Multi-source redundancy fusion temporary tide observation system and water level calibration method

Abstract

This invention relates to the field of temporary tide gauge water level elevation measurement and calibration technology, specifically a multi-source redundant fusion temporary tide gauge system and water level calibration method. It sets up hydrostatic pressure, shore-based radar, and visual water gauge observation sources, unifies the observation results to the same elevation benchmark, and samples them synchronously. Water level sequences and status records are formed by combining water body conditions, support posture, and image status. Temporal residuals between sequences are constructed. Abnormal observation sources are identified based on the same-direction deviation of a single sequence relative to the other two sequences, the closure relationship of the other two sequences, and the status records. Benchmark deviations are extracted from the closure period, and abnormal sections are reconstructed using non-abnormal observation sources. After the recovered sequence is reclosed, a continuous water level sequence is connected. This invention can distinguish between actual tidal level changes and single-source measurement deviations, achieving isolation of abnormal sources and continuous calibration during abnormal periods. It reduces the impact of changes in water body conditions, installation posture, and visual obstruction on water level results, maintaining the continuity and elevation consistency of temporary tide gauge data.

Description

Technical Field

[0001] This invention relates to the field of temporary tide gauge water level elevation measurement and calibration technology, specifically a temporary tide gauge system and water level calibration method with multi-source redundancy fusion. Background Technology

[0002] Existing temporary tide gauge modules mostly use underwater pressure sensors as the continuous observation source. The sensors are fixed underwater at the dock, bank, or near-shore support. At the same time, onshore atmospheric pressure sensors, temporary water gauges, and leveling benchmarks are set up. Before deployment, closed leveling is used to determine the pressure-sensing surface, the elevation difference between the zero point of the water gauge and the leveling benchmark. During observation, water pressure, atmospheric pressure, and water temperature data are recorded at a uniform period. The absolute pressure measurement value is subtracted from the atmospheric pressure at the same time, and the pressure is converted into the water depth above the sensor by combining the water density. The elevation of the pressure-sensing surface is then superimposed to obtain the water level elevation. The operator reads the water gauge at several time periods and compares the manual readings with the automatically recorded values. Based on this, the fixed zero point deviation and obvious abnormal values ​​are corrected. Before the station is removed, the leveling route is re-measured to check whether the sensor or support has changed position and to complete the tide level calculation.

[0003] The continuous tidal level results obtained using the above method essentially rely on a single hydrostatic measurement link. The water gauge only provides a discrete reference during manual reading. When the pressure conversion is affected by changes in water density, abnormal atmospheric pressure acquisition, drift of the pressure sensing element, or changes in the installation position, the deviation between the automatically recorded value and the actual water level will be superimposed on tidal changes and short-period water surface fluctuations. The existing comparison process can only detect the inconsistency between the two at the moment of reading the gauge. It is impossible to determine the source of the deviation based on a single set of continuous records, nor can it reliably delineate the sections where the deviation begins, continues, and recovers. If the manual correction value is directly used to shift adjacent data as a whole, it is easy to correct the actual water level changes along with it. If the original data is retained, the measurement link error will be carried over to the tidal level benchmark calculation result. Therefore, it is necessary to solve the problem that the source of the anomaly cannot be determined when heterogeneous observation results show discrepancies, and that the water level cannot be continuously calibrated for the corresponding time period. Summary of the Invention

[0004] The purpose of this invention is to provide a temporary tide gauge system and water level calibration method with multi-source redundancy fusion, which can solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a temporary tide gauge water level calibration method with multi-source redundancy fusion, comprising: establishing a conversion relationship between the hydrostatic tide level observation source and two non-hydrostatic tide level observation sources with different mechanisms and a temporary tide gauge elevation benchmark, and synchronizing the sampling time of each observation source.

[0006] Based on the aforementioned conversion relationship, water level sequences are generated, and measurement status data from each observation source are acquired simultaneously.

[0007] The temporal residual relationship between each water level sequence is constructed. Based on the deviation relationship formed by the fact that the residuals of one water level sequence change in the same direction relative to the other two, and the residuals between the other two do not change accordingly, the abnormal observation sources are determined in combination with the corresponding measurement status data.

[0008] The baseline deviation of each observation source is extracted from the effective period when the paired residuals remain stable. The water level of the abnormal period is reconstructed based on the non-abnormal observation sources and the baseline deviation. The calibrated continuous water level sequence is output when the residuals of the restored water level sequence are consistent with those of the other two water level sequences.

[0009] Preferably, the two non-hydrostatic tidal level observation sources are respectively designated as shore-based radar observation sources and visual water gauge observation sources.

[0010] The shore-based radar observation source generates a radar water level sequence based on the radar reference point elevation and the water surface distance after installation attitude correction.

[0011] The visual water level observation source establishes a mapping from pixel position to the zero elevation of the water level gauge based on an image containing the intersection of the water level gauge scale and the water surface, and generates a visual water level sequence.

[0012] The hydrostatic tidal level observation source generates a water density correction value based on water temperature and conductivity, and generates a hydrostatic water level sequence based on the water density correction value and the water pressure corrected by atmospheric pressure.

[0013] Preferably, the measurement status data are used to form status records according to the observation link, wherein the continuity of atmospheric pressure acquisition and the change of water density correction are correlated with the stability of pressure output to form a static pressure status record.

[0014] The radar echo quality is correlated with the attitude of the mounting bracket to form a radar status record.

[0015] By linking the integrity of the water gauge scale imaging with the continuity of the water surface intersection line, a visual state record is formed.

[0016] Each state record is appended to the corresponding water level sequence according to the sampling time.

[0017] Preferably, the construction of the temporal residual relationship includes: removing the common tidal trend from the three water level sequences within the same time window, and combining the residual change direction of any water level sequence relative to the other two water level sequences into the source deviation fingerprint of the corresponding observation source.

[0018] The source deviation fingerprint is matched with the measurement state changes of the same observation source in a time sequence, and the historical closure relationship between the other two water level sequences is used as an attribution constraint to form an abnormal source identifier and an abnormal start and end segment.

[0019] Preferably, the establishment of the conversion relationship includes: obtaining the elevation difference between the leveling benchmark and the physical benchmark of each observation source through a closed leveling route, wherein the physical benchmark includes the pressure sensor pressure surface and the non-hydrostatic benchmark composed of the radar reference point and the zero point of the water gauge.

[0020] The elevation difference and the measurement direction of each observation source are defined and written into the reference relationship table.

[0021] When the mounting bracket's posture changes, the elevation of the radar reference point is corrected according to the posture change, and the non-hydrostatic reference is verified using the image position of the visual water gauge zero point.

[0022] Preferably, each of the aforementioned status records is mapped to the corresponding water level data according to the acquisition timestamp and the inherent transmission delay of each observation link.

[0023] Extract the change segments of each state record, and combine the changes in the residual direction before and after the change segment to form a state-residual correlation segment.

[0024] For related segments where the states of multiple observation sources change simultaneously, the anomaly source unresolved marker is maintained until the subsequent closure relationship is restored or a distinguishable single-source state change segment is formed.

[0025] Preferably, candidate observation sources are selected from the source deviation fingerprints that simultaneously satisfy the following relationship: the residuals of the candidate observation source relative to the other two observation sources change continuously in the same direction, and the residuals between the other two observation sources remain within the closed interval formed by the corresponding effective time period.

