Intelligent monitoring and early warning system based on Internet of Things

Abstract

The invention relates to the field of geological disaster monitoring, and discloses an intelligent monitoring and early warning system based on the Internet of Things, which comprises a distributed sensor network, an edge computing node and a cloud computing platform. The intelligent monitoring node integrates environment, mechanics and acoustic modules, and collects temperature, relative dielectric constant and strain data. And the cloud computing platform processes the acquired data by using a long-short-term memory neural network model to output a collapse risk probability value, and triggers the node to switch to an active detection mode and execute adaptive acoustic scanning when the probability value exceeds a threshold value. The intelligent monitoring node dynamically adjusts the acoustic emission frequency according to the relative dielectric constant, and the edge calculation node calculates the sound wave propagation speed. And a physical parameter inversion module of the cloud computing platform calculates a water-induced softening factor by using the relative dielectric constant, and performs numerical compensation on the apparent shear modulus to obtain an effective skeleton stiffness value. According to the invention, softening caused by non-structural water and structural damage are effectively distinguished, and the false alarm rate and the system power consumption are reduced.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, specifically to an intelligent monitoring and early warning system based on the Internet of Things. Background Technology

[0002] Foundation collapse and settlement are common forms of geological hazards in the maintenance of urban infrastructure and transportation projects, characterized by their suddenness, high degree of concealment, and wide range of damage. In order to ensure the safe and stable operation of critical facilities such as roads, bridges, and underground pipelines, and to prevent major safety accidents caused by soil structural instability, real-time monitoring and early warning of the stress state and structural integrity of the foundation have become a key aspect of disaster prevention and mitigation.

[0003] Existing technologies typically employ a distributed sensor network deployment within the monitoring area for fixed-point monitoring. Common implementation methods include using strain gauges and displacement sensors to collect surface mechanical deformation data, or using acoustic emission detection equipment to capture elastic wave signals generated by internal fractures in soil and rock. These monitoring devices generally transmit the collected raw data to a central monitoring center via wireless communication networks. The monitoring system then performs statistical analysis on the amplitude or frequency changes of the signals based on preset empirical thresholds to determine whether the foundation exhibits a settlement trend or collapse risk.

[0004] However, existing monitoring technologies often overlook the nonlinear fluctuations in the physical parameters of soil as a multiphase medium, caused by changes in environmental humidity. This makes it difficult for detection systems to adapt to complex hydrogeological environments in the field. In particular, commonly used acoustic detection technologies typically transmit signals at a fixed excitation frequency, failing to consider the damping effect of increased soil moisture content on the propagation characteristics of elastic waves. When the monitored area experiences rainfall that causes the soil to become saturated, high-frequency sound waves undergo severe energy attenuation and waveform distortion in the moist medium. This leads to a significant reduction in the effective transmission distance of the detection signal and a sharp drop in the signal-to-noise ratio. Consequently, the monitoring system struggles to obtain accurate information about the internal structure of the foundation under critical and severe weather conditions, seriously affecting the reliability and continuity of disaster early warning. Summary of the Invention

[0005] The first aspect of this invention provides an intelligent monitoring and early warning system based on the Internet of Things.

[0006] The IoT-based intelligent monitoring and early warning system includes a distributed sensor network, a wireless data transmission network, edge computing nodes, and a cloud computing platform. The distributed sensor network consists of several intelligent monitoring nodes deployed in the monitoring area. Each intelligent monitoring node is configured to collect ambient temperature data, relative permittivity data, and strain data of the ground surface. It is also configured to transmit and receive elastic wave signals upon receiving an active detection trigger command. The edge computing nodes communicate with the intelligent monitoring nodes through the wireless data transmission network. These edge computing nodes are configured to aggregate data from the intelligent monitoring nodes and calculate the sound wave propagation speed. The cloud computing platform communicates with the edge computing nodes and includes a high-performance computing server configured to run a long short-term memory neural network model and a physical parameter inversion unit.

[0007] In the IoT-based intelligent monitoring and early warning system, the intelligent monitoring node includes a microprocessor module, an environmental sensing module, a mechanical sensing module, an acoustic detection module, and a power management module. The environmental sensing module includes a frequency-domain reflectoelectric dielectric sensor, configured to measure the resonant frequency shift of emitted electromagnetic waves in the soil to obtain the relative permittivity data. The microprocessor module is configured to read the relative permittivity data and use it as an input parameter for calculating the acoustic detection frequency to adjust the frequency of the emitted acoustic signal. The acoustic detection module includes a piezoelectric ceramic transducer and a transceiver isolation circuit, configured to perform bidirectional conversion between electrical and acoustic signals.

[0008] The edge computing node includes a data aggregation module and a time synchronization module. The time synchronization module is configured to send time synchronization data packets to the intelligent monitoring nodes via a precise time protocol to unify the clock reference of the distributed sensor network. The edge computing node is configured to send cooperative reception commands to adjacent intelligent monitoring nodes located within a preset radius of the transmitting source. The edge computing node is also configured to perform cross-correlation calculations on the standard waveform of the transmitted signal and the received sampled waveform data, calculate the absolute propagation time of the elastic wave in the soil based on the maximum modulus of the cross-correlation function, and then calculate the sound wave propagation velocity data.

[0009] The cloud computing platform includes a deep learning analysis module, a physical parameter inversion module, and an early warning decision module. The deep learning analysis module is configured to use the long short-term memory neural network model to output a collapse risk probability value based on the ambient temperature, the relative permittivity data, and the strain data. The physical parameter inversion module is configured to receive the relative permittivity data and the sound wave propagation velocity data, and perform dielectric-acoustic joint correction and refined inversion calculations to output an effective skeleton stiffness value. The early warning decision module is configured to generate graded alarm signals based on the collapse risk probability value and the effective skeleton stiffness value.

