Real-time data integration method and system based on ground observation data

Abstract

The invention relates to the technical field of data processing, and discloses a real-time data integration method and system based on ground observation data. The method comprises the following steps: performing multi-dimensional preprocessing on original marine observation data to obtain time-space standardized observation data; performing multi-parameter cross comparison analysis on the time-space standardized observation data to obtain stratification deviation characteristic data; constructing an adaptive multi-factor correction rule based on the layered deviation characteristic data to obtain a dynamic correction parameter; based on the dynamic correction parameters, performing nonlinear correction on the real-time temperature profile data, and performing intelligent fusion with a preset reference data set to obtain preliminary integrated observation data; performing multi-level quality control and probability uncertainty evaluation on the preliminarily integrated observation data to obtain refined observation data with a credibility index; and performing closed-loop feedback processing on the refined observation data to obtain target integration correction observation data. According to the invention, the efficiency and precision of real-time data integration based on ground observation data are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a real-time data integration method and system based on ground observation data. Background Art

[0002] Real-time integration of ocean observation data is of great significance for marine scientific research and marine environmental monitoring. Traditional ocean data integration methods mainly rely on simple statistical models and linear interpolation techniques, which are difficult to effectively handle complex marine environmental changes and multi-source heterogeneous data. In recent years, with the development of machine learning and artificial intelligence technology, some new data fusion methods have been proposed, such as multi-source data fusion based on neural networks, state estimation based on particle filters, etc., which have improved the integration accuracy of ocean observation data to a certain extent.

[0003] However, existing technologies still have some shortcomings. First, most methods lack sufficient consideration of the spatiotemporal multi-scale characteristics of ocean data, making it difficult to capture the complex dynamic changes in the marine environment. Second, in dealing with outliers and uncertainties, existing methods often use simple threshold screening or statistical models, which cannot accurately quantify and process systematic deviations and random errors in multi-source data. In addition, existing data integration methods are usually static, lacking real-time feedback and adaptive optimization mechanisms for the integration process, and are difficult to adapt to the rapid changes in the marine environment and the dynamic characteristics of the observation system. Summary of the invention

[0004] The present application provides a real-time data integration method and system based on ground observation data, which are used to improve the efficiency and accuracy of real-time data integration based on ground observation data.

[0005] In a first aspect, the present application provides a real-time data integration method based on ground observation data, the real-time data integration method based on ground observation data comprising: performing multi-dimensional preprocessing on pre-collected original ocean observation data to obtain spatiotemporal standardized observation data;

[0006] Performing multi-parameter cross-comparison analysis on the spatiotemporal standardized observation data to obtain stratified deviation characteristic data;

[0007] Constructing an adaptive multi-factor correction rule based on the stratified deviation characteristic data to obtain a dynamic correction parameter;

[0008] Based on the dynamic correction parameters, nonlinear correction is performed on the real-time temperature profile data and intelligently integrated with the preset reference data set to obtain preliminary integrated observation data;

[0009] Performing multi-level quality control and probabilistic uncertainty assessment on the preliminary integrated observation data to obtain refined observation data with a credibility index;

[0010] The refined observation data is subjected to closed-loop feedback processing to obtain target integrated corrected observation data.

[0011] In combination with the first aspect, in a first implementation of the first aspect of the present application, the pre-collected original ocean observation data is subjected to multi-dimensional preprocessing to obtain spatiotemporal standardized observation data, including:

[0012] Reconstructing the original ocean observation data in time series to obtain dynamic time-standardized data;

[0013] Performing non-uniform grid spatial interpolation on the dynamic time normalized data to obtain multi-resolution spatial distribution data;

[0014] Performing outlier identification on the multi-resolution spatial distribution data to obtain robust cleaned data;

[0015] Performing multi-scale noise separation on the robust cleaned data to obtain fine noise reduction data;

[0016] Performing deep reconstruction on the refined noise reduction data to obtain high-precision depth data;

[0017] Performing dynamic temperature response characteristic compensation on the high-precision depth data to obtain temperature correction data;

[0018] Performing salinity inversion and correction on the temperature-corrected data to obtain high-reliability salinity data;

[0019] Performing layered density gradient analysis on the high-confidence salinity data to obtain layered density structure data;

[0020] Performing adaptive modal decomposition on the layered density structure data to obtain multi-scale characteristic components;

[0021] The multi-scale feature components are reconstructed in a spatiotemporal consistency to obtain the spatiotemporal standardized observation data.

[0022] In combination with the first aspect, in a second implementation of the first aspect of the present application, the multi-parameter cross-comparison analysis is performed on the spatiotemporal standardized observation data to obtain the layered deviation feature data, including:

[0023] Performing dynamic depth registration on the spatiotemporal standardized observation data to obtain depth alignment data;

[0024] Performing multi-scale temperature gradient analysis on the depth alignment data to obtain temperature change characteristic data;

[0025] Adaptively segmenting the temperature change characteristic data into layers to obtain multi-layer structure characteristic data;

[0026] Performing modal extraction on the multi-layer structure feature data by using a variational modal decomposition algorithm to obtain inter-layer modal features;

[0027] Performing nonlinear correlation analysis on the inter-layer modal characteristics to obtain parameter coupling relationship data;

[0028] Performing uncertainty quantitative analysis on the parameter coupling relationship data to obtain a probability deviation distribution;

[0029] Multi-dimensional clustering is performed on the probability deviation distribution to obtain the layered deviation feature data.

[0030] In combination with the first aspect, in a third implementation of the first aspect of the present application, constructing an adaptive multi-factor correction rule based on the stratified deviation feature data to obtain a dynamic correction parameter includes:

[0031] Performing principal component analysis on the stratified deviation characteristic data to obtain key influencing factors;

[0032] Conduct nonlinear sensitivity analysis on the key influencing factors to obtain factor weight distribution;

[0033] Generating rules for the factor weight distribution to obtain an initial correction rule set;

[0034] Cross-validating the initial correction rule set to obtain rule effectiveness evaluation results;

[0035] Performing genetic algorithm optimization based on the rule effectiveness evaluation results to obtain an optimized rule set;

[0036] Fine-tuning parameters of the optimization rule set to obtain refined correction rules;

[0037] Performing spatiotemporal hierarchical mapping on the refined correction rules to obtain a multi-level correction matrix;

[0038] The dynamic correction parameters are obtained by solving the dynamic parameters of the multi-level correction matrix through a particle swarm optimization algorithm.

[0039] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, based on the dynamic correction parameter, the real-time temperature profile data is nonlinearly corrected and intelligently integrated with a preset reference data set to obtain preliminary integrated observation data, including:

[0040] Performing wavelet multi-resolution analysis on the real-time temperature profile data to obtain multi-scale temperature characteristics;

[0041] Performing nonlinear mapping on the multi-scale temperature features to obtain preliminary corrected temperature data;

[0042] Performing adaptive depth registration on the preliminary corrected temperature data to obtain depth calibrated temperature data;

[0043] Performing local regression smoothing on the depth calibration temperature data to obtain denoised temperature data;

[0044] Performing dynamic time warping on the denoised temperature data and the preset reference data set to obtain a time-aligned data set;

[0045] Performing multi-source data fusion on the time-aligned data set to obtain a fused temperature field;

[0046] Performing layered feature extraction on the fused temperature field to obtain a layered temperature structure;

[0047] The layered temperature structure is reconstructed in time and space by tensor decomposition technology to obtain a high-dimensional temperature field;

[0048] The high-dimensional temperature field is subjected to boundary processing to obtain the preliminary integrated observation data.

