A water area multi-parameter intelligent monitoring method based on Beidou fusion

Abstract

The application discloses a water area multi-parameter intelligent monitoring method based on Beidou fusion, and particularly relates to the field of water area monitoring communication, and is used for solving the problems that monitoring is inaccurate and it is difficult to timely upload and alarm under the conditions of extreme wind and wave, turbid water phase state variation and network interruption; through receiving global navigation satellite system space radio frequency signals and Beidou satellite meteorological element downlink signals, the reflection signal component and the scattering signal component are synchronously separated, and the solving resources are adjusted according to the regional meteorological macroscopic evolution trend, then the water surface microwave scattering cross section physical area and the water body surface layer dielectric constant mutation variable are formed respectively, the isolated forest machine learning model is used to output the hydrological phase state variation coefficient, then the switching between the normal clear water level analysis and the turbid water phase state correction tracking is carried out, the real-time water level elevation data and the local disaster danger evaluation report are generated, and the structured disaster semantic code is output when the public network link is interrupted and the alarm dictionary matching condition is hit.

Description

Technical Field

[0001] This invention relates to the field of water monitoring and communication, and more specifically, to a multi-parameter intelligent monitoring method for water areas based on BeiDou fusion. Background Technology

[0002] Integrated multi-parameter monitoring of water bodies is typically deployed in long-term monitoring scenarios such as rivers, reservoirs, lakes, and flash flood gullies. Existing solutions mostly rely on geometric ranging of liquid level, GPS reflectance measurement, or a single communication link to complete water level monitoring. However, existing technologies have begun to utilize direct and reflected signals from GPS to obtain water level data, indicating that current technologies can already perform water level measurement using satellite reflectance information.

[0003] However, existing processing methods still have significant limitations under extreme conditions such as strong winds and waves, torrential rains, flash floods, and mudslides. On the one hand, once the specular reflection conditions of the water surface are disrupted, water level calculations relying solely on reflected signals are prone to instability. When a large amount of suspended sediment is engulfed in the water, the dielectric properties of the water surface change, making it difficult to directly apply the ranging relationships established under conventional clear water conditions. On the other hand, remote hydrological stations typically rely on solar energy for communication with the public network. If monitoring results are still sent in the original long message format after a physical interruption of the public network, congestion can easily occur in the narrowband channel of the BeiDou short message system. Therefore, a processing solution that can simultaneously address phase discrimination, water level calculation, and transmission during network outages is needed.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a multi-parameter intelligent monitoring method for water bodies based on BeiDou fusion. This method receives space radio frequency signals from the Global Navigation Satellite System and downlink signals from BeiDou meteorological elements, simultaneously separates the reflected signal components and the scattered signal components, adjusts the calculation resources according to the regional meteorological macro-evolution trend, and then separately generates the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer. An isolated forest machine learning model is used to output the hydrological phase variation coefficient, and the method switches between normal clear water level analysis and turbid water phase correction tracking to generate real-time water level elevation data and local disaster risk assessment reports. Furthermore, when the public network link is interrupted and the alarm dictionary matching conditions are met, structured disaster semantic code is output to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: It receives space radio frequency signals from the Global Navigation Satellite System and downlink signals from meteorological elements of the BeiDou satellite, and separates the reflected signal components and the scattered signal components. The physical area of ​​the microwave scattering cross section of the water surface is determined based on the scattered signal components, and the abrupt change in the dielectric constant of the water surface is determined based on the reflected signal components. The hydrological phase variation coefficient is obtained by comprehensively analyzing the two. Real-time water level elevation data and local disaster risk assessment reports are generated based on the hydrological phase variation coefficient. When the public network communication link is interrupted and the local disaster risk assessment report matches the alarm dictionary conditions, the real-time water level elevation data and the local disaster risk assessment report are truncated and recombined to generate structured disaster semantic code, which is then uploaded through the Beidou short message communication module.

[0007] Furthermore, the reflected signal component and the scattered signal component are separated, including: performing multi-channel filtering on the space radio frequency signal to form the reflected signal component from the relevant echo falling into the delay window of the main peak of specular reflection, and forming the scattered signal component from the scattered echo deviating from the delay window of the main peak of specular reflection; adjusting the resources of both according to the regional meteorological trend, including writing the control register status code to the GNSS baseband receiver chip according to the slope of the wind and wave intensity fitting curve.

[0008] Furthermore, the physical area of ​​the microwave scattering cross section of the water surface is determined, including calculating the total received power of the scattered signal components and subtracting the inherent thermal noise of the hardware, and then performing inversion in combination with the spatial geometric distance; the abrupt change in the dielectric constant of the water surface is determined, including extracting the difference between the left and right rotational polarization phase delays of the reflected signal components, combining the Fresnel reflection boundary equation to obtain the equivalent surface dielectric constant of the current epoch, and comparing it with the dielectric constant of the pre-stored clear water baseline.

[0009] Furthermore, the hydrological phase variation coefficient is output, including simultaneously inputting the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface into a pre-trained isolated forest machine learning model, and determining the hydrological phase variation coefficient based on the average path length; real-time water level elevation data is generated, including performing normal clear water level analysis when the hydrological phase variation coefficient is in the normal value range, and performing turbid water phase correction tracking when the hydrological phase variation coefficient is in the abnormal isolated range.

[0010] Furthermore, the turbid water phase state correction tracking is performed, including calling the microwave propagation delay compensation dictionary for high turbidity water bodies using the abrupt change in the dielectric constant of the water surface as an index, reading the corresponding beam delay correction factor, and incorporating the beam delay correction factor into the geometric ranging calculation to generate real-time water level elevation data.

[0011] Furthermore, public network link interruption is determined by at least one of the following detection results: continuous handshake failure, link unreachability, and physical carrier loss; the local disaster risk assessment report hits the alarm dictionary matching conditions, including the internal security log process of the device matching the water level deviation value and disaster risk assessment score with the high-risk disaster entry library in the local database.

[0012] Furthermore, the truncation and reassembly process involves organizing real-time water level elevation data and local disaster risk assessment reports into a byte stream to be sent, with measurement results preceding semantic descriptions, and rearranging them in the order of numerical change segments, disaster type segments, hazard level segments, and link status segments, and then truncating them using the limit of the single physical packet length of BeiDou short messages as the boundary.

[0013] Furthermore, the generation of structured disaster semantic code includes mapping numerical change segments, disaster type segments, danger level segments, and link status segments to fixed enumeration values, forming feature mask indexes in a predefined bit-width order, and then outputting structured disaster semantic code based on a pre-burned hexadecimal hash mapping table.

[0014] Furthermore, the water monitoring equipment includes an edge computing node, a GNSS baseband receiver chip, a signal separation unit, a public network communication link interface, and a BeiDou short message communication module; the edge computing node is connected to the GNSS baseband receiver chip, the signal separation unit, the public network communication link interface, and the BeiDou short message communication module, respectively.

[0015] Furthermore, the inherent thermal noise of the hardware is obtained through silent calibration after installation, and the dielectric constant of the clear water baseline and the local water level baseline are obtained through clear water calibration under normal weather conditions; the water monitoring equipment is deployed at remote hydrological stations that rely on solar energy and battery power, and is transmitted via the public network when the public network link is available, and via the Beidou short message communication module when the public network link is interrupted.

[0016] The technical effects and advantages of the present invention, a multi-parameter intelligent monitoring method for water bodies based on BeiDou fusion, are as follows: This invention introduces two complementary processing objects at the sensing and calculation levels: the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface. It uses the hydrological phase variation coefficient as a branching basis, ensuring that water level calculation no longer relies solely on reflection measurements from a single path. Therefore, a corresponding processing chain can be formed between normal clear water scenarios and turbid water phase variation scenarios, ensuring that the generation logic of real-time water level data remains consistent with the current water state.

[0017] This invention first adjusts the computational resources for reflected and scattered signal components based on the macroscopic evolution trend of regional meteorology at the resource scheduling level, so that hardware processor instruction cycles are prioritized for computational channels more suitable for the current operating conditions. This reduces invalid high-frequency sampling and repeated computation during periods of deteriorating specular reflection conditions, maintains clear channel calling relationships, and facilitates edge computing nodes in maintaining a stable data processing rhythm during continuous monitoring.

