Marine physical data intelligent fusion and optimization processing method

Abstract

The invention relates to the technical field of marine environment monitoring, in particular to a marine physical data intelligent fusion and optimization processing method, which comprises the following steps: collecting marine physical data in real time through various sensors, including temperature, salinity, flow velocity, tide and sound wave characteristic data; and spatial-temporal feature extraction and spatial correlation analysis are carried out by using a depth residual-graph convolution hybrid model. And performing parameter optimization on the multi-source fusion data set by using a quantum annealing optimization algorithm to generate an optimal parameter set so as to improve the fusion precision of the data. And abnormal data nodes are identified through a self-adaptive threshold detection algorithm, so that the accuracy and reliability of the data are further ensured. And finally, constructing a three-dimensional visual marine environment model according to the optimized parameter set, and realizing a dynamic mapping relationship between model parameters and physical characteristics so as to realize real-time updating and visual display. According to the method, the processing precision of the marine environment data is effectively improved, accurate marine environment prediction is provided, and the method has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of marine environment monitoring, and in particular to a method for intelligent fusion and optimization processing of marine physical data. Background Art

[0002] Changes in the ocean environment are affected by many factors, including physical characteristics such as temperature, salinity, current velocity, tides, and sound wave propagation. To obtain accurate ocean environment data, modern ocean research relies on the collaborative work of multi-source sensors, such as temperature sensors, salinity sensors, current meters, tide gauges, and sonar systems, which can collect various physical data in the ocean in real time.

[0003] At present, some methods have been proposed for the fusion and correction of multi-source ocean data, but there are still many technical problems. First, most existing methods are limited to the processing of a single data source, and lack efficient algorithms for multi-source data fusion. Especially when the data volume is large and the sources are complex, the existing algorithms are difficult to ensure the accuracy and consistency of data fusion. Secondly, the correction problem of spatiotemporal data has not been effectively solved, especially how to eliminate the spatiotemporal deviation between sensors and ensure the accuracy of each data point in the time and space dimensions. Although some methods use deep learning and machine learning for feature extraction, in practical applications, the training and optimization of the model still face high computational costs and low accuracy. Furthermore, existing anomaly detection methods often use fixed thresholds, lack adaptability and flexibility, resulting in serious false alarms and missed reports. Summary of the invention

[0004] The present invention provides a method for intelligent fusion and optimization processing of ocean physical data.

[0005] The method for intelligent fusion and optimization processing of ocean physical data includes the following steps: S1: Standardize and preprocess the ocean physical data collected by multi-source sensors to generate a standardized matrix containing temperature, salinity, current velocity, tide and acoustic wave characteristics; S2: construct a spatiotemporal correction matrix based on a dynamic sliding window, perform spatiotemporal benchmark dynamic correction on the standardized matrix, and generate a spatiotemporal correction matrix; S3: Input the spatiotemporal correction matrix into the deep residual-graph convolution hybrid model for feature fusion and output a multi-source fusion dataset; S4: Construct a multi-dimensional abnormal feature association matrix based on multi-source fusion data sets, and use an adaptive threshold detection algorithm to identify abnormal data nodes; S5: Use the quantum annealing optimization algorithm to iterate the parameters of the multi-source fusion data set and generate an optimized parameter set; S6: Construct a three-dimensional visual ocean environment model based on the optimized parameter set, and establish a dynamic mapping relationship between model parameters and physical characteristics.

[0006] Optionally, the S1 includes: S11, data collection: by setting up multiple different types of ocean sensors, real-time collection of ocean physical data, including temperature, salinity, current velocity, tide and acoustic wave characteristic data; S12, data formatting: formatting the collected ocean physical data to form a data set that matches time and space; S13, missing value filling: For sensor data with missing or abnormal data, the missing values ​​are filled by interpolation or smoothing method based on adjacent data; S14, normalization processing: normalizing the collected ocean physical data; S15, feature extraction: extracting temporal features and spatial features from the standardized data; S16, constructing a standardized matrix: combining the normalized ocean physical data and the extracted eigenvalues ​​to form a standardized matrix.

[0007] Optionally, S2 includes: S21, sliding window setting and data division: according to the acquisition frequency and spatial resolution of ocean physical data, the size and step length of the dynamic sliding window are set; The window size is adjusted according to the time span and spatial range; According to the set sliding window, the standardized matrix is ​​divided into multiple spatiotemporal sub-matrices according to the time and space dimensions, and each sub-matrix contains a data set within a time period and a preset spatial range; S22, spatiotemporal benchmark data selection: Select representative benchmark data from multiple spatiotemporal sub-matrices as reference benchmarks for spatiotemporal correction.

