Ice buoy heterogeneous measurement data compression and reconstruction method and system

Abstract

The application discloses an ice buoy heterogeneous measurement data compression and reconstruction method and system, and belongs to the technical field of polar ocean observation data processing. The drift speed and the observation parameter change rate of the ice buoy are extracted at the compression end to construct a polar environment state feature, and a dynamic spatio-temporal observation window is reconstructed according to the feature to perform adaptive spatio-temporal prediction. The feature is used as a feedforward penalty term to adjust a quantization step and Golomb-Rice coding parameters, and high compression ratio packaging and transmission of data under a limited bandwidth are realized. After decoding and spatio-temporal registration at the reconstruction end, a trust coefficient based on the environment state feature is introduced to perform multi-source weighted fusion. A continuous three-dimensional data field is constructed by using a physical guided neural network, error attribution is finally calculated, and reverse control signaling is generated and issued to update the front-end control hyperparameters. The application overcomes the bottleneck of narrowband communication in the polar region, realizes high-fidelity reconstruction and long-term closed-loop self-correction.

Description

Technical Field

[0001] This invention relates to the field of polar marine observation and data processing technology. Specifically, it relates to a method and system for compressing and reconstructing heterogeneous measurement data from ice buoys, which can be applied to scenarios such as polar marine environmental monitoring, disaster early warning, and scientific research on climate change. Background Technology

[0002] As a core cold source and driving engine of the global climate system, the polar oceans exert a profound influence on global climate change through ice sheet movement, seawater temperature and salinity structure, and sea ice-ocean-atmosphere interactions. Ice buoys, as key anchoring and drifting nodes in the polar ocean environmental observation system, can continuously collect multi-dimensional data covering ice thickness, seawater temperature, salinity, chlorophyll concentration, and sea surface meteorological elements. With the advancement of the polar ocean three-dimensional observation network, the number of ice buoys of different models and manufacturers deployed in polar ice areas and marginal seas has surged, generating an exponentially increasing amount of heterogeneous and heterogeneous observation data. This places extremely high demands on the efficient transmission and accurate processing of polar ocean data.

[0003] However, the unique geographical and climatic environment of the polar regions imposes severe constraints on the operation and communication of buoys. On the one hand, the extreme cold and high ice cover of the polar environment limit the edge computing power of buoy equipment due to the scarcity of battery capacity and solar recharge. On the other hand, the polar seas have wide blind spots for conventional radio communication, and the data transmission of ice buoys relies heavily on the polar low-Earth orbit satellite communication network. This channel exhibits characteristics such as extremely narrow bandwidth, short and alternating communication windows, and extreme susceptibility to polar cyclone interference. Under these harsh boundary conditions, the full real-time transmission of high-precision raw data collected at high frequencies by multiple sensors is no longer engineering feasible. The massive amount of multi-source data faces severe communication congestion and transmission bottlenecks at the edge of the buoy, necessitating the introduction of a data compression mechanism.

[0004] Existing conventional compression techniques for marine monitoring data face significant limitations in polar scenarios. Traditional lossless compression algorithms struggle to overcome bandwidth limitations, resulting in extremely low compression ratios. Meanwhile, common lossy compression techniques often neglect the physical context of observational data when processing ice buoy data. The polar ocean exhibits typical non-stationary spatiotemporal evolution characteristics. For example, beneath thick ice sheets, water temperature and salinity parameters show extremely gradual steady-state evolution, while in ice-edge retreat zones or areas disturbed by polar cyclones, environmental parameters exhibit dramatic abrupt changes on a spatiotemporal scale. Traditional compression methods typically employ pre-set static prediction models and fixed quantization thresholds for independent data truncation and encoding, failing to fully exploit the continuous evolution of the marine environment. This not only easily leads to computational and bandwidth redundancy during periods of parameter stability but also significantly increases the risk of permanent loss of crucial physical characteristics such as stratification changes during periods of environmental upheaval.

[0005] Furthermore, due to the extremely sparse distribution of polar observation nodes, the fragility of equipment, and communication packet loss, the decompressed data received by shore-based data centers typically exhibits high spatial discreteness and temporal discontinuity. Existing ocean data field reconstruction schemes mostly rely on conventional single statistical interpolation or basic numerical assimilation models. While these methods are applicable in data-dense areas, in polar regions with extremely scarce data, the lack of sufficient constraints from ocean dynamics and thermodynamic mechanisms often leads to significant distortion in the reconstructed three-dimensional data field, and even produces physical artifacts that violate fluid continuity and the principles of temperature and salinity evolution. More critically, in existing data observation systems, the compression and transmission process on the edge side and the reconstruction processing process on the shore-based side are completely disconnected in an open-loop manner. When faced with distorted reconstruction results, shore-based data centers cannot conduct reverse error source assessment, nor can they feed back global reconstruction error information to the front-end ice buoys to intervene in their data sampling and compression strategies. This lack of two-way linkage and closed-loop feedback architecture ultimately leads to severe systematic error accumulation in the long-term independent operation of polar ice buoys, making it difficult to meet the urgent need of polar marine science for the establishment of high-fidelity, continuous three-dimensional data fields. Summary of the Invention

[0006] This invention provides a method for compressing heterogeneous measurement data from polar ice buoys, the method comprising: Obtain heterogeneous raw data from ice buoys and preprocess them; extract polar environmental state characteristics based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. Based on the characteristics of polar environment, the spatiotemporal observation window of the prediction algorithm is dynamically reconstructed. Adaptive spatiotemporal prediction is performed on the preprocessed data, and the residual and prediction confidence deviation between the actual observation value and the final spatiotemporal prediction value are extracted. The residual is quantized by dynamically calculating the quantization step size based on the prediction confidence deviation, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. Golomb-Rice coding is performed, and the generated bit stream sequence, together with the polar environment state characteristics, is encapsulated into a data frame for uplink transmission. Receive the reverse control signaling sent by the reconfiguration end, and update the control hyperparameters according to the parsed gear adjustment amount.

[0007] Polar environmental state characteristics are extracted based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. The specific calculation formula is as follows: in, Characteristics of polar environment; The drift speed at the current moment; The maximum drift velocity of the polar preset baseline is set based on historical polar climate background data; The time-series rate of change of the observed parameters; The maximum rate of change of the polar preset baseline for the corresponding observation parameters; and These represent the drift feature weights and the fluctuation feature weights, respectively, and satisfy the following conditions: .

[0008] The spatiotemporal observation window of the prediction algorithm is dynamically reconstructed based on the characteristics of the polar environment, specifically including calculating the adaptive backtracking step size for the time dimension. And the adaptive search radius for spatial dimensions : in, and These are the preset upper and lower limits of the time backtracking step size; and These are the preset upper and lower limits of the search radius for adjacent buoy spaces; This indicates a round-down operation; This describes the characteristics of the polar environment.

[0009] The residual is quantized based on the dynamic calculation of the quantization step size according to the prediction confidence bias, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. The specific calculation formula is as follows: in, For adaptive quantization step size; To predict confidence bias; and These are the preset lower and upper limits for the quantization step size, respectively; This is the preset prediction confidence level deviation tolerance limit; This indicates the operation of finding the minimum value; These are prior coding parameters; The empirical probability coefficients are set; Let $\mathbf{a}$ be the expected value of the absolute value of the current residual. The penalty coefficient for drastic environmental changes; This describes the characteristics of the polar environment.

