Model construction method and system applied to integrated analysis of marine water environment

Abstract

This invention provides a model construction method and system for integrated analysis of marine water environment, belonging to the field of marine environmental analysis technology. First, dynamic triggering rules for the correlation and evolution of marine water environment elements are generated. Based on these dynamic triggering rules, historical monitoring data undergoes multi-dimensional trajectory reconstruction to generate spatiotemporally coupled trajectories. A cross-dimensional feature collaboration module, comprising a spatiotemporal analysis layer, a feature derivation layer, and a state mapping layer, is then generated. A self-optimizing feedback loop is constructed to transmit the state mapping results back to adjust the feature extraction and generation rules. All parts are integrated to form an integrated marine water environment analysis model. Parameters are adjusted in real time for newly acquired data to dynamically adapt to different marine environmental conditions. This invention can accurately capture dynamic changes in elements, achieve multi-dimensional feature collaborative processing, possess self-optimization capabilities, and improve the accuracy of marine water environment analysis.

Description

Technical Field

[0001] This invention relates to the field of marine environmental analysis technology, and more specifically, to a model construction method and system for integrated analysis of marine water environment. Background Technology

[0002] In the field of marine water environment research, comprehensive, accurate, and dynamic analysis of the marine water environment is crucial, as it relates to several key aspects such as marine ecological protection, marine resource development and utilization, and marine disaster early warning. Currently, existing marine water environment analysis methods have many limitations.

[0003] On the one hand, traditional methods, when dealing with the correlations between marine environmental elements, often employ fixed rules and models, making it difficult to fully consider the complex spatiotemporal variations in monitoring data. The fluctuations of marine environmental elements exhibit significant spatiotemporal differences; the patterns of element change vary across different sea areas and time periods. Fixed rules cannot accurately capture these dynamic changes, leading to inaccurate analysis of the correlations between elements. On the other hand, existing technologies lack the ability to collaboratively process multi-dimensional features in marine environmental analysis. The marine environment involves numerous elements, such as water temperature, salinity, and dissolved oxygen, which have complex correlations that change over time and space. Traditional methods struggle to comprehensively extract and effectively coordinate the correlation characteristics of these elements from multiple dimensions, failing to establish an accurate mapping between element characteristics and the state of the marine environment, thus affecting the accurate judgment and prediction of the marine environmental state. Furthermore, existing models lack self-optimization mechanisms; when faced with constantly changing marine environmental data, they cannot adjust model parameters in real time to adapt to new data characteristics, leading to a gradual decrease in the model's applicability and accuracy. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a model construction method for integrated analysis of marine water environments, the method comprising: A dynamic triggering rule for the correlation and evolution of marine water environment elements is generated. The dynamic triggering rule includes the spatiotemporal stratification threshold of element monitoring data and the gradient conditions for initiating the correlation between elements. Based on dynamic triggering rules, the historical monitoring data of marine water environment elements are reconstructed in multiple dimensions to generate spatiotemporal coupling trajectories of element correlation evolution. The spatiotemporal coupling trajectories include the temporal changes in the correlation strength of elements at different sea area levels and the spatial diffusion path of the correlation range. A cross-dimensional feature collaboration module for generating spatiotemporal coupled trajectories is provided. This cross-dimensional feature collaboration module includes a spatiotemporal analysis layer, a feature derivation layer, and a state mapping layer. The spatiotemporal analysis layer decomposes the spatiotemporal coupled trajectory into spatiotemporal dimensions and extracts the initial coupling features. The feature derivation layer generates multi-dimensional associated derived features based on the initial coupling features. The state mapping layer establishes a dynamic mapping relationship with the water environment state through the multi-dimensional associated derived features. A self-optimizing feedback link for generating cross-dimensional feature collaboration modules is established. This self-optimizing feedback link transmits the water environment state mapping results output by the state mapping layer back to the feature derivation layer and the spatiotemporal analysis layer, dynamically adjusting the extraction dimensions of the initial coupled features and the generation rules of multi-dimensional associated derived features. By integrating dynamic triggering rules, spatiotemporal coupling trajectories, cross-dimensional feature collaboration modules, and self-optimizing feedback links, an integrated marine water environment analysis model is formed. The newly collected multi-ocean-area element monitoring data are processed through the integrated marine water environment analysis model. Based on the self-optimizing feedback link, the parameters of the cross-dimensional feature collaboration module are adjusted in real time to realize the dynamic adaptation of the integrated marine water environment analysis model to the water environment status of different sea areas, thus completing the construction of the integrated marine water environment analysis model.

[0005] Furthermore, embodiments of the present invention also provide a model building system for integrated analysis of marine water environments, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described model building method for integrated analysis of marine water environments by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions stored in a computer-readable storage medium, a processor of a model building system for integrated analysis of marine water environment reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the model building system for integrated analysis of marine water environment to execute the above-described model building method for integrated analysis of marine water environment.

[0007] Based on the above, by generating dynamic triggering rules for the evolution of marine environmental elements, setting spatiotemporal stratification thresholds based on the spatiotemporal characteristics of historical fluctuations of elements, and setting gradient conditions according to the gradual change law of the correlation strength between elements, the dynamic changes of marine environmental elements and the correlation initiation conditions under different spatiotemporal conditions can be captured. Then, based on the dynamic triggering rules, multi-dimensional trajectory reconstruction of historical monitoring data is performed. The generated spatiotemporal coupling trajectory can clearly show the temporal changes of the correlation strength of elements at different sea area levels and the spatial diffusion path of the correlation range, thus reflecting the correlation evolution process of marine environmental elements. The generation of the cross-dimensional feature collaboration module, through the collaborative work of the spatiotemporal analysis layer, feature derivation layer, and state mapping layer, realizes the multi-dimensional feature extraction, derivation, and dynamic mapping of spatiotemporal coupling trajectory with the water environment state, effectively solving the problem of multi-dimensional feature collaborative processing. The construction of the self-optimizing feedback link transmits the water environment state mapping results in reverse, dynamically adjusting the initial coupling feature extraction dimension and correlation derivation feature generation rules, enabling the model to have self-optimization capabilities. The integrated marine water environment analysis model formed by combining various components can adjust parameters in real time to dynamically adapt to different marine water environment conditions, greatly improving the model's analytical capabilities and accuracy in dealing with complex and ever-changing marine water environments. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the model construction method for integrated analysis of marine water environment provided in this embodiment of the invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of a model building system for integrated analysis of marine water environment provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a model construction method for integrated analysis of marine water environment provided in one embodiment of the present invention. The following is a detailed description of this model construction method for integrated analysis of marine water environment.

[0011] Step S110: Generate dynamic triggering rules for the correlation evolution of marine water environment elements. These dynamic triggering rules include the spatiotemporal stratification threshold of element monitoring data and the gradient conditions for initiating correlation between elements. The spatiotemporal stratification threshold is set according to the sea area level based on the spatiotemporal characteristics of historical fluctuations of elements, and the gradient conditions are set based on the gradual change law of correlation strength between elements.

[0012] Step S111: Collect real-time and historical monitoring data of physical, chemical and biological elements in the marine environment at different sea level. Different sea level includes nearshore waters, coastal waters and offshore waters. The monitoring data includes numerical records of each element in a continuous time series, the corresponding sea area location coordinates and marine environmental background information, including ocean current direction and seabed topography type.

[0013] When generating dynamic trigger rules, it is first necessary to comprehensively collect relevant monitoring data, including real-time and historical monitoring data. Physical elements include seawater temperature, salinity, current velocity, and current direction; chemical elements include dissolved oxygen, pH value, ammonia nitrogen content, nitrate content, and phosphate content; biological elements involve phytoplankton cell density, zooplankton abundance, and chlorophyll a concentration. The classification of different sea areas is based on their distance from land and ecological environment characteristics. Nearshore sea areas generally refer to the sea areas extending a certain distance from the coastline, while offshore sea areas are the areas extending outward from the nearshore sea areas, and open sea areas are the open sea areas further away from land. Monitoring data needs to be continuous, and the time series can be recorded hourly, daily, or weekly. The sea area location coordinates are represented by latitude and longitude, the ocean current direction in the marine environmental background information is obtained through long-term observation by marine monitoring stations, and the seabed topography type is determined through seabed topography survey data.

[0014] Step S112: Divide the real-time monitoring data of each sea area level into segments according to the time series, extract the fluctuation range and fluctuation frequency of the element values ​​in each segment, calculate the difference coefficient of the fluctuation range of the elements between the current sea area level and other sea area levels, and calculate the difference coefficient of the fluctuation frequency of the elements between the current sea area level and other sea area levels.

[0015] Real-time monitoring data for each sea area level is segmented according to time series, with segment length determined based on the time scale of element changes. For example, shorter time segments are used for rapidly changing biological elements, while longer time segments are used for slowly changing physical elements. Within each segment, the fluctuation range of the element's value is extracted, which is the difference between the maximum and minimum values ​​within that segment. The fluctuation frequency refers to the number of times the element's value completes a full fluctuation from minimum to maximum and back to minimum within that segment. The difference coefficient of the element's fluctuation range between the current sea area level and other sea area levels is calculated using the formula: |Fluctuation Range_Current - Fluctuation Range_Other| / (Fluctuation Range_Current + Fluctuation Range_Other). The difference coefficient of the fluctuation frequency is calculated using the same method.

[0016] Step S113: Based on the difference coefficient of fluctuation range, set the initial value of spatial stratification threshold for element monitoring data according to sea area level, and based on the difference coefficient of fluctuation frequency, set the initial value of temporal stratification threshold for element monitoring data according to time series.

[0017] Based on the calculated fluctuation range difference coefficient, initial values ​​for spatial stratification thresholds are set according to sea area level. Taking nearshore waters as the benchmark, an initial spatial benchmark threshold is set; for nearshore waters, the initial spatial stratification threshold = initial nearshore spatial benchmark threshold × (1 + fluctuation range difference coefficient - nearshore - nearshore); for offshore waters, the initial spatial stratification threshold = initial nearshore spatial stratification threshold × (1 + fluctuation range difference coefficient - offshore - nearshore). The initial values ​​for temporal stratification thresholds are set using a similar method, taking a certain time segment as the benchmark and setting initial stratification threshold values ​​for other time segments based on the fluctuation frequency difference coefficient.

[0018] Step S114: Integrate the initial values ​​of the spatial stratification threshold and the initial values ​​of the temporal stratification threshold to form the initial values ​​of the spatiotemporal stratification threshold for the element monitoring data.

[0019] The initial spatial stratification threshold for each sea area level is combined with the initial temporal stratification threshold for each time segment of that sea area level to form the initial spatiotemporal stratification threshold for element monitoring data under specific spatiotemporal conditions. For example, the initial spatiotemporal stratification threshold for nearshore sea areas in the first quarter is a combination of the initial spatial stratification threshold for nearshore sea areas and the initial temporal stratification threshold for the first quarter.

[0020] Step S115: Extract historical correlation strength data of physical and chemical elements, chemical and biological elements, and physical and biological elements within different sea area levels, and arrange them in chronological order to form a correlation strength sequence.

[0021] The correlation strength data among various elements at different sea area levels are extracted from historical monitoring data. For physical and chemical elements, the correlation strength is determined by calculating their correlation coefficients within the same time series; the correlation coefficients are also calculated for chemical and biological elements, and physical and biological elements. The calculated correlation strength data are arranged sequentially in chronological order to form a correlation strength sequence.

[0022] Step S116: Analyze the correlation strength difference between adjacent time points in the correlation strength sequence, statistically analyze the distribution pattern of the correlation strength difference, and determine multiple gradual change stages from the initial value to the significant correlation value. The significant correlation value is an empirical threshold that can reflect the clear interaction between elements.

[0023] The difference in association strength between adjacent time points in the association strength sequence is calculated to obtain an association strength difference sequence. The mean, variance, quantiles, and other distribution characteristics of this difference sequence are statistically analyzed. Based on these distribution characteristics, multiple gradual change stages are identified as the association strength increases from its initial value to a significant association value. The significant association value is an empirical threshold determined through analysis of a large amount of historical data; when the association strength reaches this value, it indicates a clear interaction between the elements. Each gradual change stage corresponds to a characteristic interval in the association strength growth process.

[0024] Step S117: For each gradual change stage, set a stage threshold for the correlation strength. The difference between the stage thresholds of adjacent stages is distributed in a gradient according to the gradual change law, forming the initial value of the gradient condition for the initiation of the correlation between elements.

[0025] For each identified gradual change stage, a corresponding stage threshold is set. The difference in stage thresholds between adjacent stages follows a gradient distribution according to the gradual change pattern. For example, in the initial stage of increasing correlation strength, the difference in stage thresholds is small, and as the correlation strength increases, the difference in stage thresholds gradually increases. These stage thresholds together constitute the initial values ​​of the gradient conditions for initiating correlations between elements.

[0026] Step S118: Integrate the initial value of the spatiotemporal layering threshold with the initial value of the gradient condition to generate the initial version of the dynamic triggering rule.

[0027] The initial values ​​of the spatiotemporal stratification threshold and gradient conditions are integrated to clarify their combination logic. For example, when feature monitoring data simultaneously meets both the spatiotemporal stratification threshold and gradient conditions, feature correlation evolution analysis is triggered. Through this integration, an initial version of the dynamic triggering rule is generated.

[0028] Step S119: Set the adaptive update cycle of the dynamic triggering rule. When each update cycle arrives, acquire the newly collected multi-sea area element monitoring data within that cycle. Based on the newly collected data, recalculate the fluctuation range, fluctuation frequency and correlation strength between elements at each sea area level, update the spatiotemporal stratification threshold and gradient conditions, and form an updated version of the dynamic triggering rule, so that the dynamic triggering rule continues to evolve with the new data.