[0010] A second aspect of the present invention provides an intelligent monitoring and early warning method based on the Internet of Things.

[0011] The IoT-based intelligent monitoring and early warning method is executed based on the IoT-based intelligent monitoring and early warning system. The IoT-based intelligent monitoring and early warning method includes: a full-cycle multi-dimensional data acquisition step, a neural network-based primary trend prediction step, a triggered adaptive acoustic scanning step, a dielectric-acoustic multi-parameter joint correction step, and a dual verification and graded early warning step.

[0012] In the full-cycle multidimensional data acquisition step, the intelligent monitoring node acquires the ambient temperature data, the relative permittivity data, and the strain data according to a preset sampling frequency. The edge computing node receives the above data and constructs a multidimensional feature vector containing a unified timestamp.

[0013] In the initial trend prediction step based on the neural network, the cloud computing platform inputs the multidimensional feature vector into the pre-trained long short-term memory neural network model and outputs the collapse risk probability value. The cloud computing platform compares the collapse risk probability value with a preset initial warning threshold and calculates the time rate of change of the relative permittivity data. When the collapse risk probability value is greater than or equal to the initial warning threshold, or the time rate of change is greater than or equal to a preset seepage mutation threshold, the cloud computing platform generates the active detection trigger command and sends it to the intelligent monitoring node.

[0014] In the triggered adaptive acoustic scanning step, the intelligent monitoring node switches from passive waiting mode to active detection mode in response to receiving the active detection trigger command. The intelligent monitoring node uses a pre-stored dielectric-frequency mapping algorithm based on the current relative permittivity data to calculate the target excitation center frequency. When the relative permittivity data indicates an increase in soil moisture content, the intelligent monitoring node reduces the value of the target excitation center frequency. The intelligent monitoring node uses direct digital frequency synthesis logic to generate a frequency-modulated continuous wave digital sequence centered on the target excitation center frequency and drives the piezoelectric ceramic transducer to emit a shear wave signal.

[0015] In the dielectric-acoustic multi-parameter joint correction step, the physical parameter inversion module receives the relative permittivity data and the sound wave propagation velocity data. The physical parameter inversion module uses the relative permittivity data to map and calculate the volumetric water content of the soil, and combines this with a preset dry soil density benchmark to calculate the wet soil density value. The physical parameter inversion module calculates the product of the wet soil density value and the square of the sound wave propagation velocity data to obtain the apparent shear modulus. The physical parameter inversion module substitutes the relative permittivity data into a pre-stored saturation-stiffness correction algorithm to calculate the water softening factor. The physical parameter inversion module divides the apparent shear modulus by the water softening factor to obtain the effective skeleton stiffness value.

[0016] In the dual verification and graded early warning steps, the early warning decision module performs a logical AND operation on the collapse risk probability value and the effective skeleton stiffness value. When the collapse risk probability value is greater than or equal to the primary early warning threshold and the effective skeleton stiffness value is lower than the structural failure threshold, the early warning decision module determines that a structural damage event has occurred in the foundation and generates a graded alarm data packet. When the collapse risk probability value is greater than or equal to the primary early warning threshold but the effective skeleton stiffness value is greater than or equal to the structural failure threshold, the early warning decision module determines it as a non-structural water softening event and generates an environmental interference filtering log.

[0017] This invention provides an intelligent monitoring and early warning system based on the Internet of Things (IoT). It has the following beneficial effects: 1. This invention utilizes the relative permittivity data collected by the environmental sensing module as the control variable for the acoustic detection frequency, and dynamically adjusts the excitation center frequency of the piezoelectric transducer based on a preset dielectric-frequency mapping algorithm. When the soil moisture content is detected to increase, the transmission frequency is automatically reduced to reduce the energy dissipation of elastic waves in the wet medium. This overcomes the signal attenuation problem caused by excessive medium damping in complex hydrogeological environments caused by traditional fixed-frequency acoustic wave detection technology. It ensures that the distributed sensor network can maintain a stable effective detection distance and signal-to-noise ratio under high humidity conditions such as rainfall, and improves the system's adaptability to harsh environments.

[0018] 2. This invention performs joint correction calculations of dielectric and acoustic multi-parameters through the physical parameter inversion module of a cloud computing platform. It calculates the water softening factor by mapping relative permittivity and performs numerical compensation on the apparent shear modulus calculated based on acoustic wave velocity. It removes the non-destructive modulus attenuation component caused by fluctuations in soil moisture content from a physical mechanism perspective. This allows for accurate differentiation between the physical softening of shallow soil caused by rainfall infiltration and the structural stiffness loss event caused by the development of internal cavities. It solves the technical problem that single monitoring indicators are prone to misjudgment when environmental humidity changes, and reduces the false alarm rate of intelligent monitoring and early warning systems.

[0019] 3. This invention employs a dual verification mechanism that combines primary trend prediction using a long short-term memory neural network model with triggered active acoustic scanning. The high-power acoustic detection module is only activated for physical verification when the collapse risk probability value output by the neural network exceeds a threshold or when the dielectric constant changes abruptly. This ensures early identification of minor deformation trends in the foundation while avoiding energy waste caused by sensor nodes being in a high-frequency active detection state for extended periods. It achieves an optimized balance between monitoring timeliness, early warning accuracy, and system operating power consumption, effectively extending the continuous working cycle and maintenance interval of the field distributed sensor network in battery-powered mode. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

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

[0022] Example: Please see the appendix Figure 1This invention provides an intelligent monitoring and early warning system based on the Internet of Things, comprising: a distributed sensor network, a wireless data transmission network, an edge computing node, a cloud computing platform, and a mobile terminal.