[0010] This invention also provides a method for reconstructing heterogeneous measurement data from polar ice buoys, used to reconstruct data transmitted by the compression method, the method comprising: The system receives data frames sent by the front-end ice buoy, extracts polar environment state features, decodes them to reconstruct the observation values, performs unified spatiotemporal registration on the observation values, and generates unified weighted fusion feature values ​​based on multi-source weighted fusion of polar environment state features. An initial analysis field is constructed based on unified weighted fusion eigenvalues. A physical-guided neural network constrained by the ocean dynamics conservation equation is used to perform fine reconstruction of the three-dimensional data field. Based on the calculation of the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output a physical violation check identifier. By combining the comprehensive fusion weights and physical violation inspection indicators during multi-source weighted fusion, the error attribution degree of the front-end ice buoy is calculated. For ice buoys that are determined to be faulty, an adaptive correction penalty multiplier is calculated. Based on this, the gear adjustment amount for the control hyperparameter is generated and encapsulated as a reverse control signal and sent to the corresponding ice buoy.

[0011] Based on the multi-source weighted fusion of polar environmental state characteristics, the formula for calculating the comprehensive fusion weight is as follows: in, Trust coefficient for polar scenarios; These are the polar environment state features attached to the corresponding ice buoy data frames. The polar environment sensitivity coefficients corresponding to different sensor physical types; For comprehensive integration and weighting; Spatial distance mapping weights; This is the nominal precision weight.

[0012] Based on the calculation of the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output physical violation detection indicators. Specifically, this includes extracting time slices and calculating the short-time evolution rate at deep-water grid points. : Short-time evolution rate and polar deep-water thermal capacity extreme change rate threshold Compare and output physical violation inspection marks. : in, and Adjacent time steps and Reconstructing the three-dimensional temperature field; The characterization determines the presence of non-physical jump artifacts.

[0013] The formulas for calculating the error attribution degree of the front-end ice buoy and the adaptive correction penalty multiplier for ice buoys deemed to have failed are as follows: in, The error attribution degree of the front-end ice buoy; This provides the set of all legitimate grid nodes that provide observational data for the ice buoy within the current standard analysis time and participate in the fusion process. For target grid nodes; For comprehensive integration and weighting; For physical violation inspection markings; The penalty multiplier is adaptively adjusted and used to update the control hyperparameters in reverse. This is a penalty amplification factor; The system's preset tolerance for tracing and assigning responsibility, when At that time, the front-end ice buoy was determined to be faulty.

[0014] The present invention also provides a compression system for heterogeneous measurement data from polar ice buoys, the system comprising: Acquisition module: Acquires heterogeneous raw data from ice buoys and performs preprocessing; extracts polar environmental state characteristics based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. Prediction module: Based on the characteristics of polar environment, dynamically reconstruct the spatiotemporal observation window of the prediction algorithm, perform adaptive spatiotemporal prediction on the preprocessed data, and extract the residual and prediction confidence deviation between the actual observation value and the final spatiotemporal prediction value. Encoding module: Based on the prediction confidence deviation, the quantization step size is dynamically calculated to quantize the residual, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. Golomb-Rice coding is performed, and the generated bit stream sequence, together with the polar environment state characteristics, is encapsulated into a data frame for uplink transmission. Update module: Receives reverse control signaling from the reconfiguration end and updates the control hyperparameters based on the parsed gear adjustment amount.

[0015] The present invention also provides a reconstruction system for heterogeneous measurement data from polar ice buoys, used to reconstruct data transmitted by the compression method described above, the system comprising: The restoration module receives data frames sent by the front-end ice buoy, extracts polar environment state features, decodes them to restore the observation values, performs unified spatiotemporal registration on the observation values, and generates unified weighted fusion feature values ​​based on multi-source weighted fusion of polar environment state features. Verification module: Based on the unified weighted fusion feature value, an initial analysis field is constructed. A physical guided neural network constrained by the ocean dynamics conservation equation is used to perform fine reconstruction of the three-dimensional data field. Based on the calculation of the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output a physical violation check label. Correction module: Combining the comprehensive fusion weights and physical violation inspection marks during multi-source weighted fusion, calculate the error attribution degree of the front-end ice buoy, calculate the adaptive correction penalty multiplier for the ice buoy that is determined to be failed, and generate the gear adjustment amount for the control hyperparameters accordingly, which is then encapsulated as a reverse control signal and sent to the corresponding ice buoy.

[0016] This invention provides a method and system for compressing and reconstructing heterogeneous measurement data from polar ice buoys. Based on the characteristics of the non-stationary polar marine environment, this invention constructs an end-cloud collaborative closed-loop system that integrates forward environmental status and backward error verification. This system effectively overcomes the physical bottlenecks of extremely low satellite communication bandwidth and limited node power consumption in polar regions, achieving efficient backhaul and accurate three-dimensional physical field reconstruction of heterogeneous multi-source polar marine data, and significantly improving the long-term adaptive operation capability of the polar marine monitoring network.

[0017] In the compression phase at the edge observation end, this invention, by sensing the dynamic characteristics of the ice buoy's habitat in real time, incorporates polar conditions such as ice sheet consolidation or cyclonic disturbances into the spatiotemporal prediction model and quantization encoding process. During periods of stable polar environment, the system proactively expands the spatiotemporal coordination range to maximize the elimination of data redundancy; during periods of dramatic change at the polar ice edge, the system automatically shrinks the prediction window and improves quantization fidelity, dynamically allocating encoding word lengths. This polar-environment-adaptive compression mechanism, while ensuring that key features such as the thermohaline are not lost, transforms multi-source observation data into an encoded stream with extremely low information entropy, effectively adapting to the transmission constraints of ultra-narrowband communication links for low-Earth orbit satellites in polar regions.

[0018] In the shore-based reconstruction phase, addressing the objective challenges of extremely sparse distribution of polar observation nodes and their susceptibility to harsh habitat interference, this invention introduces a polar scene perception trust assessment mechanism. The system can dynamically weight and fuse data based on the reliability of different types of sensors under specific polar physical conditions, effectively shielding transient noise caused by ice fragmentation or frost formation. To overcome the systematic error accumulation problem faced by long-term unattended buoy systems, by extracting the evolution rate of the reconstructed water field over time and cross-validating it with the physical limits of deep-water heat capacity in polar regions, the system can accurately trace and pinpoint the source buoy node where implicit data distortion occurs. Based on this, the shore-based end generates discretized control signaling with extremely low byte overhead, resetting the compressed control parameters of the corresponding buoy via satellite downlink. This feedback strategy, where global errors guide the adjustment of parameters for individual edge devices, enables the polar ice buoy array to achieve long-term self-correction and continuous self-evolution, ensuring the long-term robustness and high scientific fidelity of the overall heterogeneous observation data link in the extremely harsh and highly dynamic polar ocean monitoring scenario. Attached Figure Description

[0019] Figure 1This is a flowchart illustrating the ice buoy heterogeneous measurement data compression process of the present invention. Figure 2 This is a flowchart illustrating the reconstruction of heterogeneous measurement data from the ice buoy in this invention. Figure 3 This is a distribution diagram of heterogeneous data acquisition in this invention; Figure 4 This is an adjustment diagram of the spatiotemporal prediction window based on environmental state characteristics in this invention; Figure 5 The system error convergence timing diagram of the reverse correction process provided by this invention shows the change in error attribution degree of the ice buoy before and after receiving the reverse control signal. Detailed Implementation

[0020] This embodiment provides a system for compressing and reconstructing heterogeneous measurement data from ice buoys. This system serves as the hardware and infrastructure for implementing the data compression and reconstruction method of this invention. The system is deployed in a polar marine monitoring application scenario and specifically includes edge observation terminals distributed in polar sea areas, low-Earth orbit satellite communication links in polar regions, and shore-based data center reconstruction terminals located inland.