[0029] An adaptive update cycle is set for the dynamic triggering rules. This cycle is determined based on the severity of marine environmental changes; for example, it is set to update monthly in areas with frequent seasonal changes, and quarterly in areas with relatively stable hydrological conditions. At the end of each update cycle, newly collected multi-ocean-area monitoring data is acquired. Based on the newly collected data, the fluctuation range, fluctuation frequency, and inter-element correlation strength of each ocean-area level are recalculated. A weighted average method is used to update the spatiotemporal stratification thresholds and gradient conditions, forming an updated version of the dynamic triggering rules. Through this method, the dynamic triggering rules can continuously evolve with the addition of new data, adapting to new time types and new feature change patterns.

[0030] Step S1191: When each update cycle arrives, acquire the newly collected monitoring data of nearshore waters, offshore waters and offshore waters in that cycle to form the real-time monitoring dataset for the current cycle.

[0031] At the start of each update cycle, the system retrieves all newly collected monitoring data for that cycle from the multi-ocean-area data access module. This includes continuous time-series records of physical, chemical, and biological elements in nearshore, offshore, and open-ocean waters, which are then compiled into a real-time monitoring dataset for the current cycle.

[0032] Step S1192: Clean the real-time monitoring dataset for the current period, remove abnormal data points caused by sensor failure or communication anomalies, and obtain effective real-time monitoring data.

[0033] Data cleaning was performed on the real-time monitoring dataset. For missing values ​​in the detection data, those with consecutive missing durations not exceeding a set threshold were filled using linear interpolation; those with consecutive missing durations exceeding the threshold were marked as invalid. Outliers in the detection data were identified using box plots, identifying data points exceeding a certain multiple of the interquartile range as outliers. Combined with sensor operating status records, if an outlier was confirmed to be caused by sensor malfunction or communication failure, it was removed; if an outlier might reflect actual abrupt changes in the marine environment, it was retained and specially marked. This cleaning process yielded effective real-time monitoring data.

[0034] Step S1193: Group the effective real-time monitoring data of each sea area level according to the element type to form real-time datasets of physical elements, chemical elements, and biological elements.

[0035] The cleaned, effective real-time monitoring data were categorized according to sea area level and element type. For each sea area level, the monitoring data of physical elements were integrated into a real-time dataset of physical elements, the monitoring data of chemical elements were integrated into a real-time dataset of chemical elements, and the monitoring data of biological elements were integrated into a real-time dataset of biological elements.

[0036] Step S1194: For the real-time dataset of physical elements at each sea area level, divide it into segments according to the time series, calculate the fluctuation range and fluctuation frequency of the physical element values ​​in each segment, and obtain the real-time values ​​of the fluctuation range and fluctuation frequency of the physical elements at that sea area level.

[0037] For the real-time dataset of physical elements at each sea area level, the data is segmented according to the time series, with the segment length adaptively determined based on the time scale of element changes. For each time period, the difference between the maximum and minimum values ​​of the physical element within that period is calculated to obtain the real-time fluctuation range; the number of fluctuations in the element value within that period is counted and divided by the time period length to obtain the real-time fluctuation frequency. Using the same method, the real-time fluctuation range and fluctuation frequency values ​​for chemical and biological elements are calculated separately.

[0038] Step S1195: Based on the real-time values ​​of the fluctuation range of physical elements at the current sea level and the real-time values ​​of the fluctuation range of physical elements at other sea levels, calculate the real-time values ​​of the difference coefficients of the fluctuation range of physical elements; using the same method, calculate the real-time values ​​of the difference coefficients of the fluctuation range of chemical elements, the real-time values ​​of the difference coefficients of the fluctuation range of biological elements, and the real-time values ​​of the difference coefficients of the fluctuation frequency of each element.

[0039] For physical elements, calculate the difference coefficient between the real-time fluctuation range value of the current sea area level and the real-time fluctuation range values ​​of other sea area levels, using the formula: |Fluctuation Range_Current - Fluctuation Range_Other| / (Fluctuation Range_Current + Fluctuation Range_Other). Using the same method, calculate the real-time difference coefficients for the fluctuation ranges of chemical elements and biological elements. For fluctuation frequencies, use the same formula to calculate the real-time difference coefficients for the fluctuation frequencies of each element.

[0040] Step S1196: Based on the real-time value of the difference coefficient of the fluctuation range, update the spatial stratification threshold of the element monitoring data according to the sea area level. The update formula is: new spatial stratification threshold = α × historical spatial stratification threshold + (1-α) × candidate value of spatial stratification threshold calculated based on the real-time value of the difference coefficient, where α is the historical weight coefficient, which is dynamically adjusted according to the stability of marine environmental changes. The more drastic the environmental changes, the smaller the value of α.

[0041] The spatial stratification threshold is updated using a weighted average method. The historical weight coefficient α is dynamically adjusted based on the stability of marine environmental changes; the more drastic the environmental changes, the lower the reference value of historical data, and the smaller the value of α. α is calculated as: α = 1 / (1 + β × ΔE), where ΔE is the comprehensive index of environmental changes between the current and previous periods, calculated by weighting the fluctuation rates of multiple factors, and β is the adjustment coefficient. The method for calculating candidate values ​​for the spatial stratification threshold based on real-time differences is as follows: taking nearshore waters as the benchmark, the candidate values ​​for the spatial stratification threshold are set based on historical benchmark values; for nearshore waters, the candidate value for the spatial stratification threshold = candidate value for the spatial stratification threshold of nearshore waters × (1 + fluctuation range difference coefficient - nearshore - nearshore); for offshore waters, the candidate value for the spatial stratification threshold = candidate value for the spatial stratification threshold of nearshore waters × (1 + fluctuation range difference coefficient - offshore - nearshore). The time stratification threshold is updated using the same method.

[0042] Step S1197: Based on the real-time value of the difference coefficient of the fluctuation frequency, update the time stratification threshold of the element monitoring data according to the time series, and adopt the same weighted update method as the spatial stratification threshold.

[0043] The time stratification threshold is updated using the same weighted average method as the spatial stratification threshold. Based on the real-time value of the fluctuation frequency difference coefficient, combined with the historical time stratification threshold and the fluctuation frequency characteristics observed in the current period, the updated time stratification threshold is calculated.

[0044] Step S1198: Integrate the updated spatial stratification threshold with the temporal stratification threshold to form the updated spatiotemporal stratification threshold.

[0045] The updated spatial stratification threshold for each sea area level is combined with the updated temporal stratification threshold for each time segment of that sea area level to form the updated spatiotemporal stratification threshold.

[0046] Step S1199: Extract the correlation strength between physical and chemical elements in the real-time monitoring data of the current period, and calculate the real-time sequence of the correlation strength between physical and chemical elements; use the same method to calculate the real-time sequence of the correlation strength between chemical and biological elements and the real-time sequence of the correlation strength between physical and biological elements.

[0047] For each sea area level, monitoring data of the same time series are extracted from real-time datasets of physical and chemical elements, and the correlation strength is calculated using a sliding window correlation analysis method. A sliding window length is set, and the Pearson correlation coefficient between physical and chemical elements is calculated within each window, serving as the correlation strength value at the center of that window. The sliding window is moved point by point to obtain the real-time series of physical-chemical element correlation strength. The real-time series of chemical-biological element correlation strength and the real-time series of physical-biological element correlation strength are calculated using the same method.

[0048] Step S11910: Analyze the correlation strength difference between adjacent time points in the real-time sequence of each correlation strength, statistically analyze the distribution pattern of the difference, and identify the performance characteristics of multiple gradual change stages in the current period when the correlation strength rises from the initial value to the significant correlation value.

[0049] For each real-time correlation strength sequence, the difference in correlation strength between adjacent time points is calculated to form a difference sequence. The distribution characteristics of the difference sequence, such as mean, variance, and quantiles, are statistically analyzed. Based on these distribution characteristics, the performance characteristics of the gradual change stages in the correlation strength from the initial value to the significant correlation value within the current period are identified, and the correlation strength range, stage duration, and transition characteristics between stages are recorded for each stage.

[0050] Step S11911: For each identified progressive change stage, update the stage threshold of the association strength so that the difference in stage thresholds between adjacent stages is distributed in a gradient according to the progressive change pattern observed in the current period, thus forming the updated gradient condition.

[0051] Based on the identified characteristics of the progressively changing stages, the association strength threshold for each stage is updated using a weighted average method that combines historical stage thresholds with the stage boundary values ​​observed in the current period. The updated stage thresholds should ensure that the threshold differences between adjacent stages exhibit a gradient distribution, matching the observed growth rate of association strength within the current period, thus forming the updated gradient conditions.

[0052] Step S11912: Integrate the updated spatiotemporal stratification threshold with the updated gradient conditions to form a new version of the dynamic triggering rule in the current update cycle; compare the new version of the dynamic triggering rule with the previous version of the dynamic triggering rule, calculate the rate of change of the spatiotemporal stratification threshold and the rate of change of the gradient conditions; if the rate of change exceeds the preset rule stability threshold, trigger a rule update alarm to indicate that the marine environment has changed significantly.

[0053] The updated spatiotemporal stratification threshold and gradient conditions are integrated to form a new version of the dynamically triggered rule for the current update cycle. The new version rule is compared with the previous version, and the rates of change of the spatiotemporal stratification threshold and gradient conditions are calculated. If the rates of change exceed a preset rule stability threshold, a rule update alarm is triggered, indicating a significant change in the marine environment; if the rates do not exceed the threshold, the new version rule is stored in the core control module for data processing in the next cycle.

[0054] Step S120: Based on dynamic triggering rules, multi-dimensional trajectory reconstruction is performed on historical monitoring data of marine water environment elements to generate spatiotemporal coupling trajectories of element correlation evolution. The spatiotemporal coupling trajectories include the temporal changes in correlation intensity and spatial diffusion path of correlation range of elements at different sea area levels.

[0055] After generating dynamic triggering rules, these rules are used to process historical monitoring data of marine environmental elements. Through multi-dimensional trajectory reconstruction, a spatiotemporal coupled trajectory of element correlation evolution is obtained. This spatiotemporal coupled trajectory can comprehensively reflect the changes in element correlation in time and space.

[0056] Step S121: Filter the real-time monitoring data of different sea areas according to the spatiotemporal stratification threshold in the current version of the dynamic triggering rules, and retain the monitoring data whose numerical fluctuations exceed the spatiotemporal stratification threshold of the corresponding sea area level to obtain the filtered monitoring data.

[0057] Based on the spatiotemporal stratification thresholds defined in the current version of the dynamic triggering rules, real-time monitoring data for different sea area levels are filtered. For each sea area level, the element values ​​at each time point are checked one by one to determine whether they exceed the spatiotemporal stratification threshold for that sea area level in the corresponding time segment. If they exceed the threshold, the monitoring data is retained; otherwise, it is discarded. The filtered monitoring data obtained through this process effectively reflects situations where element values ​​fluctuate significantly.

[0058] Step S122: Group the screened monitoring data according to the element type to form physical element screening dataset, chemical element screening dataset and biological element screening dataset.

[0059] The filtered monitoring data were grouped according to element type. Monitoring data belonging to physical elements were integrated into a physical element screening dataset, monitoring data belonging to chemical elements were integrated into a chemical element screening dataset, and monitoring data belonging to biological elements were integrated into a biological element screening dataset.

[0060] Step S123: For each sea area level, extract monitoring data of the same time series from the physical element screening dataset and the chemical element screening dataset. Based on the gradient conditions in the current version of the dynamic triggering rule, calculate the correlation strength of the two elements in the time series and generate the time series curve of the physical-chemical element correlation strength of the sea area level.

[0061] For each sea area level, monitoring data of the same time series are extracted from both the physical element screening dataset and the chemical element screening dataset. Based on the gradient conditions in the current version of the dynamic triggering rule, correlation analysis is used to calculate the correlation strength between physical and chemical elements within the time series. The calculated correlation strengths are arranged in chronological order to generate the time series curve of the physical-chemical element correlation strength for that sea area level.

[0062] Step S124: Using the same method, calculate the time series curves of the correlation strength between chemical and biological elements and the correlation strength between physical and biological elements at each sea area level.

[0063] Using the same method as for calculating the time series curves of the correlation strength between physical and chemical elements, the time series curves of the correlation strength between chemical and biological elements and the time series curves of the correlation strength between physical and biological elements were calculated for each sea area level.

[0064] Step S125: Align the time series curves of each correlation intensity at the same sea area level according to the time axis, extract the peak points, valley points and inflection points of each curve, determine the synchronous change characteristics of each correlation intensity time series curve in the time dimension, the synchronous change characteristics include the time difference of the peak point and the consistency of the inflection point change trend, and generate the correlation intensity time series change map of the sea area level.

[0065] Align the time-series curves of the correlation intensity of physical-chemical elements, chemical-biological elements, and physical-biological elements at the same sea area level along the time axis. Extract peak points, trough points, and inflection points from each curve. Compare the occurrence times of the peak points of each curve and calculate the time difference; analyze the changing trends of the inflection points of each curve to determine whether they are consistent. Generate a time-series variation map of the correlation intensity at this sea area level based on the analysis results.

[0066] Step S126: After screening the monitoring data for each sea area level, extract the sea area location coordinates corresponding to the element monitoring data, and construct the spatial distribution probability field of the element. The spatial distribution probability field is generated by the kernel density estimation method, and the probability value of each spatial location point represents the possibility of the element appearing at that location.

[0067] For each sea area level of filtered monitoring data, the sea area location coordinates corresponding to each monitoring data point are extracted to form a spatial location point set of the element at that time. The kernel density estimation method is used to estimate the probability density of the point set, generating a spatial distribution probability field. The kernel density estimation formula is f(x)=(1 / (nh))∑K((x-xi) / h), where f(x) is the probability density at location x, n is the number of sample points, h is the bandwidth parameter, and K is the kernel function. Kernel density estimation transforms discrete location points into a continuous probability field, where the probability value of each spatial location point represents the likelihood of the element appearing at that location.