[0008] S23, spatiotemporal deviation analysis: by comparing the data of each spatiotemporal submatrix with the benchmark data, the deviation of the spatiotemporal data is analyzed.

[0009] S24, dynamic correction model construction: according to the results of spatiotemporal deviation analysis, a spatiotemporal correction model is established using statistical modeling methods; S25, spatiotemporal correction calculation: applying the spatiotemporal correction model to the data of each spatiotemporal submatrix, obtaining the correction parameters of each spatiotemporal window through model calculation, and performing spatiotemporal correction on the data of each submatrix; S26, generating a space-time correction matrix: merging the corrected multiple space-time sub-matrices to generate a complete space-time correction matrix; S27, correction result evaluation and optimization: perform quality evaluation on the generated spatiotemporal correction matrix.

[0010] Optionally, the S3 includes: S31, deep residual network module construction and feature extraction: construct a deep residual network module and extract the temporal features in the spatiotemporal correction matrix through the residual learning mechanism; S32, graph convolutional network module construction and spatial feature extraction: Construct a graph convolutional network module and use the graph convolution method to capture the spatial features in the spatiotemporal correction matrix.

[0011] S33, weighted fusion and multi-source fusion dataset generation: The temporal features and spatial features extracted by the deep residual network module and the graph convolutional network module are weightedly fused to form a comprehensive feature representation.

[0012] The comprehensive feature representation formed after fusion will be output as a multi-source fusion dataset, which contains the temporal and spatial features fused from multiple sensors.

[0013] S34, fusion result evaluation and optimization: perform quality evaluation on the output multi-source fusion data set. The evaluation indicators include data accuracy, stability and fusion effect.

[0014] Optionally, the S4 includes: S41, constructing a multi-dimensional abnormal feature association matrix: constructing a multi-dimensional abnormal feature association matrix according to multiple features in the multi-source fusion data set; S42, using an adaptive threshold detection algorithm: using an adaptive threshold detection algorithm to analyze the data in the multi-dimensional abnormal feature association matrix and identify abnormal data nodes.

[0015] S43, abnormal data node identification and output: after being processed by the adaptive threshold detection algorithm, all abnormal data nodes are identified and output; S44, anomaly detection result evaluation and optimization: evaluate the abnormal data identification results.

[0016] Optionally, the S5 includes: S51, define optimization objective function: determine the optimization objective function, which is used to measure the performance of each parameter in the optimization process; S52, quantum annealing algorithm initialization: Use the quantum annealing algorithm to optimize parameters, initialize the parameter set, and represent the initial state of each parameter on the quantum bit. The goal of quantum annealing is to gradually search for the optimal parameter combination through the evolution of the quantum bit state; S53, quantum annealing evolution process: The quantum annealing evolution process continuously reduces the "temperature" of the system and gradually reaches the optimization goal by annealing the quantum bit state; S54, measurement and parameter update of quantum bits: After the quantum system has undergone the annealing process, each time the temperature drops to the lowest, the quantum bits are measured to obtain the optimized parameter set.

[0017] Optionally, the S6 includes: S61, input and initialization of optimization parameter set: using the generated optimization parameter set as the basic parameters for constructing a three-dimensional visualized ocean environment model, and initializing the basic information required for the ocean environment model, the basic information including the boundary range of the ocean area and gridding parameters; S62, 3D grid construction: constructing a 3D grid according to the boundary information of the ocean area; S63, dynamic mapping of physical characteristics and model parameters: assigning corresponding physical characteristic values ​​to each grid node according to the optimized parameter set; S64, Construction of a three-dimensional visual ocean environment model: Based on the established grid and physical eigenvalues, a three-dimensional visual ocean environment model is constructed.

[0018] S65, visualization result display and analysis: Display the generated 3D visualization ocean environment model to the user and support interactive operation.

[0019] Beneficial effects of the present invention: The present invention collects ocean physical data through multi-source sensors, and processes and optimizes the data using deep learning models, quantum annealing optimization algorithms, and adaptive threshold detection algorithms. A hybrid model of deep residual networks and graph convolutional networks is used to achieve deep feature extraction of spatiotemporal data and capture of spatial dependencies, and a quantum annealing algorithm is used to dynamically optimize parameter sets, thereby achieving efficient multi-source data fusion. This method effectively improves the accuracy and consistency of data, can make full use of the data characteristics of different sensors, and achieve more accurate analysis of the ocean physical environment.