[0049] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, the performing of multi-level quality control and probabilistic uncertainty assessment on the preliminary integrated observation data to obtain refined observation data with a credibility index includes:

[0050] Performing multi-dimensional outlier detection on the preliminary integrated observation data to obtain abnormal labeled data;

[0051] Performing adaptive threshold segmentation on the abnormal labeled data to obtain graded abnormal data;

[0052] Performing spatiotemporal consistency analysis on the graded abnormal data to obtain abnormal pattern characteristics;

[0053] Classifying the abnormal pattern features by a support vector machine to obtain abnormal type distribution;

[0054] Performing a probability analysis on the abnormal type distribution based on a Bayesian network to obtain an abnormal cause probability map;

[0055] Performing Monte Carlo simulation based on the abnormal cause probability map to obtain an uncertainty distribution estimate;

[0056] Performing fuzzy comprehensive evaluation and analysis on the uncertainty distribution estimation to obtain multi-level quality indicators;

[0057] Performing hierarchical analysis on the multi-level quality indicators to obtain weighted quality scores;

[0058] The weighted quality score is normalized by information entropy theory to obtain a standardized credibility index;

[0059] The preliminary integrated observation data and the standardized credibility index are associated and mapped to obtain the refined observation data with the credibility index.

[0060] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, performing closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data includes:

[0061] Performing spatiotemporal hierarchical analysis on the refined observation data to obtain a multi-scale feature sequence;

[0062] Performing wavelet packet transform on the multi-scale feature sequence to obtain frequency domain feature distribution;

[0063] Performing adaptive threshold segmentation on the frequency domain feature distribution to obtain key frequency band information;

[0064] Performing modal extraction on the key frequency band information through empirical mode decomposition to obtain an intrinsic mode function set;

[0065] Performing Hilbert spectrum analysis on the eigenmode function set to obtain instantaneous frequency characteristics;

[0066] Constructing a nonlinear state space model based on the instantaneous frequency characteristics to obtain a dynamic system representation;

[0067] Performing Kalman filtering on the dynamic system representation to obtain a state estimation sequence;

[0068] Performing covariance analysis on the state estimation sequence to obtain an error correction matrix;

[0069] Optimizing the parameters of the error correction matrix by adaptive particle filtering to obtain an optimized correction factor;

[0070] Feedback correction is performed on the refined observation data to obtain the target integrated corrected observation data.

[0071] In a second aspect, the present application provides a real-time data integration system based on ground observation data, the real-time data integration system based on ground observation data comprising:

[0072] A processing module is used to perform multi-dimensional preprocessing on the pre-collected raw ocean observation data to obtain spatiotemporal standardized observation data;

[0073] A comparison module, used for performing multi-parameter cross comparison analysis on the spatiotemporal standardized observation data to obtain layered deviation characteristic data;

[0074] A construction module, used to construct an adaptive multi-factor correction rule based on the stratified deviation characteristic data to obtain a dynamic correction parameter;

[0075] A fusion module, used to perform nonlinear correction on the real-time temperature profile data based on the dynamic correction parameters and intelligently fuse it with a preset reference data set to obtain preliminary integrated observation data;

[0076] An evaluation module, used for performing multi-level quality control and probabilistic uncertainty evaluation on the preliminary integrated observation data to obtain refined observation data with a credibility index;

[0077] The feedback module is used to perform closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data.

[0078] In the technical solution provided by this application, by multi-dimensional preprocessing of the original ocean observation data, spatiotemporal standardization is achieved, the data quality and consistency are effectively improved, and a solid foundation is laid for subsequent analysis. The introduction of multi-parameter cross-comparison analysis enables the layered deviation characteristics to be accurately captured, greatly improving the accuracy of data integration. The construction of adaptive multi-factor correction rules not only takes into account the complexity of the marine environment, but also realizes the real-time update of dynamic correction parameters, significantly enhancing the adaptability and robustness of the method. In the nonlinear correction and intelligent fusion stage, this method makes full use of advanced signal processing technology and data fusion algorithms, such as wavelet multi-resolution analysis, tensor decomposition, etc., to achieve effective extraction and integration of multi-scale features, greatly improving the accuracy and representativeness of temperature profile data. The introduction of multi-level quality control and probabilistic uncertainty assessment has fully guaranteed the reliability of the data. By introducing advanced algorithms such as support vector machines and Bayesian networks, refined processing of anomaly detection and uncertainty quantification is achieved, providing data users with more reliable observation results. The design of the closed-loop feedback processing mechanism, especially the combination of advanced time-frequency analysis methods such as empirical mode decomposition and Hilbert spectrum analysis, as well as state estimation techniques such as Kalman filtering and adaptive particle filtering, has significantly improved the real-time and accuracy of data integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0080] Figure 1 A schematic diagram of an embodiment of a real-time data integration method based on ground observation data in an embodiment of the present application;

[0081] Figure 2 Schematic diagram of an embodiment of a real-time data integration system based on ground observation data in an embodiment of the present application. DETAILED DESCRIPTION

[0082] Embodiments of the present application provide a kind of real-time data integration method and system based on ground observation data.Term " first ", " second ", " third ", " fourth " etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or precedence.It should be understood that the data used in this way can be interchangeable in appropriate circumstances, so that the embodiment described here can be implemented in a sequence other than the content illustrated or described here.In addition, the term "comprises" or "has" and any variation thereof, is intended to cover non-exclusive inclusion, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0083] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the real-time data integration method based on ground observation data includes:

[0084] Step S101, performing multi-dimensional preprocessing on the pre-collected original ocean observation data to obtain spatiotemporal standardized observation data;

[0085] Step S102, performing multi-parameter cross-comparison analysis on the spatiotemporal standardized observation data to obtain stratified deviation characteristic data;

[0086] Step S103, constructing an adaptive multi-factor correction rule based on the stratified deviation feature data to obtain a dynamic correction parameter;

[0087] Step S104: Based on the dynamic correction parameters, nonlinear correction is performed on the real-time temperature profile data and intelligently integrated with the preset reference data set to obtain preliminary integrated observation data;

[0088] Step S105, performing multi-level quality control and probabilistic uncertainty assessment on the preliminary integrated observation data to obtain refined observation data with a credibility index;

[0089] Step S106: performing closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data.