[0068] Step S127: Based on the spatial distribution probability fields of adjacent time series, calculate the bulldozer distance between the probability fields. The bulldozer distance measures the degree of difference between the two probability distributions and reflects the overall change in the spatial distribution of elements between time series.

[0069] Bulldozer distance is used to measure the overall change in the spatial distribution of elements between adjacent time series. Let P and Q be the spatial probability fields of adjacent times t and t+1. Discretize P and Q into probability values ​​at grid points. The bulldozer distance is obtained by solving a linear programming problem: EMD(P,Q)=min∑∑f_ij×d_ij, satisfying f_ij≥0, ∑_jf_ij=p_i, ∑_if_ij=q_j, where f_ij is the flow rate from grid i of P to grid j of Q, d_ij is the spatial distance between grid i and j, and p_i and q_j are the probability values ​​of P and Q at grid i and j, respectively. The obtained EMD value is the bulldozer distance between the two probability fields.

[0070] Step S128: Based on the displacement vector of the centroid of the probability field, determine the main spatial diffusion direction of the element association range. The centroid displacement vector is obtained by calculating the weighted average of all position coordinates in the probability field with their probability values ​​as weights. Based on the ratio of the bulldozer distance to the corresponding time series duration, obtain the spatial diffusion intensity index of the element association range and generate the spatial diffusion curve of the association range at this sea area level.

[0071] For the spatial probability field, calculate its centroid position: C = (∑w_i × x_i) / ∑w_i, where w_i is the probability value of grid point i, and x_i is the spatial coordinate vector of grid point i. For the probability fields P and Q of adjacent time series, calculate the displacement vector ΔC = C_Q - C_P from the centroid to C_Q. The direction angle of this vector is the main spatial diffusion direction of the element association range. The spatial diffusion intensity index is defined as D_intensity = EMD(P,Q) / Δt, where Δt is the time interval between adjacent time series. Arrange the spatial diffusion intensity indices of different time series in chronological order to generate the spatial diffusion curve of the association range at this sea area level.

[0072] Step S129: Couple the temporal variation map of the correlation intensity of different sea area levels with the spatial diffusion curve of the correlation range. Based on the spatial positional relationship between sea area levels, adjust the correlation intensity value and spatial diffusion intensity of each sea area level so that the coupled trajectory reflects the element correlation and transmission characteristics between sea area levels.

[0073] Step S1291: Extract spatial location relationship data for different sea area levels. The spatial location relationship data includes the boundary distance between nearshore sea areas and offshore sea areas, the boundary distance between offshore sea areas and open sea areas, and hydrological data such as water flow velocity, water flow direction, and turbulence diffusion coefficient between each sea area level.

[0074] Spatial location relationship data at different sea level are extracted, including the boundary distance between nearshore and offshore sea areas, the boundary distance between offshore and open sea areas, and hydrological data such as water flow velocity, water flow direction, and turbulence diffusion coefficient between each sea level, to provide a basis for subsequent correlation and transfer calculations.

[0075] Step S1292: Calculate the correlation transfer coefficient between adjacent sea areas. The correlation transfer coefficient is determined based on the water exchange efficiency and turbulent mixing intensity between the two sea areas. The water exchange efficiency is calculated using flow monitoring data at the sea area boundary, and the turbulent mixing intensity is estimated using the turbulent kinetic energy dissipation rate output by the ocean numerical model.

[0076] The correlation transfer coefficient at the adjacent sea area level was calculated. Water exchange efficiency was calculated using flow monitoring data at the sea area boundary, and turbulent mixing intensity was estimated using the turbulent kinetic energy dissipation rate output from the ocean numerical model. The correlation transfer coefficient is positively correlated with both water exchange efficiency and turbulent mixing intensity, and its specific value was determined by establishing empirical relationships.

[0077] Step S1293: Extract the correlation intensity values ​​at each time point from the temporal variation map of the correlation intensity of the nearshore sea area, multiply them by the correlation transmission coefficient between the nearshore and offshore sea areas to obtain the correlation intensity correction value transmitted to the offshore sea area, and adjust the corresponding time point values ​​of the temporal variation map of the correlation intensity of the offshore sea area based on the correlation intensity correction value.

[0078] The correlation strength values ​​E_near(t) for each time node are extracted from the temporal variation map of the correlation strength in nearshore waters. Multiplying these values ​​by the nearshore-offshore correlation transmission coefficient C_trans_near_offshore yields the corrected correlation strength value E_trans(t) = E_near(t) × C_trans_near_offshore. A weighted average is then performed between the original value E_offshore_orig(t) and the corrected value E_trans(t) for the corresponding time node in the temporal variation map of the correlation strength in nearshore waters, resulting in the adjusted correlation strength value E_offshore_adj(t) = w_orig × E_offshore_orig(t) + w_trans × E_trans(t), where w_orig + w_trans = 1.

[0079] Step S1294: Using the same method, multiply the adjusted correlation strength value of the nearshore sea area by the correlation transmission coefficient between the nearshore and offshore sea areas to obtain the correlation strength correction value transmitted to the offshore sea area, and adjust the corresponding time node values ​​of the temporal change map of the correlation strength of the offshore sea area.

[0080] Multiplying the adjusted correlation strength value E_offshore_adj(t) from the nearshore area by the correlation transmission coefficient C_trans_offshore_far from the nearshore to the offshore area yields the corrected correlation strength value E_trans_far(t) = E_offshore_adj(t) × C_trans_offshore_far. Then, weighted averaging and merging the original value E_far_orig(t) and the corrected value E_trans_far(t) at the corresponding time point in the temporal variation map of the offshore correlation strength yields the adjusted correlation strength value E_far_adj(t) for the offshore area.

[0081] Step S1295: Extract the spatial diffusion intensity index for each time node of the spatial diffusion curve of the nearshore sea area association range. Based on the water flow direction and turbulence diffusion coefficient of the nearshore-oceanic sea area, adjust the diffusion direction probability distribution to make the diffusion direction probability distribution consistent with the water flow direction distribution. At the same time, multiply the diffusion intensity index by the association transmission coefficient to obtain the initial diffusion intensity of the nearshore sea area association range, which is used to adjust the initial segment value of the spatial diffusion curve of the nearshore sea area association range.

[0082] The spatial diffusion intensity index D_intensity_near(t) for each time node is extracted from the spatial diffusion curve of the nearshore waters. The current direction distribution and turbulence diffusion coefficient K_turb in the nearshore-offshore waters are obtained. The diffusion direction is expanded from a single centroid displacement direction to a probability distribution of diffusion directions consistent with the current direction distribution. The spatial diffusion intensity index is multiplied by the correlation transfer coefficient to obtain the initial diffusion intensity D_initial_offshore(t) = D_intensity_near(t) × C_trans_near_offshore, which is used to adjust the initial segment values ​​of the spatial diffusion curve of the nearshore waters.

[0083] Step S1296: Based on the adjusted spatial diffusion curve of the associated range in the nearshore waters, adjust the initial segment value of the spatial diffusion curve of the associated range in the farshore waters in the same way to make the diffusion of the associated range between the sea areas continuous.

[0084] Based on the spatially diffused curve of the adjusted correlation range in the offshore area, the spatial diffusion intensity index D_intensity_offshore_adj(t) is extracted. Combining the water flow direction distribution and turbulent diffusion coefficient in the offshore-far sea areas, the probability distribution of the diffusion direction is adjusted. Multiply D_intensity_offshore_adj(t) by the correlation transfer coefficient to obtain the initial diffusion intensity D_initial_far(t) of the correlation range in the far sea area, D_initial_far(t)=D_intensity_offshore_adj(t)×C_trans_offshore_far, which is used to adjust the numerical value of the starting segment of the spatially diffused curve of the correlation range in the far sea area.

[0085] Step S1297: Extract the peak point times of the temporal variation spectra of the correlation intensities at each hierarchical sea area after coupling, calculate the difference between the peak point times of adjacent hierarchical sea areas, and compare this difference with the theoretical time range for the transfer of elements between hierarchical sea areas, which is calculated based on the boundary distance, water flow velocity range, and turbulent diffusion time scale. If the difference exceeds the theoretical time range, readjust the correlation transfer coefficient until the difference falls within the theoretical time range.

[0086] Extract the peak point times T_peak_near, T_peak_offshore, and T_peak_far from the temporal variation spectra of the correlation intensities at each hierarchical sea area after coupling and adjustment. Calculate the differences ΔT_near_offshore and ΔT_offshore_far between the peak point times of adjacent hierarchical sea areas. Calculate the theoretical time range [T_min, T_max], where the lower limit is based on the minimum boundary distance divided by the maximum water flow velocity, and the upper limit is based on the maximum boundary distance divided by the minimum water flow velocity plus the turbulent diffusion time scale. Compare ΔT with the theoretical time range. If ΔT < T_min, decrease the correlation transfer coefficient; if ΔT > T_max, increase the correlation transfer coefficient, and repeat the adjustment until ΔT falls within the theoretical time range.

[0087] Step S1298: Extract the diffusion intensity sequence of the spatially diffused curve of the correlation range at each hierarchical sea area after coupling, calculate the cross-correlation function of the diffusion intensity sequences of adjacent hierarchical sea areas, determine the delay time corresponding to the maximum cross-correlation, and compare this delay time with the theoretical time range for the transfer of elements between hierarchical sea areas. If the delay time exceeds the theoretical time range, adjust the transfer relationship of the spatial diffusion intensity index until the delay time conforms to the theoretical time range.

[0088] Extract the diffusion intensity sequences D_intensity_near_adj(t), D_intensity_offshore_adj(t), and D_intensity_far_adj(t) from the spatial diffusion curves of the hierarchical correlation ranges in each sea area after coupling adjustment. Calculate the cross-correlation function of the adjacent sea area hierarchical diffusion intensity sequences to determine the time delay τ_max when the cross-correlation function reaches the maximum value. Compare τ_max with the theoretical time range [T_min, T_max]. If τ_max < T_min, reduce the diffusion intensity transfer gain. If τ_max > T_max, increase the diffusion intensity transfer gain. Recalculate after adjustment until τ_max falls within the theoretical time range.

[0089] Step S1299: Integrate the time-series change maps of the correlation intensities and the spatial diffusion curves of the correlation ranges in each sea area after adjustment to form a spatio-temporal coupling trajectory, which includes the correlation transfer path, transfer time range, and transfer intensity probability distribution between sea area hierarchies.

[0090] Integrate the time-series change maps of the correlation intensities and the spatial diffusion curves of the correlation ranges of the nearshore sea area, offshore sea area, and far-sea area after coupling adjustment, and add the correlation relationship data between sea area hierarchies, including the correlation transfer path, transfer time range, and transfer intensity probability distribution, to form a complete spatio-temporal coupling trajectory.

[0091] Step S130: Generate a cross-dimensional feature collaboration module for the spatio-temporal coupling trajectory. This cross-dimensional feature collaboration module includes a spatio-temporal analysis layer, a feature derivation layer, and a state mapping layer. The spatio-temporal analysis layer disassembles the spatio-temporal coupling trajectory in spatio-temporal dimensions and extracts initial coupling features. The feature derivation layer generates multi-dimensional correlation-derived features based on the initial coupling features. The state mapping layer establishes a dynamic mapping relationship with the water environment state through the multi-dimensional correlation-derived features.

[0092] After obtaining the spatio-temporal coupling trajectory, a cross-dimensional feature collaboration module capable of processing and analyzing this trajectory needs to be generated. This cross-dimensional feature collaboration module extracts features from the spatio-temporal coupling trajectory through different levels of processing and establishes a mapping relationship with the water environment state.

[0093] Step S131: Set a trajectory disassembly unit for the spatio-temporal analysis layer. This trajectory disassembly unit receives the spatio-temporal coupling trajectory and disassembles the spatio-temporal coupling trajectory into nearshore sea area sub-trajectories, offshore sea area sub-trajectories, and far-sea area sub-trajectories according to sea area hierarchies. Each sub-trajectory contains the time-series change map of the correlation intensity and the spatial diffusion curve of the corresponding sea area hierarchy.

[0094] The spatiotemporal analysis layer is the first layer of the cross-dimensional feature collaboration module, and a trajectory decomposition unit is first set up for it. The main function of this trajectory decomposition unit is to receive the input spatiotemporally coupled trajectory and decompose it according to the sea area level. Specifically, it separates the nearshore sea area, offshore sea area, and offshore sea area sub-trajectories from the spatiotemporally coupled trajectory. Each sub-trajectory contains the temporal variation map of the association strength and the spatial diffusion curve of the association range for the corresponding sea area level. This decomposition allows subsequent feature extraction to be performed separately for different sea area levels, improving the targeting and accuracy of feature extraction.

[0095] Step S132: Set up a dimension extraction unit for the spatiotemporal analysis layer. This dimension extraction unit extracts features in the time and space dimensions for each sub-trajectory. In the time dimension, it extracts the peak duration, valley interval duration and inflection point change rate of the correlation strength temporal change spectrum. In the space dimension, it extracts the diffusion direction change frequency, diffusion rate fluctuation amplitude and spatial coverage area growth rate of the correlation range spatial diffusion curve. The extracted features are integrated into the initial coupling features.

[0096] Following the trajectory decomposition unit, a dimension extraction unit is set up for the spatiotemporal analysis layer. This dimension extraction unit extracts features in both the temporal and spatial dimensions for each sub-trajectory. In the temporal dimension, it extracts the peak duration (the duration of the association strength near a peak point), valley interval duration (the time interval between two adjacent valley points), and inflection point rate of change (the rate of change of association strength at an inflection point, calculated as the ratio of the change in association strength to time within a set time period before and after the inflection point). In the spatial dimension, it extracts the diffusion direction change frequency (the number of times the diffusion direction changes per unit time), diffusion rate fluctuation amplitude (the difference between the maximum and minimum diffusion rates), and spatial coverage area growth rate (the ratio of the increase in spatial coverage area per unit time to the initial area) from the spatial diffusion curve of the association range. These features extracted from the temporal and spatial dimensions are integrated to form the initial coupling features.

