A deep-buried long diversion tunnel off-line monitoring and early warning system and method

Abstract

The application discloses a kind of deep burying long diversion tunnel offline monitoring early warning system and method, the present application discards underwater robot offline two-dimensional code+photosensitive switch to wake up display shooting transmission mode, using underwater acoustic beacon guidance positioning and three color LED grading early warning, signal easy to identify, not affected by underwater turbidity, light scattering etc., realizes the simplified transmission of risk information in the complex environment of diversion tunnel, solves the problem of poor transmission stability of prior art.Inbuilt edge computing unit, real-time prediction of osmotic pressure can be completed in data acquisition terminal, safety level determination and automatic early warning, predict future 7 days data trend in advance, not dependent on cloud computing power and artificial processing and prediction cycle and transmission cycle are synchronized, ensure the continuity of tunnel risk investigation, solve the problem of data transmission lag of existing monitoring device.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy engineering technology, specifically relating to an offline monitoring and early warning system and method for deep-buried long water diversion tunnels, used for long-term offline monitoring of the seepage pressure values ​​of the lining and surrounding rock of deep-buried long water diversion tunnels and for safety assessment and early warning. Background Technology

[0002] As a key component of long-distance water conveyance projects, hydraulic tunnels are situated in complex underground environments characterized by high ground stress, high external water pressure, and high humidity. Safety issues such as lining cracking, seepage and water inrush, and deformation and instability of surrounding rock require rapid investigation and real-time monitoring.

[0003] Currently, to address the challenges of unstable power supply and difficulty in transmitting monitoring data during the operation of hydraulic tunnels, Chinese patent CN112923980B discloses a wireless long-term monitoring device for deeply buried long water diversion tunnels. This device includes an energy module, a monitoring module, and a transmission module. The energy module comprises a hydroelectric generator and a battery. The monitoring module includes detection sensors. The transmission module includes a memory, a chip, a photosensitive switch, a waterproof display, and an underwater robot. This wireless long-term monitoring device for deeply buried long water diversion tunnels achieves long-term power supply for the monitoring equipment through self-generated hydroelectric power, stores the monitoring data as offline QR codes, and transmits the data using an underwater robot. While this device successfully solves the problems of unstable power supply and difficulty in transmitting monitoring data, it still has some shortcomings:

[0004] First, due to the harsh underwater optical environment, water strongly absorbs and scatters light, resulting in underwater images generally having a bluish-green tint, low contrast, and blurred details. Furthermore, the turbidity of the water and the movement of suspended particles in the water diversion tunnel exacerbate light scattering, creating noise in the image and severely interfering with the extraction of black-and-white features from the QR code. In addition, underwater robots are easily affected by the water flow in the tunnel, causing them to sway and potentially leading to problems locating the photosensitive switch, motion blur, and excessive energy consumption. In summary, existing patented solutions relying on underwater robot decoding cannot guarantee the stability of data transmission.

[0005] Second, existing patented data transmission methods require a considerable period of continuous data collection and integration until the data set reaches the capacity of a single offline QR code. Only then are new offline QR codes generated and sequentially stored in the memory. When an underwater robot retrieves a QR code, it must wait for all offline QR codes to be displayed sequentially through a waterproof display. This excessively long single-point operation time makes the endurance of underwater robots for long-distance operations a challenge.

[0006] Third, in existing monitoring methods, after the underwater robot collects all offline QR codes, it identifies and decodes all the QR codes to generate historical monitoring data. This data is then uploaded and manually processed for analysis. While this approach overcomes the problems of offline storage and transmission of monitoring data, it heavily relies on cloud computing power. Uploading large amounts of raw data generated by long-term sensor monitoring to the cloud for unified analysis can lead to excessive computing load, prolonged scheduling time, and significantly reduced overall system efficiency. Furthermore, existing monitoring devices require waiting for data units collected within the same time period to be merged into a dataset, and a new offline QR code can only be generated after the dataset reaches the capacity of a single QR code. This results in a significant time lag in the received data, preventing real-time updates and analysis of the latest security status.

[0007] Therefore, to solve the above problems, this invention will make reasonable improvements on the existing offline monitoring system and develop an offline monitoring device for hydraulic tunnels that integrates edge data analysis, real-time risk warning, and simplified offline transmission, thereby facilitating risk investigation, earthquake prevention and disaster reduction in hydraulic tunnels. Summary of the Invention

[0008] The purpose of this invention is to solve the problems in the prior art, such as the inability to guarantee the stability of data transmission, the long single-point operation time of underwater robots, the delay of data scanning and monitoring, and the imperfection of safety assessment and early warning systems. The invention provides an offline monitoring and assessment system based on water flow self-generated power, which integrates real-time monitoring, safety assessment and early warning systems.