[0090] It is understandable that the execution subject of the present application may be a real-time data integration system based on ground observation data, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0091] Specifically, the pre-collected raw ocean observation data is preprocessed in multiple dimensions to obtain spatiotemporal standardized observation data. This process includes time series reconstruction, non-uniform grid spatial interpolation, outlier identification, multi-scale noise separation, depth reconstruction, temperature response characteristic compensation, salinity inversion and correction, layered density gradient analysis, adaptive modal decomposition, and spatiotemporal consistency reconstruction. For example, for time series reconstruction, adaptive time window technology is used to dynamically adjust the window size according to the changing characteristics of the data, so as to more accurately capture the temporal variation characteristics of ocean parameters. Multi-parameter cross-comparison analysis is performed on the spatiotemporal standardized observation data to obtain stratified deviation feature data. This step involves dynamic depth registration, multi-scale temperature gradient analysis, adaptive stratified segmentation, variational modal decomposition, nonlinear correlation analysis, uncertainty quantification analysis, and multi-dimensional clustering. Among them, the variational modal decomposition algorithm is used to extract inter-layer modal features from multi-layer structural feature data, which can effectively separate ocean dynamic processes of different scales. Based on the stratified deviation feature data, an adaptive multi-factor correction rule is constructed to obtain dynamic correction parameters. This process includes principal component analysis, nonlinear sensitivity analysis, rule generation, cross-validation, genetic algorithm optimization, parameter fine-tuning, spatiotemporal hierarchical mapping, and particle swarm optimization. For example, the particle swarm optimization algorithm is used to dynamically solve the multi-level correction matrix, which can adjust the correction parameters in real time when the marine environment changes rapidly.

[0092] Based on the dynamic correction parameters, the real-time temperature profile data is nonlinearly corrected and intelligently fused with the preset reference data set to obtain preliminary integrated observation data. This step includes wavelet multi-resolution analysis, nonlinear mapping, adaptive depth registration, local regression smoothing, dynamic time warping, multi-source data fusion, hierarchical feature extraction, tensor decomposition and boundary processing. Among them, the tensor decomposition technology is used to reconstruct the layered temperature structure in time and space, which can effectively capture the spatial correlation of the three-dimensional temperature field. Subsequently, the preliminary integrated observation data is subjected to multi-level quality control and probabilistic uncertainty assessment to obtain refined observation data with a credibility index. This process involves multi-dimensional outlier detection, adaptive threshold segmentation, spatiotemporal consistency analysis, support vector machine classification, Bayesian network probability analysis, Monte Carlo simulation, fuzzy comprehensive evaluation analysis, hierarchical analysis, information entropy theory and association mapping. For example, the support vector machine is used to classify abnormal pattern features and can accurately identify different types of data anomalies. The refined observation data is closed-loop feedback processed to obtain the target integrated corrected observation data. This step includes spatiotemporal hierarchical analysis, wavelet packet transform, adaptive threshold segmentation, empirical mode decomposition, Hilbert spectrum analysis, nonlinear state space model construction, Kalman filter processing, covariance analysis, adaptive particle filter and feedback correction. Among them, empirical mode decomposition is used to extract the intrinsic mode function set from the key frequency band information, which can effectively separate the ocean dynamic processes of different scales.

[0093] For example, in a current simulation, the original observation data contained temperature, salinity and density data at 10,000 time points and 100 depth layers. Through multi-dimensional preprocessing, these data were standardized into a three-dimensional matrix of 1000×50×3. In the multi-parameter cross-comparison analysis, it was found that there were significant temperature and salinity anomalies in the depth range of 40-60 meters. Based on this, a dynamic correction rule containing 20 correction factors was constructed. These correction parameters were applied to perform nonlinear correction on the real-time temperature profile and fused with historical data to obtain preliminary integrated current data. Through multi-level quality control, 3% of the data points were identified to be abnormal, and a credibility index between 0 and 1 was assigned to each data point. Finally, through closed-loop feedback processing, these data were further optimized to obtain the three-dimensional temperature and salinity field and current simulation results.

[0094] In the embodiment of the present application, by multi-dimensional preprocessing of the original ocean observation data, spatiotemporal standardization is achieved, the data quality and consistency are effectively improved, and a solid foundation is laid for subsequent analysis. The introduction of multi-parameter cross-comparison analysis enables the layered deviation characteristics to be accurately captured, greatly improving the accuracy of data integration. The construction of adaptive multi-factor correction rules not only takes into account the complexity of the marine environment, but also realizes the real-time update of dynamic correction parameters, significantly enhancing the adaptability and robustness of the method. In the nonlinear correction and intelligent fusion stage, this method makes full use of advanced signal processing technology and data fusion algorithms, such as wavelet multi-resolution analysis, tensor decomposition, etc., to achieve effective extraction and integration of multi-scale features, greatly improving the accuracy and representativeness of temperature profile data. The introduction of multi-level quality control and probabilistic uncertainty assessment has fully guaranteed the reliability of the data. By introducing advanced algorithms such as support vector machines and Bayesian networks, refined processing of anomaly detection and uncertainty quantification is achieved, providing data users with more reliable observation results. The design of the closed-loop feedback processing mechanism, especially the combination of advanced time-frequency analysis methods such as empirical mode decomposition and Hilbert spectrum analysis, as well as state estimation techniques such as Kalman filtering and adaptive particle filtering, has significantly improved the real-time and accuracy of data integration.

[0095] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0096] (1) Reconstruct the time series of the original ocean observation data to obtain dynamic time-standardized data;

[0097] (2) Perform non-uniform grid spatial interpolation on the dynamic time-normalized data to obtain multi-resolution spatial distribution data;

[0098] (3) Identify outliers in multi-resolution spatial distribution data to obtain robust cleaned data;

[0099] (4) Perform multi-scale noise separation on the robust cleaned data to obtain fine noise reduction data;

[0100] (5) Deeply reconstruct the refined noise reduction data to obtain high-precision depth data;

[0101] (6) Compensating the high-precision depth data for dynamic temperature response characteristics to obtain temperature-corrected data;

[0102] (7) Perform salinity inversion and correction on the temperature-corrected data to obtain high-reliability salinity data;

[0103] (8) Performing layered density gradient analysis on high-confidence salinity data to obtain layered density structure data;

[0104] (9) Adaptive modal decomposition is performed on the layered density structure data to obtain multi-scale characteristic components;

[0105] (10) Reconstruct the spatiotemporal consistency of multi-scale feature components to obtain spatiotemporal standardized observation data.

[0106] Specifically, the original ocean observation data is reconstructed in time series. Adaptive time window technology is used to dynamically adjust the window size according to the changing characteristics of the data, so as to more accurately capture the temporal variation characteristics of the ocean parameters and obtain dynamic time standardized data. Then, the dynamic time standardized data is subjected to non-uniform grid spatial interpolation. Here, the Kriging interpolation method is used. This method takes into account spatial autocorrelation and can perform optimal estimation between irregularly distributed observation points, thereby obtaining multi-resolution spatial distribution data. Outliers are identified for multi-resolution spatial distribution data. The local outlier factor (LOF) algorithm is used. This algorithm detects outliers by comparing the local density of data points with their neighboring points, effectively identifying data points that are significantly different from the surrounding environment, and obtaining robust cleaning data. Multi-scale noise separation is performed on the robust cleaning data. Here, the dual-tree complex wavelet transform (DTCWT) is used. DTCWT can provide good positioning in both the time and frequency domains, effectively separate noise components of different scales, and obtain fine noise reduction data.