[0018] At the communication output level, this invention links real-time water level data and local disaster risk assessment reports with public network link status and alarm dictionary matching conditions. Protocol-level truncation and structured reassembly are only performed when disaster transmission conditions are met. The structured disaster semantic code formed after conversion by a hexadecimal hash mapping table can directly adapt to the narrowband transmission constraints of BeiDou short messages. This is beneficial for preserving core disaster semantics and compressing pending content in scenarios of loss of connection, avoiding the continuous accumulation of long messages in the underlying queue. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of a multi-parameter intelligent monitoring method for water areas based on BeiDou fusion according to the present invention; Figure 2 This is a functional architecture block diagram of a water monitoring device in a BeiDou-based intelligent multi-parameter water monitoring method according to the present invention. Figure 3 This is a schematic diagram illustrating the synchronous separation of reflected and scattered signal components in a BeiDou-based intelligent monitoring method for multiple parameters of water areas according to the present invention. Figure 4 This is a schematic diagram of the generation and path switching of hydrological phase variation coefficient in a multi-parameter intelligent monitoring method for water areas based on BeiDou fusion according to the present invention. Figure 5 This is a schematic diagram of disaster semantic compression and BeiDou short message uploading in a multi-parameter intelligent monitoring method for water areas based on BeiDou fusion according to the present invention. Detailed Implementation

[0020] The technical solutions of 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.

[0021] Please see Figures 1-5 This invention provides a multi-parameter integrated monitoring method for water bodies based on multi-protocol fusion, comprising: S1: Receives space radio frequency signals from the Global Navigation Satellite System and downlink signals from BeiDou satellite meteorological elements, separates the reflected signal components and scattered signal components, and adjusts both resources according to regional meteorological trends; S2: Determine the physical area of ​​the microwave scattering cross section of the water surface based on the scattering signal component, determine the abrupt change in the dielectric constant of the water surface based on the reflection signal component, and obtain the hydrological phase variation coefficient by comprehensively analyzing the two. S3: Generate real-time water level elevation data and local disaster risk assessment reports based on the hydrological phase variation coefficient; S4: When the public network communication link is interrupted and the local disaster risk assessment report matches the alarm dictionary conditions, the real-time water level elevation data and the local disaster risk assessment report are truncated and recombined to generate structured disaster semantic code, which is then uploaded through the Beidou short message communication module.

[0022] This invention utilizes a single water monitoring device for sensing, discrimination, and data transmission. Edge computing nodes first separate satellite echoes into reflected and scattered signal components. Then, they characterize the macroscopic morphology and microscopic material changes of the water body using the physical area of ​​the microwave scattering cross-section and the abrupt change in the dielectric constant of the water surface. The hydrological phase variation coefficient output by the isolated forest machine learning model serves as the basis for path switching. When the monitored water body is in a normal clear water state, normal clear water level analysis is performed; when the monitored water body experiences turbid water phase variation, turbid water phase correction tracking is performed; and when the public network communication link is interrupted and the local disaster risk assessment report matches the alarm dictionary conditions, the monitoring results are compressed into structured disaster semantic code and sent via the BeiDou short message communication module.

[0023] At the initial stage of the monitoring link, the water monitoring equipment first encounters space radio frequency signals from the Global Navigation Satellite System and downlink meteorological element signals from the BeiDou satellite. These two types of inputs respectively carry water surface echo observation information and external meteorological evolution information. However, before entering the physical quantity inversion stage, they are still in a primitive state without unified organization and functional separation. The corresponding step S1 revolves around this initial input. By establishing a receiving basis at the same epoch, synchronously separating the reflected signal components and the scattered signal components, and combining the regional meteorological macro-evolution trend, the underlying sampling frequency and solution resources are adjusted so that the monitoring equipment can first form a clear signal boundary and resource allocation relationship within a single monitoring cycle.

[0024] S101 multi-source signal reception and epoch binding.

[0025] Once the water monitoring equipment enters a monitoring cycle, the edge computing node first takes over the original input channel. Since subsequent processing requires both multi-channel filtering of the space radio frequency signals from the Global Navigation Satellite System and inference of the macro-evolution trend of regional meteorological elements from the BeiDou satellite downlink signals, the edge computing node first establishes a joint input basis for the same epoch in this sub-step.

[0026] Each of the following sub-steps is executed by an edge computing node deployed within the water monitoring equipment. The edge computing node sends the space radio frequency (RF) signal from the Global Navigation Satellite System (GNSS) received at the antenna front end to the RF sampling input of the signal separation unit, and sends the downlink meteorological element signal from the BeiDou satellite to the meteorological decoding buffer. The edge computing node appends the same epoch number to both inputs using a unified time reference for the current monitoring period, establishing a one-to-one correspondence between the space RF sampling segment and the BeiDou meteorological element decoding results for the corresponding time period on the time axis, thereby generating a joint input frame for the current monitoring period. The joint input frame contains at least two types of subsequent necessary objects: one is the original space RF sampling sequence that has not yet been stripped, and the other is a sequence of BeiDou meteorological elements arranged in chronological order. The BeiDou meteorological element sequence is directly fed into the physical mechanism driving model in this step, while the original space RF sampling sequence continues to be fed into the signal separation unit.

[0027] In one embodiment, the water monitoring equipment along the valley river is in a continuous monitoring state before heavy rainfall. The antenna simultaneously receives a segment of space radio frequency (RF) signals from the Global Navigation Satellite System (GNSS) and a set of downlink meteorological element signals from the BeiDou satellites within the same monitoring cycle. The edge computing node first decodes the BeiDou meteorological elements, and then writes the decoding result and the space RF sampling segment collected in the same cycle into a joint input frame. Therefore, subsequent multi-channel filtering no longer corresponds to isolated RF samples, and the regional meteorological macro-evolution trend prediction no longer corresponds to independent meteorological reports detached from the RF field conditions. Both types of inputs are locked at the same epoch as a directly referable processing starting point.

[0028] Synchronous separation of reflected and scattered signal components in S102.

[0029] After the joint input frame is formed, the edge computing node immediately calls the signal separation unit to enter parallel separation processing. Since step S2 needs to extract the total received power from the scattered signal component and the left and right rotation polarization phase delay difference from the reflected signal component, this sub-step not only needs to distinguish the two types of components, but also needs to keep the phase and power relationships of the two types of components from being mixed at the same epoch.

[0030] The signal separation unit performs multi-channel filtering on the original spatial radio frequency (RF) sampling sequence in the joint input frame. Specifically, the signal separation unit constructs parallel filtering channel groups based on the same local oscillation reference and the same sampling window. Each channel group performs synchronous processing on the original RF sampling sequence of the same epoch. The edge computing node first performs local pseudocode correlation processing on the original RF sampling sequence to form a correlation peak sequence sorted by time delay. Then, based on the center time delay of the specular reflection main peak obtained from the water-calibrated epoch statistics, a specular reflection main peak time delay window is formed within one sampling interval before and after the center time delay. Correlation peaks that fall within the specular reflection main peak time delay window and whose left-right rotation polarization amplitude ratio is within the fluctuation range of the water-calibrated amplitude ratio are written into the reflection signal component sampling sequence; correlation peaks that are outside the specular reflection main peak time delay window and whose peak width is greater than the width of the water-calibrated main peak are written into the scattering signal component sampling sequence. The filtering channel for the reflected signal component retains the correlated echo dominated by specular reflection and maintains continuous output of the phase observation link corresponding to the left and right rotation polarizations, thus forming a sampled sequence of the reflected signal component that can be used to extract the phase delay difference between the left and right rotation polarizations. The filtering channel for the scattered signal component suppresses the correlated component dominated by specular reflection and retains the diffuse scattering energy distribution formed after the water surface is broken, thus forming a sampled sequence of the scattered signal component that can be used to extract the total received power and construct the power delay spectrum. Since both types of filtering channels correspond to the same epoch number, the sampled sequence of the reflected signal component and the sampled sequence of the scattered signal component are synchronized in time and can be directly used as the input object for step S2.