[0023] The distributed sensor network consists of several intelligent monitoring nodes deployed in the monitoring area. The intelligent monitoring nodes are distributed in an array along the monitored object, with a deployment density of at least 20 nodes per kilometer to cover the ground deformation area and form a high-density sensing grid.

[0024] The intelligent monitoring node adopts a modular integrated design, specifically including an environmental sensing module, a mechanical sensing module, an acoustic detection module, a microprocessor module, and a power management module.

[0025] The environmental sensing module is equipped with a temperature sensor and a frequency domain reflective humidity sensor. The frequency domain reflective humidity sensor is used to simultaneously collect data on soil volumetric water content and dielectric constant.

[0026] The mechanical sensing module is equipped with strain sensors or displacement sensors, which are used to collect deformation data of the foundation surface.

[0027] The acoustic detection module is equipped with a piezoelectric transducer and a signal excitation circuit. The acoustic detection module is used to transmit and receive elastic wave signals propagating in the soil medium.

[0028] The microprocessor module is electrically connected to the environmental sensing module, the force sensing module, the acoustic detection module, and the power management module via either a Serial Peripheral Interface (SPI) or a Universal Asynchronous Receiver / Transmitter (UART) bus. The microprocessor module controls the data acquisition timing of each sensor and executes local trigger logic operations.

[0029] The power management module is connected to the above modules. The power management module is used to provide operating voltage for the intelligent monitoring node and respond to the sleep control command of the microprocessor module to realize the low power operation of the node.

[0030] The wireless data transmission network communicates with the intelligent monitoring nodes. The wireless data transmission network uses LoRaWAN, NB-IoT, or ZigBee communication protocols and is responsible for transmitting temperature, humidity, dielectric constant, strain, and acoustic velocity data collected by the intelligent monitoring nodes.

[0031] Edge computing nodes are deployed at the monitoring site and connected to the wireless data transmission network via wireless communication links. These nodes receive and aggregate data from the intelligent monitoring nodes and perform preprocessing operations. Preprocessing includes data cleaning, time synchronization calibration between nodes, and preliminary feature extraction of acoustic signals. Edge computing nodes also issue control commands to the intelligent monitoring nodes, coordinating the active detection and collaborative work among them.

[0032] The cloud computing platform and edge computing nodes are connected via fiber optic or 4G / 5G communication links. The cloud computing platform includes data storage servers and high-performance computing servers. The high-performance computing servers are equipped with Long Short-Term Memory (LSTM) neural network models. These LSTM models are trained on historical collapse case datasets and are used to output foundation deformation trend predictions based on received time-series monitoring data. The cloud computing platform also includes a physical parameter inversion unit, which calculates the effective skeleton stiffness of the soil based on dielectric constant and acoustic velocity data, and combines this with the prediction results from the LSTM neural network model to generate graded early warning signals.

[0033] The mobile terminal connects to the cloud computing platform via a mobile communication network. The mobile terminal receives real-time alarm information and system status reports from the cloud computing platform, enabling remote monitoring of the security status of the monitored area.

[0034] The intelligent monitoring node provided by this invention includes: a microprocessor module, an environmental sensing module, a mechanical sensing module, an acoustic detection module, and a power management module.

[0035] The microprocessor module is electrically connected to the environmental sensing module, the mechanical sensing module, the acoustic detection module, and the power management module, respectively. The microprocessor module includes an embedded microcontroller (MCU), which integrates an analog-to-digital converter (ADC), a digital-to-analog converter (DAC), and direct digital frequency synthesis logic (DDS). The embedded microcontroller is configured to perform synchronous acquisition of multi-channel data and waveform generation of excitation signals.

[0036] The environmental sensing module includes a temperature sensor and a frequency-domain reflectoelectric dielectric sensor. The temperature sensor is thermally coupled inside the housing of the intelligent monitoring node and is used to collect ambient temperature data. The frequency-domain reflectoelectric dielectric sensor comprises a metal probe array that extends outside the intelligent monitoring node and is inserted into the soil medium to be measured. The frequency-domain reflectoelectric dielectric sensor obtains the volumetric water content and relative permittivity data of the soil by measuring the impedance spectrum or reflection frequency of the emitted electromagnetic wave in the soil. The microprocessor module reads the relative permittivity data and uses it as an input parameter for calculating the acoustic detection frequency.

[0037] The mechanical sensing module includes a group of resistive strain gauges. These strain gauges are connected via a Wheatstone bridge circuit and fixed to the stress points on the foundation surface. The microprocessor module acquires the differential voltage signal output from the strain gauge group via an instrumentation amplifier circuit and converts this signal into strain values ​​characterizing the foundation deformation.

[0038] The acoustic detection module includes a piezoelectric ceramic transducer, a high-voltage power amplifier circuit, a transceiver isolation circuit, and a signal conditioning and receiving circuit. The piezoelectric ceramic transducer is made of piezoelectric ceramic material and is configured to perform bidirectional conversion between electrical and acoustic signals.

[0039] When the intelligent monitoring node is in transmit mode, the microprocessor module generates a frequency-modulated continuous wave signal through the digital-to-analog converter unit based on the calculated optimal excitation frequency. The high-voltage power amplifier circuit boosts and amplifies the voltage of the frequency-modulated continuous wave signal, driving the piezoelectric ceramic transducer to generate elastic shear waves or longitudinal waves that propagate into the soil medium. At this time, the transceiver isolation circuit is in a blocked state, protecting the signal conditioning receiver circuit from damage caused by the high-voltage signal.