[0107] The fine noise reduction data is deeply reconstructed. A nonlinear depth correction model is used, which takes into account the influence of temperature and salinity on the sound velocity. The depth value is corrected by iterative calculation to obtain high-precision depth data. Then, the high-precision depth data is compensated for the dynamic temperature response characteristics. This step takes into account the response time and thermal inertia of the temperature sensor. By establishing a dynamic response model of the sensor, the real temperature change is inverted to obtain temperature-corrected data. The temperature-corrected data is inverted and corrected for salinity. The improved UNESCO equation is used, which takes into account the nonlinear relationship between temperature, conductivity and pressure. High-confidence salinity data is obtained by iterative calculation. Subsequently, a layered density gradient analysis is performed on the high-confidence salinity data. An adaptive layering algorithm is used to automatically determine the layer interface according to the changing characteristics of the density gradient to obtain layered density structure data.

[0108] Adaptive mode decomposition is performed on the layered density structure data, and the empirical mode decomposition (EMD) algorithm is used here. EMD can decompose complex signals into a finite number of intrinsic mode functions (IMFs), each of which represents a physical process at a different scale, thereby obtaining multi-scale characteristic components. Finally, the multi-scale characteristic components are reconstructed in a spatiotemporal consistency. The tensor decomposition technology is used, which can simultaneously consider the correlation of time, space and parameter dimensions, and obtain spatiotemporal standardized observation data through low-rank approximate reconstruction.

[0109] For example, in an ocean observation, the original data contains temperature and salinity observations at 1,000 time points and 50 depth layers. First, the original 1,000 time points are reconstructed into 500 standardized time points through adaptive time window technology. Then, the data of 50 irregular depth layers are interpolated to 100 uniformly distributed depth layers using the Kriging interpolation method. Next, about 2% of outliers are identified and processed using the LOF algorithm. The data is denoised using DTCWT, effectively removing high-frequency noise. During the depth reconstruction process, the original depth data is corrected by about 0.5-1% considering the change of sound speed with temperature and salinity. The temperature response characteristic compensation corrects the temperature deviation of about 0.1-0.2°C. During the salinity inversion and correction process, the accuracy of the salinity value is improved to 0.01PSU through iterative calculation. The hierarchical density gradient analysis identifies five main density layers. EMD decomposition obtains four main IMFs, representing ocean dynamic processes at different scales. Finally, all processed data are reconstructed into a standardized three-dimensional dataset of 500×100×3 (time×depth×parameter) through tensor decomposition technology.

[0110] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0111] (1) Perform dynamic depth registration on spatiotemporal standardized observation data to obtain depth-aligned data;

[0112] (2) Perform multi-scale temperature gradient analysis on the depth alignment data to obtain temperature change characteristic data;

[0113] (3) Adaptively segmenting the temperature change characteristic data into layers to obtain multi-layer structural characteristic data;

[0114] (4) The modal features of the multi-layer structure are extracted by using the variational modal decomposition algorithm to obtain the inter-layer modal features;

[0115] (5) Perform nonlinear correlation analysis on inter-layer modal characteristics to obtain parameter coupling relationship data;

[0116] (6) Perform uncertainty quantitative analysis on parameter coupling relationship data to obtain probability deviation distribution;

[0117] (7) Perform multi-dimensional clustering on the probability deviation distribution to obtain layered deviation feature data.

[0118] Specifically, dynamic depth registration is performed on the spatiotemporal standardized observation data. This process uses the dynamic time warping (DTW) algorithm, which aligns data sequences of different depths by finding the optimal time alignment path, thereby obtaining depth-aligned data. Next, multi-scale temperature gradient analysis is performed on the depth-aligned data using wavelet transform technology. Wavelet transform can analyze the changes in temperature gradients at different scales, extract temperature change characteristics in different depth ranges, and thus obtain temperature change characteristic data. The temperature change characteristic data is adaptively layered and segmented. An adaptive threshold segmentation algorithm is used here, which dynamically adjusts the segmentation threshold according to the local statistical characteristics of the temperature gradient, realizes adaptive stratification of the temperature profile, and obtains multi-layer structural characteristic data. Next, the multi-layer structural characteristic data is modally extracted by the variational mode decomposition (VMD) algorithm. The VMD algorithm is an adaptive signal decomposition method that can decompose complex signals into a series of intrinsic mode functions, each of which represents a physical process at a different scale, thereby obtaining inter-layer modal characteristics.

[0119] The nonlinear correlation analysis of inter-layer modal features is performed, and the mutual information (MI) method is used here. MI can capture the nonlinear dependency between variables and is not restricted by linear assumptions. By calculating the mutual information between different modal features, the parameter coupling relationship data is obtained. Then, the uncertainty quantification analysis of the parameter coupling relationship data is performed. The Monte Carlo method is used to estimate the probability distribution and uncertainty range between parameters through multiple random sampling and simulations to obtain the probability deviation distribution. The probability deviation distribution is clustered in multiple dimensions. The Gaussian mixture model (GMM) clustering algorithm is used here. GMM can process multidimensional data and consider the probability distribution characteristics of the data. The optimal number of clusters and cluster centers are determined by maximum likelihood estimation, thereby obtaining layered deviation feature data.

[0120] For example, in an ocean observation, the spatiotemporal standardized observation data includes temperature, salinity and density data at 500 time points and 100 depth layers. First, the data of these 100 depth layers are dynamically aligned using the DTW algorithm to standardize the originally irregular depth intervals into a uniform depth profile with a 1-meter interval. Then, the temperature data is multi-scale analyzed using wavelet transform to extract temperature gradient features at 5 different scales (such as 1-5 meters, 5-10 meters, 10-20 meters, 20-50 meters, and 50-100 meters). During the adaptive stratification segmentation process, the entire water column is divided into 6 main levels according to the changing characteristics of the temperature gradient. The VMD algorithm performs modal decomposition on the data of these 6 levels, extracts 3-5 main modes from each level, and obtains a total of about 25 inter-layer modal features. Through mutual information analysis, the nonlinear correlation between these 25 modal features is calculated, and a 25×25 correlation matrix is ​​constructed. The Monte Carlo method was used to perform 10,000 simulations to generate probability distributions for each parameter coupling relationship. Finally, these probability distributions were clustered using the GMM clustering algorithm, and four main deviation feature categories were identified, each representing a characteristic deviation of a different depth range and a different physical process.