[0031] In one embodiment, after the river surface is disturbed by crosswinds, the original spatial radio frequency sampling sequence simultaneously contains stable specular reflections and echo energy scattered by broken wave surfaces. The signal separation unit opens two filtering paths in parallel for the same segment of the original spatial radio frequency sampling sequence: one filtering path retains the echo dominated by specular reflections and outputs the reflected signal component sampling sequence; the other filtering path retains the non-specular scattering energy and outputs the scattered signal component sampling sequence. After this synchronous stripping, step S2 will not mix in the main reflection peak when processing the scattered signal component, nor will it be covered by the diffuse scattering power caused by the water surface breaking when processing the reflected signal component. The boundary between the two types of physical observation objects has been fixed in this sub-step.

[0032] S103 Regional Meteorological Macro-evolution Trend Prediction and Channel Computing Power Scheduling.

[0033] After the reflected signal component sampling sequence and the scattered signal component sampling sequence are separated, the edge computing node does not directly end step S1, but continues to call the BeiDou meteorological element sequence in the joint input frame to predict the short-term wind and wave evolution after the current monitoring period. Only after the severity determination is completed first can the reduction of the underlying sampling frequency of the reflected signal component and the reallocation of the hardware processor instruction cycle have clear triggering conditions.

[0034] Edge computing nodes concatenate the BeiDou meteorological element sequence in the current joint input frame with the BeiDou meteorological element sequence of several consecutive epochs cached locally in chronological order to form a rolling prediction sequence, and input the rolling prediction sequence into the physical mechanism driven model. The physical mechanism driven model is a set of basic equations for regional numerical weather prediction built into the edge computing nodes, used to process large-scale BeiDou meteorological elements and extrapolate the probability of wind and waves occurring in the near future. As an example, the physical mechanism driven model writes the near-ground wind speed, precipitation, air pressure, and temperature obtained by decoding the downlink signals of BeiDou satellite meteorological elements into the local grid around the monitoring section in epochs, with each grid cell storing the state values ​​of the most recent six epochs; the edge computing nodes first update the wind speed field and precipitation field of each grid cell in chronological order, and then form the wind and wave intensity sequence for the next three prediction times using the weighted average result of the grid where the monitoring section is located and its adjacent grids, and then perform a three-point sliding fit on the wind and wave intensity sequence to obtain the wind and wave intensity fitting curve, and calculate the slope of the wind and wave intensity fitting curve based on the difference relationship between adjacent prediction times.

[0035] To ensure that the severity level has executable hardware control significance, the edge computing node compares the slope of the current wind and wave intensity fitting curve with a preset slope boundary. The preset slope boundary is taken from the high quantile boundary of the slope sequence corresponding to the historical short-term meteorological forecast samples. When the slope of the current wind and wave intensity fitting curve does not exceed the preset slope boundary, the edge computing node maintains the predetermined channel configuration of the reflected signal component sampling sequence and the scattered signal component sampling sequence. When the slope of the current wind and wave intensity fitting curve exceeds the preset slope boundary and shows a steep increasing trend, the edge computing node determines that the severity of the regional meteorological macro-evolution trend has entered an enhanced state and generates a sampling frequency scheduling instruction corresponding to the severity level.

[0036] After the sampling frequency scheduling command is generated, the data processing module sends a low-level hardware interrupt command to the control register of the GNSS baseband receiver chip. This low-level hardware interrupt command, by overwriting the control register status code, directly modifies the clock division relationship of the GNSS-IR solution channel corresponding to the reflected signal component, causing the low-level sampling frequency of the reflected signal component to be pulled down proportionally to its severity. Specifically, the edge computing node uses the ratio of the slope of the current wind and wave intensity fitting curve to the preset slope boundary as the scheduling ratio, and converts this scheduling ratio into the value written to the control register based on the pre-programmed frequency division mapping table. The frequency division mapping table is pre-statistically formed from historical short-term weather forecast samples and corresponding echo instability samples. Specifically, the scheduling ratio is divided into five intervals from low to high: not higher than 1, greater than 1 and not higher than 1.2, greater than 1.2 and not higher than 1.5, greater than 1.5 and not higher than 2, and greater than 2. The corresponding control register values ​​are 0, 1, 2, 3, and 4, respectively. A control register value of zero indicates that the current clock frequency division is maintained, and control register values ​​of 1 to 4 indicate that the clock frequency division multiple of the GNSS-ir solution channel corresponding to the reflected signal component is increased sequentially. When the scheduling ratio falls between two adjacent frequency division intervals, the control register value corresponding to the higher-level frequency division interval is taken, so that the underlying sampling frequency of the reflected signal component decreases monotonically according to the severity.

[0037] The higher the severity, the larger the clock division value written to the control register, and the longer the sampling trigger interval of the reflected signal component in subsequent monitoring cycles. Due to the conservation characteristics of the processor's computing power bus, the digital signal processing operation cycle released after the sampling frequency of the reflected signal component decreases is no longer kept in an idle state. Instead, it is rebound by the edge computing node to the GNSS-r solution channel corresponding to the scattered signal component within the same interrupt service flow, in order to increase the solution time slot and filtering iteration opportunities of the scattered signal component. After this processing, the output of the current step is not only the time-synchronized sampling sequence of the reflected signal component and the sampling sequence of the scattered signal component, but also the severity of the regional meteorological macro-evolution trend corresponding to the current monitoring cycle and the effective channel scheduling status. Among them, the first two are directly called by step S2 to perform power extraction and phase delay difference extraction, while the channel scheduling status constrains the sampling density and computing resource conditions when step S2 calls the two types of components.

[0038] In one embodiment, the edge computing node infers that the short-term wind and waves will continue to rise based on the downlink meteorological signals from the BeiDou satellite. The slope of the wind and wave intensity fitting curve increases sharply between adjacent prediction times. The data processing module then writes a new status code to the control register of the GNSS baseband receiver chip. The clock baud rate of the GNSS-ir solution channel corresponding to the reflected signal component is lowered, and the digital signal processing operation cycle originally used for high-frequency tracking of specular reflection is transferred to the GNSS-r solution channel. At this time, even if wave crests have broken in the river channel, the scattered signal components received in step S2 still receive more sufficient solution resources, while the reflected signal components continue to retain the necessary observation links at a contracted frequency. The processing chain will not be interrupted during the wind and wave enhancement phase.

[0039] In step S1, the synchronous reception and epoch binding of the global navigation satellite system space radio frequency signal and the downlink signal of BeiDou satellite meteorological elements have been completed. The sampling sequences of the reflected signal component and the scattered signal component are obtained through the signal separation unit. Simultaneously, the edge computing node dynamically reduces the underlying sampling frequency of the reflected signal component based on the severity of the regional meteorological macro-evolution trend, and schedules the released hardware processor instructions to the solution channel of the scattered signal component. At this point, this step has formed a dual-channel observation object with a clearly defined lower boundary and fixed scheduling relationship within the same epoch.

[0040] After signal separation and resource adjustment are completed in step S1, the monitoring equipment has obtained the sampling sequences of the reflected signal components and the scattered signal components of the same epoch, as well as the matching channel scheduling status. However, these objects remain at the signal level and channel level and have not yet been converted into discriminant quantities that can characterize the macroscopic morphology and microscopic material of the water body. Step S2 proceeds on this basis, extracting physically meaningful observation results from the scattered signal components and the reflected signal components respectively, and constructing feature representations for phase state recognition.

[0041] S201 scattering signal component power purification and physical area inversion of water surface microwave scattering cross section.

[0042] Step S1 has already output the sampled sequence of the scattered signal components, and through the channel scheduling state, more digital signal processing operation cycles are bound to the solution channel of the scattered signal components. Therefore, this sub-step takes the sampled sequence of the scattered signal components as the only processing starting point, first completes the quantization at the power level, and then converts the quantization result into a representation object with physical area meaning.

[0043] Edge computing nodes read amplitude values ​​point by point from the sampling sequence of scattered signal components in the current epoch, constructing a scattered power sequence according to the sampling time order. Then, the scattered power sequence is accumulated within the processing window corresponding to the current epoch to obtain the total received power of the scattered signal components. To ensure that subsequent area inversion only corresponds to the water surface echo and does not carry the background thermal disturbance of the receiving link, the edge computing node continues to read the hardware-inherent thermal noise baseline of the receiving front end under the current channel scheduling state, and subtracts the total received power from this hardware-inherent thermal noise baseline in situ to form the net scattered echo power. The hardware-inherent thermal noise baseline is obtained during the silent calibration phase after equipment installation. Specifically, when there is no effective echo at the antenna input or the external input is closed by the receiving protection window, the background power of multiple sampling windows is continuously collected and the average value is calculated. During equipment operation, when the temperature state of the receiving front end changes, the edge computing node performs periodic updates to the hardware-inherent thermal noise baseline in the same manner. Once the net scattered echo power of the current epoch is formed, it is bound to the spatial geometric distance of the same epoch. The spatial geometric distance is the geometric propagation distance between the current monitoring device and the current water surface echo action area, which is derived from the spatial positioning and propagation path calculation results of the current epoch, and maintains the same epoch number as the sampling sequence of the scattered signal components.