[0026] The overlapping sections of the measurement state change of the candidate observation source with the continuously changing sections are identified as abnormal sections, and the sections that do not overlap are retained as sections to be calibrated.

[0027] Preferably, for each non-abnormal observation source relative to the common elevation benchmark, the deviation trajectory is extracted from the effective closed segments before and after the anomalous section. The deviation trajectory is extended backward from the front and traced back from the rear to correct the water level of the non-abnormal observation source and form two candidate reconstruction sequences within the anomalous section.

[0028] When two candidate reconstruction sequences remain mutually closed, the reconstruction water level of the anomalous segment is generated based on the common change components of the two sequences.

[0029] After the restored observation source and the two non-abnormal water level sequences are reclosed, the corresponding water level sequences are continued according to the restored baseline deviation.

[0030] Preferably, the camera of the visual water level gauge observation source and the radar of the shore-based radar observation source are fixed on the same mounting bracket. Multiple elevation control markers that maintain a fixed elevation difference with the leveling benchmark are set in the imaging area of ​​the camera. The visual imaging relationship is updated according to the position changes of the elevation control markers in continuous images.

[0031] The changes in the visual imaging relationship and the changes in the attitude of the mounting bracket are respectively converted into radar reference point elevation corrections. When the two corrections are mutually closed, the reference relationship table is updated. When the two corrections are not mutually closed, the original reference relationship table is retained and a reference to be reviewed identifier is generated.

[0032] The temporary tide gauge system with multi-source redundancy fusion places the restored observation source in a bypass verification state and calculates the residuals between the restored water level sequence and the two non-abnormal water level sequences during the rising and falling tide segments, respectively.

[0033] The direction of change of the two types of residuals and their continuity on both sides of the tide level inflection point are compared with the baseline deviation before recovery.

[0034] After both types of residuals satisfy the closure relation again, the baseline deviation of the recovered observation source is updated, and the recovered water level sequence is connected to the continuous water level sequence from the most recent tide inflection moment.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The three water level sequences formed by hydrostatic pressure, shore-based radar and visual water gauge are unified to a temporary tide gauge elevation benchmark and corresponding sampling time. The source of the anomaly is determined by the residual direction of any sequence relative to the other two sequences, the closure relationship between the other two sequences and the status record of the corresponding observation link. This separates the common component of the real tide level change in each measurement mechanism from the deviation component generated by a single measurement link. When a certain observation source drifts, is blocked or has a distance measurement anomaly, the abnormal section is reconstructed by the benchmark deviation formed by the other observation sources that maintain closure and the closure period. After the recovery sequence meets the closure condition again, a continuous water level sequence is connected. This shortens the duration of erroneous data in the time series, reduces the probability of single-source anomalies entering the tide level calibration results, and maintains the continuity and elevation consistency of the water level record during the anomaly period.

[0036] 2. A water density correction is generated based on water temperature and conductivity, and a static pressure level is corrected in conjunction with atmospheric pressure. A benchmark relationship is established between the pressure-sensing surface, radar reference point, water gauge zero point, and leveling benchmark point through a closed leveling route. When the camera and radar share a support and their attitude changes, the elevation control markers in the image and the support attitude are used to generate a radar reference point elevation correction. The benchmark relationship table is updated only after the two corrections are closed, so that changes in water state, support attitude, and visual imaging relationship are entered into the corresponding calibration links. At the same time, the state record and water level data are time-mapped according to the link transmission delay. For abnormal sections, forward extension and backward backtracking are used to form candidate reconstruction sequences. Recovery source bypass verification is completed in the rising and falling tide segments to reduce the impact of time sequence misalignment, benchmark drift, and accidental consistency of a single tide stage on the water level continuity results. Attached Figure Description

[0037] Figure 1 A flowchart for multi-source redundant observation and unified benchmark water level calibration.

[0038] Figure 2 This is a summary diagram of the core relationships.

[0039] Figure 3 This is an extended diagram of the residual relationship.

[0040] Figure 4 A flowchart for identifying abnormal observation sources, reconstructing sections, and restoring access.

[0041] Figure 5 To reconstruct the weighted extended graph. Detailed Implementation

[0042] In one embodiment, please refer to Figure 1 A temporary tide gauge calibration method using multi-source redundancy fusion is adopted, specifically including: establishing a conversion relationship between the hydrostatic tide gauge observation source and two non-hydrostatic tide gauge observation sources with different mechanisms and a temporary tide gauge elevation benchmark, and synchronizing the sampling time of each observation source; generating each water level sequence based on the conversion relationship, and simultaneously acquiring the measurement status data of each observation source; constructing the temporal residual relationship between each water level sequence, and determining the abnormal observation source based on the deviation relationship formed by the residuals of one water level sequence changing in the same direction relative to the other two and the residuals between the other two not changing accordingly, combined with the corresponding measurement status data; extracting the benchmark deviation of each observation source from the effective time period where the paired residuals remain stable, reconstructing the water level of the abnormal time period based on the non-abnormal observation source and the benchmark deviation, and outputting the calibrated continuous water level sequence when the residuals of the restored water level sequence are consistent with those of the other two water level sequences again.

[0043] Please refer to Figure 2 , Figure 2The horizontal axis represents the data support quantity, using a logarithmic scale to represent the record size of different calculation relationships in the same job data. The vertical axis represents the core relationship number, corresponding in turn to paired residuals, source deviation indications, segment reconstruction, recovery access, time alignment, and benchmark correction. Figure 2 R1, D1, H1, A1, T1, and B1 are used as symbolic indexes for residuals, direction indicators, reconstructed water levels, restored access, time delays, and baseline relationships, respectively, to merge the scattered multi-source observations, time mappings, anomaly attributions, and results continuation relationships described in the text into the same technical framework.

[0044] from Figure 2 It can be seen that the data support for both pairwise residual relationships and time alignment relationships reaches 25,920 records, indicating that the continuous observation processing in this embodiment is not only corrected at a few manual reading moments, but also retains the comparability between each observation source point by point on a unified sampling axis. The source deviation indication corresponds to 8,640 unified moments, indicating that at each moment, it is possible to determine whether there is a same-direction deviation in the candidate source based on the direction combination of the two related residuals. The segment reconstruction record is 548 records, which is less than the number of full sampling points, consistent with the processing logic that H1 related reconstruction results are only generated in abnormal segments.

[0045] In this embodiment, the temporary tide gauge system is deployed near the wharf wall, pier abutment, or near-shore fixed structures. Observation locations are selected that cover the expected tidal range and allow for clear visibility of water surface echoes. Pressure sensors are fixed underwater at locations less susceptible to ship collisions and floating debris impacts. The radar and camera are mounted on the same rigid support above the water surface. A temporary water gauge is placed within the camera's field of view. An atmospheric pressure sensor, data acquisition unit, and clock reference are set up on the shore. The pressure sensor's sensing surface, radar reference point, water gauge zero point, and elevation control markers on the support are all connected to a stable leveling benchmark via a closed leveling route. The data acquisition unit receives water pressure, water temperature, conductivity, atmospheric pressure, radar ranging, radar echo quality, support attitude, water gauge images, and acquisition timestamps. The observation source, original values, benchmark version, and status records are written into the same data file, ensuring that each water level sequence retains the conversion path from the original sensor readings to the temporary tide gauge elevation benchmark.