[0097] Step S133: Set up a cross-sea area feature association unit for the feature derivation layer. The cross-sea area feature association unit receives the initial coupling features of each sea area level. Based on the association transmission coefficient between sea area levels, it extracts the common features of the initial coupling features of different sea area levels. The common features include the similarity of diffusion rate fluctuation amplitude in the initial coupling features of nearshore sea areas and offshore sea areas, and the consistency of inflection point change rate in the initial coupling features of nearshore sea areas and offshore sea areas.

[0098] The feature derivation layer is the processing layer following the spatiotemporal analysis layer. First, a cross-oceanic feature association unit is established for it. This unit receives initial coupling features from each oceanic level of the spatiotemporal analysis layer. Then, based on the previously determined correlation transmission coefficients between oceanic levels, it analyzes the relationships between the initial coupling features of different oceanic levels. By comparing the feature values ​​of the initial coupling features of different oceanic levels, their common features are extracted. For example, the similarity of diffusion rate fluctuation amplitudes in the initial coupling features of nearshore and offshore waters is calculated, and the correlation coefficient between the two fluctuation amplitudes is used as a measure. The consistency of the inflection point change rate in the initial coupling features of nearshore and offshore waters is analyzed to determine whether they have the same trend. These common features reflect the commonalities in the association of elements at different oceanic levels.

[0099] Step S134: Set up multi-dimensional derivation units for the feature derivation layer. These multi-dimensional derivation units are based on common features and combined with the correlation characteristics of element types. The correlation characteristics of element types include the correlation sensitivity between the temperature of physical elements and the salinity of chemical elements, and the correlation responsiveness between the nutrients of chemical elements and the phytoplankton of biological elements. Multi-dimensional correlation derivation features are generated, including spatiotemporal synergy features, cross-element coupling features, and marine transfer features. Spatiotemporal synergy features reflect the degree of synergy between the temporal changes and spatial diffusion of correlation strength. Cross-element coupling features reflect the comprehensive strength of the correlation between physical elements, chemical elements, and biological elements. Marine transfer features reflect the transfer efficiency of element correlation at different marine levels.

[0100] Following the cross-oceanic feature association unit, a multi-dimensional derivation unit is set up for the feature derivation layer. This multi-dimensional derivation unit generates multi-dimensional association-derived features based on extracted common features and the association characteristics of element types. The association characteristics of element types are determined based on historical data and oceanographic knowledge. For example, there is a certain correlation sensitivity between the physical element temperature and the chemical element salinity; changes in temperature will cause corresponding changes in salinity. There is a correlation response between the chemical element nutrients and the biological element phytoplankton; changes in nutrient content will affect the growth and reproduction of phytoplankton. Based on these association characteristics, spatiotemporal synergy features, cross-element coupling features, and oceanic transfer features are generated. Spatiotemporal synergy features are determined by analyzing the correlation between temporal changes in association strength and spatial diffusion; the higher the correlation coefficient, the stronger the synergy. Cross-element coupling features are obtained by comprehensively considering the association strengths between physical, chemical, and biological elements, for example, by weighted summing of the association strengths between each pair of the three elements. Oceanic transfer features are determined by calculating the transfer efficiency of element associations across different oceanic levels; the transfer efficiency can be represented by the ratio of association strength before and after transfer.

[0101] Step S135: Set up a mapping rule construction unit for the state mapping layer. This mapping rule construction unit collects historical water environment state records at different sea area levels. The historical water environment state records include water quality level, ecological balance status, and potential risk level. Extract multi-dimensional related derived features corresponding to each state record and establish a correspondence database between features and states.

[0102] The state mapping layer is the final layer of the cross-dimensional feature collaboration module. First, a mapping rule construction unit is set up for it. This unit collects historical water environment state records at different marine levels. Water quality levels can be categorized as excellent, good, medium, and poor; ecological balance status can be categorized as balanced, basically balanced, and unbalanced; and potential risk levels can be categorized as low risk, medium risk, and high risk. For each historical water environment state record, its corresponding multi-dimensional associated derived features are extracted. Then, features are mapped one-to-one with state records to establish a feature-state correspondence database.

[0103] Step S136: Set up a dynamic mapping unit for the state mapping layer. The dynamic mapping unit constructs a mapping function between multi-dimensional associated derived features and water environment state based on the corresponding relation library. The input of the mapping function is the feature value of the multi-dimensional associated derived features, and the output is the corresponding water environment state probability distribution. The water environment state probability distribution includes the probability percentage of excellent water quality, the probability percentage of good water quality, and the probability percentage of poor water quality.

[0104] For example, step S1361: Extract multi-dimensional related derived feature samples and corresponding water environment status samples from the corresponding relation database, and divide the samples into nearshore sample set, near-shore sample set and offshore sample set according to the sea area level.

[0105] After the mapping rule construction unit establishes a database of correspondences between features and states, the dynamic mapping unit first extracts data from this database. The extracted data includes multi-dimensional associated derived feature samples and their one-to-one corresponding water environment state samples. The multi-dimensional associated derived feature samples cover previously generated spatiotemporal collaborative features, cross-element coupling features, and marine transfer features; the water environment state samples include specific state records such as water quality level, ecological balance status, and potential risk level. Then, based on the corresponding marine area level, the samples are divided into nearshore sample sets, nearshore sample sets, and offshore sample sets. The division is based on the marine location coordinates recorded in the samples, using these coordinates to determine the marine area level to which the sample belongs. This division aims to improve the accuracy of subsequent mapping by constructing mapping functions for the characteristics of different marine areas, as the water environment state characteristics differ across marine areas, and separate processing better reflects the actual situation of each marine area.

[0106] Step S1362: For each sample set, the multi-dimensional associated derived feature samples are divided into spatiotemporal collaborative feature samples, cross-element coupling feature samples, and marine transfer feature samples. Each feature sample contains values ​​of multiple feature dimensions.

[0107] For each predefined sea area level sample set (nearshore, nearshore, and offshore), the multi-dimensional correlation-derived feature samples need further subdivision. These multi-dimensional correlation-derived feature samples are categorized by feature type into spatiotemporal coordination feature samples, cross-element coupling feature samples, and sea area transfer feature samples. Spatiotemporal coordination feature samples include multiple feature dimensions reflecting the temporal variation of correlation strength and the degree of spatial diffusion coordination, such as coordination coefficients and lag times. Cross-element coupling feature samples include multiple feature dimensions reflecting the comprehensive strength of correlations among physical, chemical, and biological elements, such as comprehensive coupling index and contribution weights of each element. Sea area transfer feature samples include multiple feature dimensions reflecting the efficiency of element correlation transfer at different sea area levels, such as transfer rate and attenuation coefficient. The multiple feature dimension values ​​in each feature sample are derived from the original multi-dimensional correlation-derived feature samples through data extraction and separation, ensuring that each feature type of sample has complete feature dimension information.

[0108] Step S1363: Set a weight coefficient for each feature dimension of each feature sample. The weight coefficient is determined based on the correlation between the feature dimension and the water environment status. The correlation is determined by the degree of influence of the feature dimension changes on the water environment status through historical data statistics. The higher the correlation, the larger the weight coefficient.

[0109] For each feature dimension in each feature sample, a corresponding weight coefficient needs to be set. The magnitude of the weight coefficient depends on the correlation between the feature dimension and the water environment state. The higher the correlation, the greater the influence of the feature dimension on the water environment state, and the larger its weight coefficient. The correlation is determined through statistical analysis of historical data. Specifically, the method involves observing the relationship between changes in feature dimension values ​​and changes in the water environment state. When the feature dimension value changes to a certain extent, the probability and degree of change in the water environment state are statistically analyzed. Based on these statistical results, the correlation between the feature dimension and the water environment state is quantified. For example, if an increase in the value of a certain feature dimension significantly increases the probability of a deterioration in the water environment state, and the degree of deterioration is large, then the correlation between that feature dimension and the water environment state is high, and the corresponding weight coefficient is also large. In this way, reasonable weight coefficients are set for all feature dimensions of each feature sample for subsequent calculation of the feature composite value.

[0110] Step S1364: Standardize the values ​​of each feature dimension of each feature sample to generate a formula for calculating the feature comprehensive value. Multiply the standardized values ​​of each feature dimension by the corresponding weight coefficients and sum them to obtain the feature comprehensive value of the feature sample.

[0111] Since the values ​​of different feature dimensions may have different dimensions and orders of magnitude, it is necessary to standardize the values ​​of each feature dimension for each feature sample in order to eliminate the impact of these differences on subsequent calculations. The standardization method involves subtracting the average value of each feature dimension from the value of that feature dimension across all samples, and then dividing by the standard deviation of that feature dimension, so that the standardized value has a mean of 0 and a standard deviation of 1. After standardization, a formula for calculating the feature composite value is generated. This formula involves multiplying each standardized feature dimension value by its corresponding weighting coefficient, and then summing all the products. Calculating the value for each feature sample using this formula yields the feature composite value for that sample, which integrates information from multiple feature dimensions and can more comprehensively reflect the relationship between the feature sample and the state of the aquatic environment.

[0112] Step S1365: For each sample set, calculate the range of characteristic comprehensive values ​​corresponding to different water environment states, including excellent water quality, good water quality, and poor water quality, and determine the characteristic comprehensive value interval for each water environment state.

[0113] For each marine stratum sample set (nearshore, nearshore, and offshore), statistical analysis was performed on the comprehensive characteristic values ​​corresponding to different water environment states (excellent, good, and poor water quality). For the excellent water environment state, the comprehensive characteristic values ​​of all characteristic samples belonging to this state were collected, and the minimum and maximum values ​​of these comprehensive values ​​were identified to determine the range of comprehensive characteristic values ​​corresponding to the excellent water quality state. Using the same method, the range of comprehensive characteristic values ​​for other water environment states such as good and poor water quality was determined. These ranges of comprehensive characteristic values ​​constitute the comprehensive characteristic value intervals for each water environment state. The boundaries of the intervals were determined by extreme values ​​or quantiles in the statistical analysis to ensure that each interval can well represent the distribution of comprehensive characteristic values ​​for the corresponding water environment state.

[0114] Step S1366: Construct the basic framework of the mapping function. The input is the feature value of the multi-dimensional associated derived features. First, calculate the feature comprehensive value, then determine the feature comprehensive value interval to which the feature comprehensive value belongs, and output the basic probability of the water environment state corresponding to the feature comprehensive value interval.

[0115] The basic framework of the mapping function is constructed as follows: the input of the function is the feature values ​​of each feature dimension of the multi-dimensional associated derived features. First, the input feature values ​​are standardized according to the method in step S1364. Then, based on the weight coefficients set in step S1363, the feature comprehensive value is calculated using the feature comprehensive value calculation formula. Next, the calculated feature comprehensive value is compared with the feature comprehensive value intervals for each water environment state determined in step S1365 to determine the feature comprehensive value interval to which the feature comprehensive value belongs. Finally, the basic probability of the water environment state corresponding to the feature comprehensive value interval is used as the output of the mapping function. The initial setting of the basic probability can be the frequency of the water environment state corresponding to the interval in historical samples. For example, if the frequency of a certain feature comprehensive value interval corresponding to the excellent water quality state in historical samples is a certain proportion, then this proportion is set as the basic probability of the excellent water quality state.

[0116] Step S1367: Based on the deviation data between the comprehensive feature value and the water environment state in the sample set, the deviation data is sample data where the comprehensive feature value is in a certain range but the actual water environment state is another state. The basic probability is corrected by introducing a deviation correction coefficient, which is determined based on the proportion of the number of deviation samples to the total number of samples.

[0117] In practice, there may be situations where the comprehensive feature value falls within a certain range, but the actual water environment state is different. These data constitute biased data. For example, a comprehensive feature value might fall within the range of "excellent" water quality, but the actual water environment state might be "good." Based on these biased data, the base probability output by the mapping function needs to be corrected. A bias correction coefficient is introduced, calculated based on the proportion of biased samples to the total number of samples. Specifically, for a certain comprehensive feature value range and its corresponding target water environment state, the number of biased samples within that range whose actual water environment state is another state is counted. The number of biased samples is divided by the total number of samples within that range to obtain the bias proportion. The bias correction coefficient is then 1 minus the bias proportion. Finally, the base probability is multiplied by the bias correction coefficient to obtain the corrected water environment state probability. This correction reduces the impact of biased data on the mapping results and improves the accuracy of the probability.

[0118] Step S1368: Multiply the base probability by the deviation correction factor to obtain the final water environment state probability distribution.

[0119] After obtaining the base probability in step S1366 and determining the deviation correction coefficient in step S1367, multiplying the two yields the final probability distribution of the water environment state. For example, the base probability of excellent water quality corresponding to a certain feature's comprehensive value is one value, and the deviation correction coefficient is another value; multiplying the two gives the final probability of excellent water quality. Similarly, the final probabilities of other states such as good water quality and poor water quality are calculated. These final probabilities together constitute the probability distribution of the water environment state. This probability distribution of the water environment state reflects the likelihood of different water environment states corresponding to the input multi-dimensional associated derived features.

[0120] Step S1369: Optimize the mapping function for each sea area level. Based on the typical water environment status characteristics of that sea area level, the typical water environment status characteristics include poor water quality in nearshore sea areas and good water quality in offshore sea areas. Adjust the weight coefficients of the feature dimensions and the range of feature comprehensive values ​​to make the mapping function more accurately reflect the water environment status of that sea area level.