[0044] Once the net scattered echo power and spatial geometric distance are both available, the edge computing nodes use radar meteorological equations to convert the net scattered echo power into the physical area of ​​the water surface microwave scattering cross section. Specifically, using the net scattered echo power of the current epoch... Spatial geometric distance Transmission power Transmit antenna gain Receiver antenna gain and operating wavelength Constructing the inversion relationship: ; In the formula, The physical area of ​​the microwave scattering cross section of the water surface represents the macroscopic geometric area of ​​the actual microwave diffuse reflection caused by the water surface being torn apart by strong winds, and its dimension is square meters. The net scattered echo power is derived from the total received power minus the inherent thermal noise of the hardware. The spatial geometric distance bound to the current epoch; , , and All of these are known calibration parameters of the current monitoring equipment's RF link. After the edge computing node completes the area inversion for the current epoch according to the above formula, it outputs the physical area of ​​the water surface microwave scattering cross section that uniquely corresponds to that epoch, and writes the result into the feature buffer of this step for subsequent synchronous assembly with the abrupt change in the dielectric constant of the water surface.

[0045] In one embodiment, when the river channel is affected by continuous rainfall and broken waves occur, the echo fluctuations in the sampling sequence of the scattered signal components increase significantly. The edge computing node first accumulates the scattered power within this epoch point by point, and then subtracts the hardware-inherent thermal noise baseline recorded by the receiving front end under the same channel scheduling state to obtain the net scattered echo power corresponding only to the diffuse scattering of the water surface. Subsequently, it performs area inversion in combination with the spatial geometric distance from the current monitoring device to the water surface, and outputs a physical area of ​​the microwave scattering cross section of the water surface bound to this epoch. This object will not return to the original power level in the future, but will be directly used as the first dimension input of the isolated forest machine learning model.

[0046] S202 reflected signal component polarization phase inversion and acquisition of the abrupt change in the dielectric constant of the water surface.

[0047] After the physical area of ​​the microwave scattering cross section of the water surface is formed, step S2 continues to use the sampling sequence of the reflected signal components output in step S1 to solve the material level. This sub-step no longer processes the scattered energy, but extracts the left and right rotation polarization phase delay difference from the reflected signal components, and uses this to inversely deduce the change in the dielectric constant of the water surface layer in the current epoch, so that step S2 forms another physical feature that is orthogonal to the macroscopic morphology.

[0048] The edge computing node separates the complex echo samples of the left-hand and right-hand circular polarization branches in the current epoch from the reflected signal component sampling sequence, and calculates the phase trajectories of the two branches respectively. The reflected signal component sampling sequence is already labeled with polarization channels by the dual circular polarization receiving antenna and the orthogonal demodulation link during writing. The edge computing node first splits the reflected signal component sampling sequence into a left-hand circular polarization branch sample queue and a right-hand circular polarization branch sample queue according to the polarization channel labels. Then, it performs in-phase branch and orthogonal branch pairing on the two sample queues respectively to reconstruct the complex echo samples of the current epoch. After the two branches complete phase unrolling within the same processing window, the edge computing node subtracts them point by point according to the same sampling index to obtain the left and right circular polarization phase difference sequences. Then, it aggregates these left and right circular polarization phase difference sequences within the current epoch and outputs the left and right circular polarization phase delay difference values. Since the difference in left-hand and right-hand polarization phase delay reflects the difference in polarization propagation when electromagnetic waves cross the water surface boundary, the edge computing node sends this difference along with the current monitoring geometric boundary conditions into the inverse solution process of the Fresnel reflection boundary equation to solve for the equivalent surface dielectric constant corresponding to the current epoch. During the inverse solution, the incident angle formed by the current monitoring device and the water surface normal is used as a known geometric quantity. The incident angle is determined by the installation height of the monitoring device and the geometric position of the current water surface echo zone. The edge computing node substitutes the Fresnel reflection boundary equation point by point within the preset dielectric constant search interval to calculate the theoretical left-hand and right-hand polarization phase delay difference under the corresponding candidate dielectric constant. The theoretical value is then compared with the measured left-hand and right-hand polarization phase delay difference one by one, and the candidate dielectric constant with the smallest error is taken as the equivalent surface dielectric constant of the current epoch.

[0049] To ensure that the material changes at the current epoch can be directly invoked in subsequent step S3, the edge computing node does not directly output the equivalent surface dielectric constant. Instead, it further subtracts the equivalent surface dielectric constant of the current epoch from the pre-stored clear water baseline dielectric constant to form the abrupt change in the water surface dielectric constant. The pre-stored clear water baseline dielectric constant is a local reference value established under normal weather, clear water, or slightly turbid conditions after the water monitoring equipment is installed, and it is permanently bound to the current monitoring section. The edge computing node invokes one thousand sets of clear water calibration samples continuously collected after the equipment installation, calculates the 95th percentile boundary of the absolute difference between the equivalent surface dielectric constant of each sample and the pre-stored clear water baseline dielectric constant, and uses this boundary as the fluctuation boundary. When the absolute value of the difference at the current epoch is not higher than the fluctuation boundary, the abrupt change in the water surface dielectric constant is written into the low-value region; when the absolute value of the difference at the current epoch is higher than the fluctuation boundary, the abrupt change in the water surface dielectric constant is written into the high-value region. This object is a dimensionless parameter, and its value directly represents the degree to which the microscopic material structure inside the current water body deviates from that of clear water.

[0050] In one embodiment, after a flash flood carrying sediment enters the monitoring section, the left-hand and right-hand polarization branches in the sampled sequence of the reflected signal components exhibit a continuously widening phase response. The edge computing node first unfolds the phase trajectories of the two branches, and then calculates the difference between the phase values ​​at the same sampling index point by point to form a left-hand and right-hand polarization phase difference sequence. When this sequence is aggregated into the left-hand and right-hand polarization phase delay difference within the current epoch, the inverse solution result of the Fresnel reflection boundary equation will have a significant offset relative to the pre-stored clear water baseline dielectric constant. The edge computing node then outputs a change in the dielectric constant of the water surface layer bound to the current epoch. This object is not passed along with the original phase value in the subsequent steps, but is directly used as the second dimension input of the isolated forest machine learning model.

[0051] S203 feature vector synchronous assembly and isolated forest machine learning model output hydrological phase variation coefficient.

[0052] Once the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer have been formed, step S2 enters the final discrimination stage. This sub-step uses the physical area of ​​the microwave scattering cross section of the water surface output from S201 and the abrupt change in the dielectric constant of the water surface layer output from S202 as a joint starting point, assembles the input vector according to the principle of synchronization at the same epoch, and calls the pre-trained isolated forest machine learning model to output the hydrological phase variation coefficient. The isolated forest machine learning model is constructed using the isolated forest algorithm. This sub-step selects the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer as model inputs because the former corresponds to the macroscopic geometric failure state of the water surface, and the latter corresponds to the change state of the material structure of the water surface layer; when there is only wind and wave disturbance, the former changes while the latter remains close to the clear water baseline; when there is sediment entrainment, the former and the latter shift synchronously; the total received scattering power and the difference in left and right rotation polarization phase delay have been used to form the above two features respectively, and continuing to input them in parallel will introduce homogeneous repetition.