[0020] The present invention uses a spatiotemporal correction method based on a dynamic sliding window to accurately correct the spatiotemporal deviation in sensor data and ensure the reliability and consistency of the data. Combined with an adaptive threshold detection algorithm, it effectively identifies abnormal nodes in the data and ensures the high-quality construction of the marine environment model. This method can adaptively adjust the detection threshold and improve the sensitivity of detection through local anomaly detection, reduce the false alarm rate and missed alarm rate, further improve the recognition accuracy of abnormal data, and ensure the stability and reliability of data processing.

[0021] The present invention provides an intuitive and efficient marine environment display platform by constructing a three-dimensional visual marine environment model and dynamically establishing a mapping relationship between model parameters and physical characteristics. The optimized parameter set is obtained through a quantum annealing optimization algorithm and mapped to a three-dimensional grid, supporting real-time updates and interactive operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention; Figure 2 Schematic diagram of S3 process of an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0025] like Figure 1-Figure 2 As shown, the method for intelligent fusion and optimization processing of ocean physical data includes the following steps: S1: Standardize and preprocess the ocean physical data collected by multi-source sensors to generate a standardized matrix containing temperature, salinity, current velocity, tide and acoustic wave characteristics; S2: construct a spatiotemporal correction matrix based on a dynamic sliding window, perform spatiotemporal benchmark dynamic correction on the standardized matrix, and generate a spatiotemporal correction matrix; S3: Input the spatiotemporal correction matrix into the deep residual-graph convolution hybrid model for feature fusion and output a multi-source fusion dataset; S4: Construct a multi-dimensional abnormal feature association matrix based on multi-source fusion data sets, and use an adaptive threshold detection algorithm to identify abnormal data nodes; S5: Use the quantum annealing optimization algorithm to iterate the parameters of the multi-source fusion data set and generate an optimized parameter set; S6: Construct a three-dimensional visual ocean environment model based on the optimized parameter set, and establish a dynamic mapping relationship between model parameters and physical characteristics.

[0026] S1 includes: S11, data collection: by setting up multiple different types of ocean sensors, real-time collection of ocean physical data, including temperature, salinity, current velocity, tide and acoustic wave characteristic data; Temperature data is collected through a CTD sensor (Conductivity, Temperature, Depth sensor), which can simultaneously measure the temperature, conductivity, and depth of seawater and calculate salinity based on conductivity; Salinity data is collected through CTD sensors or conductivity sensors, which use conductivity values ​​to estimate seawater salinity; The flow velocity data is collected by an acoustic Doppler current profiler, which uses the acoustic Doppler effect to calculate the water flow velocity and can provide flow velocity information at different depths; Tidal data is collected through pressure sensors or tide gauges. Pressure sensors measure changes in water levels, while tide gauges are specifically used to monitor changes in sea levels to obtain tidal data; Acoustic wave characteristic data is collected by sonar systems or ocean acoustic sensors, which can measure the propagation speed, reflection intensity and other acoustic characteristics of sound waves in water; S12, data formatting: formatting the collected ocean physical data to ensure that different data sources have unified timestamps and spatial location tags to form a data set that matches time and space; S13, missing value filling: For missing or abnormal sensor data, fill the missing values ​​through interpolation or smoothing method based on adjacent data to ensure the continuity and integrity of the data; S14, normalization processing: normalize the collected ocean physical data so that the value range of each data indicator is unified between [0,1]. The specific method is: Temperature: Perform minimum-maximum normalization on the collected temperature data. The formula is: ; Where T is the original temperature data value, The minimum temperature in the data set, The maximum temperature in the data set, The normalized temperature data ranges from [0, 1]; Salinity: The salinity data is normalized by Z-score, the formula is: ; Where S is the original salinity data value, is the mean value of salinity data, is the standard deviation of salinity data, is the normalized salinity data; Flow rate: Standard deviation normalization method is used, the formula is: ; Where V is the original flow rate data value, is the standard deviation of the velocity data, is the normalized flow velocity data; Tide: Periodically normalize the tidal data and adjust the normalization based on the day, month, or year to ensure that the impact of periodic changes is eliminated; Acoustic wave characteristics: After logarithmic transformation of the acoustic wave characteristics, normalization is performed to reduce the impact of extreme values ​​of the data; S15, feature extraction: extracting temporal and spatial features from the standardized data, such as the mean, standard deviation, maximum, minimum, and rate of change of the data; S16, constructing a standardized matrix: combining the normalized ocean physical data and the extracted eigenvalues ​​to form a standardized matrix as input for subsequent spatiotemporal correction processing.