[0021] In this embodiment, the polar sea area is divided into solidified ice zones, sea ice edge zones, and open water zones. The edge observation terminals consist of a sparsely grid-like array of ice buoys, which includes measurement nodes of various sizes and deployment methods. Preferably, the ice buoy array includes at least subglacial moored buoys (ITPs), surface ice weather stations (POPSs), and drifting unmanned ice stations. Due to the differences in their specific polar environments, these different types of observation nodes exhibit significant heterogeneity in their observation focus and data characteristics.

[0022] In the specific device limitations of the edge observation terminal (ice buoy node), a single ice buoy node includes a sensor array module, an edge computing control module, a satellite communication module, and a polar power supply module.

[0023] The sensor array module is used to collect multi-dimensional heterogeneous data of the polar environment. Preferably, for sub-ice moored buoys, the sensor array mounted on them includes temperature, salinity, and depth meters (CTDs) distributed along different depth profiles to collect seawater temperature, salinity, and depth parameters, and is equipped with a fluorometer to collect sub-ice chlorophyll concentration; for ice surface meteorological observation stations, the sensors they carry include anemometers, barometers, and ice surface temperature sensors. In this embodiment, heterogeneous measurement data are manifested in the differences in sampling frequency, data accuracy, and dimensions of different sensors. For example, the sampling frequency of the CTD sensor is preferably set to collect vertical profile data once every 12 hours, with a temperature measurement accuracy of ±0.01 degrees Celsius; while the sampling frequency of the anemometer is preferably set to collect data once every 6 hours, with a wind speed measurement accuracy of ±0.1 meters per second.

[0024] The edge computing control module is a device that performs front-end data quality control, environmental state perception, and compression algorithms. Due to the strict limitations on the overall power consumption of the buoy in polar environments, the edge computing control module preferably employs a low-power microcontroller or an embedded microprocessor based on an ARM architecture, and is equipped with non-volatile flash memory for short-term data caching. This module embeds a polar environment state perception unit, an adaptive prediction unit, and an adaptive quantization encoding unit. Because of the buoy's limited computing power, these units are configured at the hardware level to perform only lightweight integer operations and iterative operations with finite step sizes.

[0025] The polar power supply module is used to maintain the system operation of the ice buoy under polar night and extreme cold conditions. Preferably, the polar power supply module is composed of a low-temperature resistant high-energy disposable lithium battery pack and a freeze-resistant solar photovoltaic panel, and provides differentiated power to the edge computing control module and sensor module through a power management chip.

[0026] The satellite communication module is configured as a two-way communication terminal, preferably using an Iridium communication module or a BeiDou short message communication module, to achieve cross-regional data interaction between the edge observation end and the shore-based data center reconstruction end. The polar low-Earth orbit satellite communication link constructs the system's uplink and downlink transmission channels: the uplink channel is used to send the compressed and encoded data frames from the edge computing control module to the shore-based data center, and the transmission rate of the uplink channel is preferably limited to between 1kbps and 5kbps; the downlink channel is used to receive closed-loop reverse control commands issued by the shore-based data center.

[0027] In the specific device limitations of the shore-based data center reconstruction terminal, the terminal is configured as a server cluster with high-performance computing capabilities, specifically including a data receiving server, a heterogeneous computing server, and a spatiotemporal database system.

[0028] The data receiving server is connected to the satellite communication link via a satellite ground station. It is responsible for receiving compressed data frames from various ice buoy nodes in real time and performing frame parsing and decoding operations. The heterogeneous computing server is preferably configured with a multi-core CPU and a graphics processing unit array to support subsequent high-concurrency, high-computing-power reconstruction operations. The heterogeneous computing server internally contains multiple functional processing entities, including a spatiotemporal registration engine, a multi-source data fusion engine, a 3D ocean reconstruction engine, and a physical error verification unit.

[0029] The physical error verification unit is connected to the 3D ocean reconstruction engine and is used to extract time-series slices of the reconstruction results and perform verification based on the thermodynamic principles and fluid dynamic boundary conditions of the polar ocean. When this unit determines that the rate of change of the local reconstructed data field deviates from the physical conventional threshold, it triggers the instruction generation unit. The instruction generation unit is configured to calculate the correction parameters of the front-end prediction algorithm or compression quantization algorithm based on the deviation, and encapsulate the correction parameters into control signals through the data receiving server. These signals are then fed back to the edge computing control module of the corresponding numbered ice buoy node via the downlink channel, thereby resetting its compression and prediction parameters and realizing system-level closed-loop hardware linkage.

[0030] Next, based on the architecture of the ice buoy heterogeneous measurement data compression and reconstruction system, this embodiment will describe in detail the S1 performed at the edge observation end in the method of this invention: heterogeneous data preprocessing and polar environment state perception.

[0031] This step is deployed and runs within the edge computing control module inside the ice buoy. This module first acquires raw data from heterogeneous sources through the sensor array module, and then performs data cleaning tailored to the characteristics of the extreme polar environment, extracting dynamic environmental features to guide subsequent compression and encoding strategies.

[0032] Step S1: Heterogeneous Data Preprocessing and Polar Environment Status Awareness S11: Polar Anomaly Observation Detection and Temporary Completion Based on Dynamic Benchmarks: In polar marine environments, suspended CTDs are frequently affected by the adhesion of ice algae unique to polar regions or the scraping of ice debris. Meteorological observation stations on ice surfaces often experience transient sensor probe malfunctions due to surface frost, resulting in transient anomalous values ​​in the observation data stream that violate thermodynamic or kinetic principles. To prevent these anomalous values ​​from contaminating subsequent residual calculations and prediction models, the edge computing control module performs a dynamic boundary anomaly detection algorithm based on exponentially weighted moving averages on the input heterogeneous raw data.

[0033] Set discrete timestamp as ,in For time series indexing. Definition for The original values ​​of the observed parameters at that time. Calculated by the edge computing control module. Dynamic threshold benchmark at time and dispersion index The calculation formula is as follows: in, The exponentially weighted decay coefficient has a range of values. This is used to adjust the rate at which historical data is forgotten relative to the current baseline. Because the state of water beneath deep polar ice caps changes extremely slowly, this is particularly relevant for deep-water CTD data. Take the smaller value to maintain the long-term stability of the baseline; for ice surface meteorological data, A larger value is chosen to accommodate the rapid decay of wind speed.

[0034] When satisfied At that time, among them The current anomaly is determined by the preset anomaly judgment tolerance constant. Outliers caused by sudden disturbances in the polar environment are removed, and the time of the outlier is marked as missing; otherwise, if the outlier does not exceed the tolerance limit, the data is retained.