[0053] The edge computing node first verifies the epoch number of the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer. When both belong to the same epoch, the physical area of ​​the microwave scattering cross section of the water surface is placed in the first dimension and the abrupt change in the dielectric constant of the water surface layer is placed in the second dimension in a fixed order to construct the two-dimensional feature vector of the current epoch. The pre-trained isolated forest machine learning model uses these two types of parameters collected under normal weather conditions as the training set. During the training phase, multiple random trees have been built in the multi-dimensional feature space. Each random tree constructs a random cutting hyperplane by randomly selecting the feature dimension and randomly selecting the splitting threshold, thereby forming a spatial separation structure for the normal sample distribution. After the two-dimensional feature vector of the current epoch is input, the edge computing node drives the two-dimensional feature vector to pass through each random tree in sequence, records the number of spatial cutting levels that the two-dimensional feature vector traverses from the tree root to being separated, and calculates the average path length of the number of spatial cutting levels on all random trees.

[0054] For example, the pre-trained isolated forest machine learning model can be constructed using a tree ensemble structure. This model consists of 200 isolated trees, with a maximum tree depth of 8 for each tree, a training subsampling size of 256 for each tree, and a minimum number of samples per leaf node. During each node split, only one feature is randomly selected from two input features as the current splitting feature, and a splitting threshold is randomly drawn between the minimum and maximum values ​​of that feature in the current node's samples. The two input features are the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer. The training dataset can be taken from continuously collected normal weather samples after the equipment is installed, such as collecting 12,000 sets of clear or slightly turbid water condition samples, of which 9,600 sets are used as the training set and 2,400 sets as the validation set. Each sample set consists of the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer at the same epoch, and samples of manually confirmed flash floods, debris flows, and severe turbid water phase variations are removed. The training method employs unsupervised training. Specifically, for each isolated tree, 256 samples are randomly drawn from the training set without replacement. Then, random feature selection and random threshold segmentation are recursively performed until the maximum tree depth is reached or only one sample remains at the current node. The model is determined by fixing the tree structure after all isolated trees have grown. During the validation phase, the average path length of each sample in the validation set is calculated, and the 95th percentile value of the hydrological phase variation coefficient obtained from the training set is set as the upper limit of the normal numerical range. Output intervals exceeding this upper limit are considered abnormal isolated intervals. The number of trees, maximum tree depth, subsampling size, and quantile boundaries can be obtained by repeatedly training on historical normal samples and selecting the set of parameters with the smallest output fluctuation.

[0055] After the average path length is determined, the edge computing nodes perform an inverse normalization transformation on it, outputting the hydrological phase variation coefficient. The inverse normalization transformation can be expressed as: ; In the formula, This is the hydrological phase variation coefficient, used to quantify the degree of variation in the current water body; This represents the average path length of the current epoch's two-dimensional feature vector across all random trees. The baseline path length is obtained from the path length statistics of normal samples in the training set. Since normal wind and rain usually only cause changes in the physical area of ​​the microwave scattering cross section of the water surface, while the abrupt change in the dielectric constant of the water surface layer remains close to the pre-stored clear water baseline dielectric constant, the two-dimensional feature vector will fall deep into the distribution of normal samples, resulting in a larger average path length and a lower hydrological phase variation coefficient after inverse normalization. When debris flows or high-turbidity floods cause both the macroscopic morphology and microscopic material to deviate from the normal distribution, the two-dimensional feature vector will be rapidly isolated by randomly cut hyperplanes at shallower levels, resulting in a significantly shorter average path length and a correspondingly higher hydrological phase variation coefficient after inverse normalization. While outputting the hydrological phase variation coefficient, the edge computing node also writes the physical area of ​​the water surface microwave scattering cross section, the change in the dielectric constant of the water surface layer, and the hydrological phase variation coefficient bound to that epoch into the call cache area of ​​step S3. The hydrological phase variation coefficient serves as the direct input for step S3 to perform the branch discrimination between the normal numerical range and the abnormal isolated range, while the change in the dielectric constant of the water surface layer serves as the index basis when step S3 calls the microwave propagation delay compensation dictionary for high turbidity water.

[0056] In one embodiment, after continuous rainfall, the riverbed simultaneously experiences broken wave surfaces and sediment intrusion. The edge computing node constructs a two-dimensional feature vector from the physical area of ​​the water surface microwave scattering cross section and the abrupt change in the dielectric constant of the water surface layer in the current epoch, and then feeds it into the isolated forest machine learning model. If this two-dimensional feature vector is separated in most random trees after only a few levels, the edge computing node obtains a shorter average path length and outputs a higher hydrological phase variation coefficient after inverse normalization transformation. This result, along with the abrupt change in the dielectric constant of the water surface layer in the same epoch, is stored in the call cache in step S3, and can then be directly used to determine whether the monitored water body has entered a turbid water phase variation state.

[0057] In step S2, the net scattered echo power has been extracted from the scattered signal components, and the physical area of ​​the microwave scattering cross section of the water surface has been deduced by combining it with the spatial geometric distance. Simultaneously, the difference in left- and right-handed polarization phase delay is extracted from the reflected signal components, and the abrupt change in the dielectric constant of the water surface layer is deduced in reverse using the Fresnel reflection boundary equation. Further, this step synchronously inputs the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer into an isolated forest machine learning model, outputting a hydrological phase variation coefficient to quantify the current degree of water body variability, thus transforming the monitoring object from a raw signal representation into a discernible phase result.

[0058] After processing in step S2, the monitoring equipment possesses the physical area of ​​the microwave scattering cross section of the water surface, the abrupt change in the dielectric constant of the water surface, and the hydrological phase variation coefficient. This means that the water body is not only being observed but has also been endowed with a basis for distinguishing between normal clear water and turbid water phase variations. However, at this point, water level results and disaster descriptions that directly reflect the cross-sectional state have not yet been generated. Step S3, based on this discrimination, selects the appropriate analytical path according to the hydrological phase variation coefficient, transforming the phase identification results into measurement results and risk results.

[0059] S301 Hydrological Phase Variation Coefficient Branch Determination and Analysis Path Switching.

[0060] Step S2 has written the hydrological phase variation coefficient of the current epoch into the call cache of step S3. This sub-step takes the hydrological phase variation coefficient as the starting point for processing and simultaneously calls the reflection signal component sampling sequence, the scattering signal component sampling sequence and the change in dielectric constant of the water surface layer of the same epoch. First, the interval determination is completed, and then it is decided which water level analysis path to enter at the current epoch.

[0061] The edge computing node retrieves the hydrological phase variation coefficient of the current epoch from the cache in step S3, and compares it with the pre-stored boundary of the normal numerical interval. The boundary of the normal numerical interval is taken from the upper confidence limit of the statistical distribution of the hydrological phase variation coefficient in normal weather samples. When the hydrological phase variation coefficient of the current epoch is lower than or equal to the upper confidence limit, the edge computing node generates a normal analysis label and switches the current epoch to the normal clear water level analysis path. When the hydrological phase variation coefficient of the current epoch is higher than the upper confidence limit, the edge computing node generates a turbid water phase variation label and switches the current epoch to the turbid water phase correction tracking path. Since the hydrological phase variation coefficient has been inversely normalized by the isolated forest machine learning model based on the average path length, the high value state already corresponds to an extremely short separation path, and there is no need to repeatedly return to the training stage to recalculate the number of spatial cutting levels.

[0062] After the path switching is completed, the edge computing node binds the normal analysis marker or the turbid water phase variation marker with the current epoch's reflection signal component sampling sequence, scattering signal component sampling sequence, and the change in the dielectric constant of the water surface layer, forming the analysis task frame within step S3. The normal analysis marker only allows the call to the phase feature extraction link of the reflection signal component and the power delay spectrum feature extraction link of the scattering signal component. The turbid water phase variation marker directly triggers the call to the microwave propagation delay compensation dictionary for high turbidity water and freezes the subsequent execution of the normal clear water level analysis path in the current epoch to prevent two sets of analysis results from coexisting in the same epoch.

[0063] In one embodiment, after a rainstorm, the mountain river enters a continuous monitoring cycle. The edge computing node first retrieves the hydrological phase variation coefficient of the current epoch from the cache area in step S2, and then compares it with the confidence upper limit formed by normal weather samples one by one. If the comparison result falls within the normal value range, the edge computing node writes a normal analysis mark in the analysis task frame, and only calls the normal clear water level analysis path thereafter. If the comparison result enters an abnormal isolated range, the edge computing node immediately writes a turbid water phase variation mark in the analysis task frame, and brings the change in the dielectric constant of the water surface layer in the same epoch into the turbid water phase correction tracking path. The normal analysis result is no longer used in the current epoch.