[0027] S2 includes: S21, sliding window setting and data division: according to the acquisition frequency and spatial resolution of ocean physical data, the size and step length of the dynamic sliding window are set; The window size is adjusted according to the time span (such as hours, days, months) and spatial range (such as longitude and latitude range or water depth range); According to the set sliding window, the standardized matrix is ​​divided into multiple spatiotemporal sub-matrices according to the time and space dimensions. Each sub-matrix contains a data set within a time period and a preset spatial range. The preset spatial range includes the latitude range, longitude range and water depth range. For example, the outer edge of the study area is delineated based on the coastline of the actual sea area, and then the area is divided into multiple adjacent local units according to established rules; these local units can be several grid blocks of similar size, or they can be divided according to the distribution of sensors in the sea area, the homogeneity of marine elements or administrative divisions. Regardless of the division method used, it is necessary to ensure that the marine physical data monitored in each local unit has internal consistency, so that the local unit can represent the environmental status of a local sea area; when further divided in combination with the time dimension, a "spatiotemporal sub-matrix" can be formed in each local unit for subsequent dynamic sliding window analysis and data correction; S22, spatiotemporal benchmark data selection: Select representative benchmark data from multiple spatiotemporal sub-matrices as the reference benchmark for spatiotemporal correction. The benchmark data is stable and representative ocean physical data, usually selected from time periods and spatial locations where the sensor is in good condition and the environment changes little. The purpose of selecting benchmark data is to ensure the consistency and accuracy of the subsequent correction process.

[0028] S23, spatiotemporal deviation analysis: By comparing the data of each spatiotemporal submatrix with the baseline data, the spatiotemporal data deviation is analyzed. The specific method is to use correlation analysis, regression analysis or error matrix calculation to find out the spatiotemporal deviation caused by sensor errors, environmental changes or data noise. This analysis helps to identify the specific source of data deviation and provides a basis for the construction of a dynamic correction model.

[0029] S24, dynamic correction model construction: Based on the results of spatiotemporal deviation analysis, a spatiotemporal correction model is established using statistical modeling methods (such as least squares method, Kalman filtering method, etc.). The model automatically adjusts the spatiotemporal data based on the spatiotemporal data and deviation information to eliminate the deviation in the data, and can dynamically adjust the correction parameters based on the actual collected environmental data to ensure the real-time and accuracy of the correction; S25, spatiotemporal correction calculation: applying the spatiotemporal correction model to the data of each spatiotemporal submatrix, obtaining the correction parameters of each spatiotemporal window through model calculation, and performing spatiotemporal correction on the data of each submatrix. The correction process includes adjusting the data such as temperature, salinity, flow velocity, tide, acoustic wave characteristics, etc., to ensure the consistency of the data in time and space, and eliminate the deviation caused by factors such as sensor error and environmental fluctuation; S26, generation of spatiotemporal correction matrix: merging multiple spatiotemporal sub-matrices after correction to generate a complete spatiotemporal correction matrix, which contains corrected ocean physical data (such as temperature, salinity, current velocity, tide, etc.), ensuring that each data point is dynamically corrected in the spatiotemporal dimension and meets the unified benchmark standard; S27, Correction result evaluation and optimization: Perform quality evaluation on the generated spatiotemporal correction matrix. The evaluation criteria include data smoothness, consistency, precision and accuracy. The correction effect is evaluated by calculating indicators such as the standard deviation, deviation, and error range of the data. If the correction result does not meet expectations, adjust the sliding window parameters, the algorithm of the correction model, or the selection of benchmark data to further optimize the correction effect.

[0030] S3 includes: S31, deep residual network module construction and feature extraction: construct a deep residual network module, extract the temporal features in the spatiotemporal correction matrix through the residual learning mechanism, and the residual network adopts a residual block structure, which can alleviate the gradient vanishing problem and extract more meaningful features. The deep residual network module includes the following steps: Residual block design: Each residual block contains two convolutional layers and skip connections. The input is directly added to the output through the skip connection to enhance the learning ability of the network.

[0031] Feature extraction: By stacking multiple residual blocks, high-level temporal and spatial features are extracted from the spatiotemporal correction matrix.

[0032] Activation function application: After each convolutional layer, a nonlinear activation function (such as ReLU) is applied to enhance the expressiveness of the features.

[0033] The features processed by the deep residual network module will be used for subsequent graph convolution processing; S32, graph convolutional network module construction and spatial feature extraction: Construct a graph convolutional network module and use the graph convolution method to capture the spatial features in the spatiotemporal correction matrix. The specific steps are as follows: Graph structure construction: The spatial information in the spatiotemporal correction matrix (such as the location of ocean sensors, spatial dependencies between data, etc.) is converted into a graph structure. Each spatiotemporal data point is a node in the graph, and the edges between nodes represent the spatial associations between data.