[0009] The present invention adopts the following technical solution:

[0010] An offline monitoring and early warning system for a deep-buried long water diversion tunnel includes multiple monitoring and early warning units deployed along the tunnel and an underwater robot; the monitoring and early warning unit includes a power supply module, a pressure monitoring module, an edge computing module, and an early warning module;

[0011] The power supply module includes a micro hydroelectric generator, a battery, and a voltage regulator module;

[0012] The pressure monitoring module includes an osmotic pressure sensor;

[0013] The edge computing module includes a memory card, a processing chip, and a Raspberry Pi;

[0014] The early warning module includes an LED warning light, a waterproof sealed shell, and an underwater acoustic beacon;

[0015] The micro hydroelectric generator is electrically connected to the battery, and the battery supplies power to the Raspberry Pi and the osmotic pressure sensor via a voltage regulator module.

[0016] The osmotic pressure sensor is communicatively connected to the Raspberry Pi and is used to collect osmotic pressure data according to a preset sampling frequency. The Raspberry Pi is connected to the memory card, processing chip, LED warning light and underwater acoustic beacon respectively.

[0017] The waterproof sealed shell is equipped with a transparent waterproof observation window, and the LED warning light and underwater acoustic beacon are located inside the shell and close to the observation window;

[0018] The underwater robot is used to receive underwater acoustic beacon signals, locate anomaly monitoring and early warning units, collect LED warning light status data, and transmit monitoring and early warning data back to the shore.

[0019] By replacing QR codes and photosensitive switches with underwater acoustic beacons and LED tri-color lights, the robot completely eliminates interference from the underwater optical environment. Positioning is achieved using specific low-frequency acoustic pulses, unaffected by water turbidity or light attenuation. In warning mode, the LED remains constantly lit while continuously emitting acoustic signals, eliminating the need for complex image capture. The robot no longer requires precise positioning or prolonged image capture, resolving issues of unstable underwater optical transmission, recognition failures, and high power consumption, significantly improving data transmission reliability. Simultaneously, local edge computing is implemented, eliminating reliance on the cloud and the need to upload massive amounts of raw data, thus avoiding computing power congestion and data lag.

[0020] Specifically, the underwater robot is equipped with an acoustic signal receiving module, an image acquisition module, a data storage module, and a wireless communication module. The underwater robot can determine the distance to abnormal points based on the intensity of the acoustic signal and inspect multiple abnormal monitoring and early warning units in order from near to far. After completing the acquisition of each abnormal point, the underwater robot removes the signal of that point and continues to inspect the next abnormal point in order from near to far until there is no specific acoustic signal.

[0021] The robot calculates the distance between itself and each abnormal point by receiving the intensity difference of specific low-frequency sound signals, and plans the inspection path accordingly. After completing the collection of single-point data, the corresponding signal is removed to avoid repeated inspections.

[0022] The robot is unaffected by water flow and does not require precise positioning of photosensitive switches, thus avoiding motion blur. Its extremely short dwell time at a single point significantly reduces robot energy consumption and improves endurance and operational efficiency.

[0023] The preset sampling frequency is 10 minutes / time, the single monitoring cycle adapted to edge computing is 7 days, the sampling data in a single cycle is 1008, and the data collection is continuous and unaffected by edge computing. After each data collection is completed, it is transmitted to the edge computing module.

[0024] Specifically, the edge computing module uses a 7-day prediction window, first uses Ridge regression to reduce noise, then uses the ARIMA model to predict the osmotic pressure for the next 7 days, calculates the root mean square error (RMSE) of the prediction results, and triggers an early warning when the RMSE exceeds 5% three times in a row.

[0025] Data noise reduction, time-series prediction, and accuracy verification are performed locally on the monitoring terminal, enabling real-time assessment of seepage pressure safety status without waiting for data to accumulate or requiring cloud processing. This achieves real-time data processing and prediction, eliminating data lag, avoiding reliance on cloud computing power, and preventing system efficiency reduction caused by the centralized uploading of large amounts of raw data. Furthermore, it can predict seepage pressure anomalies up to 7 days in advance, providing early warning of potential hazards.

[0026] Specifically, the early warning module has three states:

[0027] Safety: LED is always green, underwater acoustic beacon is off;

[0028] Warning: The LED is constantly lit in yellow, indicating that the underwater acoustic beacon is continuously emitting specific low-frequency acoustic pulses;

[0029] Danger: LED is constantly red, indicating that the underwater acoustic beacon is continuously emitting specific low-frequency acoustic pulses.