[0121] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0122] (1) Perform principal component analysis on the stratified deviation characteristic data to obtain key influencing factors;

[0123] (2) Conduct nonlinear sensitivity analysis on key influencing factors to obtain factor weight distribution;

[0124] (3) Generate rules for factor weight distribution to obtain an initial correction rule set;

[0125] (4) Cross-validating the initial correction rule set to obtain rule effectiveness evaluation results;

[0126] (5) Perform genetic algorithm optimization based on the rule effectiveness evaluation results to obtain an optimized rule set;

[0127] (6) fine-tuning the parameters of the optimized rule set to obtain refined correction rules;

[0128] (7) Performing spatiotemporal hierarchical mapping on the refined correction rules to obtain a multi-level correction matrix;

[0129] (8) The dynamic parameters of the multi-level correction matrix are solved by particle swarm optimization algorithm to obtain dynamic correction parameters.

[0130] Specifically, principal component analysis (PCA) is performed on the stratified deviation feature data. PCA transforms variables that may be correlated into linearly independent variables through orthogonal transformation, thereby identifying the main direction of change in the data. In this method, PCA is used to reduce the dimension of the stratified deviation feature data and extract key influencing factors. Then, nonlinear sensitivity analysis of these key influencing factors is performed using the Sobol method. The Sobol method quantifies the degree of influence of each factor on the model output by calculating the global sensitivity index, thereby obtaining the factor weight distribution.

[0131] The fuzzy inference system (FIS) is used to generate rules for the distribution of factor weights. Based on fuzzy logic theory, FIS converts quantitative factor weights into qualitative language rules, such as "if the weight of factor A is high and the weight of factor B is low, then adopt correction strategy X", thereby obtaining an initial correction rule set. This initial correction rule set is cross-validated using the K-fold cross-validation method. This method divides the data set into K subsets, uses K-1 subsets to train the model each time, and uses the remaining subset for verification. After K cycles, the rule effectiveness evaluation results are obtained.

[0132] Based on the rule effectiveness evaluation results, a genetic algorithm (GA) is used for optimization. GA simulates natural selection and genetic processes, iteratively optimizes the rule set through selection, crossover and mutation operations, and obtains the optimized rule set. The gradient descent method is used to fine-tune the parameters of this optimized rule set. The gradient descent method calculates the gradient of the loss function to the parameter, iteratively adjusts the parameter value, and minimizes the error to obtain a refined correction rule. The refined correction rule is mapped in time and space. This step uses adaptive grid technology to dynamically adjust the grid density according to the spatiotemporal distribution characteristics of the data, maps the correction rules to different spatiotemporal scales, and obtains a multi-level correction matrix. Finally, the particle swarm optimization (PSO) algorithm is used to dynamically solve the parameters of the multi-level correction matrix. The PSO algorithm simulates the intelligent behavior of the group and finds the optimal solution by moving particles in the solution space, thereby obtaining dynamic correction parameters.

[0133] For example, in an ocean observation data processing, the stratified deviation feature data contained 100 feature variables. Through PCA analysis, these 100 variables were reduced to 10 principal components, explaining 95% of the data variance. Sobol method analysis showed that 3 of the principal components contributed more than 80% to the model output. Based on these results, 50 initial correction rules were generated using FIS. These 50 rules were evaluated using 10-fold cross validation, and it was found that 35 of them performed stably on all validation sets. The genetic algorithm used these 35 rules as the initial population, and after 100 generations of evolution, 20 optimized rules were obtained. The gradient descent method fine-tuned the parameters of these 20 rules, reducing the average error from 5% to 3%.

[0134] In the process of spatiotemporal hierarchical mapping, the observation area is divided into four time scales (hours, days, weeks, and months) and five spatial scales (surface, subsurface, middle, deep, and bottom) according to the spatiotemporal characteristics of the data, forming a 4×5 multi-level correction matrix. Finally, the PSO algorithm is used to optimize the parameters of each element in this matrix. The particle swarm size is set to 100, and it is iterated 500 times to obtain 20 dynamic correction parameters, which can adaptively adjust the correction strategy according to different spatiotemporal scales.

[0135] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0136] (1) Perform wavelet multi-resolution analysis on real-time temperature profile data to obtain multi-scale temperature characteristics;

[0137] (2) Perform nonlinear mapping on multi-scale temperature features to obtain preliminary corrected temperature data;

[0138] (3) performing adaptive depth registration on the preliminary corrected temperature data to obtain deep calibration temperature data;

[0139] (4) performing local regression smoothing on the depth calibration temperature data to obtain denoised temperature data;

[0140] (5) Dynamically time warping the denoised temperature data and a preset reference data set to obtain a time-aligned data set;

[0141] (6) Perform multi-source data fusion on the time-aligned data set to obtain the fused temperature field;

[0142] (7) extracting hierarchical features of the fused temperature field to obtain a layered temperature structure;

[0143] (8) The layered temperature structure is reconstructed in time and space by tensor decomposition technology to obtain a high-dimensional temperature field;

[0144] (9) Perform boundary processing on the high-dimensional temperature field to obtain preliminary integrated observation data.

[0145] Specifically, wavelet multi-resolution analysis is performed on real-time temperature profile data. This process uses discrete wavelet transform (DWT) to decompose the temperature profile data into approximate coefficients and detail coefficients of different scales, thereby obtaining multi-scale temperature features. Then, these multi-scale temperature features are nonlinearly mapped using the support vector regression (SVR) algorithm. SVR maps the data to a high-dimensional feature space and finds the optimal nonlinear functional relationship to obtain preliminary corrected temperature data. The preliminary corrected temperature data is adaptively deep aligned. The dynamic time warping (DTW) algorithm is used here. DTW can find the best alignment between two time series, so that the temperature data at different depths are aligned to the standard depth grid to obtain deep calibrated temperature data. The depth calibrated temperature data is locally regressed and smoothed using LOESS (local weighted scatter point smoothing). LOESS smoothes the data by fitting a low-order polynomial around each data point, effectively removing high-frequency noise and obtaining denoised temperature data.

[0146] Dynamic time warping is performed on the denoised temperature data and the preset reference data set. The DTW algorithm is used again, but this time to align the time series of the real-time observation data and the historical reference data to obtain a time-aligned data set. Multi-source data fusion is performed on the time-aligned data set using the Kalman filter algorithm. The Kalman filter optimally combines data from different sources through two steps, prediction and update, to obtain a fused temperature field. Hierarchical feature extraction is performed on the fused temperature field using the empirical mode decomposition (EMD) method. EMD can decompose complex signals into a finite number of intrinsic mode functions (IMFs), each of which represents a physical process at a different scale, thereby obtaining a layered temperature structure. The layered temperature structure is reconstructed in time and space by tensor decomposition technology, where Tucker decomposition is used. Tucker decomposition represents high-dimensional data as the product of a core tensor and a factor matrix, which can capture the multidimensional correlation of the data and obtain a high-dimensional temperature field. The high-dimensional temperature field is subjected to boundary processing using the conditional random field (CRF) model. CRF takes into account the spatial dependency between data points and can better handle the discontinuity and uncertainty of the temperature field boundary, and finally obtains preliminary integrated observation data.