[0034] Graph convolution operation: In a graph convolutional network, the information of each node’s neighborhood is aggregated through convolution operations. Specifically, for each node, the graph convolution layer updates the feature representation of the node by aggregating the information of neighboring nodes (such as the correlation of physical quantities such as temperature and salinity).

[0035] Adjacency matrix and weight matrix: The core operation of graph convolutional networks is to perform convolution based on the adjacency matrix (representing the connection relationship between nodes) and the weight matrix (representing the strength of each connection). This process can learn complex spatial relationships between nodes.

[0036] The spatial features processed by the graph convolutional network module will be fused with the temporal features extracted by the deep residual network module.

[0037] S33, weighted fusion and multi-source fusion data set generation: weighted fusion of temporal features and spatial features extracted by the deep residual network module and graph convolutional network module to form a comprehensive feature representation. The specific steps include: Weighted fusion: The temporal features and spatial features output by the deep residual network and graph convolutional network are weighted and summed. The weights are learned through training. Weighted fusion can dynamically adjust the contribution ratio of temporal features and spatial features in the final fusion result according to their importance.

[0038] Feature representation after fusion: The feature representation obtained by weighted fusion can simultaneously retain the temporal information and spatial information of the spatiotemporal correction matrix, has stronger representation ability, and can better reflect the comprehensive characteristics of the marine physical environment.

[0039] The comprehensive feature representation formed after fusion will be output as a multi-source fusion dataset, which contains temporal and spatial features fused from multiple sensors (such as temperature, salinity, flow velocity, tide, and acoustic wave characteristics), providing high-quality input data for subsequent anomaly detection, predictive modeling or other tasks.

[0040] S34, Fusion result evaluation and optimization: The quality of the output multi-source fusion data set is evaluated. The evaluation indicators include data accuracy, stability and fusion effect. The evaluation is performed by calculating the error between the fused data and the real data, and using clustering or classification algorithms for verification. If the evaluation result does not meet the requirements, the fusion effect is further improved by optimizing the network structure (such as adjusting the number of layers of the residual block, modifying the parameters of the graph convolution layer, etc.).

[0041] S4 includes: S41, constructing a multi-dimensional abnormal feature association matrix: constructing a multi-dimensional abnormal feature association matrix based on multiple features (such as temperature, salinity, flow velocity, tide and acoustic wave features) in the multi-source fusion data set. The specific steps include: Feature correlation analysis: The Pearson correlation coefficient is used to calculate the linear correlation between the features. The formula is as follows: ; in, is the correlation coefficient between feature i and feature j, is the measurement data of the i-th sensor at the k-th moment, is the mean of feature i, and n is the total number of data points.

[0042] By calculating the correlation, the relationship between the features is determined, and a symmetrical correlation matrix R is constructed. Each element in the matrix Represents the correlation between feature i and feature j; Construct a correlation matrix: Organize the correlation calculation results of multiple features into an m×m matrix, where m is the number of features. Each matrix element Represents the correlation between feature i and feature j; Feature weighting: assign a weight to each feature , according to the importance of the feature and its correlation with other features, the specific method is to use the average correlation of the features to determine the weight, the formula is as follows: ; in, is the weight of feature i, is the correlation between feature i and feature j; S42, using an adaptive threshold detection algorithm: using an adaptive threshold detection algorithm to analyze the data in the multi-dimensional abnormal feature association matrix and identify abnormal data nodes. The specific steps include: Adaptive threshold determination: For each feature data point , by calculating its mean and standard deviation , and determine the dynamic threshold based on the 3σ rule: ; in, is the threshold of feature i, is the adaptive factor, which adjusts the sensitivity of the threshold. Usually, The value is between 2 and 4; Abnormal determination: If the data point Exceeding the threshold , it is determined to be an abnormal data node, and the judgment conditions are as follows: ; Local anomaly detection: For local anomalies, the local weighted method is used for adjustment. Specifically, the outliers in the local area are weighted and the local mean is used. and standard deviation , Update threshold calculation: ; in, is the local threshold, is the local abnormal sensitivity factor, usually taken as 2.