[0121] Because different sea area levels have different typical water environment characteristics—for example, nearshore sea areas are more affected by human activities and are prone to poor water quality; offshore sea areas are less affected by human activities and are mostly in a good water quality state; the water environment state of nearshore sea areas is somewhere in between—it is necessary to optimize the mapping function for each sea area level separately. The optimization includes adjusting the weight coefficients of feature dimensions and the feature composite value range. For nearshore sea areas, the weight coefficients of feature dimensions related to water pollution (such as feature dimensions related to nutrient content in chemical elements) are increased, and the feature composite value range for poor water quality is adjusted to better reflect the actual data distribution of nearshore sea areas. For offshore sea areas, the weight coefficients of feature dimensions reflecting good water quality are increased, and the feature composite value range for good water quality is adjusted. Through the above optimization, the mapping function for each sea area level can better adapt to the water environment characteristics of that sea area, improving the accuracy of the mapping.

[0122] Step S13610: Integrate the mapping functions of different sea area levels to form a unit algorithm for dynamic mapping units. This unit algorithm automatically calls the mapping function of the corresponding sea area level according to the sea area level to which the multi-dimensional correlation derived features of the input belong, and outputs the probability distribution of the water environment state.

[0123] After constructing and optimizing the mapping functions for each marine area level, these functions are integrated to form the unit algorithm of the dynamic mapping unit. This unit algorithm has the function of determining the marine area level to which the input multi-dimensional associated derived features belong, based on the marine location coordinate information contained in the features. Upon receiving the input multi-dimensional associated derived features, the unit algorithm first extracts its marine location coordinates, determines the marine area level (nearshore, coastal, or offshore) based on the coordinates, and then automatically calls the mapping function corresponding to that marine area level. The called mapping function processes the input features, calculates and outputs the probability distribution of the water environment state. Through the above integration, the dynamic mapping unit can process multi-dimensional associated derived features from different marine area levels and output the corresponding probability distribution of the water environment state, realizing the dynamic mapping of the entire marine water environment state.

[0124] Step S137: Establish a feature transmission channel between the spatiotemporal resolution layer and the feature derivation layer. This feature transmission channel transmits the initial coupling features of each sea area level output by the spatiotemporal resolution layer to the feature derivation layer in sea area level order. During the transmission process, the spatiotemporal dimension markers of the initial coupling features are retained. The spatiotemporal dimension markers include time nodes and sea area coordinates.

[0125] To facilitate data transmission between the spatiotemporal resolution layer and the feature derivation layer, a feature transmission channel is established. This channel transmits the initial coupled features of each sea area level output by the dimension extraction unit of the spatiotemporal resolution layer to the cross-sea area feature association unit of the feature derivation layer in the order of nearshore sea area, near-shore sea area, and offshore sea area. During transmission, the spatiotemporal dimension markers of the initial coupled features need to be preserved, the time node markers need to be marked with the specific time corresponding to the initial coupled features, and the sea area coordinates need to be marked with the sea area location corresponding to the features. This ensures that the feature derivation layer can clearly understand the spatiotemporal context when processing the initial coupled features.

[0126] Step S138: Establish a feature transmission channel between the feature derivation layer and the state mapping layer. This feature transmission channel transmits the multi-dimensional associated derived features output by the feature derivation layer to the state mapping layer according to feature type. Feature types include spatiotemporal collaborative features, cross-element coupling features, and marine area transfer features. During the transmission process, the marine area level information corresponding to the multi-dimensional associated derived features is retained.

[0127] Similarly, a feature transmission channel is established between the feature derivation layer and the state mapping layer. This channel classifies the multi-dimensional associated derived features output by the multi-dimensional derivation units of the feature derivation layer according to their types: spatiotemporal collaborative features, cross-element coupling features, and marine transfer features, and transmits them to the dynamic mapping unit of the state mapping layer. During transmission, the marine hierarchical information corresponding to the multi-dimensional associated derived features is retained, that is, it is clear which marine hierarchical level the feature belongs to. Thus, when performing dynamic mapping, the state mapping layer can combine the marine hierarchical information to improve the accuracy of the mapping.

[0128] Step S139: Set up a feature update unit for the cross-dimensional feature collaboration module. This feature update unit receives newly generated spatiotemporal coupled trajectories in real time, triggers the spatiotemporal parsing layer to re-extract the initial coupled features, and the feature derivation layer to re-generate multi-dimensional associated derived features, so that the cross-dimensional feature collaboration module can dynamically update feature data based on the latest trajectory.

[0129] To ensure the cross-dimensional feature collaboration module can process new spatiotemporal coupled trajectories and update feature data, a feature update unit is set up for it. This feature update unit receives newly generated spatiotemporal coupled trajectories in real time. Upon receiving a new trajectory, it immediately triggers the trajectory decomposition unit and dimension extraction unit of the spatiotemporal parsing layer to decompose and extract features from the new trajectory, obtaining new initial coupled features. Then, it triggers the cross-ocean-area feature association unit and multi-dimensional derivation unit of the feature derivation layer to generate new multi-dimensional associated derived features based on the new initial coupled features. In this way, the cross-dimensional feature collaboration module can dynamically update feature data based on the latest spatiotemporal coupled trajectories, ensuring that subsequent state mapping can reflect the latest water environment conditions.

[0130] Step S140: Generate a self-optimizing feedback link for the cross-dimensional feature collaboration module. This self-optimizing feedback link transmits the water environment state mapping result output by the state mapping layer back to the feature derivation layer and the spatiotemporal parsing layer, dynamically adjusting the extraction dimension of the initial coupled features and the generation rules of the multi-dimensional associated derived features.

[0131] To improve the performance and accuracy of the cross-dimensional feature collaboration module, a self-optimizing feedback loop is generated. This self-optimizing feedback loop can adjust the parameters of the feature derivation layer and the spatiotemporal analysis layer in reverse based on the output of the state mapping layer, thereby achieving self-optimization of the module.

[0132] Step S141: Extract the probability distribution data from the water environment state mapping results output by the state mapping layer, determine the peak state and the distribution dispersion of the probability distribution. The peak state is the water environment state with the highest probability ratio, and the distribution dispersion is the variance of the probability distribution.

[0133] The self-optimizing feedback loop first extracts probability distribution data from the water environment state mapping results output by the state mapping layer, i.e., the probability percentage of water quality states such as excellent, good, and poor. Then, it determines the peak state of the probability distribution, which is the water environment state with the highest probability percentage. The distribution dispersion is obtained by calculating the variance of the probability distribution. The larger the variance, the more dispersed the probability distribution is, and the weaker the ability of the features to distinguish the states.

[0134] Step S142: Based on the difference between the peak state and the preset target state, calculate the state deviation value. The preset target state includes the ideal water quality level and the healthy ecological balance state. The greater the difference, the greater the state deviation value. Based on the distribution dispersion, calculate the feature sensitivity value. The greater the distribution dispersion, the weaker the ability of multi-dimensional correlation-derived features to distinguish the water environment state, and the smaller the feature sensitivity value.

[0135] The preset target state is set according to the requirements of marine water environment management, such as an ideal water quality level of "excellent" and a healthy ecological balance state of "balanced". The difference between the peak state and the preset target state is calculated. This difference can be determined by comparing the state levels of the two; for example, if the peak state is "good" and the preset target state is "excellent", the difference is one level. A state deviation value is calculated based on the magnitude of the difference; the larger the difference, the larger the state deviation value. Simultaneously, a characteristic sensitivity value is calculated based on the distribution dispersion; the larger the distribution dispersion, the smaller the characteristic sensitivity value, and vice versa.

[0136] Step S143: Based on the state deviation value, generate a first adjustment parameter for adjusting the feature extraction dimension and derivation rules. The larger the state deviation value, the larger the adjustment range of the first adjustment parameter. Based on the feature sensitivity value, generate a second adjustment parameter for optimizing feature selection and weight allocation. The smaller the feature sensitivity value, the larger the adjustment range of the second adjustment parameter. Based on the first adjustment parameter and the second adjustment parameter, generate the spatiotemporal analysis layer adjustment parameter and the feature derivation layer adjustment parameter.

[0137] A first adjustment parameter is generated based on the state deviation value. This parameter is used to adjust the feature extraction dimension and derivation rules. The larger the state deviation value, the greater the adjustment is required to more significantly change the feature extraction and derivation process, thereby reducing the state deviation. A second adjustment parameter is generated based on the feature sensitivity value. This parameter is used to optimize feature selection and weight allocation. The smaller the feature sensitivity value, the weaker the discriminative ability of the current feature, and the greater the adjustment is required to improve the sensitivity of the feature. Then, the first and second adjustment parameters are combined to generate spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters according to the set weight ratio. The weight ratio can be determined based on the degree of influence of state deviation and feature sensitivity on model performance.

[0138] Step S144: Establish a feedback sub-link from the state mapping layer to the spatiotemporal analysis layer. This feedback sub-link transmits the adjustment parameters of the spatiotemporal analysis layer to the dimension extraction unit of the spatiotemporal analysis layer, and adjusts the feature extraction dimension of the dimension extraction unit of the spatiotemporal analysis layer. The adjustment methods include adding periodic change feature extraction of the correlation strength temporal change spectrum and supplementing boundary smoothness feature extraction of the correlation range spatial diffusion curve.

[0139] A feedback sub-link is established from the state mapping layer to the spatiotemporal analysis layer. This link transmits the generated spatiotemporal analysis layer adjustment parameters to the dimension extraction unit of the spatiotemporal analysis layer. The dimension extraction unit adjusts the feature extraction dimensions based on the adjustment parameters. For example, it adds the extraction of periodic variation features of the correlation strength temporal change spectrum, extracting features such as period length and trend within the period by analyzing the periodic fluctuations of the spectrum; it also supplements the extraction of boundary smoothness features of the spatial diffusion curve of the correlation range, reflecting the regularity of the diffusion range by calculating the smoothness of the curve boundaries.

[0140] Step S145: Establish a feedback sub-link from the state mapping layer to the feature derivation layer. This feedback sub-link transmits the adjustment parameters of the feature derivation layer to the multi-dimensional derivation unit of the feature derivation layer, and adjusts the derivation rules of the multi-dimensional derivation unit of the feature derivation layer. The adjustment methods include improving the screening accuracy of cross-sea area feature association units for common features and optimizing the weight allocation of multi-dimensional derivation units for cross-element coupling features.

[0141] A feedback sub-link is established from the state mapping layer to the feature derivation layer, transmitting the adjustment parameters of the feature derivation layer to the multi-dimensional derivation units within the feature derivation layer. The multi-dimensional derivation units adjust the derivation rules based on the adjustment parameters, improving the accuracy of cross-oceanic feature association units in selecting common features. For example, by setting stricter selection criteria, only highly correlated common features are retained. The weight allocation of cross-element coupling features by the multi-dimensional derivation units is optimized; for correlations between elements that have a greater impact on the water environment state, their weight in the cross-element coupling features is increased.

[0142] Step S146: Establish a secondary feedback sub-link from the feature derivation layer to the spatiotemporal analysis layer. After the feature derivation layer updates the multi-dimensional associated derivation features based on the adjustment parameters, it extracts the difference between the new multi-dimensional associated derivation features and the old multi-dimensional associated derivation features. Based on the difference, it generates secondary adjustment parameters and transmits them to the spatiotemporal analysis layer through the secondary feedback sub-link to further optimize the extraction dimension of the initial coupled features.

[0143] After updating the multi-dimensional associated derived features based on adjustment parameters at the feature derivation layer, the difference between the new and old features is calculated. The difference can be determined by comparing the differences in feature values ​​across each dimension of the new and old features, for example, by calculating the absolute difference of each dimension's feature value and taking the average. Secondary adjustment parameters are generated based on the difference. A larger difference indicates a more significant change in the adjusted features, and the adjustment range of the secondary adjustment parameters can be appropriately reduced; a smaller difference indicates a less significant adjustment effect, and the adjustment range of the secondary adjustment parameters can be appropriately increased. The secondary adjustment parameters are transmitted to the spatiotemporal resolution layer through a secondary feedback sub-link to further optimize the extraction dimensions of the initial coupled features, such as increasing or decreasing the extraction of certain feature dimensions.

[0144] Step S147: Based on historical feedback data, establish a correlation model between feedback parameters and state deviation values ​​and feature sensitivity values. This correlation model automatically generates optimal feedback parameters based on the real-time output state deviation values ​​and feature sensitivity values.

[0145] For example, step S1471: Collect historical feedback data during the operation of the self-optimized feedback link. Each piece of historical feedback data includes a state deviation value, a feature sensitivity value, the corresponding spatiotemporal analysis layer adjustment parameters, the feature derivation layer adjustment parameters, and the state deviation improvement rate after adjustment. The state deviation improvement rate is the ratio of the state deviation value after adjustment to the state deviation value before adjustment.

[0146] To establish a correlation model between feedback parameters and state deviation values ​​and feature sensitivity values, it is necessary to collect historical feedback data during the operation of the self-optimizing feedback loop. This historical feedback data was accumulated during previous feedback adjustments, and each data point contains multiple key information items. The state deviation value is the deviation value before adjustment, reflecting the difference between the model output and the target state before adjustment; the feature sensitivity value is also the value before adjustment, reflecting the ability of the feature to distinguish the state before adjustment; the corresponding spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters are the adjustment parameters actually used in this feedback; the improved state deviation rate after adjustment is the ratio of the adjusted state deviation value to the state deviation value before adjustment. The smaller the ratio, the better the adjustment effect and the more significant the improvement. When collecting this historical feedback data, it is necessary to ensure the completeness and accuracy of the data. Each piece of information in each data point must correspond one-to-one, and the time sequence must be clear to facilitate subsequent analysis of the relationship between feedback parameters and state deviation values ​​and feature sensitivity values.

[0147] Step S1472: Preprocess the historical feedback data, delete data with abnormal state deviation improvement rates (abnormal state deviation improvement rates include improvement rates greater than 1 or less than 0), retain data with improvement rates between 0 and 1, and obtain valid historical feedback data.