[0040] When the intelligent monitoring node is in receive mode, the transceiver isolation circuit is in the conducting state. The piezoelectric ceramic transducer senses the elastic wave vibration propagating through the soil medium and converts it into an analog electrical signal. The signal conditioning receiving circuit performs bandpass filtering and low-noise amplification on the analog electrical signal. The microprocessor module digitally samples the processed analog electrical signal through the analog-to-digital converter unit.

[0041] The power management module includes a DC-DC conversion circuit and an energy storage unit. It regulates the operating voltage of the intelligent monitoring nodes. The microprocessor module controls the power supply to the environmental sensing module, mechanical sensing module 2, and acoustic detection module via the power management module to achieve intermittent operation.

[0042] The edge and cloud architecture provided by this invention mainly consists of edge computing nodes deployed on the monitoring site and a cloud computing platform deployed on a remote server.

[0043] Edge computing nodes are built on industrial-grade embedded computers and include a central processing unit, memory, and communication interfaces. They are configured to run a real-time operating system and are logically divided into a data aggregation module, a time synchronization module, and an edge inference module.

[0044] The data aggregation module connects to the wireless data transmission network via a wired bus or wireless communication interface. It receives raw sensor data packets from intelligent monitoring nodes. The module performs communication protocol conversion, transforming heterogeneous sensor data into a standard data exchange format. Furthermore, it is configured to execute data cleaning algorithms, using amplitude limiting or median filtering to remove outliers from the raw sensor data packets.

[0045] The time synchronization module maintains a unified clock reference for the local network. During active acoustic detection missions, the time synchronization module sends time synchronization data packets to relevant intelligent monitoring nodes based on the Precision Time Protocol (IEEE 1588 PTP). The module corrects network transmission delays using hardware timestamps, keeping the clock synchronization error between the transmitting and receiving nodes within a preset microsecond-level accuracy range.

[0046] The edge inference module's storage space contains a trigger threshold rule table. The edge inference module extracts humidity change rate data and strain change rate data output by the data aggregation module and compares this data with the trigger threshold rule table. When the monitored data exceeds a preset threshold, the edge inference module generates a control command and sends it to the intelligent monitoring node. This control command drives the intelligent monitoring node to switch from passive monitoring mode to active detection mode.

[0047] The cloud computing platform is built on a distributed server cluster and includes physically connected computing and storage units. Functionally, the cloud computing platform includes a time-series database module, a deep learning analysis module, a physical parameter inversion module, and an early warning and decision-making module.

[0048] The time-series database module stores historical monitoring data uploaded by the intelligent monitoring nodes. It establishes timestamp indexes for temperature, humidity, strain, and acoustic wave velocity data.

[0049] The deep learning analytics module deploys the executable code of a Long Short-Term Memory (LSTM) neural network model. It is configured to train the LSTM neural network model by reading historical collapse case datasets from the time-series database module. During the real-time monitoring phase, the deep learning analytics module inputs real-time temperature, humidity, and strain sequences into the trained LSTM neural network model, outputting predicted deformation trends and collapse risk probabilities for the foundation.

[0050] The physical parameter inversion module is the processing unit that implements multi-physics coupling correction. It receives relative permittivity data and acoustic wave velocity data uploaded by the intelligent monitoring nodes. The module calls a pre-stored saturation-stiffness correction algorithm, which uses the relative permittivity to calculate the water softening factor and compensates for the apparent shear modulus corresponding to the acoustic wave velocity, outputting the effective soil skeleton stiffness value after removing the influence of water.

[0051] The early warning decision module is connected to the deep learning analysis module and the physical parameter inversion module. The early warning decision module receives the collapse risk probability value and the effective skeleton stiffness value. It performs a logical AND operation on the collapse risk probability value and the effective skeleton stiffness value according to a preset logical decision strategy. When the collapse risk probability value is greater than a probability threshold and the effective skeleton stiffness value is lower than a structural failure threshold, the early warning decision module generates a graded alarm signal and sends it to the mobile terminal via a mobile communication network gateway.

[0052] This invention provides an IoT-based intelligent monitoring and early warning method, which is executed based on the aforementioned IoT-based intelligent monitoring and early warning system. The IoT-based intelligent monitoring and early warning method includes the following steps: Step S1: Full-cycle multi-dimensional data acquisition. The intelligent monitoring nodes in the distributed sensor network perform environmental parameter acquisition and mechanical parameter acquisition according to a preset sampling frequency. The sampling frequency is set to at least once per minute. The intelligent monitoring nodes use the environmental sensing module to collect environmental temperature data and relative permittivity data corresponding to the soil volumetric water content of the monitoring area. The intelligent monitoring nodes use the mechanical sensing module 2 to collect strain data of the foundation surface. The edge computing node receives the environmental temperature data, relative permittivity data, and strain data, and constructs a multi-dimensional feature vector containing a unified timestamp.

[0053] Step S2: Primary Trend Prediction Based on Neural Networks. The cloud computing platform inputs the multidimensional feature vectors into a pre-trained Long Short-Term Memory (LSTM) neural network model. Based on the multidimensional feature vectors from the current and historical moments, the LSM model outputs a predicted value for the foundation's deformation trend and a probability value of collapse risk. The cloud computing platform compares the probability value of collapse risk with a preset primary warning threshold. When the probability value of collapse risk is less than the primary warning threshold, the system maintains the operation of Step S1; when the probability value of collapse risk is greater than or equal to the primary warning threshold, the cloud computing platform generates an active detection trigger command and sends it to the intelligent monitoring node, executing Step S3.