[0030] Based on different safety levels, the LED lights and sound beacons are configured with different operating states: the sound beacons are turned off in a safe state to save power, and the sound beacons are turned on in an abnormal state to trigger robot inspections. At the same time, the risk level is intuitively distinguished by the color of the lights.

[0031] In abnormal situations, the lights and sounds immediately and continuously illuminate, without waiting for QR code generation and display, without QR code recognition, and unaffected by underwater optical environment interference; the robot does not need to wait for display, it can take pictures upon arrival and leave immediately after taking pictures, significantly shortening inspection time and reducing energy consumption, and using the simplest sound and light signals to convey the risk level.

[0032] Specifically, the rules for determining the early warning are as follows:

[0033] If the predicted seepage pressure is less than 80% of the design allowable value, then it is safe.

[0034] If the predicted seepage pressure value is between 80% × the design allowable value and the design allowable value, an early warning will be issued;

[0035] If the predicted seepage pressure is greater than or equal to the design allowable value, then there is a danger.

[0036] If the RMSE exceeds 5% three times in a row, it will directly enter the warning state.

[0037] A dual-dimensional early warning judgment logic, combining seepage pressure threshold and prediction accuracy, is employed. This logic assesses structural risk by comparing predicted seepage pressure with design allowable values, and also checks the operational status of the model and sensors based on anomalies in prediction accuracy. This dual verification of early warning triggering conditions effectively avoids missed and false alarms caused by single-dimensional judgment, improves early warning accuracy, further ensures local real-time analysis, and prevents data lag and misjudgment.

[0038] Specifically, the data collected by the osmotic pressure sensor is timestamped and stored in a memory card to form a traceable time-series dataset.

[0039] When collecting seepage pressure data, a collection timestamp is added to each data point simultaneously, and the time-series data is stored in chronological order on a memory card to ensure the temporal correlation of the data. This ensures the traceability of the monitoring data, provides an accurate time benchmark for subsequent time-series analysis and forecasting, and facilitates maintenance personnel in tracing historical anomalies and identifying patterns in the development of potential hazards.

[0040] Specifically, the voltage regulator module is a DC-DC step-down module that converts the 12V output voltage from the battery to 5V, specifically for use by the Raspberry Pi, processing chip, LED warning light, and underwater acoustic beacon.

[0041] A DC-DC step-down module is used to convert the 12V high voltage output from the battery into the 5V operating voltage required by low-voltage devices such as Raspberry Pi and processing chips, while simultaneously achieving stable voltage output and suppressing voltage fluctuations. This ensures long-term stable power supply to the system and improves the operational reliability of the device in deeply buried, high-humidity environments.

[0042] An offline monitoring and early warning method for deeply buried long water diversion tunnels, implemented based on the aforementioned system, includes the following steps:

[0043] S1 deploys multiple monitoring and early warning units along the tunnel, with each unit independently monitoring, performing edge computing, and determining the early warning level;

[0044] The S2 anomaly monitoring and early warning unit's LED is constantly lit in yellow / red, and the underwater acoustic beacon continuously emits specific low-frequency acoustic pulses;

[0045] The S3 underwater robot receives specific acoustic signals and determines the inspection sequence from near to far based on the signal strength.

[0046] The S4 robot arrives at the abnormal locations in sequence, photographs the LED status, and records it.

[0047] After collecting data from all abnormal locations, the robot will transmit the data back to the shore.

[0048] Each monitoring unit independently completes local monitoring, calculation, and early warning judgment. It only emits a specific low-frequency sound signal to summon a robot when an anomaly is detected. The robot then completes the centralized collection and transmission of abnormal data, realizing an offline working mode of distributed sensing and centralized collection. This solves the problems of existing technologies such as underwater optical interference, waiting for display, and reliance on the cloud, achieving stable, low-power, and real-time tunnel safety monitoring and early warning, and is suitable for long-distance, multi-point tunnel monitoring scenarios.

[0049] Furthermore, the monitoring and early warning unit in step S1 is arranged and installed according to the following structure and method:

[0050] A monitoring and early warning unit is installed at fixed intervals along the tunnel axis. The battery, voltage stabilization module, edge computing module, LED warning lights of the early warning module, and underwater acoustic beacon are all integrated and installed inside the same waterproof sealed shell. The waterproof sealed shell is fixed to the inner wall of the tunnel lining with expansion bolts or epoxy resin adhesive. The waterproof sealed shell has a transparent waterproof observation window, and the LED warning lights and underwater acoustic beacon are arranged close to the transparent waterproof observation window and facing the inside of the tunnel. The micro-hydroelectric generator is independently and fixedly installed on the inner wall of the tunnel, above the tunnel floor, and charges the lithium-ion battery inside the waterproof sealed shell via wires. The permeability pressure sensor is buried in a monitoring borehole between the lining and the surrounding rock. The borehole diameter is 10-12 cm, the borehole is filled with sand and gravel, and the borehole opening is sealed with cement mortar. The permeability pressure sensor is connected to the edge computing module inside the waterproof sealed shell via a sealed cable.