[0148] After collecting historical feedback data, preprocessing is required to remove outliers and ensure the accuracy of subsequent modeling. Outliers mainly refer to data with abnormal state deviation improvement rates, specifically those with improvement rates greater than 1 and less than 0. An improvement rate greater than 1 means that the adjusted state deviation value is larger than before the adjustment, indicating that the model performance has not only failed to improve but has actually decreased. This data may be due to incorrect parameter settings or other interfering factors. An improvement rate less than 0 is unreasonable because the state deviation value is non-negative, and the ratio before and after the adjustment cannot be negative. These outliers are deleted. Data with improvement rates between 0 and 1 are retained. These data indicate that the adjusted state deviation value has decreased or remained unchanged, and the model performance has improved or remained stable. This data is used as valid historical feedback data for subsequent correlation model building.

[0149] Step S1473: Classify the valid historical feedback data according to the range of state deviation values ​​and the range of feature sensitivity values, and divide them into multiple data intervals. Each data interval contains multiple sets of state deviation values, feature sensitivity values ​​and corresponding feedback parameters.

[0150] The valid historical feedback data is classified based on the range of state deviation values ​​and the range of feature sensitivity values. First, the intervals for dividing the state deviation values ​​are determined, for example, by dividing the state deviation values ​​from 0 to their maximum value into several intervals; similarly, the feature sensitivity values ​​are also divided into several intervals from 0 to their maximum value. Then, based on the state deviation value and feature sensitivity value in each valid historical feedback data point, it is assigned to the data interval formed by the intersection of the corresponding state deviation value interval and feature sensitivity value interval. Each data interval contains multiple sets of data, each set containing the state deviation value, feature sensitivity value, and the corresponding spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters (feedback parameters). Through this classification, continuous state deviation values ​​and feature sensitivity values ​​are discretized into different data intervals, facilitating subsequent analysis of the patterns of feedback parameters within each interval.

[0151] Step S1474: For each data interval, calculate the average value of the feedback parameters within that data interval as the baseline feedback parameters for that data interval; calculate the average value of the state deviation improvement rate within that data interval as the baseline improvement rate for that data interval.

[0152] For each defined data interval, the baseline feedback parameter and baseline improvement rate need to be calculated. The baseline feedback parameter is calculated by averaging all feedback parameters (including spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters) within the data interval; the average value is the baseline feedback parameter for that data interval. For example, if there are multiple spatiotemporal analysis layer adjustment parameter values ​​within a data interval, the arithmetic mean of these values ​​is calculated as the baseline spatiotemporal analysis layer adjustment parameter for that interval; the baseline value for the feature derivation layer adjustment parameter is calculated using the same method. The baseline improvement rate is calculated by averaging all state deviation improvement rates within the data interval. The baseline feedback parameter and baseline improvement rate represent the average level of the feedback parameters and improvement effects within the data interval, and are the foundational data for establishing the correlation model.

[0153] Step S1475: Standardize the state deviation value and feature sensitivity value; construct the input layer, hidden layer and output layer of the association model. The input layer consists of the standardized state deviation value and feature sensitivity value. The hidden layer contains multiple neurons. Each neuron corresponds to the baseline feedback parameter and baseline improvement rate of a data interval. The output layer consists of the optimal feedback parameter.

[0154] First, the state bias values ​​and feature sensitivity values ​​are standardized using a method similar to the standardization of the feature dimensions: subtracting the mean from the value and then dividing by the standard deviation. This ensures that the standardized state bias values ​​and feature sensitivity values ​​have a uniform dimension and distribution range, facilitating the training and computation of the association model. Next, the network structure of the association model is constructed, consisting of an input layer, hidden layers, and an output layer. The input layer has two neurons, receiving the standardized state bias values ​​and feature sensitivity values ​​respectively. The hidden layer contains multiple neurons, the number of which is equal to the number of data intervals. Each neuron corresponds to the baseline feedback parameter and baseline improvement rate for a data interval, storing this information internally. The output layer has the same number of neurons as the number of feedback parameters, used to output the optimal feedback parameters (including parameters adjusted by the spatiotemporal analysis layer and the feature derivation layer).

[0155] Step S1476: Based on effective historical feedback data, the association model is trained using the gradient descent method. The weights of the hidden layer neurons are adjusted so that when the association model inputs the state deviation value and the feature sensitivity value, the state deviation improvement rate corresponding to the optimal feedback parameters output is close to the benchmark improvement rate.

[0156] The constructed association model is trained using valid historical feedback data. The training process employs gradient descent, aiming to adjust the weights of hidden layer neurons so that the model can output optimal feedback parameters based on the input state bias and feature sensitivity values, with the state bias improvement rate corresponding to these optimal feedback parameters approaching the baseline improvement rate. The specific training steps are as follows: Standardized state bias and feature sensitivity values ​​are input into the model; the model calculates the output feedback parameters through hidden layer neurons (multiplying the input by the neuron weights and processing them in conjunction with the baseline feedback parameters and baseline improvement rate); the output feedback parameters are applied to model adjustment, calculating the actual state bias improvement rate; the actual improvement rate is compared with the baseline improvement rate, and the error between the two is calculated; based on the magnitude and direction of the error, the weights of the hidden layer neurons are adjusted using gradient descent to reduce the error; this process is repeated until the error between the state bias improvement rate corresponding to the model's output feedback parameters and the baseline improvement rate reaches a preset threshold, at which point training ends.

[0157] Step S1477: Validate the trained association model by inputting new state deviation values ​​and feature sensitivity values, and comparing the state deviation improvement rate corresponding to the optimal feedback parameters output by the association model with the manually adjusted feedback parameters. If the difference between the two is less than the preset difference threshold, the association model is validated. If the difference is greater than the preset difference threshold, the association model is trained until the difference meets the requirements.

[0158] After training, the association model needs to be validated to verify its generalization ability and accuracy. During validation, new state bias values ​​and feature sensitivity values ​​(which were not used in model training) are input, allowing the association model to output optimal feedback parameters. Simultaneously, domain experts manually adjust these parameters based on the new state bias values ​​and feature sensitivity values, obtaining manually adjusted feedback parameters. The optimal feedback parameters output by the association model and the manually adjusted parameters are then applied to the model, and their respective state bias improvement rates are calculated. These two improvement rates are compared, and the difference between them (e.g., absolute difference or relative difference) is calculated. If the difference is less than a preset difference threshold, it indicates that the feedback parameters output by the association model are close to the effect of the manually adjusted parameters, and the model validation is successful. If the difference is greater than the preset difference threshold, it indicates that the model performance still needs improvement, and more effective historical feedback data needs to be used to train the model and adjust the model parameters until the difference between the model's output improvement rate and the manually adjusted improvement rate meets the requirements.

[0159] Step S1478: Embed the validated association model into the self-optimizing feedback loop. When the self-optimizing feedback loop outputs the state deviation value and feature sensitivity value in real time, the association model automatically inputs the state deviation value and feature sensitivity value, outputs the optimal spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters, and transmits them to the corresponding level module for parameter adjustment.

[0160] The validated association model is formally embedded into the self-optimizing feedback loop, making it a component of the feedback loop. During the operation of the self-optimizing feedback loop, after the state mapping layer outputs the water environment state mapping result, the state deviation value and feature sensitivity value are calculated in real time, and these two values ​​are transmitted as input data to the association model. Upon receiving the input, the association model automatically processes it and outputs the optimal spatiotemporal analysis layer adjustment parameters and feature derivation layer adjustment parameters. These optimal adjustment parameters are transmitted to the corresponding spatiotemporal analysis layer and feature derivation layer modules through the feedback sub-links in the self-optimizing feedback loop to adjust the module parameters. In this way, the self-optimizing feedback loop can automatically generate optimal feedback parameters based on real-time state deviation and feature sensitivity, realizing dynamic adjustment of cross-dimensional feature collaboration modules and improving the model's performance and adaptability.

[0161] Step S150: Integrate dynamic triggering rules, spatiotemporal coupling trajectories, cross-dimensional feature collaboration modules, and self-optimizing feedback links to form an integrated marine water environment analysis model. Process the newly collected multi-ocean-area element monitoring data through the integrated marine water environment analysis model. Adjust the parameters of the cross-dimensional feature collaboration modules in real time based on the self-optimizing feedback links to achieve dynamic adaptation of the integrated marine water environment analysis model to the water environment status of different sea areas, and complete the construction of the integrated marine water environment analysis model.

[0162] Finally, the various parts generated above are integrated to form a complete integrated analysis model of the marine water environment. By processing new data and self-optimizing adjustments, the model is dynamically adapted to the water environment conditions of different sea areas, thus completing the model construction.

[0163] Step S151: Set up the core control module of the integrated analysis model of marine water environment. This core control module is used to store the parameters of dynamic triggering rules, the generation algorithm of spatiotemporal coupling trajectory, the configuration of each level of cross-dimensional feature collaboration module, and the associated model parameters of the self-optimization feedback link. The parameters of dynamic triggering rules include spatiotemporal layering threshold and gradient conditions.

[0164] The core control module is the central hub of the integrated analysis model of the marine water environment, responsible for storing and managing various core parameters and algorithms of the model. Parameters of dynamic triggering rules, such as spatiotemporal stratification thresholds and gradient conditions, the generation algorithm for spatiotemporal coupled trajectories, the configuration of each level of units in the spatiotemporal analysis layer, feature derivation layer, and state mapping layer in the cross-dimensional feature collaboration module, and the associated model parameters of the self-optimizing feedback loop are all stored in this module for use by other modules of the model.

[0165] Step S152: Set up a multi-ocean-area data access module for the integrated analysis model of marine water environment. This multi-ocean-area data access module is used to receive newly collected monitoring data of elements in nearshore waters, offshore waters and offshore waters, classify and store the data according to the sea area level, and convert the data format so that the converted data meets the input requirements of dynamic triggering rules.

[0166] The multi-ocean-area data access module is responsible for receiving newly collected marine environmental element monitoring data from monitoring stations or equipment in nearshore, offshore, and open-ocean areas. The module categorizes and stores the received data according to the ocean-area hierarchy for easy subsequent processing. Simultaneously, it converts data formats, for example, uniformly converting data from different monitoring devices into the standard format specified by the model, ensuring that the converted data meets the input requirements of dynamic triggering rules, such as the time series format and numerical units.

[0167] Step S153: Set up the trajectory generation control module of the integrated analysis model of marine water environment. This trajectory generation control module calls the dynamic triggering rules and spatiotemporal coupling trajectory generation algorithm in the core control module to process the transformed data output by the multi-sea area data access module, generate a new spatiotemporal coupling trajectory, and transmit the new spatiotemporal coupling trajectory to the cross-dimensional feature collaboration module.

[0168] The trajectory generation control module invokes dynamic triggering rules and a spatiotemporally coupled trajectory generation algorithm from the core control module, and then processes the transformed data output from the multi-sea area data access module. The data is filtered according to the dynamic triggering rules, and then multi-dimensional trajectory reconstruction is performed based on the spatiotemporally coupled trajectory generation algorithm to generate a new spatiotemporally coupled trajectory. After generation, the new spatiotemporally coupled trajectory is transmitted to the cross-dimensional feature collaboration module as its input data.

[0169] Step S154: Set up the feature co-processing module of the integrated analysis model of marine water environment. This feature co-processing module loads the spatiotemporal analysis layer, feature derivation layer and state mapping layer of the cross-dimensional feature co-processing module, receives the new spatiotemporal coupled trajectory output by the trajectory generation control module, and sequentially completes the initial coupled feature extraction, multi-dimensional correlation derived feature generation and water environment state mapping, and outputs the water environment state probability distribution result.

[0170] The feature co-processing module loads various layers of the cross-dimensional feature co-processing module, including the spatiotemporal analysis layer, the feature derivation layer, and the state mapping layer. After receiving the new spatiotemporally coupled trajectory output by the trajectory generation and control module, the spatiotemporal analysis layer first decomposes the trajectory and extracts initial coupled features; then, the feature derivation layer generates multi-dimensional associated derived features based on the initial coupled features; finally, the state mapping layer establishes a dynamic mapping with the water environment state through the multi-dimensional associated derived features and outputs the probability distribution result of the water environment state.

[0171] Step S155: Set up the parameter self-optimization module of the integrated analysis model of marine water environment. The parameter self-optimization module loads the correlation model of the self-optimization feedback link, receives the water environment state probability distribution results output by the feature collaborative processing module, calculates the state deviation value and feature sensitivity value, inputs the correlation model to obtain the optimal feedback parameters, and transmits the optimal feedback parameters to the feature collaborative processing module and the trajectory generation control module.

[0172] The parameter self-optimization module loads the correlation model of the self-optimization feedback loop and receives the water environment state probability distribution results output by the feature collaborative processing module. Based on the results, it calculates the state deviation value and feature sensitivity value, and inputs these two values ​​into the correlation model to obtain the optimal feedback parameters. Then, the optimal feedback parameters are transmitted to the feature collaborative processing module and the trajectory generation and control module for adjusting the parameters of these two modules.

[0173] Step S156: The feature collaborative processing module adjusts the feature extraction dimension of the spatiotemporal parsing layer and the generation rules of the feature derivation layer based on the optimal feedback parameters, and the trajectory generation control module adjusts the correlation transfer coefficient in the spatiotemporal coupled trajectory generation algorithm based on the optimal feedback parameters.

[0174] After receiving the optimal feedback parameters, the feature co-processing module adjusts the feature extraction dimensions of the spatiotemporal analysis layer according to the parameters, such as increasing or decreasing the feature extraction of certain time or spatial dimensions; at the same time, it adjusts the generation rules of the feature derivation layer, such as changing the screening conditions for common features or the weight allocation of cross-element coupling features. The trajectory generation control module adjusts the correlation and transmission coefficients in the spatiotemporal coupling trajectory generation algorithm according to the optimal feedback parameters to more accurately reflect the feature correlation and transmission characteristics between sea area levels.