[0046] Specifically, after the system is activated, it first reads the difference between the internal clock of each observation link and the unified clock, and determines the inherent transmission delay of the link based on the acquisition time and arrival time recorded during the installation and inspection phase. Then, it converts the original observation values ​​to the unified time axis, performs local interpolation on continuous data between adjacent unified moments according to their measurement change direction, keeps missing marks on observation values ​​that cross data interruption sections without pre-filling them, and backtracks the physical observation moments corresponding to camera exposure, radar echo generation, and pressure sampling to the unified clock. This avoids mistaking the communication arrival time for the water surface state occurrence time. After completing the time correction, it performs elevation conversion so that the three water level sequences at the same moment correspond to the same tidal state, and ensures that the measurement state record can maintain time correspondence with the original observation segment that caused the residual change.

[0047] Furthermore, the hydrostatic tidal level observation source receives the underwater absolute pressure output by the pressure sensor and reads the observation value of the onshore atmospheric pressure sensor at the corresponding moment, and determines the water column pressure by the difference between the two. For the ventilated pressure sensor, the gauge pressure value that has been balanced with atmospheric pressure through the ventilator is read, while the onshore atmospheric pressure record is retained as the basis for ventilation status verification. Water temperature and conductivity are converted into water density correction values. The conversion process uses a density relationship table that matches the type of water body on site. The density value corresponding to the current water body state is obtained by interpolating between adjacent table entries according to the direction of water temperature and conductivity. Then, the vertical water column height above the pressure sensor is calculated by converting the water column pressure and density value, and the elevation of the pressure-sensing surface is added to the vertical water column height to generate the hydrostatic water level sequence. During the conversion process, status information such as whether the atmospheric pressure is continuous, whether the density correction value changes abruptly, and whether the pressure output has unidirectional drift is also saved.

[0048] In a preferred embodiment, the shore-based radar transmits ranging signals to the water surface and receives echoes returned from the free water surface. The data acquisition unit separates the reflections from the fixed structure, the near-field reflections from the support structure, and the water surface according to the echo distance. It selects the reflection clusters that move along the direction of tidal change and whose echo contours remain coherent during continuous sampling as water surface echoes. Based on the support attitude record, the tilt range measured by tilt is converted into the vertical distance from the radar reference point to the water surface. Then, the vertical distance is subtracted from the elevation of the radar reference point to form the radar water level sequence. The camera simultaneously acquires images containing water gauge scales, water surface intersection lines, and elevation control marks. The camera's viewing angle is first corrected based on the control marks. Then, the grayscale and texture boundaries between the wet and non-wet areas are identified within the water gauge area. The boundary positions are screened for consistency along the width of the water gauge. Finally, a visual water level sequence is generated based on the correspondence between the image coordinates of the water gauge scale and the actual elevation.

[0049] In this embodiment, after the three water level sequences are unified in terms of time and elevation, they first enter the common tidal trend extraction process. In each continuous time slice, the processor sorts the three water level values ​​according to elevation, uses the continuous change path of the middle water level value as the initial result of the common trend, and then uses the tidal change direction of adjacent time slices to constrain the initial result to avoid a single jump value from changing the common trend. When a certain data path is missing, the common trend is not replaced by the remaining single data path, but only the common change path of the two paths is used when the other two water levels remain closed. Subsequently, the same common trend is subtracted from each water level sequence to obtain a relative change sequence containing the observation source bias, and pairwise residuals are formed between hydrostatic pressure and radar, hydrostatic pressure and vision, and radar and vision, respectively. The pairwise residuals save the current value, change direction, continuous segment and its corresponding observation link status in chronological order.

[0050] Specifically, paired residuals and source-offset fingerprints are calculated using the following formula:

[0051] in, and These represent the water level elevations generated by different observation sources at a unified time t. This represents the paired residuals between observation source i and observation source j. This represents the change in the paired residuals relative to the previous unified time. sgn represents the change in the observed source k at the previous unified moment. sgn represents the sign operation, taking 1 when the input is positive, -1 when the input is negative, and 0 when the input has not changed. This indicates the consistency of the deviation direction of observation source i. For example, if i is taken as the hydrostatic pressure observation source, j as the radar observation source, and k as the visual observation source, and the water levels of the three sources were 2.400 meters, 2.410 meters, and 2.400 meters at the previous moment, and are 2.460 meters, 2.420 meters, and 2.410 meters at the current moment, then the residual change of hydrostatic pressure relative to the radar is 0.050 meters, and the residual change of hydrostatic pressure relative to the visual observation is also 0.050 meters. We set the value to 1, while keeping the residual between radar and vision at 0.010 meters. Based on this, we form a candidate deviation fingerprint for the hydrostatic observation source and use the pressure link status as the basis for subsequent attribution.

[0052] Please refer to Figure 3 , Figure 3 The horizontal axis represents the unified sampling sequence number, and the vertical axis represents the water level residual, with the unit being meters. The three rows of planes represent the paired residual relationships between the three sets of observation sources: hydrostatic pressure-radar, hydrostatic pressure-visual, and radar-visual. Figure 3 Corresponding to the text , and etc. symbols, among which This represents the difference between the two water levels at the same time. This represents the change in residual between adjacent time points. Used to determine whether the deviation direction of a certain observation source is consistent with that of the other two sources.

[0053] from Figure 3 As can be seen from the upper and middle curves, the residuals of hydrostatic pressure relative to radar and relative to vision fluctuate slightly around the zero reference line for most sampling sequences. However, a significant rise and peak occur in the sampling interval of approximately 200 to 250, with peak values ​​reaching approximately 0.13 meters and 0.14 meters, respectively. This phenomenon indicates that the hydrostatic link deviates in the same direction relative to the two non-hydrostatic observation sources in the corresponding interval, and the deviation magnitude is significantly higher than that in the preceding and following stable periods, providing direct residual evidence for the identification of candidate anomaly sources.

[0054] Preferably, the anomaly attribution is not determined directly by the deviation fingerprint at a single uniform moment, but by checking whether the deviation fingerprint maintains the same direction in continuous time slices, checking whether the paired residuals of the other two water level sequences are still within the residual range formed in the previous closed period, and checking whether the state records of candidate observation sources show changes consistent with the measurement mechanism in the corresponding time slices; static pressure candidate sources correspond to atmospheric pressure interruption, density correction jump or pressure output drift, radar candidate sources correspond to echo profile change or support attitude change, and visual candidate sources correspond to missing scale imaging or discontinuous water surface intersection. Only when the source deviation fingerprint, the closure relationship of the other two sources and the state records of candidate sources form a consistent chain in time are an anomaly source identifier written. The segment that cannot form a consistent chain is marked as a segment to be calibrated and all original observations are retained.

[0055] Furthermore, the reference deviation is extracted from the closed segment where the paired residuals of the three water levels remain stable and the status of each observation link remains unchanged. The processor calculates the deviation sequence of each water level relative to the common water level of the three water levels, and uses the middle position value of the deviation sequence within the closed segment as the current reference deviation of the observation source, while retaining the direction of the deviation changing slowly over time. When a certain water level is identified as an abnormal observation source, the reference deviations of the other two water levels are deducted respectively to form two candidate correction sequences. Only when the two candidate correction sequences continue to remain closed within the abnormal segment are their common change components extracted to generate the reconstructed water level. If the two candidate correction sequences also separate, the generation of the determined water level is paused and the corresponding data is kept in a state of pending verification to avoid forcibly filling continuous data when the available observation constraints are insufficient.