[0035] To ensure the data temporal continuity requirement of subsequent time series prediction algorithms, the edge computing control module performs unidirectional extrapolation linear interpolation based on previous valid historical observations to calculate temporary filler values ​​for outliers that have been removed and physically missing values ​​caused by satellite communication blind spots. in, The output after preprocessing Observe the effective value of the parameters at all times; and Distance The two most recent preceding historical moments (i.e.) and (Legitimate and valid observations. Generated through provisional completion.) It only resides in the memory of the edge computing control module to participate in timing prediction, is temporarily identified, and does not participate in subsequent quantization coding and satellite downlink transmission, thus eliminating the contamination of the shore-based reconstruction field by falsified data.

[0036] S12: Heterogeneous data normalization and alignment: After anomaly removal and completion, the effective value sequence of the output observation parameters is aligned in terms of units and time scale by the edge computing control module. Data from different sensors with different sampling frequencies are uniformly resampled and mapped onto a globally preset standard discrete-time grid, and different units of measurement are normalized to standard SI data through a linear mapping function to generate a standardized observation parameter sequence.

[0037] Step S13: Extraction of Polar Environment State Features: The polar environment exhibits highly non-stationary characteristics. When ice buoys are in areas where thick ice sheets have been consolidated for many years (i.e., during stable periods), the bottom water data they measure have extremely high temporal redundancy. However, when ice buoys drift with ocean currents to areas where ice edges melt, or encounter strong polar cyclones (i.e., during periods of dramatic change), the structure of the water and weather is disrupted, and the data exhibits high-frequency oscillations. This step constructs a measurement index by introducing the kinematic characteristics of the buoys and the temporal derivative characteristics of the data to achieve a perception of the current state of the polar environment.

[0038] The edge computing control module extracts data including the current time. Historical observation data, including those included, were used to construct a system with a length of [length missing]. A state-aware backtracking window. First, the temporal rate of change of the observed parameters is calculated. : In the formula, and The effective values ​​of the observed parameters at two adjacent time points within the state-aware backtracking window. and This is the corresponding timestamp. This characterizes the degree of drastic root mean square fluctuation of the observed parameter in the recent time period.

[0039] Meanwhile, the edge computing control module uses its built-in positioning receiver to obtain information from the edge observation terminal. drift speed at any moment In polar oceanography, a high buoy drift velocity means that it is crossing the boundary of a water mass with a strong front, which will cause spatial abrupt changes in observational data.

[0040] The edge computing control module calculates the state characteristics of the polar environment. : in, The maximum drift velocity of the polar preset baseline is set based on historical polar climate background data; The maximum rate of change of the polar preset baseline for the corresponding observation parameters. Both are used as normalization denominators to eliminate differences between different physical dimensions. and These represent the drift feature weights and the fluctuation feature weights, respectively, and satisfy the following conditions: .

[0041] Polar environmental characteristics It is a dimensionless constant whose range converges to . Between. The edge computing control module is based on the calculated... Determine the current habitat status: when Approaching At that time, the system confirmed that the polar ice cap was in a stable, consolidated state; when Approaching At that time, the system confirmed that the ice periphery was undergoing dramatic changes or cyclonic disturbances.

[0042] Next, this embodiment elaborates on the second step S2 performed by the edge observation end in the method of the present invention: adaptive spatiotemporal prediction and residual extraction controlled by environmental state characteristics. Based on the polar environmental state characteristics extracted in the previous step, this step calculates high-precision predicted values ​​and extracts residuals under limited computing power, transforming the original data into a residual sequence with extremely low information entropy.

[0043] S2: Adaptive spatiotemporal prediction and residual extraction controlled by environmental state characteristics: Polar ocean buoy data exhibits significant physical inertia and spatiotemporal correlation, but the extent of this correlation is not a static constant. In vast solidified ice areas, deep water masses are homogeneous, and the data remains highly stable over long periods and large spatial spans. However, in ice-edge retreat zones or at the edges of mesoscale eddies, water masses converge to form fronts, causing the spatiotemporal correlation of the data to decay rapidly. Therefore, the edge computing control module utilizes the polar environmental state characteristics output in step S1. This imposes restrictions on the boundary conditions of the prediction algorithm.

[0044] S21: Dynamic reconstruction of the spatiotemporal observation window for the prediction algorithm: The edge computing control module receives the current time. Polar environmental characteristics ( Based on this, the adaptive backtracking step size for the time dimension is dynamically calculated. And the adaptive search radius for spatial dimensions The calculation formula is as follows: in, and These are the preset upper and lower limits of the time backtracking step size, in units of the number of time nodes; and These are the preset upper and lower limits of the search radius in adjacent buoy spaces, in kilometers; This indicates the floor function.

[0045] when Approaching (When the polar environment is stable) and Approaching the maximum value, the edge computing control module optimizes within a broader spatiotemporal domain to maximize the extraction of data redundancy; when Approaching During dramatic changes in the polar environment, fronts increase spatial heterogeneity, and the edge computing control module forcibly shrinks the search window to... and The purpose is to eliminate weakly correlated remote interference nodes and ancient historical data to prevent the prediction model from diverging.

[0046] S22: Prediction based on lightweight adaptive Kalman-differential pulse coding (Kalman-DPCM): To address the limited computing power of edge computing control modules, this step improves the standard high-order Kalman filter by reducing its dimensionality, constructing a lightweight first-order trend-following prediction model. Definition The effective observation value at time is The evolution trend of the observed parameters is the state variable. The edge computing control module calculates the current time. Time dimension prediction value : when The actual observed value at time After it is generated, the edge computing control module needs to process the trend state quantity. Iterative updates are performed. Traditional fixed-gain systems are prone to hysteresis errors under drastic polar conditions. This application introduces a controlled... Adaptive Kalman gain equivalent coefficient : in, and These are the preset upper and lower bounds of the gain, respectively. In the polar periphery of dramatic change ( (larger) Automatically increasing in size, the model assigns higher confidence to the slope of the latest data, enabling the tracker to respond quickly to sudden changes in thermo-salinity transition layers or cyclone wind speeds; in the consolidation plateau region ( (smaller) As the size decreases, the model becomes more reliant on historical trends, thus acting as a smoothing filter.

[0047] S23: Spatial Dimensional Co-prediction Based on Distance Attenuation: Polar ice buoys typically rely on VHF or underwater acoustic communication to broadcast lightweight status data between nodes. The edge computing control module retrieves the current location within the adaptive search radius. The set of adjacent legal buoys within the range is denoted as . If the set is not empty, the edge computing control module extracts each float within the set. ( ) exist Effective observations broadcast at any time And calculate the spatial dimension co-predicted value. : Among them, spatial coordination weight Based on the spherical distance between buoys It exhibits a Gaussian decay distribution, consistent with the physical law that the correlation of ocean water masses decreases with distance: By adapting the search radius Directly introducing the variance term into the Gaussian kernel function, when the polar environment undergoes drastic changes ( (Smaller), the weight of buoys at slightly greater distances will be reduced exponentially, effectively cutting off data mismatches between water masses with different properties.

[0048] S24: Residual Calculation: The edge computing control module performs a weighted fusion of the temporal and spatial prediction results to generate... Final spatiotemporal prediction of time : in, This refers to the spatiotemporal fusion weighting coefficient. When encountering extreme abrupt disturbances such as ice fragmentation or strong upwelling, the continuity of the spatial physical field is disrupted before its temporal coherence. Therefore, The value of is also controlled by the characteristics of the polar environment. : when Rise to near hour, Approaching The system will automatically cut off unreliable spatial dimension collaborative predictions, completely degrading to single-node time dimension predictions, thus ensuring the robustness of the prediction system under extreme sea conditions. If adjacent buoy sets... If it is empty, then directly force an order. .