[0064] Analysis of normal clear water level under the normal numerical range of S302.

[0065] When S301 outputs the normal analysis mark, the current epoch is determined to meet the hydrodynamic properties of clear or slightly turbid. This sub-step uses the normal analysis mark as the entry condition, calls the reflection signal component sampling sequence and the scattering signal component sampling sequence of the same epoch, and forms the water level calculation result facing the normal water surface.

[0066] The edge computing node first performs phase unrolling processing on the sampled sequence of reflected signal components, eliminating phase jumps to form a continuous phase trajectory. Then, it extracts the phase features of the current epoch from this continuous phase trajectory. These phase features include at least the phase change of the main reflection path and its direction of change within several consecutive sampling windows. Subsequently, the edge computing node uses the current sampling interval as the time delay bin width and superimposes the power of the scattered signal components within thirty-two consecutive sampling windows to obtain a power delay spectrum. Then, it selects the three consecutive time delay bins with the highest cumulative power from the power delay spectrum as the main energy cluster, and uses the interval containing the center time delay of this main energy cluster as the effective water surface echo time delay window. In this sub-step, the power delay spectrum features do not output the water level separately, but rather serve as a constraint condition for the phase features, used to determine whether the main reflection path remains consistent with the water surface echo position within the current epoch.

[0067] After the phase feature and power delay spectrum feature are extracted simultaneously, the edge computing node first locks the effective water surface echo delay window for the current epoch based on the power delay spectrum feature. Then, it reads the corresponding phase feature within this effective water surface echo delay window, converts the phase change into the propagation path difference change, and converts the propagation path difference change into the water level change according to the incident angle of the current epoch. The calculation relationship is as follows: ; in, The change in the path difference during propagation. This is the current operating wavelength of the RF link. This refers to the phase change extracted within the effective water surface echo time delay window. This represents the change in water level relative to a reference water level. The angle of incidence at the current epoch. This provides real-time water level elevation data for the current epoch. This is the reference water level position after the equipment installation is completed; the incident angle is determined by the monitoring geometry of the current epoch, and the reference water level position is taken from the initial calibration result. Thus, the normal analysis path only forms unique real-time water level elevation data within the same epoch, no longer calling the microwave propagation delay compensation dictionary for high-turbidity water bodies, nor introducing corrections for abrupt changes in the dielectric constant of the water surface.

[0068] In one embodiment, the monitoring section is in a stage of continuous rainfall but the water body has not yet carried a large amount of sediment. Under normal analytical labeling, the edge computing node first recovers the continuous phase trajectory from the sampled sequence of reflected signal components, and then forms the power delay spectrum of the current epoch from the sampled sequence of scattered signal components. When the power delay spectrum shows that the main energy cluster of the water surface is still concentrated in a stable time delay interval, the edge computing node uses this time delay interval as the effective water surface echo time delay window, uses it to constrain the phase calculation of the main reflection path, and finally obtains the real-time water level elevation data of the current epoch. This real-time water level elevation data is directly incorporated into the subsequent local disaster risk assessment report generation process within the current epoch.

[0069] Turbid water phase correction tracking in the S303 abnormal isolated interval.

[0070] When S301 outputs the turbid water phase change marker, the current epoch has been determined to be a turbid water phase change in the internal material structure of the monitored water body. This sub-step takes the turbid water phase change marker and the change in the dielectric constant of the water surface layer in the same epoch as the core input, cuts off the normal clear water level analysis path, and switches to the turbid water phase correction tracking algorithm.

[0071] Upon receiving a turbid water phase change marker, the edge computing node first freezes any unsubmitted intermediate results in the current epoch's normal clear water level analysis path. Then, it calls a pre-stored high-turbidity water microwave propagation delay compensation dictionary. This dictionary uses the change in the dielectric constant of the water surface as an index axis, establishing a lookup table based on the degree of dielectric constant deviation and the corresponding propagation hysteresis. The dictionary is calibrated using multiple sets of known turbidity water samples before equipment deployment. Specifically, for each set of samples, the change in the dielectric constant of the water surface and the additional propagation path difference relative to clear water propagation conditions are measured, and index intervals are established from low to high based on the change in the dielectric constant of the water surface. Each index interval in the dictionary corresponds to a beam delay correction factor, which characterizes the additional propagation path difference that should be deducted from the original propagation path difference within that index interval. The edge computing node sends the change in the dielectric constant of the water surface layer in the current epoch to the microwave propagation delay compensation dictionary for high turbidity water. After matching it with its corresponding interval, it reads out the corresponding beam delay correction factor. When the change in the dielectric constant of the water surface layer in the current epoch is between two adjacent index intervals, the edge computing node first locates the lower-side interval and the higher-side interval in the two adjacent index intervals, then reads the beam delay correction factor corresponding to the lower-side interval and the beam delay correction factor corresponding to the higher-side interval. Subsequently, based on the position ratio of the change in the dielectric constant of the water surface layer between the two interval boundaries, it determines the transition beam delay correction factor for the current epoch and writes the transition beam delay correction factor into the correction parameter area of ​​the current epoch.

[0072] After obtaining the beam delay correction factor, the edge computing node re-calls the sampled sequences of the reflected and scattered signal components for the current epoch. The reflected signal component sampled sequence is used to form the basic propagation path difference, and the scattered signal component sampled sequence is used to confirm whether the main energy distribution under high turbidity conditions still falls within the current echo zone. The edge computing node first forms the original propagation path difference based on the current epoch's reflected signal component sampled sequence, then converts the beam delay correction factor into an additional propagation path difference, and subtracts the additional propagation path difference from the original propagation path difference to obtain the corrected propagation path difference. Subsequently, based on the current epoch's incident angle and the geometric relationship of the monitoring section, the corrected propagation path difference is converted into the corrected water level change and the corrected real-time water level elevation data. Since this correction path uses the change in the dielectric constant of the water surface as a direct index, when debris flows or high-turbidity floods advance over several consecutive epochs, the edge computing node can successively re-recall the beam delay correction factor epoch by epoch, maintaining the water level calculation link continuously tracking the changes in the turbid water phase.

[0073] In one embodiment, after the debris flow front enters the monitoring section, the edge computing node receives a turbid water phase change marker in the current epoch. The normal clear water level analysis result, which was originally awaiting submission, is immediately frozen. Subsequently, the edge computing node sends the change in the dielectric constant of the water surface layer in the current epoch to the microwave propagation delay compensation dictionary for high turbidity water to find the corresponding beam delay correction factor. After the beam delay correction factor is written into the geometric ranging calculation process, the original propagation path difference caused by the reflection path is first corrected, and then the real-time water level elevation data of the current epoch is output. Thus, even if a large amount of suspended sediment has appeared inside the water body, only a corrected real-time water level elevation data is generated in the current epoch.

[0074] S304 water level results are normalized and a local disaster risk assessment report is generated.

[0075] Regardless of whether the current epoch originates from the normal clear water level analysis path of S302 or the turbid water phase correction tracking path of S303, after obtaining the unique real-time water level elevation data, the edge computing node needs to organize the result into an alarm object that can be directly called in subsequent steps. This sub-step takes the real-time water level elevation data output by the previous sub-step as its starting point and combines the path markers of S301 and the hydrological phase anomaly coefficient of S2 to generate a local disaster risk assessment report.

[0076] The edge computing node first compares the real-time water level elevation data of the current epoch with the pre-stored local water level baseline to obtain the water level deviation value. The local water level baseline is a reference elevation established and continuously stored under normal hydrological conditions for the monitoring section, used to characterize the degree of deviation of the water level in the current epoch relative to the section. Subsequently, the edge computing node sends the water level deviation value, hydrological phase anomaly coefficient, and normal analysis marker or turbid water phase variation marker output by S301 into the local disaster classification rule table in a fixed order. The water level deviation value is used to reflect the degree of rise or fall of the section water level, the hydrological phase anomaly coefficient is used to reflect the degree of material structure anomaly, and the normal analysis marker or turbid water phase variation marker is used to limit the interpretation path of the hazard type. The local disaster classification rule table is pre-established based on the flood warning water level of the monitoring section location, historical disaster samples, and preset risk level thresholds. The rule table includes at least the water level deviation value classification interval, the hydrological phase anomaly coefficient classification interval, and the disaster type interpretation item corresponding to the path marker. As an example, the local disaster classification rule table divides water level deviation values ​​into four levels, hydrological phase variation coefficients into four levels, and path markers into two categories: normal analytical markers and turbid water phase variation markers. The edge computing node first determines the water level level based on the water level deviation value, then determines the phase level based on the hydrological phase variation coefficient, and then reads the basic score according to the intersection of the water level level and the phase level. When the path marker is a turbid water phase variation marker, a high-risk score is added to the basic score to obtain a disaster risk assessment score. Subsequently, based on the score range into which the disaster risk assessment score falls, four levels of text descriptions (ordinary, attention, alarm, and emergency) are output, and a local disaster risk assessment report is generated based on the disaster risk assessment score.