[0147] For example, in an ocean observation, the real-time temperature profile data contains measurements of 100 depth layers. Through DWT, these data are decomposed into 5 scale levels, each level contains an approximation coefficient and a detail coefficient. SVR uses these coefficients as input to train a nonlinear mapping model to correct the original temperature value to a range closer to the true value. The DTW algorithm aligns the data of these 100 depth layers to 50 standard depth layers, solving the problem of inconsistent depth sampling at different observation times. LOESS uses 20% of the data points as the local regression window to effectively remove short-term fluctuations in the temperature profile. DTW is used again to align the real-time data with the historical data of the past 7 days in the time dimension. Kalman filtering fuses real-time observation data and numerical model forecast data. The state vector contains temperature and estimation errors, and the observation matrix is ​​dynamically adjusted according to data availability. EMD decomposes the fused temperature field into 5 IMFs and a residual, representing different scale processes from daily changes to seasonal changes.

[0148] Tucker decomposition reconstructs the layered temperature structure into a 4-dimensional tensor (time, longitude, latitude, depth), and the size of the core tensor is set to [10, 8, 8, 5], which effectively compresses the original data while retaining the main features. Finally, the CRF model uses an 8-neighborhood structure to process the boundaries of the temperature field, especially the temperature gradient at the junction of land and sea.

[0149] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0150] (1) Perform multi-dimensional outlier detection on the preliminary integrated observation data to obtain abnormal labeled data;

[0151] (2) Adaptively threshold segment the abnormal labeled data to obtain graded abnormal data;

[0152] (3) Perform spatiotemporal consistency analysis on the classified abnormal data to obtain abnormal pattern characteristics;

[0153] (4) Classify the abnormal pattern features through support vector machine to obtain the abnormal type distribution;

[0154] (5) Perform a probability analysis of the abnormal type distribution based on the Bayesian network to obtain the abnormal cause probability map;

[0155] (6) Monte Carlo simulation is performed based on the abnormal cause probability diagram to obtain uncertainty distribution estimation;

[0156] (7) Perform fuzzy comprehensive evaluation analysis on uncertainty distribution estimation to obtain multi-level quality indicators;

[0157] (8) Perform hierarchical analysis on multi-level quality indicators to obtain weighted quality scores;

[0158] (9) The weighted quality score is normalized by information entropy theory to obtain a standardized credibility index;

[0159] (10) The preliminary integrated observation data are correlated with the standardized credibility index to obtain refined observation data with credibility index.

[0160] Specifically, multi-dimensional outlier detection is performed on the preliminary integrated observation data. This process uses the isolation forest algorithm, which constructs multiple decision trees and calculates the anomaly score of each data point to obtain anomaly labeled data. Then, the abnormal labeled data is adaptively thresholded and segmented using the OTSU method. The OTSU algorithm automatically determines the optimal threshold by maximizing the inter-class variance, divides the abnormal data into different levels, and obtains graded abnormal data. The spatiotemporal consistency analysis of the graded abnormal data is performed using the spatial autocorrelation analysis method. Here, the Moran's I index is used to quantify the spatial aggregation degree of abnormal data, and combined with time series analysis, the abnormal pattern characteristics are obtained. The abnormal pattern characteristics are classified by the support vector machine (SVM). The SVM divides the abnormal patterns into different types by constructing the optimal separation hyperplane in the high-dimensional space to obtain the abnormal type distribution.

[0161] The probability analysis of the abnormal type distribution is based on the Bayesian network. The Bayesian network graphically represents the conditional dependency between variables and calculates the conditional probability between different abnormal types, thereby obtaining the abnormal cause probability map. Monte Carlo simulation is performed based on the abnormal cause probability map. Through a large number of random sampling and simulations, the probability distribution of various abnormal situations is estimated to obtain the uncertainty distribution estimate. The uncertainty distribution estimate is subjected to fuzzy comprehensive evaluation analysis using fuzzy set theory. This step converts the quantitative uncertainty estimate into a qualitative language evaluation, such as "high confidence", "medium confidence", etc., to obtain multi-level quality indicators. The multi-level quality indicators are subjected to hierarchical analysis using the analytic hierarchy process (AHP). AHP constructs a judgment matrix, calculates the eigenvector, obtains the weight of each quality indicator, and thus obtains a weighted quality score.

[0162] The weighted quality scores are normalized by information entropy theory. Information entropy is used to measure the uncertainty of data. The standardized credibility index is obtained by calculating the information entropy of each score and normalizing it. Finally, the association mapping between the preliminary integrated observation data and the standardized credibility index is performed using the association rule mining algorithm. The algorithm calculates the support and confidence, establishes the mapping relationship between the data point and the credibility index, and obtains the refined observation data with the credibility index.

[0163] For example, in an ocean temperature observation, the initial integrated observation data contains 1000 spatiotemporal data points. The isolation forest algorithm detected 50 potential anomalies, and the OTSU method divided these anomalies into three levels: mild (30), moderate (15) and severe (5). Spatial autocorrelation analysis found that severe anomalies showed significant clustering in the depth range of 40-60 meters (Moran's I = 0.75). The SVM classifier divided the anomaly patterns into four categories: instrument failure, environmental disturbance, data transmission error and unknown cause. Bayesian network analysis showed that the conditional probability of anomalies caused by environmental disturbances was the highest (0.6) in the depth range of 40-60 meters. Monte Carlo simulation was performed 10,000 times to generate probability distributions of various anomalies. Fuzzy comprehensive evaluation converted the uncertainty distribution into five levels of quality indicators. AHP analysis determined the weights of each indicator, such as data integrity (0.3), temporal consistency (0.25), spatial coherence (0.25) and physical rationality (0.2). After information entropy calculation and normalization, a standardized credibility index between 0 and 1 was obtained. Finally, through association rule mining, a mapping relationship between observation data points and credibility index was established. For example, it was found that the credibility index of data points with a depth of 40-60 meters and a temperature gradient greater than 0.5℃ / m was generally lower than 0.7.

[0164] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0165] (1) Perform spatiotemporal hierarchical analysis on the refined observation data to obtain multi-scale feature sequences;

[0166] (2) Perform wavelet packet transform on the multi-scale feature sequence to obtain the frequency domain feature distribution;

[0167] (3) Adaptively threshold segment the frequency domain feature distribution to obtain key frequency band information;

[0168] (4) Extract the modal information of key frequency bands through empirical mode decomposition to obtain the intrinsic mode function set;

[0169] (5) Perform Hilbert spectrum analysis on the eigenmode function set to obtain instantaneous frequency characteristics;

[0170] (6) Construct a nonlinear state space model based on the instantaneous frequency characteristics to obtain a dynamic system representation;

[0171] (7) Perform Kalman filtering on the dynamic system representation to obtain a state estimation sequence;

[0172] (8) Perform covariance analysis on the state estimation sequence to obtain the error correction matrix;

[0173] (9) Optimizing the parameters of the error correction matrix through adaptive particle filtering to obtain an optimized correction factor;

[0174] (10) Feedback correction is performed on the refined observation data to obtain target integrated corrected observation data.