[0049] The edge computing control module extracts the valid observations at the current time. Compared with the final spatiotemporal prediction value The difference between them generates the residual. : The edge computing control module calculates the absolute value moving average of the residuals and outputs the prediction confidence bias of the current prediction result. : in, The smoothing coefficient represents the prediction confidence bias. Including residuals and environmental characteristics This will be output to the subsequent step S3. The extracted residuals Its numerical distribution is highly concentrated near zero, and its information entropy is greatly compressed, directly clearing the way for high compression ratio coding under polar bandwidth-limited conditions.

[0050] Next, this embodiment elaborates on the third step performed by the edge observation end in the method of the present invention: S3 encoding transmission.

[0051] Data transmission from polar ocean buoys is severely limited by narrowband communication via low-Earth orbit satellites in polar regions. Furthermore, edge-side microcontrollers lack the computational power to perform complex block matching and high-order arithmetic coding. Therefore, this step introduces the prediction confidence bias from the output of step S2. Polar environment state characteristics output in step S1 By implementing positive control over the quantization step size and Golomb-Rice encoding parameters, a high compression ratio with extremely low latency can be achieved.

[0052] S31: Adaptive dynamic quantization and nonnegativity mapping controlled by prediction confidence: In polar oceanographic monitoring, the minute temperature and salinity evolution of deep, stable waters beneath ice has extremely high scientific research value and requires extremely high quantization fidelity. However, during periods of dramatic change such as the passage of polar cyclones, the measurement itself is accompanied by extremely high turbulence noise. Appropriately reducing the quantization accuracy not only does not affect feature extraction, but can also effectively suppress the explosion of data flow.

[0053] Therefore, the edge computing control module bases its prediction confidence deviation on the preceding input. Dynamically calculate the current time. Adaptive quantization step size : in, and These are the lower limit of the quantization step size preset according to the physical dimensions of different sensors, i.e. the highest accuracy and the upper limit of the quantization step size, i.e. the lowest tolerance accuracy; The preset prediction confidence deviation tolerance limit for the system. The function is used to prevent Abnormal divergence caused the quantization step size to exceed the limit.

[0054] Quantization step size using dynamic calculation The edge computing control module processes the input floating-point residuals. Perform rounding quantization to generate integer quantization residuals. : in, This indicates a rounding operation to the nearest integer.

[0055] Since the entropy coding algorithm used subsequently is mainly optimized for the non-negative integer field, while the integer quantization residual... Theoretically, it exhibits a positive and negative symmetric Laplace distribution centered at zero. Therefore, the edge computing control module utilizes an interleaved mapping function to transform the positive and negative distributions... Convert to non-negative integer mapping value : Through the above interleaved mapping, residuals with small absolute values ​​(which have a very high probability of occurrence) are mapped to extremely small non-negative integers, providing the necessary conditions for subsequent allocation of shorter code word lengths.

[0056] S32: Delay-free adaptive Golomb-Rice coding incorporating polar environment characteristics: The standard Golomb-Rice encoding encodes non-negative integers by dividing them into quotients and remainders. Its core lies in determining the optimal exponent parameter to partition the data. Traditional methods require buffering long sequences of data and performing probabilistic exhaustive search, resulting in significant latency and computational overhead. This invention proposes a latency-free adaptive strategy that utilizes the characteristics of the polar environment for forward order determination.

[0057] Theoretically, the optimal Golomb-Rice parameters are logarithmically related to the expected value of the current data. The edge computing control module uses the quantified prediction confidence bias to estimate the expected value of the absolute value of the current residual. : However, when polar sea ice suddenly breaks apart or encounters a strong front, the prediction confidence bias increases. Due to the characteristics of moving average smoothing filtering, there may be a response lag. Step S1, however, directly extracts polar environment state features from drift velocity and first derivative. It possesses extremely strong advanced sensing capabilities. Therefore, the edge computing control module will... Introduced as a penalty term, the prior coding parameters are directly calculated in a forward manner. : in, The empirical probability coefficients, which are set based on the pre-statistical characteristics of the polar residual distribution, are preferably taken as approximate natural logarithms. The constant; This is the penalty coefficient for drastic environmental changes. When the polar environment is in a state of drastic change (…), When it approaches 1), the penalty term The significant increase forced the edge computing control module to proactively expand. This effectively avoids the risk of explosive growth in commercial value due to local forecast lag.

[0058] Obtaining prior encoding parameters Then, the edge computing control module maps the current non-negative integer values. Calculate the quotients separately remainder : The edge computing control module performs bitstream reconstruction: for the commerce Perform unary code encoding, that is, output continuous... Each binary 1 is appended to the end as a binary 0 as a truncation marker; the remainder is... Standard fixed-length binary encoding is performed, with its bit width strictly locked to 1000. The unary code of the quotient and the fixed-length binary code of the remainder are seamlessly concatenated to generate the highly compressed final bitstream sequence at the current moment. .

[0059] S33: Frame encapsulation and uplink transmission: After completing the parallel encoding of multi-channel heterogeneous sensor data, the edge computing control module encapsulates the generated bit stream sequence of parameters into frames.

[0060] The header of the encapsulated data frame is constructed as a fixed-length environment awareness header. This header includes the buoy's unique identifier (ID) and a timestamp. In addition, the edge computing control module will process the polar environment state characteristics calculated in the preceding steps. and the current quantization step size of each parameter Discretized gear mapping is performed, and the control bits are written into the information header in a minimally simplistic manner. The payload portion of the data frame then compactly concatenates the final bitstream sequence corresponding to multidimensional environmental parameters. .

[0061] A cyclic redundancy check (CRC) code is appended to the end of the data frame. The encapsulated data frame is pushed to the transmission queue of the satellite communication module. Based on the transit communication window time of the polar low-Earth orbit satellite system, it is transmitted in a low-bandwidth burst to the inland shore-based data center through the uplink channel, thereby completing the data compression and transmission workflow of the polar edge observation end.

[0062] Next, based on the architecture defined in the foregoing embodiments, this embodiment elaborates in detail the fourth step S4 executed at the inland shore-based data center reconstruction end after crossing the satellite space link in the method of the present invention: multi-source collaborative decoding and scene-aware weight fusion.

[0063] When compressed data frames transmitted from polar edge observation terminals reach shore-based data centers, the data exhibits high spatiotemporal dispersion due to intermittent satellite communication windows and node drift. Furthermore, the reliability of physical data varies greatly among sensors from different manufacturers and models in different polar environments.

[0064] S41: Synchronous Decoding and Residual Inverse Mapping of Shore-Based-Edge Models: The data receiving server captures uplink data frames in real time, first parsing the fixed-length environmental perception header to extract the unique identification code and discrete timestamp of the corresponding ice buoy. Characteristics of polar environment and the quantization step size of this parameter. .

[0065] To recreate the Golomb-Rice encoding, the shore-based server maintains a spatiotemporal prediction model locally that is completely consistent with the front-end, synchronously calculating estimates of the absolute value's mathematical expectation. The data receiving server then... Based on the prediction confidence bias maintained synchronously at the shore and base stations, the current prior coding parameters are calculated. .