[0056] In this embodiment, the recovered observation source first generates a bypass water level sequence. The bypass water level sequence does not immediately participate in the common water level calculation. The processor compares the pairwise residuals of the bypass water level sequence with the two non-abnormal water level sequences in the rising tide segment, the tide turning point region, and the falling tide segment, respectively, to check whether the fixed deviation, the direction of change, and the turning point time have closed again. When both the rising and falling segments form a closed relationship, the reference deviation of the recovered observation source is updated from the most recent common tide turning point time, and the continuous water level sequence is connected after that time. When it is closed only in a single tide change direction, the bypass verification state is maintained to distinguish between the fixed zero-point deviation and the hysteresis deviation with direction correlation. After connection, the data source identifier, reference version, and abnormal section boundary before and after recovery are still retained.

[0057] Specifically, the data acquisition unit establishes a data record for each unified moment, including the original observation record, the converted water level, paired residuals, source deviation fingerprints, status records, anomaly markers, and calibration versions. The original observation record is not overwritten during any calibration process. The converted water level can be traced back to the original sensing value through the elevation relationship and density correction records used. The reconstructed water level also saves the non-abnormal observation sources involved in the reconstruction and the closed segment indexes before and after the abnormal section. When the benchmark relationship table is updated, a new version is generated and the leveling results, attitude changes, and image control marker positions on which the update is based are recorded. This allows the original water level, calibration water level, and reconstructed water level in the same temporary tide gauge operation to be compared with each other according to a unified timeline, and retains a complete processing chain for data verification after the station is removed.

[0058] In a preferred embodiment, the two non-hydrostatic tidal level observation sources are respectively identified as a shore-based radar observation source and a visual water gauge observation source; the shore-based radar observation source generates a radar water level sequence based on the radar reference point elevation and the water surface distance corrected by the installation attitude; the visual water gauge observation source establishes a mapping from pixel position to the water gauge zero point elevation based on an image containing the intersection of the water gauge scale and the water surface and generates a visual water level sequence; the hydrostatic tidal level observation source forms a water density correction amount based on water temperature and conductivity, and generates a hydrostatic tidal level sequence based on the water density correction amount and the water pressure corrected by atmospheric pressure.

[0059] Furthermore, the water density correction relationship is initialized during setup based on the corresponding records of water temperature, conductivity, and density of the field water samples. During continuous observation, conductivity changes are not directly equated with density changes. Instead, instantaneous contact interference in the conductivity signal is first removed, and the positions of water temperature and conductivity falling between adjacent corresponding records are determined separately. After obtaining intermediate density results along the water temperature dimension, interpolation is performed along the conductivity dimension to obtain the density correction amount corresponding to the current water state. Atmospheric pressure records are first paired with underwater pressure at a unified time. Data lacking contemporaneous atmospheric pressure are marked for calibration to avoid using atmospheric pressure values ​​with large time spans as substitutes. After completing the water column pressure correction, the system uses the elevation of the pressure-sensitive surface as a fixed geometric reference, superimposes the converted water column height onto the fixed geometric reference, and writes the change process of the density correction amount into the static pressure state record.

[0060] Specifically, the raw radar data includes multiple echo distances and corresponding echo intensities generated in each ranging operation. The processor first establishes a fixed reflection position set based on the open field echo records collected during installation, eliminating reflections from the support structure and bank wall that remain at fixed distances for extended periods from the candidate echoes. Then, it searches for water surface reflection clusters that continuously move with the water level at adjacent times among the remaining echoes. For cases where multiple candidate water surface echoes appear in the same ranging operation, the processor compares the continuity of the water surface distance between each candidate echo and the previous time, as well as the integrity of the echo contour, retaining candidate values ​​that simultaneously satisfy both distance continuity and contour coherence. The support attitude is given by the tilt observation records on the installation support. The processor projects the slant range into a vertical distance based on the change of the radar measurement axis relative to the vertical direction, and writes the distance difference before and after attitude correction into the radar status record. The radar reference point elevation is read from the current reference relationship table.

[0061] In this embodiment, visual image processing uses the water level gauge scale area as the defined region. First, image edge distortion is eliminated based on lens calibration. Then, a mapping between the vertical position of the image and the water level gauge elevation is established using multiple known scale points on the water level gauge. The extraction of the water surface intersection line does not rely on a single grayscale edge. Instead, it obtains the wet texture changes on the water level gauge surface, the lateral movement boundary of water surface ripples in adjacent frames, and the comprehensive color difference of the free water surface on both sides of the water level gauge. The three types of candidate positions are projected along the width direction of the water level gauge and the intersection line position that continuously covers the main body of the water level gauge is selected. When wave occlusion occurs, the short-term gap is connected by the same direction of change on both sides of the intersection line. When continuous occlusion makes the scale or intersection line unrecognizable, no visual water level is generated. After the camera angle changes, the image mapping is first updated by the elevation control marker, and then the water surface intersection line position is converted into the water level height above the zero point of the water level gauge and the zero point elevation of the water level gauge is superimposed.

[0062] In a preferred embodiment, the measurement status data are used to form status records according to the observation link. Specifically, the continuity of atmospheric pressure acquisition and the change of water density correction are associated with the pressure output stability to form a static pressure status record; the radar echo quality is associated with the attitude of the mounting bracket to form a radar status record; the integrity of the water gauge scale imaging is associated with the continuity of the water surface intersection line to form a visual status record; and each status record is attached to the corresponding water level sequence according to the sampling time.

[0063] Specifically, the static pressure state recording uses the pressure output change within the same time slice as the main line, and correlates whether there is continuous synchronous recording of atmospheric pressure, whether the density correction changes smoothly along the direction of water temperature and conductivity changes, and whether the pressure output maintains the same tidal direction as the other two water levels; when the atmospheric pressure data is interrupted but the underwater absolute pressure is still continuous, the static pressure level does not directly use the air pressure before the interruption, but enters the air pressure waiting state; when the water temperature or conductivity causes isolated jumps due to sensor attachment, the density correction retains the continuity before and after and marks the current time as the water state to be verified; when the original pressure value deviates slowly in one direction during the period of joint stabilization of radar and vision, the pressure output drift segment is recorded; each state segment saves the start time, end time, original sensing quantity change direction and associated residual direction, so as to distinguish between pressure conversion input anomalies and pressure sensing element output anomalies when attributing anomalies.

[0064] Furthermore, the radar status record is formed by the continuous position of the water surface echo, the integrity of the echo profile, and the attitude change of the mounting bracket. When the fixed reflection is enhanced while the water surface reflection is weakened, the echo aliasing state is recorded. When the water surface echo splits in the range direction and a unique reflection cluster cannot be determined by the continuity of adjacent moments, the multi-echo state is recorded. When the bracket attitude changes, the direction and duration of the attitude change are recorded. The visual status record is formed by the imaging range of the water gauge scale, the coverage of the water surface intersection line in the width direction of the water gauge, and the position change of the elevation control mark. When the scale area is obstructed, the imaging missing state is recorded. When the water surface intersection line only appears in a local width, the incomplete intersection line state is recorded. When the control mark moves as a whole, the camera viewing angle change state is recorded. Both types of status records use the same unified timestamp as their respective water level sequences and retain independent status records when the water level has not been generated.