[0054] Step S3: Triggered Adaptive Acoustic Scanning Step. In response to receiving the active detection trigger command, the intelligent monitoring node switches from passive waiting mode to active detection mode. The intelligent monitoring node reads the relative permittivity data at the current moment and retrieves the corresponding target acoustic excitation frequency according to a preset dielectric-damping mapping table. The acoustic detection module of the intelligent monitoring node uses the target acoustic excitation frequency to drive a piezoelectric transducer to emit elastic waves into the soil medium and receives the echo signal after propagation through the soil. The edge computing node calculates the propagation speed of the elastic wave in the soil based on the cross-correlation delay between the emitted signal and the echo signal.

[0055] Step S4: Dielectric-Acoustic Multi-parameter Joint Correction Step. The physical parameter inversion module of the cloud computing platform receives the relative permittivity data and the propagation velocity. The physical parameter inversion module substitutes the relative permittivity data into a pre-stored saturation correction function to calculate the water softening factor. The physical parameter inversion module uses the propagation velocity to calculate the apparent shear modulus, and uses the water softening factor to numerically compensate for the apparent shear modulus, outputting the effective skeleton stiffness value after removing the influence of water softening.

[0056] Step S5: Dual Verification and Graded Early Warning Step. The early warning decision module of the cloud computing platform performs a logical AND operation on the collapse risk probability value and the effective skeleton stiffness value. Only when the collapse risk probability value is greater than or equal to the primary early warning threshold and the effective skeleton stiffness value is lower than the structural failure threshold, the early warning decision module determines that structural damage has occurred in the foundation. The cloud computing platform generates an alarm command and issues an alarm through a mobile terminal.

[0057] Step S1 provided by this invention: The full-cycle multi-dimensional data acquisition step specifically includes the following execution process: In the distributed sensor network, intelligent monitoring nodes read configuration parameters from their local memory during the initialization phase. The intelligent monitoring nodes then lock the sampling period based on these configuration parameters. In this embodiment, the sampling period is set to 60 seconds, meaning the intelligent monitoring node performs a wake-up and data acquisition operation once per minute. The deployment density of intelligent monitoring nodes along the monitored road section is set to 20 nodes per kilometer to form a spatially continuous monitoring coverage grid.

[0058] When the sampling time is reached, the microprocessor module of the intelligent monitoring node sends a data acquisition enable signal to the environmental sensing module and the mechanical sensing module. The microprocessor module reads the current ambient temperature value output by the temperature sensor in the environmental sensing module via the serial communication bus.

[0059] Subsequently, the microprocessor module drives the frequency-domain reflectoelectric dielectric sensor in the environmental sensing module to perform dielectric measurements. The frequency-domain reflectoelectric dielectric sensor uses an oscillating circuit to emit electromagnetic waves at a frequency of 0 MHz towards a metal probe array inserted into the soil. The frequency-domain reflectoelectric dielectric sensor detects the resonant frequency shift caused by the propagation of the electromagnetic waves in the soil medium. The microprocessor module substitutes this resonant frequency shift into a preset dielectric-frequency calibration equation to calculate the volumetric water content of the soil and the corresponding relative permittivity. The microprocessor module stores the relative permittivity as an environmental humidity characteristic parameter and marks it as a physical correction parameter for subsequent acoustic inversion.

[0060] Simultaneously, the microprocessor module activates the resistive strain gauge array in the mechanical sensing module. The microprocessor module controls the instrumentation amplifier circuit to sample the output voltage of the Wheatstone bridge circuit containing the resistive strain gauge array. The microprocessor module uses a preset sensitivity factor to convert the acquired analog voltage signal into a digital value representing the ground surface strain.

[0061] After collecting the above three sets of physical quantities, the microprocessor module encapsulates the current ambient temperature, volumetric water content, relative permittivity, and foundation surface strain into a single-frame monitoring data packet. The microprocessor module writes a globally unified timestamp synchronized by the edge computing node via the network and the unique device identifier of the intelligent monitoring node into the header of the single-frame monitoring data packet.

[0062] The intelligent monitoring node sends the single-frame monitoring data packet to the edge computing node via a wireless communication protocol. The data aggregation module of the edge computing node receives the single-frame monitoring data packet and performs integrity verification on it using a Cyclic Redundancy Check (CRC) algorithm. Once the verification passes, the edge computing node aggregates data from intelligent monitoring nodes at different locations at the same time, constructs a multi-dimensional feature vector matrix for the current moment, and stores this multi-dimensional feature vector matrix in the cache queue of the data aggregation module.

[0063] Step S2 provided by the invention: The primary trend prediction step based on a long short-term memory neural network specifically includes the following execution process: The deep learning analytics module of the cloud computing platform periodically accesses the time-series database module via its internal data bus. The deep learning analytics module extracts a multidimensional feature vector sequence within a sliding time window from the time-series database module. This multidimensional feature vector sequence consists of ambient temperature values, relative permittivity, and ground surface strain values ​​arranged in chronological order.

[0064] The deep learning analysis module performs min-max normalization on the multidimensional feature vector sequence. It linearly maps the ambient temperature, relative permittivity, and foundation surface strain values ​​to the closed interval [0, 1], eliminating differences in physical dimensions. The deep learning analysis module then reconstructs the processed data into a three-dimensional input tensor, where the dimensions are defined as (batch size, time step, feature dimension).