[0066] The data receiving server extracts a bit stream from the data frame payload, extracts consecutive binary 1s until it encounters a 0 to count the number of unary codes, and obtains the decoder. Then read the subsequent... Convert the binary stream to decimal and obtain the decoding remainder. Therefore, the decoded non-negative integer mapping value is calculated. : To reverse the interleaving mapping, the data receiving server executes the inverse mapping function to obtain the decoded integer quantization residual. : Then, an inverse quantization operation is performed to obtain the decoded floating-point residual. : The data receiving server invokes the internally maintained digital twin of the prediction model to generate spatiotemporal prediction values ​​synchronously calculated on the shore-based side. Then, the decoded floating-point residuals are accumulated and restored to obtain... The buoy number at that time was High-fidelity observations : S42: Unified spatiotemporal registration of multi-source heterogeneous data: High-fidelity observations after restoration This is a randomly drifting Lagrange observation coordinate system. To meet the computational requirements of the subsequent 3D ocean reconstruction engine (based on an Eulerian mesh), the heterogeneous computing server invokes the spatiotemporal registration engine to establish a standard polar 3D Eulerian mesh, defining the target mesh node coordinates as follows: And set a unified standard analysis time step. .

[0067] For time-dimensional registration, the spatiotemporal registration engine selects a buoy. At the target time step Perform linear time interpolation on the two most recent observation times to convert the discrete timestamps. Observations on Projections are aligned to a unified time point to generate time-registered observations. .

[0068] S43: Polar Scene Perception Trust Assessment and Multi-Source Weighted Fusion At specific grid nodes In the vicinity, data may be transmitted from both subglacial moored buoys (ITPs) and surface weather stations (POPSs). Traditional assimilation methods assign weights based solely on the sensor's factory-specified error, neglecting the dynamic impact of the polar habitat on the sensors. For example, during periods of dramatic change in the polar environment (such as ice floes breaking apart), the anemometers at surface weather stations are highly susceptible to severe noise due to icing or vibration, while subglacial deep temperature, salinity, and depth (CTD) sensors are less affected by wave disturbances, and their data remain robust.

[0069] To quantify this impact, the multi-source data fusion engine introduces features based on polar environment conditions. (Attached to the buoy) Trust coefficient for polar scenes (in data frames) : In the formula, For different sensor physical types The corresponding polar environment sensitivity coefficient. For surface sensors that are susceptible to interference from severe weather conditions on ice surfaces. The preset value is relatively large; for deep, concealed sensors under ice, The preset value is a minimum value approaching 0. Therefore, when the environment experiences severe shocks ( When the temperature rises, the confidence level of surface sensors will be significantly suppressed by the mechanism.

[0070] Based on this, the multi-source data fusion engine calculates the buoy. For target mesh nodes Comprehensive integration weight : in, The nominal accuracy weight is equal to the sensor's factory error variance. The reciprocal of, that is ; The spatial distance mapping weights follow a principle based on the grid influence radius. Gaussian decay law: in, For buoys Actual 3D coordinates and target mesh nodes The Euclidean distance between them.

[0071] Multi-source data fusion engine retrieves data falling into grid nodes The set of all legal buoys within the radius of influence Calculate the node in Unified weighted fusion eigenvalues ​​at time points : This embodiment, based on the architecture defined in the foregoing embodiments, elaborates in detail the fifth step executed by the shore-based data center reconstruction end in the method of the present invention. Step S5: Three-dimensional hybrid reconstruction and backward spatiotemporal error verification: Due to the harsh polar environment, the damage rate of edge observation terminals (ice buoys) is high and their distribution is extremely uneven. The unified weighted fusion feature value output in step S4... Numerous spatial gaps and discontinuities still exist within the global 3D Eulerian mesh. Traditional statistical spatial interpolation cannot compensate for long-distance data gaps and is prone to generating staircase-like artifacts that violate the fluid continuity assumption. This step introduces a Physically Guided Neural Network (PINN) to use the polar ocean dynamics conservation equations as soft constraints to achieve fine reconstruction, and then performs source tracing verification on the reconstruction results.

[0072] S51: Construction of the initial analysis field based on multivariate empirical orthogonal functions (MEOF) and optimal interpolation (OI): To accelerate the convergence of the neural network, the 3D ocean reconstruction engine first constructs the current standard analysis time step. The initial analysis field is determined by reconstructing and extracting historical polar climate data. The MEOF algorithm is then used for dimensionality reduction to extract the preceding data containing large-scale polar circulation and temperature-salinity structures. One main spatial mode vector And by combining the current time-series evolution coefficients, a three-dimensional temperature field is reconstructed. Three-dimensional salinity field and three-dimensional flow velocity field Spatiotemporal background field matrix .

[0073] Reconstruct the uniformly weighted fused feature values ​​at the finite grid points output in step S4. As the observation input, the optimal interpolation (OI) assimilation algorithm is executed to calculate the gain and fuse the real observation information into the background field, generating an initial analysis field matrix covering the global grid. : in, For all legal node feature values The constructed observation column vector; A linear observation operator that maps the background space to the observation space; This is the optimal interpolation gain matrix derived based on the covariance matrix of the polar feature space. It has a preliminary polar flow field morphology, but its local details still have the risk of physical divergence in areas where data is missing.

[0074] S52: Fine reconstruction of three-dimensional data fields guided by physical conservation equations: 3D ocean reconstruction invokes an in-house deployed physical-guided neural network model The input to this model is grid spatiotemporal coordinates. The output is the reconstructed target physical quantity. Its network weight parameters are denoted as .

[0075] To address the non-stationary spatiotemporal evolution characteristics of the polar ocean, the reconstruction engine constructs a joint loss function that includes a data fidelity term and a dynamics penalty term. Optimize unsupervised network training: In the formula, These are the weighting coefficients.

[0076] Data fidelity loss item The network output is forced to approximate the initial analysis field constructed above. The high-confidence region in the middle, namely: Continuity (mass conservation) penalty term Based on the incompressible fluid equations under the Boussinesq approximation, the continuity of the reconstructed velocity field is constrained under complex polar terrain: Thermodynamics and Salinity Conservation Penalty Terms The formation and melting of polar sea ice are highly dependent on the thermo-salinity convection of water bodies. The reconstruction engine incorporates the polar thermodynamic energy equation and salinity diffusion equation to constrain the spatiotemporal gradient of the output field. in, For time partial derivative operators; and These are the three-dimensional spatial gradient and the Laplacian operator, respectively. and These are the preset thermal diffusivity and salt diffusivity of polar seawater, respectively.

[0077] Minimize using the backpropagation algorithm Update network weights The physical conservation mechanism acts as a high-dimensional interpolation smoother. After network convergence, the output is a continuous three-dimensional polar ocean physical field that strictly follows fluid dynamics and polar thermodynamics.

[0078] S53: Backward Spatiotemporal Error Verification Mechanism Based on Polar Thermodynamics When the front-end edge observation terminals (S2 and S3) face sensor vibration interference caused by strong polar cyclones, resulting in the failure of their dynamic environmental state characteristic perception or excessive quantization step size truncation causing residual distortion, this distortion can insidiously penetrate the satellite link and enter the reconstruction terminal. To address this, the heterogeneous computing server calls the physical error verification unit to perform backward verification based on the inherent physical principles of the polar region.