[0043] S43, abnormal data node identification and output: After being processed by the adaptive threshold detection algorithm, all abnormal data nodes are identified and output. The output results include: Abnormal data node location and marking: Mark the specific location of each abnormal data node and associate it with the corresponding physical characteristics (such as temperature, salinity, etc.); Abnormal data type analysis: Classify the identified abnormal data and analyze its abnormal causes (such as equipment failure, environmental interference, etc.) to facilitate subsequent processing or optimization; S44, Anomaly detection result evaluation and optimization: Evaluate the abnormal data identification results. The evaluation criteria include: Anomaly detection accuracy: evaluates the accuracy of anomaly detection by comparing the error between the recognition results and the real data; False alarm rate and false negative rate: Calculate the false alarm rate and false negative rate to evaluate the sensitivity and specificity of the algorithm; Optimization and adjustment: If the evaluation results do not meet expectations, the detection results can be optimized by adjusting the parameters in the adaptive threshold algorithm or improving the construction method of the abnormal feature association matrix.

[0044] S5 includes: S51, define optimization objective function: determine the optimization objective function, which is used to measure the performance of each parameter in the optimization process. The objective function reflects the overall effect of data processing and fusion, and takes into account the adjustability of the parameters; Assume that the optimization objective function is ,in is the parameter set to be optimized, and the objective function is defined as a comprehensive indicator of fusion accuracy, data consistency, and anomaly detection accuracy in the multi-source fusion data set, specifically: Fusion accuracy Data consistency Anomaly detection accuracy ; Among them, the fusion accuracy It indicates the accuracy of the data after multi-source data fusion, data consistency (p) indicates the consistency between the fused data and the original data, and anomaly detection accuracy (p) indicates the recognition accuracy of abnormal data nodes. , , is the weighting coefficient, which controls the relative importance of each indicator. The optimization objective function can be obtained by data fitting, error minimization and other methods; S52, quantum annealing algorithm initialization: use quantum annealing algorithm to optimize parameters and initialize parameter set , and represent each parameter on a qubit The goal of quantum annealing is to gradually search for the optimal parameter combination through the evolution of the quantum bit state. The initialization steps are as follows: Initialization parameter set ,Each parameter value is randomly selected within a predetermined range. The search range of the parameter is set to ,in and are the minimum and maximum values ​​of the i-th parameter respectively.

[0045] In the quantum annealing algorithm, each parameter Mapped to the state of a quantum bit ,The initial state can be set as a uniform superposition state; ; in, and is the ground state of the quantum bit, is the superposition state of the quantum bit, indicating the random choice of the parameter between the minimum and maximum values; S53, quantum annealing evolution process: The quantum annealing evolution process continuously reduces the "temperature" of the system and gradually reaches the optimization goal by annealing the quantum bit state. The specific evolution process is as follows: Hamiltonian Design: Designing the Quantum Hamiltonian , the Hamiltonian and the objective function The Hamiltonian minimizes the objective function by guiding the qubits to find the lowest energy state in the parameter space. It can be expressed as: ; in, is the adjustment term, which controls the deviation between the objective function and the initial parameters. is the objective function, which represents the error or cost of optimization. As an additional constraint, ensure that the optimization process does not stray too far from the initial value.

[0046] Temperature scheduling: Gradually reduce the energy of the system through the "temperature" scheduling of quantum annealing. The temperature T of the quantum system gradually decreases over time. During the annealing process, the system transitions from a high temperature state (strong randomness) to a low temperature state (enhanced stability), and finally converges to the optimal solution; ; in, is the system temperature at time t, is the initial temperature, is the maximum temperature, t is the current time step; S54, quantum bit measurement and parameter update: After the quantum system has gone through the annealing process, each time the temperature drops to the lowest point, the quantum bit is measured to obtain the optimized parameter set , the specific steps are as follows: Quantum measurement: measure the quantum bits to obtain the optimized parameter set , that is, by measuring the state of the quantum bit To determine each parameter The optimal value of , the measurement of the quantum bit can be obtained by calculating the probability amplitude: ; in, is the probability of the measurement result, The quantum bit is in the ground state The probability of Parameter update: Update the optimization parameter set based on the measurement results , and input it as the new parameter value into the next round of optimization process; The optimization process continuously iterates and updates parameters through the quantum annealing algorithm until one of the following termination conditions is met: Optimization convergence: When the change of the objective function is less than the preset threshold ,Right now: ; in, and are the parameter sets for the current and previous rounds of optimization, respectively. is the convergence threshold; Maximum number of iterations: If the maximum number of iterations is reached , the optimization process is terminated; The optimization result is the generated optimization parameter set ,This parameter set can be used for subsequent feature fusion and data analysis tasks.

[0047] S6 includes: S61, input and initialization of optimization parameter set: The generated optimization parameter set is used as the basic parameters for constructing a three-dimensional visual ocean environment model. The optimization parameter set contains the optimal parameters after quantum annealing optimization on a multi-source fusion data set. These parameters are the core input for building the model and initialize the basic information required for the ocean environment model. The basic information includes the boundary range and gridding parameters of the ocean area, such as the boundary range (such as longitude, latitude, depth, etc.) and gridding parameters of the ocean area.