[0065] In a preferred embodiment, please refer to Figure 4 The construction of the time-series residual relationship includes: removing the common tidal trend from the three water level sequences within the same time window; combining the residual change direction of any water level sequence relative to the other two water level sequences to form the source deviation fingerprint of the corresponding observation source; performing time-series matching of the source deviation fingerprint with the measurement state change of the same observation source; and using the historical closure relationship between the other two water level sequences as the attribution constraint to form anomaly source identifier and anomaly start and end segments.

[0066] In this embodiment, the common tidal trend is formed by rolling continuous time windows. When all three water levels exist within the window, the intermediate water level trajectory after correction of their respective reference deviations is used as the basis, and the trend connection direction is restricted by adjacent trajectories before and after the window. When only two water levels exist within the window, the common trend is formed only under the condition that the two residuals continue their historical closed state. When only one water level remains in the window, the common trend is stopped from being updated and the original change of that path is retained. After deducting the common trend from each water level sequence, the source-related deviation trajectory is obtained. The processor combines the residual change direction of one path relative to the other two paths. If the two residuals increase or decrease at the same time, a same-direction deviation fingerprint is formed. If the two residuals are in opposite directions, a source-inter-divergence fingerprint is formed. If one of the residuals cannot be calculated due to missing data, the fingerprint remains incomplete and the abnormal source is uncertain.

[0067] Furthermore, the historical closure relationship is updated jointly by the stable observation segments and subsequent recovery segments after the station deployment is completed. The segments that enter the historical closure relationship need to meet the following conditions: the paired residuals of the three water levels do not have continuous unidirectional changes, the status records of each observation link do not have corresponding abnormal sections, and the water level turning points are consistent on a unified clock. The processor saves the median residual value, the natural fluctuation range of the residual, and the residual trajectory that changes with the direction of tide for each pair of observation sources in the closed segment. When performing attribution, it does not use fixed numerical limits to determine whether the other two paths are closed. Instead, it compares the current residual trajectory with the historical trajectory under the same direction of tide change. When the current trajectory is still covered by the historical natural fluctuation range, the closure mark is maintained. When the current trajectory is out of the historical range, the closure mark is canceled and single-source attribution is prevented to avoid misjudging the change of the common reference as a fault of a certain sensor.

[0068] In a preferred embodiment, the establishment of the conversion relationship includes: obtaining the elevation difference between the leveling benchmark and the physical benchmark of each observation source through a closed leveling route, wherein the physical benchmark includes the pressure sensing surface of the pressure sensor and a non-hydrostatic benchmark composed of a radar reference point and a water gauge zero point; writing the elevation difference and the measurement direction of each observation source into a benchmark relationship table; when the attitude of the mounting bracket changes, correcting the elevation of the radar reference point according to the attitude change, and verifying the non-hydrostatic benchmark with the image position of the visual water gauge zero point.

[0069] Specifically, the closed leveling route starts from the stable leveling benchmark on site, and sequentially observes the radar elevation transfer point, the water gauge zero-point transfer point, and the pressure sensor suspension benchmark on the mounting bracket. Then, it returns to the original leveling benchmark to form a closed observation. The elevation difference between the pressure sensor suspension benchmark and the pressure-sensing surface is transferred through a vertical connector that is fixedly connected to the sensor and has a traceable length. The fixed geometric relationships between the radar elevation transfer point and the radar reference point, and between the water gauge zero-point transfer point and the water gauge zero point, are recorded separately. During data processing, the symbols are unified according to the measurement direction convention that the water level rise corresponds to the increase in the height of the pressure water column, the decrease in the radar vertical distance, and the movement of the visual water surface intersection line in the direction of increasing elevation. The benchmark relationship table saves the leveling benchmark elevation, the benchmark elevation of each entity, the transfer path, the observation date, and the closed verification status. Before the station is removed, the same route is used for re-measurement and the results are compared with the deployment results.

[0070] Furthermore, when the support attitude record indicates that the radar measurement axis has tilted, the processor reads the rigid geometric relationship between the radar reference point and the support elevation transfer point, decomposing the attitude change into a change in the measurement axis direction and a change in the vertical position of the reference point. The former is used to correct the radar slant range, and the latter is used to correct the radar reference point elevation. The camera is fixed on the same support and continuously observes the zero point of the water gauge and the elevation control marker. After the processor updates the visual imaging relationship based on the control marker, it checks whether the converted elevation of the zero point of the water gauge is still consistent with the benchmark relationship table. If the radar attitude correction and visual verification results are mutually closed in terms of the direction and time of change, a new non-hydrostatic benchmark version is established. If the two are not closed, the original benchmark version is retained and the radar water level and visual water level are placed together in the benchmark pending verification state until the on-site leveling is re-measured or the subsequent control markers return to stability.

[0071] In a preferred embodiment, each state record is mapped to the corresponding water level data according to the acquisition timestamp and the inherent transmission delay of each observation link; the change segments of each state record are extracted, and the changes in the residual direction before and after the change segment are combined to form a state-residual correlation segment; for correlation segments where the states of multiple observation sources change simultaneously, the anomaly source unresolved mark is maintained until the subsequent closure relationship is restored or a distinguishable single-source state change segment is formed.

[0072] Specifically, the inherent transmission delay of each observation link is calibrated during the installation and inspection phase using the same observable water surface disturbance. The processor records the physical sequence of the appearance of the same turning point characteristics in the pressure water level, radar water level, and visual water level, and subtracts the arrival time difference of the data packets to obtain the link delay record from physical observation to data writing. During continuous operation, the original timestamp is first subtracted from the corresponding link delay and then mapped to a unified time axis. The status record uses the same correction time as the original observation it describes. When camera image processing backlog or radar echo recalculation occurs, the location is based on the original exposure time or the original transmission time, and the processing completion time is not used. When the link delay changes due to buffer changes, it is recalibrated using subsequent common tidal level turning point characteristics, and the data before and after the change are divided into different time-mapped versions.

[0073] Furthermore, the state-residual correlation segment takes the residual segment that was still closed before the state change began as the front boundary and the residual segment that re-entered closure after the state recovery as the back boundary. Within the boundary, the state change direction and the two residual change directions of the corresponding source are stored. If only one observation source changes state and the residual of that source changes in the same direction relative to the other two sources, a distinguishable single-source correlation segment is formed. If wind waves, ship waves, or support vibration cause the radar echo, visual intersection, and pressure output to change simultaneously, the unresolved anomaly source mark is retained, and the recovery order of the three residuals after disturbance attenuation is observed. When two of them recover closure while the other source remains deviated, a single-source correlation segment is formed. If the three recover synchronously, the segment is saved as a common environmental disturbance and is not written into the single anomaly source mark.

[0074] In a preferred embodiment, candidate observation sources that simultaneously satisfy the following relationship are selected from the source deviation fingerprint: the residuals of the candidate observation source relative to the other two observation sources change continuously in the same direction, and the residuals between the other two observation sources remain within the closed interval formed by the corresponding effective time period; the measurement state change segment of the candidate observation source and the overlapping segment of the continuous change are determined as abnormal segments, and the segments that do not overlap are retained as segments to be calibrated.