[0079] The deep seawater at the bottom of the ice sheet possesses extremely high specific heat capacity and a massive volume of water, making a sudden temperature change in this region highly unlikely within a very short timeframe. The physical error verification unit calculates the three-dimensional temperature field for the reconstructed time slices. Short-time evolution rate at deep-water grid points For all that satisfy : The testing unit compares the evolution rate with the set threshold for extreme variation of polar deep-water heat capacity. Compare and output a violation inspection label. : When the non-compliance inspection mark is within the target grid area When high-density clustering, i.e., large-area non-physical jump artifacts, is observed, the error verification unit can determine that the source data transmitted back by relevant ice buoys within the coverage area in the current time window contains systematic distortion. This violation determination result is not only used to force additional filtering and smoothing operations at the shore-based end to repair the local reconstruction field, but is also preserved as core diagnostic evidence. The backward spatiotemporal error verification mechanism enables tracing the source of the reconstruction result deviation to the compression quality of the front-end data.

[0080] It should be noted that the multivariate empirical orthogonal function (MEOF), optimal interpolation (OI) algorithm, and the basic topology and mathematical equation derivation of the Physically Guided Neural Network (PINN) involved in this embodiment are all existing technologies that have been disclosed and are highly mature in this field, and will not be repeated in the specification. Specifically, those skilled in the art can directly obtain and implement the above-mentioned basic algorithms based on the following classic academic literature: Regarding the Physically Guided Neural Network (PINN), those skilled in the art can directly refer to the literature "Physics-informed neural networks: Adeep learning framework for solving forward and inverse problems involving nonlinear partial differential equations" (authors: Raissi, M., Perdikaris, P., & Karniadakis, GE, published in the Journal of Computational Physics, 2019, Vol. 378, pp. 686-707). Regarding optimal interpolation (OI) assimilation algorithms, those skilled in the art can refer to the classic ocean data assimilation literature, "Improved global sea surface temperature analyses using optimum interpolation" (authors: Reynolds, RW, & Smith, TM, published in the authoritative climatology journal *Journal of Climate*, 1994, Vol. 7, No. 6, pp. 929-948). This literature details the complete derivation process of obtaining the optimal interpolation gain matrix based on the background field and spatial covariance matrix, and it is widely used in the oceanographic community. Regarding multivariable empirical orthogonal functions (MEOF), those skilled in the art can refer to the literature, "Empirical orthogonal functions and related techniques in atmospheric science: A review" (authors: Hannachi, A. et al., published in *International Journal of Climatology*, 2007, Vol. 27, No. 9, pp. 1119-1152). This literature utilizes MEOF to process multidimensional ocean and atmospheric variables, extracting a standardized mathematical process for large-scale spatial mode matrices and temporal evolution coefficients.

[0081] Next, based on the defined system architecture, this embodiment elaborates in detail the sixth step S6 executed by the shore-based data center reconstruction end in the method of the present invention: reverse parameter correction based on error verification.

[0082] Polar ice buoys are often unattended, and if their internal prediction or quantification models experience parameter drift, it will lead to a continuous stream of distorted data being transmitted back. This step utilizes physical verification results from a global 3D grid to locate the source buoy causing the distortion and implement targeted intervention.

[0083] S61: Buoy Error Source Tracing and Attribution Calculation Based on Fusion Weights: In S5, the error inspection unit outputs the violation inspection flag on the target mesh node. However, the reconstruction distortion of a grid node may be caused by multiple buoys falling within the influence radius of that grid. To achieve accountability, the instruction generation unit must eliminate interference from high-fidelity buoys and pinpoint the anomalous buoy that actually caused the violation.

[0084] The instruction generation unit extracts the buoy calculated in step S4. For grid nodes Comprehensive integration weight .definition For buoys At the current standard analysis time step The internal set of all legitimate grid nodes that provide observation data and participate in the fusion. The instruction generation unit calculates the buoy. Error attribution : If the buoy Contribution weight to a grid node Extremely high, and the node also triggered a violation ( If so, the buoy will bear a great responsibility. The range of values ​​is The larger the value, the stronger the buoy. The more compressed data is transmitted, the greater the suspicion that it causes non-physical jumps in the shore-based reconstruction field.

[0085] The instruction generation unit will calculate the result. Compared with the system's preset tolerance for tracing and holding accountable Compare them. When At that time, the system officially determined the buoy. The edge-end spatiotemporal prediction model or quantization coding model can no longer adapt to the current polar habitat, triggering an adaptive parameter correction mechanism.

[0086] S62: Adaptive correction of penalty multiplier calculation and front-end parameter update: For buoys that have been deemed invalid The instruction generation unit calculates an adaptive correction penalty multiplier based on the degree to which the error exceeds the limit. : in, This is a penalty amplification factor. As a dynamic multiplier greater than 1, it is used to perform reverse update calculations of the control hyperparameters in S2 and S3.

[0087] When data distortion leads to thermodynamic changes, it means that the front-end prediction model is lagging in tracking abrupt polar environmental changes, or that erroneous remote space-based collaborative data has been introduced. The instruction generation unit forcibly raises the upper limit of the adaptive Kalman gain equivalent coefficient in the first-order trend-following prediction model. And shrink the upper limit of the search radius in the space of adjacent buoys. : After correction, the buoy It will have a more sensitive ability to capture temporal mutations, while reducing reliance on unreliable water mass data in the surrounding area and preventing the prediction model from diverging.

[0088] Another major source of error is the clipping of high-frequency features caused by excessively large dynamic quantization step sizes. The instruction generation unit forcibly reduces the upper limit of the quantization step size. At the same time, increase the environmental change penalty coefficient in Golomb-Rice coding. : After correction, the buoy Even in polar environment characteristics Even during periods of violent oscillation, higher precision must be maintained, and this is achieved by increasing... Allocate a wider coding word length in advance to prevent the loss of physical information caused by brute-force truncation.

[0089] S63: Simplified Control Signaling Encapsulation and Polar Satellite Downlink Transmission The downlink bandwidth of polar low-Earth orbit satellite systems is typically only a few hundred bps, and the downlink communication window is extremely short. Therefore, shore-based data centers cannot process the calculated floating-point parameters. , , , Send the entire amount directly to the ice buoy.

[0090] To achieve simplified communication, the instruction generation unit maps the difference between the old and new parameters to a preset clutch advance gear. The increment is divided into... The system has five integer logic levels, with each parameter adjustment requiring only 3 bits. The instruction generation unit compactly concatenates the float's unique identifier (ID), the check bit, and the level adjustments of the aforementioned four parameters into a reverse control signaling string less than 4 bytes in length.

[0091] The data receiving server pushes this extremely lightweight control signaling to the ground station's launch queue. This is crucial when the low-Earth orbit satellite passes over buoys operating in polar night or rough sea conditions. At that time, the buoy's satellite communication module receives the signal. Its internal edge computing control module parses the gear adjustment amount and updates the corresponding hyperparameters in the flash memory.