[0048] Assume that the latitude and longitude range of the ocean area is: ; The depth range is: ; S62, 3D grid construction: construct a 3D grid based on the boundary information of the ocean area, and the resolution of the grid depends on the gridding parameters in the optimization parameter set (such as the horizontal and vertical resolution of the grid). This step divides the ocean space into multiple small units to construct a discrete 3D coordinate system for subsequent data mapping and visualization; Grid division: Assume that the ocean area is divided into A grid, where is the lateral resolution, is the vertical resolution, is the resolution in the depth direction; Grid node calculation: Each grid node represents a location in an ocean region, where , , is the depth interval; S63, Dynamic mapping of physical characteristics and model parameters: According to the optimized parameter set, for each grid node The corresponding physical characteristic values ​​(such as temperature, salinity, flow velocity, tide, sound wave, etc.) are assigned. The dynamic mapping relationship between model parameters and physical characteristics is established based on the sensor data characteristics and their changing rules during data collection. The specific steps are as follows: Physical feature model: Based on the optimized parameters, the physical features are mapped to the 3D grid nodes using interpolation or regression algorithms, with temperature data , then according to the changes in space and time, an interpolation method (such as Kriging interpolation) is used to calculate the temperature value for each grid node, expressed as: ; in, is the temperature data of known node m, is the weight based on spatial distance, For grid nodes The temperature value at Similarly, other physical characteristics (salinity, flow velocity, etc.) can be mapped to each grid node through similar interpolation methods; S64, construction of a three-dimensional visual ocean environment model: based on the established grid and physical eigenvalues, a three-dimensional visual ocean environment model is constructed; The model maps the physical characteristics of each mesh node into three-dimensional space and generates visual images or animations for intuitive analysis. The specific steps include: Surface drawing: Use 3D drawing tools (such as OpenGL, Matplotlib 3D, etc.) to draw the surface of the ocean environment according to the physical characteristics of each grid node, showing the distribution of data such as temperature and salinity.

[0049] Volume rendering: Using volume rendering technology, various levels of 3D grids (such as different depths, different temperature layers, etc.) are superimposed to form a complete ocean environment model.

[0050] S65, visualization result display and analysis: The generated three-dimensional visualization ocean environment model is displayed to the user, supporting interactive operations. The user can adjust the viewing angle, zoom, rotate and other operations through the interface to view the ocean environment characteristics of different areas and depths.

[0051] Interactive interface design: Design a user-friendly interactive interface that allows users to select different ocean physical characteristics for display (such as temperature, salinity, current velocity, etc.) and supports observation from different perspectives.

[0052] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0053] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The method for intelligent fusion and optimization processing of ocean physical data is characterized by: The following steps are involved: S1: Standardize and preprocess the ocean physical data collected by multi-source sensors to generate a standardized matrix containing temperature, salinity, current velocity, tide and acoustic wave characteristics; S2: construct a spatiotemporal correction matrix based on a dynamic sliding window, perform spatiotemporal benchmark dynamic correction on the standardized matrix, and generate a spatiotemporal correction matrix; S3: Input the spatiotemporal correction matrix into the deep residual-graph convolution hybrid model for feature fusion and output a multi-source fusion dataset; S4: Construct a multi-dimensional abnormal feature association matrix based on multi-source fusion data sets, and use an adaptive threshold detection algorithm to identify abnormal data nodes; S5: Use the quantum annealing optimization algorithm to iterate the parameters of the multi-source fusion data set and generate an optimized parameter set; S6: Construct a three-dimensional visual ocean environment model based on the optimized parameter set, and establish a dynamic mapping relationship between model parameters and physical characteristics.

2. The method for intelligent fusion and optimization processing of ocean physical data according to claim 1 is characterized in that: The S1 includes: S11, data collection: by setting up multiple different types of ocean sensors, real-time collection of ocean physical data, including temperature, salinity, current velocity, tide and acoustic wave characteristic data; S12, data formatting: formatting the collected ocean physical data to form a data set that matches time and space; S13, missing value filling: For sensor data with missing or abnormal data, the missing values ​​are filled by interpolation or smoothing method based on adjacent data; S14, normalization processing: normalizing the collected ocean physical data; S15, feature extraction: extracting temporal features and spatial features from the standardized data; S16, constructing a standardized matrix: combining the normalized ocean physical data and the extracted eigenvalues ​​to form a standardized matrix.