[0065] The deep learning analysis module inputs the three-dimensional input tensor into the Long Short-Term Memory (LSTM) neural network model. The LSTM neural network model is in a pre-training convergence state, and its network weight parameters are obtained through supervised training using a historical collapse case dataset. The LSTM neural network model topologically comprises an input layer, multiple stacked hidden layers, and a fully connected output layer. Each hidden layer consists of several LSTM cells, and each LSTM cell integrates a forget gate logic circuit, an input gate logic circuit, and an output gate logic circuit.

[0066] The deep learning analysis module utilizes the Long Short-Term Memory (LSTM) neural network model to perform forward inference operations. The fully connected output layer outputs a collapse risk probability value. This collapse risk probability value is a scalar value processed by the Sigmoid activation function.

[0067] The cloud computing platform compares the collapse risk probability value with a preset primary warning threshold. Simultaneously, the cloud computing platform calculates the rate of change of the relative permittivity over time and compares this rate of change with a preset seepage mutation threshold.

[0068] When the collapse risk probability value is less than the primary warning threshold and the time change rate is less than the seepage mutation threshold, the cloud computing platform determines that the current foundation is in a safe and stable state, generates a normal operation log, and ends the judgment process of the current cycle.

[0069] When the probability value of the collapse risk is greater than or equal to the primary warning threshold, or the rate of change over time is greater than or equal to the seepage mutation threshold, the cloud computing platform determines that there is a potential geological disaster risk in the monitored area. At this time, the cloud computing platform generates an active detection trigger command. The active detection trigger command contains the device network address of the intelligent monitoring node in the target area. The cloud computing platform sends the active detection trigger command to the corresponding intelligent monitoring node through the edge computing node, driving the system to enter the physical verification process described in step S3.

[0070] Step S3 of the present invention: the triggered adaptive acoustic scanning step specifically includes the following execution process: In response to receiving an active detection trigger command from a cloud computing platform or edge computing node, the microprocessor module of the intelligent monitoring node performs a working mode switching operation. The microprocessor module sends a wake-up level signal to the power management module, connects the power supply circuit of the acoustic detection module, and drives the intelligent monitoring node to switch from passive standby mode to active detection mode.

[0071] The microprocessor module reads the relative permittivity data from its local memory. It then invokes a pre-stored dielectric-frequency mapping algorithm. This algorithm contains a set of inverse proportional functions used to dynamically adjust the acoustic wave frequency based on the soil moisture content to counteract medium damping. The microprocessor module substitutes the relative permittivity data into the dielectric-frequency mapping algorithm to calculate the target excitation center frequency. In this embodiment, the target excitation center frequency is set to a range of 1 kHz to kHz. When the relative permittivity data indicates that the soil is nearing saturation, the microprocessor module adjusts the target excitation center frequency to a low-frequency range of 1 kHz to 5 kHz to ensure the penetration distance of the elastic wave in the moist soil.

[0072] The microprocessor module uses direct digital frequency synthesis logic to generate a frequency-modulated continuous wave (Chirp) digital sequence. The frequency coverage of the Chirp digital sequence is set to a scanning range of 2kHz to 5kHz centered on the target excitation center frequency. The microprocessor module converts the Chirp digital sequence into an analog excitation signal through a digital-to-analog converter.

[0073] The high-voltage power amplifier circuit in the acoustic detection module receives the analog excitation signal and performs voltage gain amplification to generate a high-voltage drive signal. This high-voltage drive signal excites the piezoelectric ceramic transducer to produce mechanical vibration, emitting shear wave signals into the surrounding soil medium. During the transmission phase, the microprocessor module controls the transceiver isolation circuit to be in a high-impedance disconnected state, protecting the signal conditioning receiving circuit.

[0074] Edge computing nodes send collaborative reception commands to adjacent smart monitoring nodes located within a 5-meter to 1-meter radius of the transmitter. These adjacent smart monitoring nodes function as receiving nodes. The microprocessor module of each receiving node controls the transceiver isolation circuit to be in a low-impedance conduction state. The piezoelectric ceramic transducer of the receiving node acquires the shear wave signal propagating through the soil medium and converts it into a weak electrical signal. The signal conditioning receiving circuit performs bandpass filtering and low-noise amplification on the weak electrical signal. The microprocessor module of the receiving node uses an analog-to-digital converter to sample the signal and records the arrival timestamp of the signal waveform. The arrival timestamp is generated based on a local clock calibrated according to the Precision Time Protocol (IEEE 1588 PTP), with a time synchronization accuracy better than microseconds.

[0075] The receiving node uploads the echo data packet containing the received timestamp and sampled waveform data to the edge computing node via a wireless data transmission network. The edge computing node extracts the transmission timestamp from the transmitting node and the reception timestamp from the receiving node. The edge computing node performs a cross-correlation operation on the standard waveform of the transmitted signal and the received sampled waveform data. The edge computing node searches for the maximum modulus point of the cross-correlation function and calculates the absolute propagation time of the elastic wave in the soil. The edge computing node, combined with the known geometric distance between the transmitting and receiving nodes, calculates the current sound wave propagation velocity in the soil medium.

[0076] The step S4 provided by this invention: the dielectric-acoustic joint correction and refined inversion step specifically includes the following execution process: the physical parameter inversion module of the cloud computing platform receives the acoustic wave propagation speed data from the edge computing node and the relative permittivity data of the same monitoring point at the current time from the time series database module through the internal data interface.

[0077] The physical parameter inversion module performs dynamic correction calculations for soil density. The physical parameter inversion module reads the system's preset dry soil density benchmark value (…). The physical parameter inversion module uses the Topp formula or a preset dielectric-water content calibration curve to convert the relative permittivity data into the volumetric water content of the soil. ).

