape system based on multi-source data deep learning, characterized by: include: The multi-source monitoring module is composed of multi-source data acquisition equipment deployed in the deep foundation pit and is used to collect multi-source data in the deep foundation pit; An early warning platform is connected to the multi-source monitoring module through a data transmission module. The early warning platform has a built-in deep learning model, which is used to integrate and analyze multi-source data, predict foundation pit deformation trends and safety risk levels, and output graded early warning signals. The escape execution module is connected to the early warning platform and is used to automatically start the escape program in response to the early warning signal.

2. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 1 is characterized by: The multi-source data acquisition equipment includes a camera, a concrete strain gauge, an earth pressure cell, a rebar meter, a displacement sensor, a water level monitor, an inclinometer, a point displacement meter and a gas detector; the multi-source data acquisition equipment transmits the collected data to the early warning platform via wireless communication.

3. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 1 is characterized by: The escape execution module includes a hanging basket type escape device, which includes a lifting basket, a vertical channel and a buffer base located at the bottom of the vertical channel; the lifting basket is arranged in the vertical channel, using an electric hoist as power and a steel wire rope for traction, and the hanger where the electric hoist is located can be adjusted left and right to ensure that the hook position is aligned with the center of the vertical channel when in use; a lighting lamp and a harmful gas detector are provided on the top of the basket; the vertical channel uses cylindrical pipe piles, and mud wall protection is made when the pipe piles are excavated to ensure the necessary strength to prevent local collapse, and the diameter of the pipe piles is between 0.8 meters and 1.2 meters.

4. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 1 is characterized by: The escape execution module is a cylindrical escape passage, in which a liftable car is provided; the cylindrical escape passage is formed by axially splicing multiple sections of steel pipes through flanges, and each section of steel pipe is provided with an escape exit; a traction machine for pulling the car is provided at the top of the cylindrical escape passage, and a fall arrester is provided on the side wall of the cylindrical escape passage, and when the falling speed is greater than 1.1m / s, the fall arrester triggers the rack brake; a buffer is provided at the bottom end of the cylindrical escape passage.

5. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 1 is characterized by: The early warning platform includes: The data fusion unit processes multi-source data using pixel-level and feature-level fusion methods to obtain time series monitoring data; The LSTM neural network prediction unit takes time series monitoring data as input and outputs the deformation prediction value and risk probability for the next 5-30 minutes; The threshold comparison unit compares the predicted value with the preset safety threshold and triggers an early warning.

6. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 5 is characterized by: The warnings include: Level 1 warning: 10 minutes in advance, the sound and light alarm will be activated; Level 2 warning: Activated 5 minutes in advance, ventilation system and escape route indication; Level 3 warning: 1 minute in advance, the escape execution module's escape mode is forcibly activated.

7. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 5 is characterized by: The LSTM neural network prediction unit is trained through the following steps: The input layer receives time series monitoring data; The hidden layer uses memory units to store long-term dependencies, and the optimization algorithm uses the Adam algorithm; The output layer generates predicted values, with average prediction accuracy σp and average relative error MAPE as evaluation indicators.

8. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 1 is characterized by: It also includes a linkage control module; the linkage control module is connected to the on-site equipment of the deep foundation pit through the Internet of Things protocol to execute: Automatically turn on exhaust fans and emergency lighting after receiving early warning signals; Push evacuation route map to construction workers' terminals; Lock the overloaded escape execution module and activate the sound and light alarm.

9. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 8 is characterized by: Also includes evacuation simulation unit: The evacuation simulation unit constructs a three-dimensional model of a deep foundation pit based on Pathfinder software, and combines it with real-time personnel positioning data to dynamically generate the optimal escape path and push it to the construction personnel terminal through the linkage control module.

10. The deep foundation pit safety warning and escape system based on multi-source data deep learning according to claim 5 is characterized by: The data fusion unit processes multi-source data using pixel-level and feature-level fusion methods to obtain time series monitoring data, specifically including: Pixel-level fusion: Multi-camera field of view stitching is performed on foundation pit image data to generate a panoramic image. The crack extension trajectory is identified by comparing pixel grayscale value differences, and the crack width change sequence and damage area change curve in pixel units are output; Feature-level fusion: Synchronously sample and align sensor raw signals in time and space, analyze the coupling relationship between displacement and stress, and output a fused joint time series curve. Extract mean, variance, trend term, and frequency domain component features from time series data to generate a multidimensional feature vector. Perform correlation analysis and dimensionality reduction on harmful gas concentration and construction load data to generate a joint time series curve of gas safety index and construction load. Collaborative application: Pixel-level fusion provides original deformation data for the feature level. When feature-level fusion is abnormal, it triggers an increase in the frequency of pixel-level image acquisition to achieve data complementarity and dynamic optimization.

Images