3. The method for intelligent fusion and optimization processing of ocean physical data according to claim 2 is characterized in that: The S2 includes: S21, sliding window setting and data division: according to the acquisition frequency and spatial resolution of ocean physical data, the size and step length of the dynamic sliding window are set; The window size is adjusted according to the time span and spatial range; According to the set sliding window, the standardized matrix is ​​divided into multiple spatiotemporal sub-matrices according to the time and space dimensions, and each sub-matrix contains a data set within a time period and a preset spatial range; S22, spatiotemporal benchmark data selection: selecting representative benchmark data from multiple spatiotemporal sub-matrices as reference benchmarks for spatiotemporal correction; S23, spatiotemporal deviation analysis: analyzing the deviation of spatiotemporal data by comparing the data of each spatiotemporal submatrix with the benchmark data; S24, dynamic correction model construction: according to the results of spatiotemporal deviation analysis, a spatiotemporal correction model is established using statistical modeling methods; S25, spatiotemporal correction calculation: applying the spatiotemporal correction model to the data of each spatiotemporal submatrix, obtaining the correction parameters of each spatiotemporal window through model calculation, and performing spatiotemporal correction on the data of each submatrix; S26, generating a space-time correction matrix: merging the corrected multiple space-time sub-matrices to generate a complete space-time correction matrix; S27, correction result evaluation and optimization: perform quality evaluation on the generated spatiotemporal correction matrix.

4. The method for intelligent fusion and optimization processing of ocean physical data according to claim 3 is characterized in that: The S3 includes: S31, deep residual network module construction and feature extraction: construct a deep residual network module and extract the temporal features in the spatiotemporal correction matrix through the residual learning mechanism; S32, Graph Convolutional Network Module Construction and Spatial Feature Extraction: Construct a graph convolutional network module and use the graph convolution method to capture the spatial features in the spatiotemporal correction matrix; S33, weighted fusion and multi-source fusion dataset generation: weighted fusion of temporal features and spatial features extracted by the deep residual network module and graph convolutional network module to form a comprehensive feature representation; The comprehensive feature representation formed after fusion will be output as a multi-source fusion dataset, which contains the temporal and spatial features fused from multiple sensors; S34, fusion result evaluation and optimization: perform quality evaluation on the output multi-source fusion data set. The evaluation indicators include data accuracy, stability and fusion effect.

5. The method for intelligent fusion and optimization processing of ocean physical data according to claim 4 is characterized in that: The S4 includes: S41, constructing a multi-dimensional abnormal feature association matrix: constructing a multi-dimensional abnormal feature association matrix according to multiple features in the multi-source fusion data set; S42, using an adaptive threshold detection algorithm: using an adaptive threshold detection algorithm to analyze the data in the multi-dimensional abnormal feature association matrix and identify abnormal data nodes; S43, abnormal data node identification and output: after being processed by the adaptive threshold detection algorithm, all abnormal data nodes are identified and output; S44, anomaly detection result evaluation and optimization: evaluate the abnormal data identification results.

6. The method for intelligent fusion and optimization processing of ocean physical data according to claim 5 is characterized in that: The S5 includes: S51, define optimization objective function: determine the optimization objective function, which is used to measure the performance of each parameter in the optimization process; S52, quantum annealing algorithm initialization: Use the quantum annealing algorithm to optimize parameters, initialize the parameter set, and represent the initial state of each parameter on the quantum bit. The goal of quantum annealing is to gradually search for the optimal parameter combination through the evolution of the quantum bit state; S53, quantum annealing evolution process: The quantum annealing evolution process continuously lowers the "temperature" of the system and gradually reaches the optimization goal by annealing the quantum bit state; S54, measurement and parameter update of quantum bits: After the quantum system has undergone the annealing process, each time the temperature drops to the lowest, the quantum bits are measured to obtain the optimized parameter set.

7. The method for intelligent fusion and optimization processing of ocean physical data according to claim 6 is characterized in that: The S6 includes: S61, input and initialization of optimization parameter set: using the generated optimization parameter set as the basic parameters for constructing a three-dimensional visualized ocean environment model, and initializing the basic information required for the ocean environment model, the basic information including the boundary range of the ocean area and gridding parameters; S62, 3D grid construction: constructing a 3D grid according to the boundary information of the ocean area; S63, dynamic mapping of physical characteristics and model parameters: assigning corresponding physical characteristic values ​​to each grid node according to the optimized parameter set; S64, construction of a three-dimensional visual ocean environment model: constructing a three-dimensional visual ocean environment model based on the established grid and physical eigenvalues; S65, visualization result display and analysis: Display the generated 3D visualization ocean environment model to the user and support interactive operation.

Images