[0077] After the local disaster risk assessment report is generated, the edge computing node writes the real-time water level elevation data, water level deviation, disaster risk assessment score, and the local disaster risk assessment report of the current epoch into the output buffer. The real-time water level elevation data serves as the core measurement result for subsequent communication; the water level deviation and disaster risk assessment score serve as direct comparison objects when matching alarm dictionary conditions in subsequent steps; and the local disaster risk assessment report serves as the source report text for subsequent byte stream truncation and structured disaster semantic code reconstructing.

[0078] In one embodiment, if the current epoch outputs compensated and corrected real-time water level elevation data by S303, the edge computing node subtracts the real-time water level elevation data from the local water level baseline to obtain the water level height deviation value. Then, the water level height deviation value, the hydrological phase variation coefficient, and the turbid water phase variation marker are sent together to the local disaster classification rule table. When the table lookup result falls into the high-risk range, the edge computing node writes the corresponding hazard level and phase description in the local disaster risk assessment report, and writes it together with the real-time water level elevation data of the current epoch into the output buffer area for direct use in subsequent steps.

[0079] In step S3, the branching determination between the normal numerical range and the abnormal isolated range has been completed based on the hydrological phase variation coefficient. Within the normal numerical range, normal clear water level analysis is performed by extracting the phase characteristics of the reflected signal component and the power delay spectrum characteristics of the scattered signal component. Within the abnormal isolated range, turbid water phase correction and tracking are performed by calling the microwave propagation delay compensation dictionary for high turbidity water bodies. Through the above processing, this step generates real-time water level elevation data, water level deviation value, disaster risk assessment score, and local disaster risk assessment report that uniquely correspond to the current epoch, completing the closed-loop transformation from phase discrimination to monitoring result expression.

[0080] After real-time water level data and local disaster risk assessment reports have been generated in step S3, the monitoring equipment has completed the measurement and identification of the current cross-section. However, in extreme natural disaster scenarios, the public network communication link may have failed, and conventional long message transmission methods are difficult to adapt to the physical bandwidth constraints of BeiDou short messages. Step S4 focuses on the established monitoring results and link status. By continuously probing the physical transmission handshake status of the public network communication link and combining alarm dictionary matching conditions to select emergency upload objects, the transmission process enters a controlled disaster semantic reorganization stage.

[0081] S401 Public Network Communication Link Detection and Alarm Dictionary Hit Determination.

[0082] Step S3 has already written real-time water level elevation data, water level deviation value, disaster risk assessment score and local disaster risk assessment report to the output buffer. This sub-step takes the output buffer as the data starting point and the physical transmission handshake status of the public network communication link on the device as the trigger starting point to first determine whether the current epoch has entered the Beidou short message relay transmission path.

[0083] Edge computing nodes periodically read the physical transmission handshake status of the public network communication link's transceiver interface and form a handshake status sequence according to the continuous monitoring order. When the handshake status sequence continuously shows handshake failure, link unreachability, or physical carrier loss within the current judgment window, the edge computing node generates a public network communication link physical interruption flag and binds this flag with the real-time water level elevation data, local disaster risk assessment report, water level deviation value, and disaster risk assessment score of the current epoch to form a disaster transmission judgment frame for the current epoch. The current judgment window consists of handshake detection cycles of no less than a preset number of judgment windows N, where the preset number of judgment windows N is taken from the minimum number of retries corresponding to the handshake timeout parameter of the public network communication module. When a valid handshake is not restored for N consecutive handshake detection cycles, the edge computing node determines that the public network communication link has been physically interrupted.

[0084] After the physical interruption marker of the public network communication link is formed, the internal security log process of the device immediately retrieves the high-risk disaster entry library from the local database and sends the water level deviation value and disaster risk assessment score of the current epoch into the alarm dictionary matching process in a fixed order. Specifically, the internal security log process first locks high-risk disaster entries related to sudden rise in water level, exceeding water level limits, or abnormal rise in cross-section based on the water level deviation value, and then locks high-risk disaster entries related to flash floods, debris flows, or turbid water phase changes based on the disaster risk assessment score. In one embodiment, when the water level level corresponding to the water level deviation value is 1.5 meters above the warning water level, the internal security log process first locks the "rapid rise in cross-section water level" entry; when the disaster risk assessment score falls into the emergency range and the path marker corresponds to the turbid water phase change marker, the "debris flow accompanied by turbid water phase change" entry is then locked; when both entries are matched simultaneously, the internal security log process writes their intersection into the alarm dictionary match marker. When two types of locking results intersect at the same epoch, the device's internal security log process generates an alarm dictionary hit flag and writes the flag back to the disaster transmission decision frame. Only when the public network communication link physical interruption flag and the alarm dictionary hit flag are both true in the same disaster transmission decision frame will the edge computing node send the current epoch into subsequent byte stream truncation and semantic reassembly processing.

[0085] In one embodiment, after a typhoon hits the valley monitoring station, the public network communication link transceiver interface returns a handshake failure state for multiple consecutive decision windows. Based on this, the edge computing node generates a physical interruption flag for the public network communication link. Within the same epoch, the device's internal security log process reads the water level deviation value and disaster risk assessment score from the output of step S3. It then matches the intersection of entries related to debris flow accompanied by rapid water level rise in the local database's high-risk disaster entry library, thus generating an alarm dictionary hit flag. The current epoch then enters the semantic compression link, no longer remaining on the conventional public network transmission path.

[0086] S402 byte stream forced truncation and disaster state semantic fragment construction.

[0087] When S401 simultaneously outputs the public network communication link physical interruption flag and the alarm dictionary hit flag, the current epoch has met the triggering conditions for entering narrowband emergency transmission. This sub-step takes the disaster transmission judgment frame as the only input, organizes real-time water level elevation data and local disaster risk assessment reports in a fixed order, and performs forced truncation at the network transmission protocol layer to form a disaster status semantic fragment that can be called by a hexadecimal hash mapping table.

[0088] The edge computing node first reads real-time water level data and local disaster risk assessment report from the disaster transmission judgment frame, organizing them into a byte stream to be sent in the order of measurement results first and semantic description last. The real-time water level data provides the core measurement results of the current section, while the local disaster risk assessment report provides the disaster type and risk level description for the current epoch. In this sub-step, these two types of objects are no longer processed separately as complete long texts, but are uniformly transcribed into a continuous byte sequence. After the byte stream to be sent is formed, the edge computing node forcibly truncates this continuous byte sequence at the network transmission protocol layer. The truncation position is bounded by the single physical packet length limit of the BeiDou short message service, prioritizing the retention of core semantic fragments that describe the current disaster situation.

[0089] In specific processing, edge computing nodes first extract numerical change segments from real-time water level elevation data, then extract disaster type segments, hazard level segments, and link status segments from local disaster risk assessment reports, and rearrange them in the order of numerical change segments, disaster type segments, hazard level segments, and link status segments. Subsequently, using the single physical packet length limit of BeiDou short messages as the boundary, the rearranged continuous byte sequence is truncated from front to back to generate a disaster state semantic segment. This disaster state semantic segment still retains the most critical state expression of the current epoch, but it has already eliminated the header overhead of redundant source address, destination address, and checksum bits in conventional network messages, and can directly enter the hexadecimal hash table mapping processing.