[0175] Step S157: Set up the model adaptation verification module of the marine water environment integrated analysis model. This model adaptation verification module receives the new water environment state probability distribution results output by the feature collaborative processing module after adjusting the parameters, and determines the water environment state with the highest probability from the new water environment state probability distribution results as the predicted state. Compare the predicted state with the historical water environment state records. If the number of times the predicted state matches the historical records exceeds the preset adaptation threshold multiple times, it means that the marine water environment integrated analysis model has achieved dynamic adaptation to the current marine water environment state.

[0176] The model adaptation and validation module is used to verify the model's adaptation after parameter adjustments. It receives the new water environment state probability distribution results output by the feature co-processing module after parameter adjustments and determines the water environment state with the highest probability as the predicted state. Then, it compares the predicted state with historical water environment state records, counting the number of times the predicted state matches the historical records. If the number of consecutive matches exceeds a preset adaptation threshold (e.g., five consecutive matches), it indicates that the model has achieved dynamic adaptation to the current marine water environment state.

[0177] Step S158: Integrate the core control module, multi-sea area data access module, trajectory generation control module, feature collaborative processing module, parameter self-optimization module, and model adaptation verification module to form an integrated marine water environment analysis model; conduct multi-sea area adaptation tests on the integrated marine water environment analysis model, input monitoring data of multi-sea area elements under different seasons and climate conditions, so that the integrated marine water environment analysis model can accurately map the water environment state through parameter self-optimization in various scenarios, and complete the final construction of the integrated marine water environment analysis model.

[0178] The various modules of the model, including the core control module, multi-sea area data access module, trajectory generation and control module, feature collaborative processing module, parameter self-optimization module, and model adaptation and verification module, are integrated to form a complete integrated marine water environment analysis model. Then, multi-sea area adaptation tests are conducted on the model, inputting multi-sea area element monitoring data under different seasons (e.g., spring, summer, autumn, winter) and different climatic conditions (e.g., sunny days, rainy days, typhoon days), observing the model's performance in various scenarios. Through adjustments to the parameter self-optimization module, the model accurately maps the water environment state under different scenarios. When the model meets the preset accuracy requirements in all test scenarios, the final construction of the integrated marine water environment analysis model is completed.

[0179] The entire model construction process involves collecting monitoring data on marine environmental elements, which may include some privacy-sensitive data, such as the specific location coordinates of monitoring stations. To protect this privacy-sensitive data, data anonymization techniques are employed to obfuscate the marine location coordinates, for example, converting specific latitude and longitude coordinates into approximate marine area ranges to avoid the leakage of precise coordinates. Simultaneously, encryption technologies, such as SSL encryption, are used during data transmission to ensure that data is not stolen or tampered with during transmission. Regarding data storage, access control mechanisms are implemented to restrict access to sensitive data to only authorized personnel, thereby achieving privacy protection and preventing the leakage of privacy-sensitive data.

[0180] In one exemplary embodiment, a model building system for integrated analysis of marine water environment is provided. This model building system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, the model building system for integrated analysis of marine water environments includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a model building method for integrated analysis of marine water environments. The display unit is used to generate a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of a model building system for integrated analysis of marine water environment, or an external keyboard, touchpad, or mouse, etc.

[0181] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A model construction method for integrated analysis of marine water environment, characterized in that, The method includes: A dynamic triggering rule for the correlation and evolution of marine water environment elements is generated. The dynamic triggering rule includes the spatiotemporal stratification threshold of element monitoring data and the gradient conditions for initiating the correlation between elements. Based on dynamic triggering rules, the historical monitoring data of marine water environment elements are reconstructed in multiple dimensions to generate spatiotemporal coupling trajectories of element correlation evolution. The spatiotemporal coupling trajectories include the temporal changes in the correlation strength of elements at different sea area levels and the spatial diffusion path of the correlation range. A cross-dimensional feature collaboration module for generating spatiotemporal coupled trajectories is provided. This cross-dimensional feature collaboration module includes a spatiotemporal analysis layer, a feature derivation layer, and a state mapping layer. The spatiotemporal analysis layer decomposes the spatiotemporal coupled trajectory into spatiotemporal dimensions and extracts the initial coupling features. The feature derivation layer generates multi-dimensional associated derived features based on the initial coupling features. The state mapping layer establishes a dynamic mapping relationship with the water environment state through the multi-dimensional associated derived features. A self-optimizing feedback link for generating cross-dimensional feature collaboration modules is established. This self-optimizing feedback link transmits the water environment state mapping results output by the state mapping layer back to the feature derivation layer and the spatiotemporal analysis layer, dynamically adjusting the extraction dimensions of the initial coupled features and the generation rules of multi-dimensional associated derived features. By integrating dynamic triggering rules, spatiotemporal coupling trajectories, cross-dimensional feature collaboration modules, and self-optimizing feedback links, an integrated marine water environment analysis model is formed. The newly collected multi-ocean-area element monitoring data are processed through the integrated marine water environment analysis model. Based on the self-optimizing feedback link, the parameters of the cross-dimensional feature collaboration module are adjusted in real time to realize the dynamic adaptation of the integrated marine water environment analysis model to the water environment status of different sea areas, thus completing the construction of the integrated marine water environment analysis model.

2. The model construction method for integrated analysis of marine water environment according to claim 1, characterized in that, The dynamic triggering rules for generating the correlation and evolution of marine water environment elements include: Real-time and historical monitoring data of physical, chemical and biological elements in the marine environment are collected at different marine levels, including nearshore, offshore and open ocean. The historical monitoring data includes numerical records of each element in a continuous time series, corresponding marine location coordinates and marine environmental background information, including ocean current direction and seabed topography type. The real-time monitoring data for each sea area level is segmented according to the time series. The fluctuation range and frequency of the element values ​​within each segment are extracted. The difference coefficient of the fluctuation range of the elements between the current sea area level and other sea area levels is calculated. The difference coefficient of the fluctuation frequency of the elements between the current sea area level and other sea area levels is also calculated. Based on the difference coefficient of fluctuation range, the initial value of the spatial stratification threshold of the element monitoring data is set according to the sea area level, and based on the difference coefficient of fluctuation frequency, the initial value of the time stratification threshold of the element monitoring data is set according to the time series. The initial values ​​of spatial stratification threshold and temporal stratification threshold are integrated to form the initial values ​​of spatiotemporal stratification threshold for element monitoring data. Historical correlation strength data of physical and chemical elements, chemical and biological elements, and physical and biological elements within different sea area levels are extracted and arranged in chronological order to form a correlation strength sequence. Analyze the correlation strength difference between adjacent time points in the correlation strength sequence, statistically analyze the distribution pattern of the correlation strength difference, and determine multiple gradual change stages of the correlation strength from the initial value to the significant correlation value. The significant correlation value is an empirical threshold that can reflect the clear interaction between elements. For each gradual change stage, a stage threshold for the correlation strength is set. The difference between the stage thresholds of adjacent stages is distributed in a gradient according to the gradual change law, forming the initial value of the gradient condition for the initiation of the correlation between elements. The initial value of the spatiotemporal layer threshold is integrated with the initial value of the gradient condition to generate the initial version of the dynamic triggering rule; Set an adaptive update cycle for the dynamic triggering rule. When each update cycle arrives, acquire newly collected monitoring data of multiple sea area elements within that cycle. Based on the newly collected data, recalculate the fluctuation range, fluctuation frequency, and correlation strength between elements at each sea area level, update the spatiotemporal stratification threshold and gradient conditions, and form an updated version of the dynamic triggering rule, so that the dynamic triggering rule continues to evolve with the new data.

3. The model construction method for integrated analysis of marine water environment according to claim 1, characterized in that, The method, based on dynamic triggering rules, reconstructs the historical monitoring data of marine water environment elements in multiple dimensions to generate spatiotemporal coupled trajectories of element correlation evolution, including: Real-time monitoring data at different sea level are filtered according to the spatiotemporal stratification threshold in the current version of the dynamic triggering rules, and monitoring data with numerical fluctuations exceeding the spatiotemporal stratification threshold of the corresponding sea level are retained to obtain the filtered monitoring data. The screened monitoring data were grouped by element type to form physical element screening dataset, chemical element screening dataset and biological element screening dataset; For each sea area level, monitoring data of the same time series are extracted from the physical element screening dataset and the chemical element screening dataset. Based on the gradient conditions in the current version of the dynamic triggering rule, the correlation strength between the two elements in the time series is calculated, and the time series curve of the correlation strength between physical and chemical elements in the sea area level is generated. Using the same method, time-series curves of the correlation intensity between chemical and biological elements and the correlation intensity between physical and biological elements were calculated for each sea area level; Align the time series curves of the correlation intensity of each sea area level according to the time axis, extract the peak points, valley points and inflection points of each curve, determine the synchronous change characteristics of each correlation intensity time series curve in the time dimension, the synchronous change characteristics include the time difference of the peak point and the consistency of the inflection point change trend, and generate the correlation intensity time series change map of the sea area level. After screening the monitoring data at each sea area level, the sea area location coordinates corresponding to the element monitoring data are extracted, and a spatial distribution probability field of the element is constructed. The spatial distribution probability field is generated by the kernel density estimation method, and the probability value of each spatial location point represents the possibility of the element appearing at that location. Based on the spatial distribution probability fields of adjacent time series, the bulldozer distance between the probability fields is calculated. The bulldozer distance measures the degree of difference between the two probability distributions and reflects the overall change in the spatial distribution of elements between time series. Based on the displacement vector of the centroid of the probability field, the main spatial diffusion direction of the element association range is determined. The centroid displacement vector is obtained by calculating the weighted average of all position coordinates in the probability field with their probability values ​​as weights. Based on the ratio of bulldozer distance to the corresponding time series duration, the spatial diffusion intensity index of the element association range is obtained, and the spatial diffusion curve of the association range at this sea area level is generated. The temporal variation map of the correlation intensity of different sea area levels is coupled with the spatial diffusion curve of the correlation range. Based on the spatial positional relationship between sea area levels, including the boundary distance between nearshore sea areas and the water flow connectivity between nearshore sea areas and offshore sea areas, the correlation intensity value and spatial diffusion intensity of each sea area level are adjusted so that the coupled trajectory reflects the element correlation and transmission characteristics between sea area levels. The temporal variation map of the correlation intensity of each sea area level and the spatial diffusion curve of the correlation range after integration and coupling processing are used to form the spatiotemporal coupling trajectory of the element correlation evolution. The spatiotemporal coupling trajectory includes the correlation intensity value of each sea area level at different time nodes, the spatial distribution probability field parameters, and the correlation transmission coefficient between sea area levels.

4. The model construction method for integrated analysis of marine water environment according to claim 1, characterized in that, The cross-dimensional feature collaboration module for generating spatiotemporally coupled trajectories includes: A trajectory decomposition unit is set up for the spatiotemporal analysis layer. This trajectory decomposition unit receives the spatiotemporal coupled trajectory and decomposes the spatiotemporal coupled trajectory into nearshore sea area sub-trajectories, near-shore sea area sub-trajectories and offshore sea area sub-trajectories according to the sea area level. Each sub-trajectory contains the temporal variation map of the correlation strength and the spatial diffusion curve of the correlation range at the corresponding sea area level. A dimension extraction unit is set up for the spatiotemporal analysis layer. This dimension extraction unit performs feature extraction in the time and space dimensions for each sub-trajectory. The time dimension extracts the peak duration, valley interval duration and inflection point change rate of the correlation strength temporal change spectrum. The space dimension extracts the diffusion direction change frequency, diffusion rate fluctuation amplitude and spatial coverage area growth rate of the correlation range spatial diffusion curve. The extracted features are integrated into the initial coupling features. A cross-sea-area feature association unit is set up for the feature derivation layer. This cross-sea-area feature association unit receives the initial coupling features of each sea-area level. Based on the association transmission coefficient between sea-area levels, common features of the initial coupling features of different sea-area levels are extracted. The common features include the similarity of diffusion rate fluctuation amplitude in the initial coupling features of nearshore sea areas and offshore sea areas, and the consistency of inflection point change rate in the initial coupling features of nearshore sea areas and offshore sea areas. A multi-dimensional derivation unit is set up for the feature derivation layer. This multi-dimensional derivation unit is based on common features and combined with the correlation characteristics of element types. The correlation characteristics of element types include the correlation sensitivity of temperature of physical elements and salinity of chemical elements, and the correlation responsiveness of nutrients of chemical elements and phytoplankton of biological elements. Multi-dimensional correlation derivation features are generated, including spatiotemporal synergy features, cross-element coupling features and marine transfer features. Spatiotemporal synergy features reflect the degree of synergy between temporal changes in correlation strength and spatial diffusion. Cross-element coupling features reflect the comprehensive strength of correlation among physical elements, chemical elements and biological elements. Marine transfer features reflect the transfer efficiency of element correlation at different marine levels. A mapping rule construction unit is set up for the state mapping layer. This mapping rule construction unit collects historical water environment state records at different sea area levels. The historical water environment state records include water quality level, ecological balance status, and potential risk level. Multi-dimensional related derived features corresponding to each state record are extracted, and a correspondence database between features and states is established. A dynamic mapping unit is set up for the state mapping layer. Based on the corresponding relation library, the dynamic mapping unit constructs a mapping function between multi-dimensional associated derived features and water environment state. The input of the mapping function is the feature value of the multi-dimensional associated derived features, and the output is the corresponding water environment state probability distribution. The water environment state probability distribution includes the probability percentage of excellent water quality, the probability percentage of good water quality, and the probability percentage of poor water quality. A feature transmission channel is established between the spatiotemporal resolution layer and the feature derivation layer. This feature transmission channel transmits the initial coupling features of each sea area level output by the spatiotemporal resolution layer to the feature derivation layer in sea area level order. During the transmission process, the spatiotemporal dimension markers of the initial coupling features are retained. The spatiotemporal dimension markers include time nodes and sea area coordinates. A feature transmission channel is established between the feature derivation layer and the state mapping layer. This feature transmission channel transmits the multi-dimensional associated derived features output by the feature derivation layer to the state mapping layer according to the feature type. The feature types include spatiotemporal collaborative features, cross-element coupling features, and marine transfer features. During the transmission process, the marine hierarchical information corresponding to the multi-dimensional associated derived features is retained. A feature update unit is set up for the cross-dimensional feature collaboration module. This feature update unit receives newly generated spatiotemporal coupled trajectories in real time, triggers the spatiotemporal parsing layer to re-extract the initial coupled features, and the feature derivation layer to regenerate multi-dimensional associated derived features, so that the cross-dimensional feature collaboration module can dynamically update feature data based on the latest trajectory.