[0092] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0093] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for compressing heterogeneous measurement data from polar ice buoys, characterized in that, The method includes: Obtain heterogeneous raw data from ice buoys and preprocess them; extract polar environmental state characteristics based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. Based on the characteristics of polar environment, the spatiotemporal observation window of the prediction algorithm is dynamically reconstructed. Adaptive spatiotemporal prediction is performed on the preprocessed data, and the residual and prediction confidence deviation between the actual observation value and the final spatiotemporal prediction value are extracted. The residual is quantized by dynamically calculating the quantization step size based on the prediction confidence deviation, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. Golomb-Rice coding is performed, and the generated bit stream sequence, together with the polar environment state characteristics, is encapsulated into a data frame for uplink transmission. Specifically, the residual is quantized by dynamically calculating the quantization step size based on the prediction confidence bias, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. The specific calculation formula is as follows: ; ; ; in, For adaptive quantization step size; To predict confidence bias; and These are the preset lower and upper limits for the quantization step size, respectively; This is the preset prediction confidence level deviation tolerance limit; This indicates the operation of taking the minimum value; These are prior coding parameters; The empirical probability coefficients are set; Let $\mathbf{a}$ be the expected value of the absolute value of the current residual. The penalty coefficient for drastic environmental changes; Characteristics of polar environment; Receive the reverse control signaling sent by the reconfiguration end, and update the control hyperparameters according to the parsed gear adjustment amount.

2. The method according to claim 1, characterized in that, Polar environmental state characteristics are extracted based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. The specific calculation formula is as follows: ; in, Characteristics of polar environment; The drift speed at the current moment; The maximum drift velocity of the polar preset baseline is set based on historical polar climate background data; The time-series rate of change of the observed parameters; The maximum rate of change of the polar preset baseline for the corresponding observation parameters; and These represent the drift feature weights and the fluctuation feature weights, respectively, and satisfy the following conditions: .

3. The method according to claim 1, characterized in that, The spatiotemporal observation window of the prediction algorithm is dynamically reconstructed based on the characteristics of the polar environment, specifically including calculating the adaptive backtracking step size for the time dimension. And the adaptive search radius for spatial dimensions : ; ; in, and These are the preset upper and lower limits of the time backtracking step size; and These are the preset upper and lower limits of the search radius for adjacent buoy spaces; This indicates a round-down operation; This describes the characteristics of the polar environment.

4. A method for reconstructing heterogeneous measurement data from polar ice buoys, characterized in that, The method for reconstructing data transmitted by the compression method of claim 1 includes: The system receives data frames sent by the front-end ice buoy, extracts polar environment state features, decodes them to reconstruct the observation values, performs unified spatiotemporal registration on the observation values, and generates unified weighted fusion feature values ​​based on multi-source weighted fusion of polar environment state features. An initial analysis field is constructed based on unified weighted fusion eigenvalues. A physical-guided neural network constrained by the ocean dynamics conservation equation is used to perform fine reconstruction of the three-dimensional data field. Based on the calculation of the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output a physical violation check identifier. By combining the comprehensive fusion weights and physical violation inspection indicators during multi-source weighted fusion, the error attribution degree of the front-end ice buoy is calculated. For ice buoys that are determined to be faulty, an adaptive correction penalty multiplier is calculated. Based on this, the gear adjustment amount for the control hyperparameter is generated and encapsulated as a reverse control signal and sent to the corresponding ice buoy.

5. The method according to claim 4, characterized in that, Based on the multi-source weighted fusion of polar environmental state characteristics, the formula for calculating the comprehensive fusion weight is as follows: ; ; in, Trust coefficient for polar scenarios; These are the polar environment state features attached to the corresponding ice buoy data frames. The polar environment sensitivity coefficients corresponding to different sensor physical types; For comprehensive integration and weighting; Spatial distance mapping weights; This is the nominal precision weight.

6. The method according to claim 4, characterized in that, Based on the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output a physical violation check identifier. Specifically, this includes extracting time slices and calculating the short-time evolution rate at deep-water grid points. : ; Short-time evolution rate and polar deep-water thermal capacity extreme change rate threshold Compare and output physical violation inspection marks. : ; in, and Adjacent time steps and The reconstructed three-dimensional temperature field; The characterization determines the presence of non-physical jump artifacts.

7. The method according to claim 4, characterized in that, The formulas for calculating the error attribution degree of the front-end ice buoy and the adaptive correction penalty multiplier for ice buoys deemed to have failed are as follows: ; ; in, The error attribution degree of the front-end ice buoy; This provides the set of all legitimate grid nodes that provide observational data for the ice buoy within the current standard analysis time and participate in the fusion process. For target grid nodes; For comprehensive integration and weighting; For physical violation inspection markings; The penalty multiplier is adaptively adjusted for inverse updates of control hyperparameters; This is a penalty amplification factor; The system's preset tolerance for tracing and assigning responsibility, when At that time, the front-end ice buoy was determined to be faulty.

8. A compression system for heterogeneous measurement data from polar ice buoys, characterized in that, The system includes: Acquisition module: Acquires heterogeneous raw data from ice buoys and performs preprocessing; extracts polar environmental state characteristics based on the temporal rate of change of observed parameters and the drift velocity of ice buoys. Prediction module: Based on the characteristics of polar environment, dynamically reconstruct the spatiotemporal observation window of the prediction algorithm, perform adaptive spatiotemporal prediction on the preprocessed data, and extract the residual and prediction confidence deviation between the actual observation value and the final spatiotemporal prediction value. Encoding module: Based on the prediction confidence deviation, the quantization step size is dynamically calculated to quantize the residual, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. Golomb-Rice coding is performed, and the generated bit stream sequence, together with the polar environment state characteristics, is encapsulated into a data frame for uplink transmission. Specifically, the residual is quantized by dynamically calculating the quantization step size based on the prediction confidence bias, and the polar environment state characteristics are used as a penalty term to calculate the prior coding parameters. The specific calculation formula is as follows: ; ; ; in, For adaptive quantization step size; To predict confidence bias; and These are the preset lower and upper limits for the quantization step size, respectively; This is the preset prediction confidence level deviation tolerance limit; This indicates the operation of taking the minimum value; These are prior coding parameters; The empirical probability coefficients are set; Let $\mathbf{a}$ be the expected value of the absolute value of the current residual. The penalty coefficient for drastic environmental changes; Characteristics of polar environment; Update module: Receives reverse control signaling from the reconfiguration end and updates the control hyperparameters based on the parsed gear adjustment amount.

9. A reconstruction system for heterogeneous measurement data from polar ice buoys, characterized in that, The system is used to reconstruct data transmitted by the compression method of claim 1, the system comprising: The restoration module receives data frames sent by the front-end ice buoy, extracts polar environment state features, decodes them to restore the observation values, performs unified spatiotemporal registration on the observation values, and generates unified weighted fusion feature values ​​based on multi-source weighted fusion of polar environment state features. Verification module: Based on the unified weighted fusion feature value, an initial analysis field is constructed. A physical guided neural network constrained by the ocean dynamics conservation equation is used to perform fine reconstruction of the three-dimensional data field. Based on the calculation of the short-time evolution rate of the reconstructed data field, a backward spatiotemporal error check is performed to output a physical violation check label. Correction module: Combining the comprehensive fusion weights and physical violation inspection marks during multi-source weighted fusion, calculate the error attribution degree of the front-end ice buoy, calculate the adaptive correction penalty multiplier for the ice buoy that is determined to be failed, and generate the gear adjustment amount for the control hyperparameters accordingly, which is then encapsulated as a reverse control signal and sent to the corresponding ice buoy.

Images