[0075] Specifically, the processor reads the source deviation fingerprint for each observation source. When the two residuals of the candidate source relative to the other two sources both move from the historical reference position to the same elevation direction, it continues to read the residual trajectory between the other two sources and selects the historical closed segment that is the same as the current tide level rising, turning or falling direction for comparison. When the residuals of the other two sources are still within the coverage of the historical closed segment, the candidate relationship is confirmed. When the residuals of the other two sources simultaneously leave the historical closed range, the single-source candidate relationship is canceled and the system enters the multi-source waiting calibration state. After the candidate relationship is established, the processor checks whether the change type in the candidate source status record is compatible with the residual direction. For example, a pressure output drifting to a higher value should correspond to a deviation of the static pressure water level relative to the other two sources in the direction of increasing elevation. An increase in the radar vertical distance conversion due to the tilt of the support should correspond to a deviation of the radar water level in the direction of decreasing elevation.

[0076] Furthermore, the starting boundary of the abnormal segment is taken from the first common unified moment of the candidate source measurement state change segment and the residual continuous unidirectional change segment, and the ending boundary is taken from the first common unified moment when the measurement state recovers and the two residuals of the candidate source stop expanding in the original direction; the preceding segment where the measurement state changes first but the residual has not yet deviated only saves the state warning, and the segment where the residual deviates first but the state record has not yet shown a corresponding change is retained as the calibration segment. If the state change can be mapped back to the segment through the link delay, the boundary will be corrected forward. If there is still no corresponding state evidence, the calibration mark will be maintained; for residual interruption caused by short-term missing measurement, the abnormal segment does not automatically connect across the missing measurement data, but forms separate segments with boundaries to be confirmed. It is determined whether they belong to the same abnormality after the other two sources continue to close and the candidate source recovers.

[0077] In a preferred embodiment, the deviation trajectory of each non-abnormal observation source relative to the common elevation benchmark is extracted from the effective closed segments before and after the anomalous section. The deviation trajectory is extended backward from the front deviation trajectory and traced back from the rear deviation trajectory to correct the water level of the non-abnormal observation source and form two candidate reconstruction sequences within the anomalous section. When the two candidate reconstruction sequences remain mutually closed, the reconstructed water level of the anomalous section is generated based on their common change components. After the recovered observation source and the two non-abnormal water level sequences are reclosed, the corresponding water level sequence is continued according to the recovered benchmark deviation.

[0078] In this embodiment, the front-side deviation trajectory is generated from the latest consecutive closed segment before the anomaly starts. The processor respectively extracts the median deviation of the two non-anomalous observation sources relative to the common elevation datum and its slow change direction, and extends them along the time into the anomaly section to form a forward correction relationship; the rear-side deviation trajectory is generated from the consecutive reclosed segments of the two non-anomalous observation sources after the anomaly ends. The processor extracts the same content in reverse chronological order and traces back into the anomaly section to form a backward correction relationship; after each path of non-anomalous water level is respectively applied with the forward correction relationship and the backward correction relationship, a forward candidate sequence and a backward candidate sequence are formed. When the difference between the candidate sequences at the same moment is still within the historical closed range of the corresponding observation source, two-way fusion is performed, and the moment when the difference deviates from the closed range is retained with a pending review mark without outputting a determined reconstructed value.

[0079] Specifically, the two-way fused water level in the anomaly section is calculated according to the following formula:

[0080] wherein, denotes the reconstructed water level corresponding to the unified moment t in the anomaly section, denotes the start moment of the anomaly section, denotes the end moment of the anomaly section, denotes the forward candidate water level obtained by extending backward from the front-side closed segment, denotes the backward candidate water level obtained by tracing back forward from the rear-side closed segment; for example, let be 0, be 4, t be 1, the forward candidate water level be 3.218 meters, and the backward candidate water level be 3.226 meters, then be 3, be 1, the calculated reconstructed water level is 3.220 meters. The values make the data near the anomaly start position inherit more of the water level datum of the front-side closed segment, and make the reconstructed trajectory transition continuously toward the anomaly end position to the rear-side closed segment.

[0081] Please refer to Figure 5 , Figure 5 the horizontal axis in the figure is the sampling serial number in the anomaly section, the vertical axis of the upper coordinate system is the reconstructed water level in meters, the vertical axis of the middle coordinate system is the candidate deviation, and the vertical axis of the bottom coordinate system is the forward and backward weights. Figure 5 corresponds to the , , , and and other symbols in the text.

[0082] from Figure 5As can be seen from the upper curve, the reconstructed water level maintains a continuous tidal fluctuation pattern throughout the entire anomaly section, with the curve varying from approximately 1.25 meters to 3.10 meters, without any abrupt jumps at the start or end points of the section or at intermediate sampling locations. This trend indicates that the reconstruction process inherited the common tidal level change component from the closed segments on both sides, rather than replacing the anomaly section with a single constant or artificially shifted value, allowing the reconstruction results to continue reflecting the periodic fluctuations of the true tidal level.

[0083] The mid-bias facet shows the deviation between the forward and backward candidates in the fusion result. The forward bias gradually expands towards the negative with the sampling sequence number, while the backward bias gradually approaches the zero-point reference line from the positive direction. Both remain on the order of millimeters. This distribution indicates that near the anomaly initiation boundary, The constraints on the fusion results are stronger; near the anomalous end boundary, The constraints gradually increase. If both deviation curves simultaneously show a step or rapidly expand, it indicates that the preceding and following closed segments cannot jointly constrain the abnormal section, and the condition should be left pending review.

[0084] Furthermore, after the anomaly termination boundary, the water level sequence of the recovered observation source is first deducted from the baseline deviation saved before recovery and compared with two non-anomaly sequences. If the paired residuals show a new fixed offset while the tidal change direction and turning point are consistent, the fixed offset is updated to the recovered baseline deviation. If the paired residuals show opposite changes in the rising and falling tidal segments, the direction-related deviation is retained and bypass observation continues. When reconnecting, the turning point where the tidal direction changes together in all three water levels is selected as the connection position. The reconstructed water level is retained before the connection position, and the recovered sequence after updating the baseline deviation is used to participate in the common water level calculation after the connection position. The adjacent water level changes on both sides of the connection position are continuously checked. If a sudden jump occurs, the connection is re-executed at the next common turning point.

[0085] In a preferred embodiment, the camera of the visual water level gauge observation source and the radar of the shore-based radar observation source are fixed on the same mounting bracket. Multiple elevation control markers that maintain a fixed elevation difference with the leveling benchmark are set in the imaging area of ​​the camera. The visual imaging relationship is updated according to the position changes of the elevation control markers in continuous images. The changes in the visual imaging relationship and the changes in the attitude of the mounting bracket are respectively converted into radar reference point elevation corrections. When the two corrections are mutually closed, the benchmark relationship table is updated. When the two corrections are not mutually closed, the original benchmark relationship table is retained and a benchmark to be verified marker is generated.

[0086] Specifically, elevation control markers are fixed on the surface of a structure that is stably connected to the on-site leveling benchmark and distributed at different elevations and lateral positions within the camera's field of view. During deployment, the elevation difference of each control marker relative to the zero point of the water gauge is recorded through leveling observations. After the camera is installed, a visual imaging relationship is established using the actual positions of the control markers and the image positions. During continuous observation, the processor tracks the common displacement, relative displacement, and shape changes of each control marker in the image. The common displacement is used to identify the rigid body movement of the camera along with the support, and the relative displacement is used to identify changes in the lens angle or partial marker occlusion. Only control markers that are not occluded and whose relative positions remain geometrically consistent participate in updating the imaging relationship. The updated imaging relationship simultaneously recalculates the water gauge scale position and the water gauge zero point position, ensuring that the visual water level remains connected to the elevation control on the stable structure.