[0051] All electronic control modules are integrated into a single waterproof sealed housing, achieving overall waterproof protection. The hydroelectric generator is installed at a high position to avoid impact from sand and gravel, and the seepage pressure sensor is buried in the borehole for internal seepage pressure monitoring. Reliable connection between the internal and external modules is achieved through sealed cables. This system achieves IP68-level waterproof protection in high-water-pressure, high-humidity tunnel environments, while preventing damage to the generator from sand and gravel accumulation and ensuring the accuracy of sensor monitoring. The modular installation method significantly reduces the difficulty of on-site construction and subsequent maintenance.

[0052] Furthermore, after the edge computing module completes the prediction, it automatically generates a curve showing the change between historical data and predicted data for the next 7 days of osmotic pressure, and stores it along with the early warning information for the underwater robot to collect and transmit back to the shore.

[0053] The edge computing module integrates historical monitoring data with future prediction data, automatically generates visualized change curves, stores the raw data and visualized charts together, and stores them synchronously with early warning information, without the need for cloud synthesis, for robots to collect and transmit back.

[0054] By transforming abstract time-series data into intuitive trend charts, operations and maintenance personnel can quickly grasp the historical changes and future trends of seepage pressure, significantly improving the efficiency of anomaly analysis and emergency response, and lowering the barrier to professional analysis. It achieves real-time data visualization with minimal data transmission, allowing operations and maintenance personnel to quickly assess the safety status without delay or consuming cloud resources.

[0055] The beneficial effects of this invention are:

[0056] 1. This invention abandons the shooting and transmission method of underwater robot offline QR code + photosensitive switch to wake up the display. It adopts underwater acoustic beacon guidance and positioning and three-color LED graded early warning. The signal is easy to identify and is not affected by underwater turbidity, light scattering, etc., realizing the simplified transmission of risk information in the complex environment of water diversion tunnels and solving the problem of poor transmission stability of existing technologies.

[0057] 2. Based on offline monitoring, this invention incorporates an edge computing unit, enabling real-time prediction of seepage pressure, determination of safety level, and automatic early warning at the data acquisition terminal. It can predict the data trend for the next 7 days in advance, without relying on cloud computing power or manual processing, and the prediction cycle is synchronized with the transmission cycle, ensuring the continuity of tunnel risk investigation and solving the problem of data transmission lag in existing monitoring devices.

[0058] 3. This monitoring system enables real-time data updates, predictions, assessments, and early warnings without relying on cloud networks and remote computing power, thus avoiding the problem of insufficient computing power caused by the massive amounts of raw data collected by sensors being processed uniformly in the cloud.

[0059] 4. This invention constructs a complete closed-loop monitoring and early warning system, forming a complete intelligent monitoring closed loop. It achieves edge data analysis, real-time risk warning, and simplified offline transmission, effectively solving the problems of stability and timeliness of data transmission and processing, and providing strong support for engineering risk investigation. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the offline monitoring device of the present invention;

[0061] Figure 2 This is a schematic diagram illustrating the steps of the entire offline monitoring and security early warning process implemented in this invention. Detailed Implementation

[0062] To make the invention clearer and easier to understand, the following describes in further detail an offline monitoring and early warning system and device for deeply buried long water diversion tunnels, in conjunction with the above-mentioned accompanying drawings and specific embodiments.

[0063] Example 1: An offline monitoring and early warning system for a deep-buried long water diversion tunnel includes multiple monitoring and early warning units deployed along the tunnel and an underwater robot. The underwater robot is used to receive underwater acoustic beacon signals emitted by the monitoring and early warning units, locate abnormal monitoring and early warning units, collect the status of LED warning lights in the monitoring and early warning units, and transmit the monitoring and early warning data back to the shore.

[0064] like Figure 1 As shown, an offline monitoring and early warning unit for a deep-buried long water diversion tunnel includes a power supply module, a pressure monitoring module, an edge computing module, and an early warning module.

[0065] The power supply module includes a micro hydroelectric generator 1, a storage battery 2, and a voltage regulator module 8.