[0078] The physical parameter inversion module calculates the density value of wet soil under the current water content state based on the principle of mixture density. The formula for calculating the density of wet soil is: ,in is the density constant of water at room temperature.

[0079] The physical parameter inversion module calculates the apparent shear modulus using the elastic wave dynamics formula. The module calculates the product of the wet soil density value and the square of the sound wave propagation velocity data to obtain the apparent shear modulus. The apparent shear modulus characterizes the overall dynamic stiffness of the soil-water composite medium under the current water content state.

[0080] The physical parameter inversion module calls the saturation-stiffness correction algorithm in local memory to calculate the water softening factor. The physical parameter inversion module defines a normalized water content index, which is determined by the current relative permittivity data, the reference value of the permittivity in the dry state, and the reference value of the permittivity in the saturated state.

[0081] The physical parameter inversion module calculates the water softening factor using the following correction function. : in, For the current relative permittivity data, The dielectric constant reference value for the dry state is given. The reference value for the dielectric constant in the saturation state is given. The softening sensitivity coefficient, The nonlinear shape index is used. The reference values ​​of the dielectric constant in the dry state, the dielectric constant in the saturated state, the softening sensitivity coefficient, and the nonlinear shape index are all fixed constants pre-entered into the physical parameter inversion module based on the geotechnical engineering investigation report of the monitoring area.

[0082] The physical parameter inversion module performs a refined inversion calculation of the skeleton stiffness. The physical parameter inversion module divides the apparent shear modulus by the water softening factor to obtain the effective skeleton stiffness value (…). The physical parameter inversion module performs a moving average filtering process on the effective skeleton stiffness value and outputs the processed effective skeleton stiffness value to the early warning decision module.

[0083] Step S5 of this invention: the dual verification and cloud-based early warning step specifically includes the following execution process: The early warning decision module of the cloud computing platform reads the collapse risk probability value from the deep learning analysis module and the effective skeleton stiffness value from the physical parameter inversion module via the internal data bus. The early warning decision module then calls the primary early warning threshold and structural failure threshold pre-stored in the storage unit. The early warning decision module is configured to perform a logical AND operation on the collapse risk probability value and the effective skeleton stiffness value.

[0084] The early warning decision module first performs a first-level numerical comparison: determining whether the collapse risk probability value is greater than or equal to the primary early warning threshold.

[0085] When the collapse risk probability value is less than the primary warning threshold, the warning decision module determines the current monitoring status as safe. The warning decision module stores the timestamp of this determination, the collapse risk probability value, and the effective skeleton stiffness value into the time series database module.

[0086] When the collapse risk probability value is greater than or equal to the primary warning threshold, the warning decision module performs a second-level numerical comparison: determining whether the effective skeleton stiffness value is less than the structural failure threshold.

[0087] The early warning decision module executes a branch operation based on the results of the second-level numerical comparison: Non-structural water softening determination: When the effective skeleton stiffness value is greater than or equal to the structural failure threshold, the early warning decision module determines that the current monitoring data anomaly is a non-structural water softening event. The early warning decision module generates an environmental interference filtering log and sends a negative sample labeling instruction to the deep learning analysis module. The cloud computing platform remains silent and does not generate alarm pushes to suppress false alarms.

[0088] Structural Damage Determination: When the effective skeleton stiffness value is less than the structural failure threshold, the early warning decision module determines that a structural damage event has occurred in the foundation. The early warning decision module generates a high-priority interruption request and constructs a graded alarm data packet. The data fields of the graded alarm data packet include: alarm level code, location coordinates of the abnormal intelligent monitoring node, current collapse risk probability value, and residual skeleton stiffness percentage.

[0089] In response to the high-priority interruption request, the cloud computing platform sends the tiered alarm data packet to the designated mobile terminal via the mobile communication network gateway. The mobile terminal's application parses the tiered alarm data packet and drives the mobile terminal's buzzer and display screen to perform an audible and visual alarm operation. The IoT-based intelligent monitoring and early warning system is configured to ensure that the total system latency from the time the intelligent monitoring node collects data to the time the mobile terminal performs the audible and visual alarm operation is less than 3 minutes.

[0090] The cloud computing platform feeds back relevant data on structural damage events confirmed by the early warning decision-making module to the deep learning analysis module. The deep learning analysis module labels the relevant data as positive samples and uses the positive samples to perform online backpropagation training on the long short-term memory neural network model, updating the synaptic weight parameters of the long short-term memory neural network model.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring and early warning system based on the Internet of Things, characterized in that, include: Distributed sensor networks, wireless data transmission networks, edge computing nodes, and cloud computing platforms; The distributed sensor network consists of several intelligent monitoring nodes deployed in the monitoring area. The intelligent monitoring nodes are configured to collect ambient temperature data, relative permittivity data, and strain data, and are configured to transmit and receive elastic wave signals when they receive an active detection trigger command. The edge computing node is connected to the intelligent monitoring node through the wireless data transmission network. The edge computing node is configured to aggregate data from the intelligent monitoring node and calculate sound wave propagation speed data. The cloud computing platform is connected to the edge computing node. The cloud computing platform includes a high-performance computing server, which is configured to run a long short-term memory neural network model and a physical parameter inversion unit. The cloud computing platform uses the long short-term memory neural network model to output a collapse risk probability value based on the data collected by the intelligent monitoring node, and uses the physical parameter inversion unit to calculate the effective skeleton stiffness value based on the relative permittivity data and the sound wave propagation speed data. The cloud computing platform generates graded alarm signals based on the collapse risk probability value and the effective skeleton stiffness value.