[0175] Specifically, the refined observation data is subjected to spatiotemporal hierarchical analysis. This process uses multi-scale decomposition technology to decompose the data at different time and space scales to obtain a multi-scale feature sequence. Next, the multi-scale feature sequence is subjected to wavelet packet transform, which is an extension of wavelet transform and can provide finer frequency resolution, thereby obtaining the frequency domain feature distribution. The frequency domain feature distribution is subjected to adaptive threshold segmentation using the OTSU method. The OTSU algorithm automatically determines the optimal threshold by maximizing the inter-class variance, divides the frequency domain features into different levels, and obtains the key frequency band information. The key frequency band information is subjected to modal extraction by empirical mode decomposition (EMD). EMD is an adaptive signal decomposition method that can decompose complex signals into a finite number of intrinsic mode functions (IMFs) to obtain an eigenmode function set. The eigenmode function set is subjected to Hilbert spectrum analysis, which can extract the instantaneous frequency and amplitude information of the signal, thereby obtaining the instantaneous frequency characteristics. A nonlinear state space model is constructed based on the instantaneous frequency characteristics. Here, the nonlinear autoregressive moving average (NARMA) model is used. This model can capture the nonlinear dynamic characteristics of the system and obtain the dynamic system characterization.

[0176] The Kalman filter is used to characterize the dynamic system. The Kalman filter is a recursive optimal estimation algorithm. Through two steps of prediction and update, the system state is estimated to obtain a state estimation sequence. The state estimation sequence is subjected to covariance analysis to calculate the correlation and variability between different state variables to obtain the error correction matrix. The error correction matrix is ​​optimized by adaptive particle filtering. The particle filter is a nonlinear filtering technology based on the sequential Monte Carlo method. It can handle non-Gaussian distribution and strong nonlinear systems. The optimized correction factor is obtained by continuously adjusting the particle weight and resampling. Finally, the refined observation data is feedback corrected, and the optimized correction factor is applied to the original data to obtain the target integrated corrected observation data.

[0177] For example, in an ocean temperature profile observation, the refined observation data contains 1,000 time points, each of which has temperature data for 100 depth layers. Spatiotemporal hierarchical analysis decomposes the data into five time scales (hours, days, weeks, months, and seasons) and four spatial scales (surface, subsurface, middle, and deep), obtaining 20 multi-scale feature sequences.

[0178] Wavelet packet transform was used to decompose these 20 sequences, using the db4 wavelet basis and 5 decomposition levels to obtain the frequency domain feature distribution. The OTSU method divided the frequency domain features into 3 levels and identified 10 key frequency bands. EMD decomposed these 10 frequency bands separately, and each frequency band obtained an average of 4-6 IMFs, for a total of about 50 IMFs. Hilbert spectrum analysis calculated the instantaneous frequency and amplitude of these 50 IMFs to obtain the instantaneous frequency characteristics. Based on these characteristics, a 5th-order NARMA model was constructed to characterize the system dynamics. The Kalman filter used this NARMA model as the state equation, and the observation equation took into account the measurement noise. After 1000 iterations, the state estimation sequence was obtained.

[0179] Covariance analysis calculated the correlation matrix between state variables and found that the correlation coefficient between temperature and depth changed significantly over time. Adaptive particle filtering used 1000 particles and optimized the parameters of the error correction matrix through 50 iterations to obtain the optimized correction factor.

[0180] Finally, the optimized correction factor is applied to the original refined observation data to correct the temperature data of each depth layer. For example, in the surface layer (0-10m) data, the original temperature value of the data point is 25.3℃, which is corrected to 25.1℃, which is closer to the actual ocean conditions.

[0181] The above describes the real-time data integration method based on ground observation data in the embodiment of the present application. The following describes the real-time data integration system based on ground observation data in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a real-time data integration system based on ground observation data includes:

[0182] The processing module 201 is used to perform multi-dimensional preprocessing on the pre-collected original ocean observation data to obtain spatiotemporal standardized observation data;

[0183] A comparison module 202 is used to perform multi-parameter cross comparison analysis on the spatiotemporal standardized observation data to obtain layered deviation characteristic data;

[0184] A construction module 203 is used to construct an adaptive multi-factor correction rule based on the stratified deviation characteristic data to obtain a dynamic correction parameter;

[0185] A fusion module 204 is used to perform nonlinear correction on the real-time temperature profile data based on the dynamic correction parameters and intelligently fuse it with a preset reference data set to obtain preliminary integrated observation data;

[0186] An evaluation module 205 is used to perform multi-level quality control and probabilistic uncertainty evaluation on the preliminary integrated observation data to obtain refined observation data with a credibility index;

[0187] Feedback module 206 is used to perform closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data.

[0188] Through the collaboration of the above components, the original ocean observation data is preprocessed in multiple dimensions to achieve spatiotemporal standardization, effectively improve the data quality and consistency, and lay a solid foundation for subsequent analysis. The introduction of multi-parameter cross-comparison analysis enables the layered deviation characteristics to be accurately captured, greatly improving the accuracy of data integration. The construction of adaptive multi-factor correction rules not only takes into account the complexity of the ocean environment, but also realizes the real-time update of dynamic correction parameters, significantly enhancing the adaptability and robustness of the method. In the nonlinear correction and intelligent fusion stage, this method makes full use of advanced signal processing technology and data fusion algorithms, such as wavelet multi-resolution analysis and tensor decomposition, to achieve effective extraction and integration of multi-scale features, greatly improving the accuracy and representativeness of temperature profile data. The introduction of multi-level quality control and probabilistic uncertainty assessment ensures the reliability of the data. By introducing advanced algorithms such as support vector machines and Bayesian networks, the refined processing of anomaly detection and uncertainty quantification is achieved, providing data users with more reliable observation results. The design of the closed-loop feedback processing mechanism, especially the combination of advanced time-frequency analysis methods such as empirical mode decomposition and Hilbert spectrum analysis, as well as state estimation techniques such as Kalman filtering and adaptive particle filtering, has significantly improved the real-time and accuracy of data integration.

[0189] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0190] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time data integration method based on ground observation data, characterized in that: The real-time data integration method based on ground observation data includes: Perform multi-dimensional preprocessing on the pre-collected raw ocean observation data to obtain spatiotemporal standardized observation data; Performing multi-parameter cross-comparison analysis on the spatiotemporal standardized observation data to obtain stratified deviation characteristic data; Constructing an adaptive multi-factor correction rule based on the stratified deviation characteristic data to obtain a dynamic correction parameter; Based on the dynamic correction parameters, nonlinear correction is performed on the real-time temperature profile data and intelligently integrated with the preset reference data set to obtain preliminary integrated observation data; Performing multi-level quality control and probabilistic uncertainty assessment on the preliminary integrated observation data to obtain refined observation data with a credibility index; The refined observation data is subjected to closed-loop feedback processing to obtain target integrated corrected observation data.