[0066] The miniature hydroelectric generator 1 is fixed to the inner wall of the tunnel with bolts to prevent damage from sand and gravel impacts when installed at the bottom. It is connected to the battery 2 via wires. Under the impact of the water flow inside the tunnel, the hydroelectric generator 1 converts kinetic energy into electrical energy and stores it in the battery 2, thus providing a stable power supply for the other modules. In this embodiment, both the battery 2 and the pressure sensor 6 have a voltage of 12V and are directly connected via wires. The voltage regulator module 8 is connected to both the battery 2 and the Raspberry Pi 7. The Raspberry Pi 7 has a voltage of 5V. Direct connection to the battery 2 is prone to burning out and short-circuiting. Connecting to the voltage regulator module 8 can stabilize the voltage and protect the device.

[0067] The pressure monitoring module includes an osmotic pressure sensor 6. The osmotic pressure sensor 6 is installed via a drilled hole, ensuring the measuring end is not blocked by mud or sand. After the sensor is placed in the hole, it is filled with sand and gravel to isolate the soil, and then sealed tightly with cement mortar to prevent pore water from seeping out along the borehole. The sensor measures the osmotic pressure data at the measuring point using pore water pressure, with a monitoring frequency of 10 minutes per measurement. The osmotic pressure sensor 6 is directly connected to the battery 2 via wires and to the Raspberry Pi 7 via a USB data transmission cable, secured with DuPont wires to prevent loosening due to water flow impact. The sensor collects data and stores it on the Raspberry Pi 7's built-in memory card 4, where edge computing is performed on the processing chip 3.

[0068] The edge computing module includes a memory card 4, a processing chip 3, and a Raspberry Pi 7. After the permeability pressure sensor 6 collects data, it inputs the actual data into the memory card 4 for storage. The complete dataset is then input into the processing chip 3 for ARMA fusion prediction calculation. During the calculation process, a fixed prediction window length limits the prediction accuracy and complexity, taking into full account the processing power and prediction accuracy requirements of the microprocessor. Currently, the prediction window length is set to 1008 data points, but it can be adjusted according to the specific needs inside the tunnel. Each time raw permeability pressure data is received, it is stored in the measured dataset, and the data count is incremented by 1. A prediction condition judgment is then performed. If the count value increases to 1008 after the previous prediction, a prediction trigger signal is generated.

[0069] The early warning module includes an LED warning light 5, a waterproof sealed housing 10, and an underwater acoustic beacon 11. The waterproof sealed housing 10 has a small transparent waterproof observation window 12 on its surface, and is secured to the inner wall of the water diversion tunnel near the sensor using waterproof tape 9. The underwater acoustic beacon 11 and the LED warning light 5 are both housed inside the waterproof sealed housing, fixed to a plastic bracket inside the housing, and positioned as close as possible to the transparent waterproof observation window 12. The underwater acoustic beacon 11 assists the underwater inspection robot in locating abnormal points, and the LED warning light 5 transmits the abnormal signal level. When the safety level is determined to be a warning, the underwater acoustic beacon emits a specific low-frequency acoustic pulse, and the LED warning light turns solid yellow. When the safety level is determined to be dangerous or the prediction accuracy is abnormal, the underwater acoustic beacon emits a pulse, and the LED warning light turns solid red. If the safety level is determined to be safe and the curve data is normal, no warning is triggered, the underwater acoustic beacon does not respond, the LED warning light remains solid green, and the device continues to operate normally.

[0070] Component specifications for each module are optional:

[0071] Power supply module:

[0072] Miniature hydroelectric generator: DC12V output, starting water flow velocity ≥0.3m / s, protection rating IP68.

[0073] Battery: 12V / 5000mAh, charge / discharge cycles ≥1000 times

[0074] Voltage regulator module: DC-DC step-down module, 12V to 5V / 3A, ripple <50mV

[0075] Pressure monitoring module:

[0076] Osmotic pressure sensor: measuring range 0~10MPa, accuracy 0.5% FS, 12V power supply, RS485 / USB output.

[0077] Monitoring frequency: 10 minutes / time, which can be modified remotely via the program.

[0078] Edge computing module:

[0079] Raspberry Pi: ARM Cortex-A72 quad-core, 5V / 3A power supply, running Linux.

[0080] Processing chip: Built-in 32-bit high-performance MCU, main frequency ≥240MHz

[0081] Storage card: 32GB high-speed TF card, with circular storage, capable of storing ≥1 year of monitoring data.

[0082] Early warning module:

[0083] LED warning light: Red / yellow / green tri-color, high brightness, waterproof, visible distance ≥5m

[0084] Underwater acoustic beacons: specific low-frequency acoustic pulses, frequency 8–16 kHz, underwater propagation distance ≥100 m

[0085] All wiring connections of this device are triple-sealed using heat shrink tubing, waterproof adhesive, and water-stopping tape; the waterproof sealing shell is made of ABS engineering plastic or 316 stainless steel with a pressure resistance of ≥10MPa; the sensor cables are Kevlar-reinforced waterproof cables, resistant to water flow tension and sand and gravel abrasion; the entire device meets the long-term operation requirements of deep-buried tunnels in high water pressure, high humidity, and high corrosion environments.