[0090] In one embodiment, the edge computing node extracts the fact of rising water level from the real-time water level elevation data of the current epoch, and extracts the facts of debris flow, high-risk level, and public network interruption from the local disaster risk assessment report, and splices them into a continuous byte sequence in a fixed order. Subsequently, the edge computing node performs forced truncation at the network transmission protocol layer according to the single physical packet length limit of BeiDou short messages, retaining only the disaster state semantic fragment that can express the core danger of the current section. At this point, the information to be transmitted in the current epoch has been compressed from complete measurement results and complete reports into short semantic fragments that can be further mapped.

[0091] S403 hexadecimal hash mapping table reassembly and BeiDou short message upload.

[0092] After the disaster state semantic fragment is formed, step S4 enters the final encoding and transmission stage. This sub-step starts with the disaster state semantic fragment output by S402, calls the hexadecimal hash mapping table embedded in the underlying communication firmware, reassembles the semantic fragment into structured disaster semantic code, and drives the Beidou short message communication module to complete the relay upload.

[0093] Edge computing nodes first segment the catastrophe state semantic fragments field by field into numerical change sub-fragments, catastrophe seed fragments, danger level sub-fragments, and link status sub-fragments. Then, each sub-fragment is fed into a predetermined unique hash function according to the field order specified in the mapping table, yielding a feature mask index that corresponds one-to-one with the current epoch's catastrophe state. Subsequently, a hexadecimal hash mapping table is retrieved based on this feature mask index, and the corresponding hexadecimal encoded result is read out. This hexadecimal encoded result is used as the structured catastrophe semantic code. Because the hexadecimal hash mapping table uses a lossless dimensionality reduction encoding rulebook, the same catastrophe state semantic fragment always corresponds to a unique structured catastrophe semantic code after mapping. Therefore, the backend server can reconstruct the combination relationship of catastrophe facts for the current epoch based on the same rulebook at the receiving end. The predetermined unique hash function is not an open collision hash. Instead, it assigns fixed enumeration values ​​to the numerical change sub-fragments, disaster seed sub-fragments, danger level sub-fragments, and link state sub-fragments, and then concatenates them into a fixed-length sequence according to a predefined bit width to form a unique feature mask index. The hexadecimal hash mapping table establishes a one-to-one correspondence with this feature mask index as the key value. Therefore, the same disaster state semantic fragments are always mapped to the same structured disaster semantic code, and different disaster state semantic fragments will not share the same encoding result.

[0094] After the structured catastrophe semantic code is generated, the edge computing node writes it into the transmission register of the BeiDou short message communication module, and simultaneously writes the current epoch identifier to maintain the upload order. Once the BeiDou short message communication module is activated, it completes framing and transmission according to the single physical packet length limit, relaying the structured catastrophe semantic code to the backend server. After the upload is complete, the edge computing node registers the structured catastrophe semantic code and transmission status of the current epoch in its local transmission log, ensuring a traceable correspondence between the backend server's received results and the real-time water level data and local disaster risk assessment report of the current epoch.

[0095] In one embodiment, the disaster state semantic fragment indicates three types of facts: sudden rise in water level, debris flow, and public network interruption at the current cross-section. The edge computing node decomposes the semantic fragment into corresponding sub-fragments and sends them to a predetermined unique hash function. The hexadecimal hash hash mapping table is retrieved to obtain a structured disaster semantic code that occupies only 1 to 2 bytes. Subsequently, the Beidou short message communication module is awakened and completes the framing and transmission of the current epoch. After receiving the structured disaster semantic code, the backend server can identify the corresponding high-risk disaster state according to the same rules.

[0096] In step S4, the physical interruption determination of the public network communication link and the alarm dictionary hit determination have been completed. When both conditions are met simultaneously, the pending byte stream formed by the real-time water level elevation data and the local disaster risk assessment report is forcibly truncated and the fields are rearranged to generate a disaster status semantic fragment. Subsequently, the edge computing node reassembles the disaster status semantic fragment into a structured disaster semantic code adapted to the single physical packet length limit of Beidou short message based on the pre-burned hexadecimal hash mapping table. This code is then sent by the Beidou short message communication module, so that the monitoring results are uploaded in the form of compressed disaster semantics.

[0097] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of multiple parameters in water areas based on BeiDou fusion, characterized in that, Including the following steps: It receives space radio frequency signals from the Global Navigation Satellite System and downlink signals from meteorological elements of the BeiDou satellite, and separates the reflected signal components and the scattered signal components. The physical area of ​​the microwave scattering cross section of the water surface is determined based on the scattered signal components, and the abrupt change in the dielectric constant of the water surface layer is determined based on the reflected signal components. The output hydrological phase variation coefficient includes simultaneously inputting the physical area of ​​the microwave scattering cross section of the water surface and the abrupt change in the dielectric constant of the water surface layer into a pre-trained isolated forest machine learning model, and determining the hydrological phase variation coefficient based on the average path length. Real-time water level elevation data and local disaster risk assessment reports are generated based on the hydrological phase anomaly coefficient. The generation of real-time water level elevation data includes performing normal clear water level analysis when the hydrological phase anomaly coefficient is in the normal value range, and performing turbid water phase correction tracking when the hydrological phase anomaly coefficient is in the abnormal isolated range. When the public network communication link is interrupted and the local disaster risk assessment report matches the alarm dictionary conditions, the real-time water level elevation data and the local disaster risk assessment report are truncated and recombined to generate structured disaster semantic code, which is then uploaded through the Beidou short message communication module.

2. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 1, characterized in that, The process of separating the reflected signal component and the scattered signal component includes: performing multi-channel filtering on the spatial radio frequency signal to form the reflected signal component from the correlated echo falling into the delay window of the main peak of the specular reflection, and forming the scattered signal component from the scattered echo deviating from the delay window of the main peak of the specular reflection.

3. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 2, characterized in that, Determine the physical area of ​​the microwave scattering cross section of the water surface, including calculating the total received power of the scattered signal components and subtracting the inherent thermal noise of the hardware, and then performing inversion in combination with the spatial geometric distance; determine the abrupt change in the dielectric constant of the water surface layer, including extracting the difference between the left and right rotational polarization phase delays of the reflected signal components, combining the Fresnel reflection boundary equation to obtain the equivalent surface dielectric constant of the current epoch, and comparing it with the dielectric constant of the pre-stored clear water baseline.

4. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 3, characterized in that, Performing turbid water phase state correction tracking includes calling the microwave propagation delay compensation dictionary for high turbidity water bodies using the abrupt change in the dielectric constant of the water surface as an index, reading the corresponding beam delay correction factor, and incorporating the beam delay correction factor into geometric ranging calculation to generate real-time water level elevation data.

5. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 1, characterized in that, Public network link interruption is determined by at least one of the following detection results: continuous handshake failure, link unreachability, and physical carrier loss; the local disaster risk assessment report hits the alarm dictionary matching conditions, including the intersection matching of water level deviation value and disaster risk assessment score with the high-risk disaster entry library in the local database by the internal security log process of the device.

6. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 5, characterized in that, The truncation and reassembly process involves organizing real-time water level data and local disaster risk assessment reports into a byte stream to be sent, with the measurement results preceding the semantic descriptions. This stream is then rearranged according to the order of numerical change segments, disaster type segments, hazard level segments, and link status segments, and finally truncated using the limit of the single physical packet length of the BeiDou short message as the boundary.

7. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 6, characterized in that, The process of generating structured disaster semantic code involves mapping numerical change segments, disaster type segments, hazard level segments, and link status segments to fixed enumeration values, forming feature mask indexes in a predefined bit-width order, and then outputting the structured disaster semantic code based on a pre-burned hexadecimal hash mapping table.

8. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 1, characterized in that, The water monitoring equipment includes an edge computing node, a GNSS baseband receiver chip, a signal separation unit, a public network communication link interface, and a BeiDou short message communication module; the edge computing node is connected to the GNSS baseband receiver chip, the signal separation unit, the public network communication link interface, and the BeiDou short message communication module respectively.

9. The intelligent monitoring method for multiple parameters of water bodies based on BeiDou fusion according to claim 8, characterized in that, The inherent thermal noise of the hardware is obtained through silent calibration after installation, and the dielectric constant of the clear water baseline and the local water level baseline are obtained through clear water calibration under normal weather conditions. The water monitoring equipment is deployed at remote hydrological stations that rely on solar energy and battery power, and is transmitted via the public network when the public network link is available, and via the Beidou short message communication module when the public network link is interrupted.

Images