2. The real-time data integration method based on ground observation data according to claim 1 is characterized in that: The multi-dimensional preprocessing of the pre-collected original ocean observation data to obtain spatiotemporal standardized observation data includes: Reconstructing the original ocean observation data in time series to obtain dynamic time-standardized data; Performing non-uniform grid spatial interpolation on the dynamic time normalized data to obtain multi-resolution spatial distribution data; Performing outlier identification on the multi-resolution spatial distribution data to obtain robust cleaned data; Performing multi-scale noise separation on the robust cleaned data to obtain fine noise reduction data; Performing deep reconstruction on the refined noise reduction data to obtain high-precision depth data; Performing dynamic temperature response characteristic compensation on the high-precision depth data to obtain temperature correction data; Performing salinity inversion and correction on the temperature-corrected data to obtain high-reliability salinity data; Performing layered density gradient analysis on the high-confidence salinity data to obtain layered density structure data; Performing adaptive modal decomposition on the layered density structure data to obtain multi-scale characteristic components; The multi-scale feature components are reconstructed in a spatiotemporal consistency to obtain the spatiotemporal standardized observation data.

3. The real-time data integration method based on ground observation data according to claim 1 is characterized in that: The multi-parameter cross-comparison analysis of the spatiotemporal standardized observation data is performed to obtain stratified deviation characteristic data, including: Performing dynamic depth registration on the spatiotemporal standardized observation data to obtain depth alignment data; Performing multi-scale temperature gradient analysis on the depth alignment data to obtain temperature change characteristic data; Adaptively segmenting the temperature change characteristic data into layers to obtain multi-layer structure characteristic data; Performing modal extraction on the multi-layer structure feature data by using a variational modal decomposition algorithm to obtain inter-layer modal features; Performing nonlinear correlation analysis on the inter-layer modal characteristics to obtain parameter coupling relationship data; Performing uncertainty quantitative analysis on the parameter coupling relationship data to obtain a probability deviation distribution; Multi-dimensional clustering is performed on the probability deviation distribution to obtain the layered deviation feature data.

4. The real-time data integration method based on ground observation data according to claim 1 is characterized in that: The step of constructing an adaptive multi-factor correction rule based on the stratified deviation feature data to obtain dynamic correction parameters includes: Performing principal component analysis on the stratified deviation characteristic data to obtain key influencing factors; Conduct nonlinear sensitivity analysis on the key influencing factors to obtain factor weight distribution; Generating rules for the factor weight distribution to obtain an initial correction rule set; Cross-validating the initial correction rule set to obtain rule effectiveness evaluation results; Performing genetic algorithm optimization based on the rule effectiveness evaluation results to obtain an optimized rule set; Fine-tuning parameters of the optimization rule set to obtain refined correction rules; Performing spatiotemporal hierarchical mapping on the refined correction rules to obtain a multi-level correction matrix; The dynamic correction parameters are obtained by solving the dynamic parameters of the multi-level correction matrix through a particle swarm optimization algorithm.

5. The real-time data integration method based on ground observation data according to claim 1, characterized in that: Based on the dynamic correction parameters, the real-time temperature profile data is nonlinearly corrected and intelligently integrated with the preset reference data set to obtain preliminary integrated observation data, including: Performing wavelet multi-resolution analysis on the real-time temperature profile data to obtain multi-scale temperature characteristics; Performing nonlinear mapping on the multi-scale temperature features to obtain preliminary corrected temperature data; Performing adaptive depth registration on the preliminary corrected temperature data to obtain depth calibrated temperature data; Performing local regression smoothing on the depth calibration temperature data to obtain denoised temperature data; Performing dynamic time warping on the denoised temperature data and the preset reference data set to obtain a time-aligned data set; Performing multi-source data fusion on the time-aligned data set to obtain a fused temperature field; Performing layered feature extraction on the fused temperature field to obtain a layered temperature structure; The layered temperature structure is reconstructed in time and space by tensor decomposition technology to obtain a high-dimensional temperature field; The high-dimensional temperature field is subjected to boundary processing to obtain the preliminary integrated observation data.

6. The real-time data integration method based on ground observation data according to claim 1, characterized in that: The multi-level quality control and probabilistic uncertainty assessment of the preliminary integrated observation data is performed to obtain refined observation data with a credibility index, including: Performing multi-dimensional outlier detection on the preliminary integrated observation data to obtain abnormal labeled data; Performing adaptive threshold segmentation on the abnormal labeled data to obtain graded abnormal data; Performing spatiotemporal consistency analysis on the graded abnormal data to obtain abnormal pattern characteristics; Classifying the abnormal pattern features by a support vector machine to obtain abnormal type distribution; Performing a probability analysis on the abnormal type distribution based on a Bayesian network to obtain an abnormal cause probability map; Performing Monte Carlo simulation based on the abnormal cause probability map to obtain an uncertainty distribution estimate; Performing fuzzy comprehensive evaluation and analysis on the uncertainty distribution estimation to obtain multi-level quality indicators; Performing hierarchical analysis on the multi-level quality indicators to obtain weighted quality scores; The weighted quality score is normalized by information entropy theory to obtain a standardized credibility index; The preliminary integrated observation data and the standardized credibility index are associated and mapped to obtain the refined observation data with the credibility index.

7. The real-time data integration method based on ground observation data according to claim 1, characterized in that: The performing closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data includes: Performing spatiotemporal hierarchical analysis on the refined observation data to obtain a multi-scale feature sequence; Performing wavelet packet transform on the multi-scale feature sequence to obtain frequency domain feature distribution; Performing adaptive threshold segmentation on the frequency domain feature distribution to obtain key frequency band information; Performing modal extraction on the key frequency band information through empirical mode decomposition to obtain an intrinsic mode function set; Performing Hilbert spectrum analysis on the eigenmode function set to obtain instantaneous frequency characteristics; Constructing a nonlinear state space model based on the instantaneous frequency characteristics to obtain a dynamic system representation; Performing Kalman filtering on the dynamic system representation to obtain a state estimation sequence; Performing covariance analysis on the state estimation sequence to obtain an error correction matrix; Optimizing the parameters of the error correction matrix by adaptive particle filtering to obtain an optimized correction factor; Feedback correction is performed on the refined observation data to obtain the target integrated corrected observation data.

8. A real-time data integration system based on ground observation data, used to execute the real-time data integration method based on ground observation data according to any one of claims 1 to 7, characterized in that: The real-time data integration system based on ground observation data includes: A processing module is used to perform multi-dimensional preprocessing on the pre-collected raw ocean observation data to obtain spatiotemporal standardized observation data; A comparison module, used for performing multi-parameter cross comparison analysis on the spatiotemporal standardized observation data to obtain layered deviation characteristic data; A construction module, used to construct an adaptive multi-factor correction rule based on the stratified deviation characteristic data to obtain a dynamic correction parameter; A fusion module, used to perform nonlinear correction on the real-time temperature profile data based on the dynamic correction parameters and intelligently fuse it with a preset reference data set to obtain preliminary integrated observation data; An evaluation module, used for performing multi-level quality control and probabilistic uncertainty evaluation on the preliminary integrated observation data to obtain refined observation data with a credibility index; The feedback module is used to perform closed-loop feedback processing on the refined observation data to obtain target integrated corrected observation data.

Images