[0086] Example 2: As Figure 2 As shown, an offline monitoring and early warning method for deeply buried long water diversion tunnels includes the following steps:

[0087] Step 1: Drill holes at the monitoring locations on the inner wall of the tunnel, embed the seepage pressure sensor 6 inside the holes, and fill the outside of the sensor with sand to isolate the seepage water inside the tunnel to the greatest extent.

[0088] Step 2: Place the waterproof sealing shell 10 into the drill hole, connect the permeation pressure sensor 6 to the battery 2 and the Raspberry Pi 7, and connect the underwater acoustic beacon 11 and the LED warning light 5 to the Raspberry Pi 7 and attach them to the transparent waterproof observation window 12 of the sealing shell.

[0089] Step 3: Install the micro hydro generator 1 on the inner wall of the tunnel, ensuring that the micro hydro generator 1 is below the water surface so that the water flow can drive the micro hydro generator to work, while preventing sand and gravel from impacting and damaging the micro hydro generator.

[0090] Step 4: After the tunnel is filled with water, the water flow drives the rotor of the micro hydroelectric generator 1 to rotate, converting kinetic energy into electrical energy, which is then transmitted to the battery 2. The battery 2 stores, stabilizes, and distributes the electrical energy. The battery 2 directly provides power to the sensor and, through the voltage regulator module 8, steps down the voltage to provide power to the Raspberry Pi 7.

[0091] Step 5: The seepage pressure sensor 6 monitors the seepage pressure value inside the tunnel lining structure and surrounding rock in real time. The monitoring frequency is set to 10 minutes / time, and the monitoring data is continuously written into the actual measurement dataset. The dataset is updated every 7 days.

[0092] Step 6: The edge computing module fixes the prediction window length to 1008 data points. After receiving the raw osmotic pressure data, it stores it in the measured dataset, increments the data count by 1, and performs prediction condition judgment. It checks whether the new data count in the dataset after the previous prediction is 1008. If so, it immediately generates a prediction trigger signal; otherwise, it continues to collect new data until the condition is met.

[0093] After receiving the prediction trigger signal, the processing chip identifies the latest monitoring dataset and fits it using an autoregressive moving average (ARMA) (p,q) model, assuming a historical monitoring data sequence over a continuous period of 7 days. The mathematical expression for the autoregressive moving average (ARMA) (p,q) model is:

[0094]

[0095] in, This is the predicted value at time t. Here, p represents the autoregressive coefficient, and p is the autoregressive order. Here, q represents the moving average coefficient, and q represents the order of the moving average. It is a white noise sequence.

[0096] Perform edge computation on the original data using the following steps:

[0097] Step 61: Data Preprocessing and Stationarity Test

[0098] Using 1008 time-series monitoring data points as the analysis sequence, stationarity was first tested:

[0099] Calculate the mean, variance, and autocorrelation coefficient of the sequence to determine whether they change significantly over time;

[0100] If the mean / variance of the sequence changes significantly over time, it is determined to be a non-stationary sequence. Then, it is subjected to first-order differencing to obtain a differencing sequence. After differencing, the stationarity is tested again until the sequence meets the stationarity requirement.

[0101] Step 62: Roughly estimate the range of p and q using ACF and PACF.

[0102] Calculations for stationary time series data:

[0103] Autocorrelation function (ACF): used to estimate the order q of a moving average.

[0104] Partial autocorrelation function (PACF): used to estimate the autoregressive order p.

[0105] Specific steps:

[0106] Draw the ACF and PACF diagrams and observe the truncation position;

[0107] If the ACF rapidly approaches 0 after order q, then q takes that order.

[0108] If PACF rapidly approaches 0 after order p, then p takes that order.

[0109] In this method, p≤5 and q≤3 are initially specified.

[0110] Step 63: Determine the optimal p and q using the AIC criterion.

[0111] Within the range of p≤5 and q≤3, traverse all possible (p,q) combinations, establish ARMA models for each combination, and calculate the AIC information criterion for each model:

[0112] AIC = −2ln(L) + 2k

[0113] Where: L is the likelihood function value of the model, and k is the total number of model parameters.

[0114] Choose the pair (p,q) with the smallest AIC value as the optimal order.

[0115] Step 64: Solve for model parameters φ, θ using maximum likelihood estimation.