[0087] Furthermore, the relative positions of the camera and radar on the same mounting bracket are recorded during deployment using rigid geometric measurements. The processor calculates the camera's attitude change relative to the stable structure based on the control markers, and converts the first elevation correction of the radar reference point according to the fixed spatial relationship between the camera and the radar reference point. Simultaneously, it converts the second elevation correction based on the bracket attitude record and the geometric relationship between the radar reference point and the bracket's rotation center. When the directions of the two corrections are consistent and their time-varying trajectories are mutually closed, their common elevation change is written into a new benchmark relationship table, and the relationship between the radar reference point elevation and visual imaging is corrected simultaneously. When the two corrections are not closed, neither result is selected for individual updating. Instead, the processor checks for control marker obstruction, local camera loosening, and attitude recording interruptions, and maintains the non-hydrostatic water level in an independent observation state while awaiting on-site level verification.

[0088] In a preferred embodiment, the restored observation source is placed in a bypass verification state, and the residuals between the restored water level sequence and the two non-abnormal water level sequences are calculated during the rising and falling tide segments, respectively. The direction of change of the two types of residuals and their continuity on both sides of the tide inflection point are compared with the reference deviation before restoration. After both types of residuals satisfy the closure relationship again, the reference deviation of the restored observation source is updated, and the restored water level sequence is connected to the continuous water level sequence from the most recent tide inflection moment.

[0089] In this embodiment, the rising and falling tide segments are divided based on the direction of change of the common water level at a continuous and uniform moment. The reciprocating changes caused by short-period waves are first separated through a common trend extraction process. The position where the continuous change direction changes from rising to falling or from falling to rising is determined as the tide turning point. During the bypass verification period, the residual trajectories of the recovered sequence relative to the two non-abnormal sequences in the rising and falling segments are summarized respectively. The fixed zero-point deviation is manifested as the relationship between the residual direction and amplitude in the two types of tide segments being consistent. The time lag is manifested as the change of residual direction on both sides of the turning point. The direction-related conversion deviation is manifested as the opposite direction of residual change in the rising and falling segments. The processor only restores the observation source to participate in the fusion when the fixed deviation can be absorbed by the new reference deviation and there is no continuous residual expansion in the turning point, and the bypass data that fails the verification continues to be stored in the independent sequence.

[0090] Preferably, before the system enters formal continuous observation, an initial closed dataset is established. After the on-site personnel complete the elevation difference verification of the leveling benchmark, pressure surface, radar reference point, and water gauge zero point, the three observation sources are continuously collected under the same water surface conditions. The processor completes time mapping according to a unified clock, and generates static pressure, radar, and visual water levels respectively. It also checks whether the three sets of paired residuals change back and forth around their respective middle positions and whether there is a continuous unidirectional expansion. The segments that meet the closure conditions are written into the initial historical closure relationship. The deviation of each water level in the segment relative to the common water level is written into the initial reference deviation. When a closure is not formed, the alignment is not forced by modifying the zero point of a single channel. Instead, the atmospheric pressure pairing, density correction relationship, radar reference point elevation, and visual imaging relationship are checked in sequence until the differences between the three channels can be explained by the original sensing quantity and elevation conversion path before the initial reference version is established.

[0091] Furthermore, when each observation source is deployed on-site, it should correspond to the same local water surface area as much as possible. If it is not possible to set them in exactly the same location, the processor uses the long-period common tidal trend of the three water levels to establish a time correspondence, and retains the short-period differences caused by ship waves or local reflections in the source correlation residuals. When a short-period wave peak passes through different observation positions in sequence, its residuals show that the directions are opposite and return to the historical closed range after the wave peak passes. The processor records this as a spatial propagation segment, without forming a continuous same-direction deviation fingerprint. Only when a certain water level continuously moves in the same direction relative to the other two in multiple common tidal change segments does it enter the candidate anomaly judgment. In this way, the short-time phase difference caused by local water surface propagation and the continuous deviation of the sensing link are processed separately, while the original high-frequency changes are retained for the verification of the water surface status in the measurement area.

[0092] In a preferred embodiment, when visual obstruction, radar maintenance, or pressure link interruption causes only two water levels to be available when formal observation begins, the system still generates water levels and calculates the residuals of the two channels according to their respective conversion relationships, but does not identify a single source of anomaly when a third independent measurement constraint is lacking. The time period is recorded as a dual-source degraded observation segment. If the two residuals continue the previous historical closed trajectory, a temporary common water level is generated with the two correction sequences and the degraded source identifier is saved. If the two residuals deviate from the historical closed trajectory, the generation of a determined calibration water level is stopped and the original data is retained. After the third observation is restored, it first enters a bypass state. The direction of the residuals formed by the third observation and the original two channels is used to determine whether there is a traceable deviation in the original dual-source segment. Subsequent calibration is only performed on segments that can be supported by the restored closed relationship.

[0093] Specifically, during the station removal phase, each elevation transfer point is observed again along the closed leveling route established during deployment, and the station removal elevation difference is matched item by item with the initial benchmark relationship table. The elevation change of the pressure-sensitive surface is determined by the re-measurement relationship of the suspension base point and vertical connector. The changes of the radar reference point and the zero point of the water gauge are cross-checked by the co-frame attitude record and the visual elevation control mark record. If the initial and station removal elevation differences remain closed, the original benchmark version is used throughout the observation period. If the elevation difference change can be located in a specific time period in the support attitude or control mark record, then two benchmark versions are established with the time period as the boundary and the corresponding water level is recalculated. If the change time period cannot be located, then the segment from the most recently confirmed closed segment to the station removal time is marked as the benchmark section to be checked, and the entire observation period is not uniformly shifted based on the station removal difference.

[0094] In this embodiment, continuous water level outputs are stored in layers according to the observation constraint status. When all three water level channels are in a closed state, the common water level after benchmark deviation unification is output, and the three sources involved in the calculation are recorded. When a single channel is identified as abnormal and the other two channels remain closed, the reconstructed water level is output, and the abnormal source identifier, abnormal boundary, and bidirectional candidate sequence are associated. When the abnormal source is unresolved or the two non-abnormal channels lose closure, only the water level to be calibrated is saved, and it is not written into the determined result sequence. When the benchmark relationship is in a state of pending verification, the temporary water level converted according to the original benchmark and the original water level without benchmark change are retained respectively. After the leveling is re-measured, the control marker is restored, or the subsequent closed segment provides a basis for determination, only the corresponding benchmark version and the affected time segment are recalculated, and the remaining closed data and original observation records remain unchanged.

[0095] Specifically, in the temporary tide gauging operation supporting harbor basin dredging survey, the pressure sensor is deployed inside the harbor basin, the radar and camera are fixed on the quay wall bracket, and the visual water gauge maintains elevation connection with the onshore benchmark. When the tidal current brings continuous changes in water temperature and conductivity, the hydrostatic pressure link first updates the water density correction value. If the corrected hydrostatic water level still deviates in the same direction relative to the radar and visual water levels, while the radar and visual measurements maintain historical closure, and a unidirectional drift appears in the pressure output stability record, the section corresponding to the hydrostatic pressure observation source is determined as an abnormal section, and the water level is reconstructed using the two-way candidate sequences of the radar and visual measurements. After the pressure output is restored, the system performs bypass verification in the subsequent rising and falling tide stages, updates the reference deviation and accesses the continuous water level sequence from the common turning moment. The height difference obtained from the re-measurement after station removal is written into the new reference version and used to verify the water level reduction chain throughout the operation period.