2. The intelligent monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The intelligent monitoring node includes: a microprocessor module, an environmental sensing module, a mechanical sensing module, an acoustic detection module, and a power management module; The environmental sensing module includes a frequency domain reflective dielectric sensor, which is configured to measure the resonant frequency offset of the emitted electromagnetic wave in the soil to obtain the relative permittivity data. The acoustic detection module includes a piezoelectric ceramic transducer and a transceiver isolation circuit. The microprocessor module is electrically connected to the environmental sensing module, the mechanical sensing module, and the acoustic detection module, respectively. The microprocessor module is configured to read the relative permittivity data and use it as an input parameter for calculating the acoustic detection frequency.

3. The intelligent monitoring and early warning system based on the Internet of Things according to claim 2, characterized in that, The microprocessor module is configured to invoke a pre-stored dielectric-frequency mapping algorithm to calculate the target excitation center frequency based on the relative permittivity data; When the relative permittivity data indicates that the soil is close to saturation, the microprocessor module adjusts the target excitation center frequency to a low frequency range of 1kHz to 5kHz. The microprocessor module uses direct digital frequency synthesis logic to generate a frequency-modulated continuous wave digital sequence centered on the target excitation center frequency, and drives the piezoelectric ceramic transducer to emit shear wave signals to the soil medium.

4. The intelligent monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The edge computing node includes a data aggregation module and a time synchronization module; The time synchronization module is configured to send time synchronization data packets to the intelligent monitoring nodes via a precise time protocol to unify the clock reference of the distributed sensor network. The edge computing node is configured to send a cooperative reception command to adjacent smart monitoring nodes located within a radius of 5 meters to 1 meter from the transmission source; The edge computing node is also configured to perform cross-correlation operations on the standard waveform of the transmitted signal and the received sampled waveform data, calculate the absolute propagation time of the elastic wave in the soil based on the maximum modulus point of the cross-correlation function, and then calculate the sound wave propagation speed data.

5. The intelligent monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The cloud computing platform includes a time-series database module, a deep learning analysis module, a physical parameter inversion module, and an early warning decision module; The deep learning analysis module is configured to extract a multi-dimensional feature vector sequence from the time series database module and input it into the long short-term memory neural network model, and output the collapse risk probability value. The physical parameter inversion module is configured to receive the relative permittivity data and the sound wave propagation velocity data, perform dielectric-acoustic joint correction and refined inversion calculation, and output the effective skeleton stiffness value.

6. The intelligent monitoring and early warning system based on the Internet of Things according to claim 5, characterized in that, The deep learning analysis module is configured to map the ambient temperature values, relative permittivity data and strain data in the multidimensional feature vector sequence to the closed interval [0, 1] and reconstruct them into a three-dimensional input tensor; The cloud computing platform is configured to compare the collapse risk probability value with a preset primary warning threshold and calculate the time change rate of the relative permittivity data. When the collapse risk probability value is greater than or equal to the primary warning threshold, or the time change rate is greater than or equal to the preset seepage mutation threshold, the cloud computing platform generates the active detection trigger command.

7. The intelligent monitoring and early warning system based on the Internet of Things according to claim 5, characterized in that, The physical parameter inversion module is configured to perform the following operations: The volumetric water content of the soil is calculated by mapping the relative permittivity data, and the wet soil density is calculated by combining the preset dry soil density benchmark value. The apparent shear modulus is obtained by multiplying the wet soil density value by the square of the sound wave propagation velocity data. The relative permittivity data is substituted into a pre-stored saturation-stiffness correction algorithm to calculate the water softening factor. The effective skeleton stiffness value is obtained by dividing the apparent shear modulus by the water softening factor.

8. The intelligent monitoring and early warning system based on the Internet of Things according to claim 7, characterized in that, The parameters used by the physical parameter inversion module to calculate the water softening factor include: the current relative permittivity data, the reference value of the permittivity in the dry state, the reference value of the permittivity in the saturated state, the softening sensitivity coefficient, and the nonlinear shape index. The reference value of dielectric constant in the dry state, the reference value of dielectric constant in the saturated state, the softening sensitivity coefficient, and the nonlinear shape index are all fixed constants pre-entered into the physical parameter inversion module.

9. The intelligent monitoring and early warning system based on the Internet of Things according to claim 5, characterized in that, The early warning decision module is configured to perform logical AND judgment operations on the collapse risk probability value and the effective skeleton stiffness value: When the collapse risk probability value is greater than or equal to the preset primary warning threshold, and the effective skeleton stiffness value is greater than or equal to the preset structural failure threshold, the warning decision module determines it as a non-structural water softening event, generates an environmental interference filtering log, and sends a negative sample labeling instruction to the deep learning analysis module. When the collapse risk probability value is greater than or equal to the primary warning threshold, and the effective skeleton stiffness value is less than the structural failure threshold, the warning decision module determines that a structural damage event has occurred in the foundation and generates a graded alarm data packet containing an alarm level code, abnormal location coordinates, the collapse risk probability value, and the residual percentage of skeleton stiffness.

10. The intelligent monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, It also includes mobile terminals; The mobile terminal is connected to the cloud computing platform via a mobile communication network. The mobile terminal is configured to receive and parse the hierarchical alarm data packets sent by the cloud computing platform and perform audible and visual alarm operations. The IoT-based intelligent monitoring and early warning system is configured to ensure that the total system delay time from the moment the intelligent monitoring node collects data to the moment the mobile terminal executes the audible and visual alarm operation is less than 3 minutes.

Images