[0116] For an ARMA model with a determined optimal order, the parameters are solved using the maximum likelihood estimation method. The main process is as follows:

[0117] Assuming the residuals follow a Gaussian white noise distribution, we construct the log-likelihood function of the samples and maximize it through numerical iterative optimization, such as gradient descent or L-BFGS. The output model parameters are: autoregressive coefficients φ, moving average coefficients θ, and noise sequence. .

[0118] Step 65: Fit the data using historical data and calculate the MSE accuracy.

[0119] Using the pre-trained ARMA model, backfit the data from 1008 historical datasets to obtain fitted values. Compare the fitted values ​​with the actual monitoring values ​​from the same period and calculate the root mean square error (RMSE). If the MSE is greater than 5%, the model is unreliable, and steps 62-64 are repeated for retraining.

[0120] If MSE ≤ 5%, the model accuracy is acceptable, and proceed to the next step.

[0121] If the MSE still exceeds the limit after three consecutive retraining attempts, it is determined to be a sensor malfunction or system failure, triggering a fault warning.

[0122] Step 66: Ridge Regression Denoising Preprocessing

[0123] The original monitoring data is input into the Ridge regression model for regularization fitting. By suppressing abnormal peaks and water flow noise, smoothed time series data is obtained, which is then input into the ARMA model.

[0124] Step 67: Model predicts 1008 future osmotic pressure data points.

[0125] Using the trained ARMA (p,q) model, point-by-point recursive prediction is performed: the most recent stationary time series data is used as the initial input, and the predicted value of the next time moment is calculated according to the model formula. For each predicted point, the point is added to the input sequence, and the prediction of the next point continues. This process is repeated until the predicted values ​​of the difference sequence for the next 7 days (1008 points) are output.

[0126] Step 68: Restore the difference

[0127] By performing inverse differential operation, the differential prediction value from step 66 is restored to the original dimensional seepage pressure data. The restoration is performed point by point to finally obtain a sequence of 1008 real seepage pressure prediction values ​​for the future.

[0128] Step 7: The early warning module identifies and analyzes the prediction dataset, comparing the model's prediction dataset with the preset safety threshold.

[0129] Predicted value < 80% × Design allowable value → Safe

[0130] 80% × Design Allowable Value ≤ Predicted Value < Design Allowable Value → Warning

[0131] Predicted value ≥ Design allowable value → Danger

[0132] Based on the comparison results, the safety level evaluation was completed according to the preset safety level classification standards (divided into three levels: safe, warning, and dangerous).

[0133] The early warning module combines the final output safety level information and real-time seepage pressure change curve to issue an early warning response. If the safety level is determined to be either warning or dangerous, or the root mean square error of the prediction result exceeds 5% three times consecutively, the early warning unit will be triggered. The underwater acoustic beacon will emit a specific low-frequency acoustic pulse, and the LED warning light will flash, quickly alerting inspection personnel to the potential safety hazard at the location. Specifically, if the safety level is determined to be warning, the underwater acoustic beacon will emit a pulse, and the LED warning light will turn solid yellow. If the safety level is determined to be dangerous, or the prediction accuracy is abnormal, the underwater acoustic beacon will emit a pulse, and the LED warning light will turn solid red. If the safety level is determined to be safe, and the curve data is normal, no early warning will be triggered, the underwater acoustic beacon will not respond, the LED warning light will remain solid green, and the device will continue to operate normally.

[0134] Step 8: The underwater inspection robot inspects the tunnel according to a preset cycle. It locates abnormal risk points by receiving specific low-frequency sound pulses. When it reaches the vicinity of the monitoring device, it takes pictures of the device installation point and records the color of the device's LED warning light and the point number.

[0135] Step 9: After receiving the abnormal signal, the tunnel maintenance personnel will inspect and repair the abnormal location.

[0136] The undescribed parts involved in this invention are the same as or implemented using existing technology.

Claims

1. An offline monitoring and early warning system for deeply buried long water diversion tunnels, characterized in that, It includes multiple monitoring and early warning units deployed along the tunnel and an underwater robot; the monitoring and early warning units include a power supply module, a pressure monitoring module, an edge computing module, and an early warning module; The power supply module includes a micro hydroelectric generator (1), a storage battery (2), and a voltage regulator module (8). The pressure monitoring module includes an osmotic pressure sensor (6); The edge computing module includes a memory card (4), a processing chip (3), and a Raspberry Pi (7); The warning module includes an LED warning light (5), a waterproof sealing shell (10), and an underwater acoustic beacon (11). The micro hydroelectric generator (1) is electrically connected to the battery (2), and the battery supplies power to the Raspberry Pi and the osmotic pressure sensor through the voltage stabilization module; The osmotic pressure sensor is connected to the Raspberry Pi for collecting osmotic pressure data at a preset sampling frequency. The Raspberry Pi has a built-in memory card and processing chip and is connected to an LED warning light and an underwater acoustic beacon. The waterproof sealed shell is equipped with a transparent waterproof observation window, and the LED warning light and underwater acoustic beacon are located inside the shell and close to the observation window; The underwater robot is used to receive underwater acoustic beacon signals, locate anomaly monitoring and early warning units, collect LED warning light status data, and transmit monitoring and early warning data back to the shore.