[0096] In another preferred embodiment, the temporary tide gauging system is deployed in a nearshore survey area jointly affected by rainfall, fog and ship-generated waves. When the visual water gauge image has missing scale imaging due to attached water droplets, the visual state record first marks the corresponding section. If the visual water level deviates in the same direction relative to the hydrostatic pressure and radar water levels while the hydrostatic pressure and radar measurements remain closed, the visual observation source is determined to be abnormal, and the water level is reconstructed with the two non-abnormal sequences. When ship-generated waves simultaneously cause short-term pressure fluctuations, multiple radar echoes and visual intersection line fragmentation, the three channels' states change at the same time, and the system keeps the undetermined mark for the abnormal source and waits for the disturbance to attenuate, without attributing the common environmental disturbance to a single observation source. After the visual imaging is restored, the processor reconstructs the imaging relationship using the elevation control mark, completes the restoration verification through the bypass residuals of the rising and falling stages, then connects the visual sequence to the continuous water level processing chain and retains the original observation records of the disturbance section.

Claims

1. A temporary tide gauge calibration method based on multi-source redundancy fusion, characterized in that, include: A conversion relationship to a temporary tide gauge elevation benchmark is established for the hydrostatic tide level observation source and two non-hydrostatic tide level observation sources with different mechanisms, and the sampling time of each observation source is synchronized. Based on the aforementioned conversion relationship, water level sequences are generated, and measurement status data from each observation source are acquired simultaneously. Construct the temporal residual relationship between each water level sequence. Based on the deviation relationship formed by the fact that the residuals of one water level sequence change in the same direction relative to the other two and the residuals between the other two do not change accordingly, combine the corresponding measurement status data to determine the source of abnormal observations. The baseline deviation of each observation source is extracted from the effective period when the paired residuals remain stable. The water level of the abnormal period is reconstructed based on the non-abnormal observation sources and the baseline deviation. The calibrated continuous water level sequence is output when the residuals of the restored water level sequence are consistent with those of the other two water level sequences.

2. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 1, characterized in that, The two non-hydrostatic tidal level observation sources were respectively identified as shore-based radar observation sources and visual water gauge observation sources; The shore-based radar observation source generates a radar water level sequence based on the radar reference point elevation and the water surface distance after installation attitude correction. The visual water level observation source establishes a mapping from pixel position to the zero elevation of the water level gauge based on an image containing the intersection of the water level gauge scale and the water surface, and generates a visual water level sequence. The hydrostatic tidal level observation source generates a water density correction value based on water temperature and conductivity, and generates a hydrostatic water level sequence based on the water density correction value and the water pressure corrected by atmospheric pressure.

3. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 2, characterized in that, The measurement status data are recorded according to the observation link. The atmospheric pressure acquisition continuity and the change of the water density correction are correlated with the pressure output stability to form a static pressure status record. The radar echo quality is correlated with the attitude of the mounting bracket to form a radar status record; By linking the integrity of the water gauge scale imaging with the continuity of the water surface intersection line, a visual state record is formed; Each state record is appended to the corresponding water level sequence according to the sampling time.

4. The temporary tide level calibration method with multi-source redundancy fusion according to claim 3, characterized in that, The construction of the time series residual relationship includes: removing the common tidal trend from the three water level series within the same time window, and combining the residual change direction of any water level series relative to the other two water level series into the source deviation fingerprint of the corresponding observation source. The source deviation fingerprint is matched with the measurement state changes of the same observation source in a time sequence, and the historical closure relationship between the other two water level sequences is used as an attribution constraint to form an abnormal source identifier and an abnormal start and end segment.

5. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 2, characterized in that, The establishment of the conversion relationship includes: obtaining the elevation difference between the leveling benchmark and the physical benchmark of each observation source through a closed leveling route. The physical benchmark includes the pressure sensor pressure surface and the non-static pressure benchmark composed of the radar reference point and the zero point of the water gauge. The elevation difference and the measurement direction of each observation source are defined in the reference relationship table; When the mounting bracket's posture changes, the elevation of the radar reference point is corrected according to the posture change, and the non-hydrostatic reference is verified using the image position of the visual water gauge zero point.

6. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 3, characterized in that, Based on the acquisition timestamp and the inherent transmission delay of each observation link, each state record is mapped to the corresponding water level data; Extract the change segments of each state record, and form a state-residual correlation segment by combining the changes in the direction of the residuals before and after the change segments; For related segments where the states of multiple observation sources change simultaneously, the anomaly source unresolved marker is maintained until the subsequent closure relationship is restored or a distinguishable single-source state change segment is formed.

7. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 4, characterized in that, Candidate observation sources that simultaneously satisfy the following relationship are selected from the source deviation fingerprints: the residuals of the candidate observation source relative to the other two observation sources change continuously in the same direction, and the residuals between the other two observation sources remain within the closed interval formed by the corresponding effective time period. The overlapping sections of the measurement state change of the candidate observation source with the continuously changing sections are identified as abnormal sections, and the sections that do not overlap are retained as sections to be calibrated.

8. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 7, characterized in that, For each non-abnormal observation source relative to the common elevation benchmark, the deviation trajectory of each non-abnormal observation source is extracted from the effective closed segments before and after the anomalous section. The deviation trajectory is extended backward from the front and traced back from the rear to correct the water level of the non-abnormal observation source and form two candidate reconstruction sequences within the anomalous section. When two candidate reconstruction sequences remain mutually closed, the reconstruction water level of the abnormal segment is generated based on the common change components of the two sequences. After the restored observation source and the two non-abnormal water level sequences are reclosed, the corresponding water level sequences are continued according to the restored baseline deviation.

9. The temporary tide level calibration method based on multi-source redundancy fusion according to claim 5, characterized in that, The camera of the visual water level gauge observation source and the radar of the shore-based radar observation source are fixed on the same mounting bracket. Multiple elevation control markers that maintain a fixed elevation difference with the leveling benchmark are set in the imaging area of ​​the camera. The visual imaging relationship is updated according to the position change of the elevation control markers in continuous images. The changes in the visual imaging relationship and the changes in the attitude of the mounting bracket are respectively converted into radar reference point elevation corrections. When the two corrections are mutually closed, the reference relationship table is updated. When the two corrections are not mutually closed, the original reference relationship table is retained and a reference to be reviewed identifier is generated.

10. A temporary tide gauge system with multi-source redundancy fusion, characterized in that, The restored observation source is placed in bypass verification state, and the residuals between the restored water level sequence and the two non-abnormal water level sequences are calculated in the rising and falling tide segments, respectively. The direction of change of the two types of residuals and their continuity on both sides of the tide level inflection point are compared with the baseline deviation before recovery. After both types of residuals satisfy the closure relation again, the baseline deviation of the recovered observation source is updated, and the recovered water level sequence is connected to the continuous water level sequence from the most recent tide inflection moment.

Images