2. The system according to claim 1, characterized in that, The underwater robot is equipped with an acoustic signal receiving module, an image acquisition module, a data storage module, and a wireless communication module. The underwater robot can determine the distance to abnormal points based on the intensity of low-frequency acoustic signals, and inspect multiple abnormal monitoring and early warning units in order from near to far. After the underwater robot completes the collection of data at each abnormal point, it removes the signal from that point and continues to inspect the next abnormal point in order from near to far until there are no low-frequency acoustic signals.

3. The system according to claim 1, characterized in that, The edge computing module uses a 7-day prediction window. It first uses Ridge regression to reduce noise, then uses the ARIMA model to predict the osmotic pressure for the next 7 days, and calculates the root mean square error (RMSE) of the prediction results. When the RMSE exceeds 5% three times in a row, an early warning is triggered.

4. The system according to claim 1, characterized in that, The early warning module has three status levels: Safety: LED is always green, underwater acoustic beacon is off; Warning: LED yellow is constantly lit, indicating that the underwater acoustic beacon is continuously emitting low-frequency acoustic pulses; Danger: LED is constantly red, underwater acoustic beacon continuously emitting low-frequency sound wave pulses.

5. The system according to claim 4, characterized in that, The rules for determining the early warning are as follows: If the predicted seepage pressure is less than 80% of the design allowable value, then it is safe. If the predicted seepage pressure value is between 80% × the design allowable value and the design allowable value, an early warning will be issued; If the predicted seepage pressure is greater than or equal to the design allowable value, then there is a danger. If the RMSE exceeds 5% three times in a row, it will directly enter the warning state.

6. The system according to claim 1, characterized in that, The data collected by the osmotic pressure sensor is timestamped and stored in a memory card to form a traceable time-series dataset.

7. The system according to claim 1, characterized in that, The voltage regulator module is a DC-DC step-down module that converts the 12V output voltage from the battery to 5V, specifically for use by Raspberry Pi, processing chips, LED warning lights, and underwater acoustic beacons.

8. A method for offline monitoring and early warning of deeply buried long water diversion tunnels, implemented based on the system described in any one of claims 1-7, characterized in that, Including the following steps: S1 deploys multiple monitoring and early warning units along the tunnel, with each unit independently monitoring, performing edge computing, and determining the early warning level; The LED of the S2 anomaly monitoring and early warning unit is constantly lit in yellow / red, and the underwater acoustic beacon continuously emits low-frequency acoustic pulses of the same frequency; The S3 underwater robot receives low-frequency acoustic signals and determines the inspection sequence from near to far based on the signal strength. The S4 robot arrives at the abnormal locations in sequence, photographs the LED status, and records it. After collecting data from all abnormal locations, the robot will transmit the data back to the shore.

9. The method according to claim 8, characterized in that, In step S1, the monitoring and early warning unit is arranged and installed according to the following structure and method: A monitoring and early warning unit is deployed at fixed intervals along the tunnel axis. The battery (2), voltage regulator module (8), edge computing module, LED warning light of the early warning module, and underwater acoustic beacon (11) are all integrated and installed inside the same waterproof sealed shell (10); The waterproof sealing shell (10) is fixed to the inner wall of the tunnel lining by expansion bolts or epoxy resin. The waterproof sealing shell (10) is provided with a transparent waterproof observation window (12). LED warning lights and underwater acoustic beacons (11) are arranged close to the transparent waterproof observation window (12) and facing the inside of the tunnel. The micro hydroelectric generator (1) is independently and fixedly installed on the inner wall of the tunnel at a position higher than the bottom of the tunnel, and charges the lithium-ion battery (2) inside the waterproof sealed shell through the wire; The permeation pressure sensor (6) is buried in the monitoring borehole between the lining and the surrounding rock. The borehole diameter is 10-12cm. The borehole is filled with sand and gravel and the opening is sealed with cement mortar. The permeation pressure sensor (6) is connected to the edge computing module inside the waterproof sealing shell (10) through a sealed cable.

10. The method according to claim 8, characterized in that, After the edge computing module completes the prediction, it automatically generates a curve showing the change between historical seepage pressure data and predicted data for the next 7 days, and stores it along with the early warning information for the underwater robot to collect and transmit back to the shore.

Images