5. The model construction method for integrated analysis of marine water environment according to claim 1, characterized in that, The self-optimizing feedback link of the cross-dimensional feature collaboration module transmits the water environment state mapping result output by the state mapping layer back to the feature derivation layer and the spatiotemporal parsing layer, dynamically adjusting the extraction dimensions of the initial coupled features and the generation rules of multi-dimensional associated derived features, including: Extract the probability distribution data from the water environment state mapping results output by the state mapping layer, determine the peak state and the distribution dispersion of the probability distribution. The peak state is the water environment state with the highest probability, and the distribution dispersion is the variance of the probability distribution. Based on the difference between the peak state and the preset target state, the state deviation value is calculated. The preset target state includes the ideal water quality level and the healthy ecological balance state. The greater the difference, the greater the state deviation value. Based on the distribution dispersion, the feature sensitivity value is calculated. The larger the distribution dispersion, the weaker the ability of multi-dimensional associated derived features to distinguish the state of the water environment, and the smaller the feature sensitivity value. Based on the state deviation value, a first adjustment parameter is generated to adjust the feature extraction dimension and derivation rules. The larger the state deviation value, the larger the adjustment range of the first adjustment parameter. Based on the feature sensitivity value, a second adjustment parameter is generated to optimize feature selection and weight allocation. The smaller the feature sensitivity value, the larger the adjustment range of the second adjustment parameter. Based on the first and second adjustment parameters, the spatiotemporal analysis layer adjustment parameters and the feature derivation layer adjustment parameters are generated in combination. A feedback sub-link is established from the state mapping layer to the spatiotemporal analysis layer. This feedback sub-link transmits the adjustment parameters of the spatiotemporal analysis layer to the dimension extraction unit of the spatiotemporal analysis layer, and adjusts the feature extraction dimension of the dimension extraction unit of the spatiotemporal analysis layer. The adjustment methods include adding periodic change feature extraction of the correlation strength temporal change spectrum and supplementing boundary smoothness feature extraction of the correlation range spatial diffusion curve. A feedback sub-link is established from the state mapping layer to the feature derivation layer. This feedback sub-link transmits the adjustment parameters of the feature derivation layer to the multi-dimensional derivation unit of the feature derivation layer, and adjusts the derivation rules of the multi-dimensional derivation unit of the feature derivation layer. The adjustment methods include improving the screening accuracy of cross-sea area feature association units for common features and optimizing the weight allocation of multi-dimensional derivation units for cross-element coupling features. A secondary feedback sub-link is established from the feature derivation layer to the spatiotemporal analysis layer. After the feature derivation layer updates the multi-dimensional associated derivation features based on the adjustment parameters, the difference between the new multi-dimensional associated derivation features and the old multi-dimensional associated derivation features is extracted. Secondary adjustment parameters are generated based on the difference and transmitted to the spatiotemporal analysis layer through the secondary feedback sub-link to further optimize the extraction dimension of the initial coupled features. Based on historical feedback data, a correlation model is established between feedback parameters and state deviation values ​​and feature sensitivity values. This correlation model automatically generates the optimal feedback parameters based on the real-time output state deviation values ​​and feature sensitivity values. By integrating the feedback sub-link, the secondary feedback sub-link, and the associated model, a self-optimizing feedback link is formed, enabling the cross-dimensional feature collaboration module to continuously optimize the feature extraction and generation process based on the water environment state mapping results.

6. The model construction method for integrated analysis of marine water environment according to claim 2, characterized in that, The dynamically triggered rule is set to adaptively update the cycle. At the end of each update cycle, newly collected multi-ocean-area element monitoring data is acquired within that cycle. Based on the newly collected data, the fluctuation range, fluctuation frequency, and inter-element correlation strength at each ocean-area level are recalculated. The spatiotemporal stratification threshold and gradient conditions are updated to form an updated version of the dynamically triggered rule, including: At the end of each update cycle, newly collected monitoring data of nearshore, offshore and offshore waters are acquired to form the real-time monitoring dataset for the current cycle. Data cleaning is performed on the real-time monitoring dataset for the current period to remove abnormal data points caused by sensor failure or communication anomalies, thereby obtaining effective real-time monitoring data. Effective real-time monitoring data for each sea area level are grouped by element type to form real-time datasets of physical elements, chemical elements, and biological elements. For the real-time dataset of physical elements at each sea area level, the data is segmented according to the time series. The fluctuation range and frequency of the physical element values ​​within each segment are calculated to obtain the real-time values ​​of the fluctuation range and frequency of the physical elements at that sea area level. Using the same method, the real-time values ​​of the fluctuation range and frequency of chemical elements and the real-time values ​​of the fluctuation range and frequency of biological elements were calculated for each sea area level. Based on the real-time values ​​of the fluctuation range of physical elements at the current sea level and the real-time values ​​of the fluctuation range of physical elements at other sea levels, calculate the real-time values ​​of the difference coefficients of the fluctuation range of physical elements; using the same method, calculate the real-time values ​​of the difference coefficients of the fluctuation range of chemical elements, the real-time values ​​of the difference coefficients of the fluctuation range of biological elements, and the real-time values ​​of the difference coefficients of the fluctuation frequency of each element. Based on the real-time value of the difference coefficient of the fluctuation range, the spatial stratification threshold of the element monitoring data is updated according to the sea area level. The update formula is: new spatial stratification threshold = α × historical spatial stratification threshold + (1-α) × candidate value of spatial stratification threshold calculated based on the real-time value of the difference coefficient, where α is the historical weight coefficient, which is dynamically adjusted according to the stability of marine environmental changes. The more drastic the environmental changes, the smaller the value of α. Based on the real-time value of the difference coefficient of fluctuation frequency, the time-series threshold of the element monitoring data is updated according to the time series, and the same weighted update method as the spatial-series threshold is adopted. The updated spatial stratification threshold is integrated with the temporal stratification threshold to form the updated spatiotemporal stratification threshold. The correlation strength between physical and chemical elements in the real-time monitoring data of the current period is extracted, and the real-time sequence of the correlation strength between physical and chemical elements is calculated. The real-time sequence of the correlation strength between chemical and biological elements and the real-time sequence of the correlation strength between physical and biological elements are calculated in the same way. Analyze the difference in correlation strength between adjacent time points in the real-time sequence of each correlation strength, statistically analyze the distribution pattern of the difference, and identify the performance characteristics of multiple gradual change stages in the current period as the correlation strength rises from the initial value to the significant correlation value. For each identified progressive change stage, the stage threshold of the association strength is updated so that the difference between the stage thresholds of adjacent stages is distributed in a gradient according to the progressive change pattern observed in the current period, thus forming the updated gradient condition. The updated spatiotemporal stratification threshold is integrated with the updated gradient conditions to form a new version of the dynamic triggering rule in the current update cycle; The new version of the dynamic triggering rules is compared with the previous version. The rate of change of the spatiotemporal stratification threshold and the rate of change of the gradient conditions are calculated. If the rate of change exceeds the preset rule stability threshold, a rule update alarm is triggered, indicating that the marine environment has changed significantly.

7. The model construction method for integrated analysis of marine water environment according to claim 3, characterized in that, The process of coupling the temporal variation maps of correlation strength at different sea level levels with the spatial diffusion curves of correlation ranges, and adjusting the correlation strength values ​​and spatial diffusion strengths of each sea level based on the spatial positional relationships between sea levels, so that the coupled trajectory reflects the element correlation and transmission characteristics between sea levels, includes: Extract spatial location relationship data at different sea level. The spatial location relationship data includes the boundary distance between nearshore sea areas and near-shore sea areas, the boundary distance between near-shore sea areas and offshore sea areas, as well as hydrological data such as water flow velocity, water flow direction, and turbulence diffusion coefficient between each sea level. The correlation transfer coefficient at the level of adjacent sea areas is calculated. The correlation transfer coefficient is determined based on the water exchange efficiency and turbulent mixing intensity between the two sea areas. The water exchange efficiency is calculated using flow monitoring data at the sea area boundary, and the turbulent mixing intensity is estimated using the turbulent kinetic energy dissipation rate output by the ocean numerical model. For the temporal variation map of the correlation intensity in nearshore waters, the correlation intensity values ​​at each time node are extracted, multiplied by the correlation transmission coefficient between nearshore and offshore waters, and the correlation intensity correction value transmitted to offshore waters is obtained. The corresponding time node values ​​of the temporal variation map of the correlation intensity in offshore waters are adjusted based on the correlation intensity correction value. Using the same method, the adjusted correlation strength value of the nearshore sea area is multiplied by the correlation transmission coefficient between the nearshore and offshore sea areas to obtain the correlation strength correction value transmitted to the offshore sea area, and the corresponding time node values ​​of the correlation strength time series change map of the offshore sea area are adjusted. For the spatial diffusion curve of the associated range in nearshore waters, the spatial diffusion intensity index at each time node is extracted. Based on the water flow direction and turbulence diffusion coefficient in nearshore-ocean waters, the probability distribution of diffusion direction is adjusted to make the probability distribution of diffusion direction consistent with the distribution of water flow direction. At the same time, the diffusion intensity index is multiplied by the association transfer coefficient to obtain the initial diffusion intensity of the associated range in nearshore waters, which is used to adjust the value of the starting segment of the spatial diffusion curve of the associated range in nearshore waters. Based on the adjusted spatial diffusion curve of the associated range in the nearshore waters, the initial segment value of the spatial diffusion curve of the associated range in the offshore waters is adjusted in the same way to make the diffusion of the associated range between the sea areas continuous. The peak time of the temporal variation spectrum of the correlation intensity of each sea area level after coupling is extracted, the difference of the peak time of the adjacent sea area level is calculated, and the difference is compared with the theoretical time range of the correlation and transmission between sea area levels. The theoretical time range is calculated based on the boundary distance, the water flow velocity range and the turbulence diffusion time scale. If the difference exceeds the theoretical time range, the correlation and transmission coefficient is readjusted until the difference falls within the theoretical time range. Extract the diffusion intensity sequence of the spatial diffusion curves of each sea area level after coupling, calculate the cross-correlation function of the diffusion intensity sequence of adjacent sea areas, determine the delay time corresponding to the maximum cross-correlation, compare the delay time with the theoretical time range of the transmission of element association between sea areas, if the delay time exceeds the theoretical time range, adjust the transmission relationship of the spatial diffusion intensity index until the delay time meets the theoretical time range. The integrated and adjusted temporal variation map of the correlation intensity of each sea area level and the spatial diffusion curve of the correlation range form a spatiotemporal coupling trajectory, which includes the correlation transmission path, transmission time range and transmission intensity probability distribution between sea area levels.

8. The model construction method for integrated analysis of marine water environment according to claim 1, characterized in that, The integrated dynamic triggering rules, spatiotemporal coupling trajectories, cross-dimensional feature collaboration modules, and self-optimizing feedback links form an integrated marine water environment analysis model. This model processes newly acquired multi-ocean-area element monitoring data and adjusts the parameters of the cross-dimensional feature collaboration modules in real time based on the self-optimizing feedback links. This enables the integrated marine water environment analysis model to dynamically adapt to different marine water environment states, thus completing the construction of the integrated marine water environment analysis model, including: The core control module of the integrated analysis model of marine water environment is set up. This core control module is used to store the parameters of dynamic triggering rules, the generation algorithm of spatiotemporal coupling trajectory, the configuration of each level of cross-dimensional feature collaboration module, and the associated model parameters of self-optimization feedback link. The parameters of dynamic triggering rules include spatiotemporal layering threshold and gradient conditions. The system is equipped with a multi-ocean-area data access module for the integrated analysis model of marine water environment. This module is used to receive newly collected monitoring data of elements in nearshore waters, offshore waters and offshore waters, classify and store the data according to the sea area level, and convert the data format so that the converted data meets the input requirements of dynamic triggering rules. The trajectory generation control module of the integrated analysis model of marine water environment is set up. This trajectory generation control module calls the dynamic triggering rules and spatiotemporal coupling trajectory generation algorithm in the core control module to process the transformed data output by the multi-sea area data access module, generate a new spatiotemporal coupling trajectory, and transmit the new spatiotemporal coupling trajectory to the cross-dimensional feature collaboration module. A feature co-processing module for the integrated analysis model of marine water environment is set up. This feature co-processing module loads the spatiotemporal analysis layer, feature derivation layer and state mapping layer of the cross-dimensional feature co-processing module, receives the new spatiotemporal coupled trajectory output by the trajectory generation control module, and sequentially completes the initial coupled feature extraction, multi-dimensional correlation derived feature generation and water environment state mapping, and outputs the water environment state probability distribution results. The parameter self-optimization module of the integrated analysis model of marine water environment is set up. This parameter self-optimization module loads the correlation model of the self-optimization feedback link, receives the water environment state probability distribution results output by the feature collaborative processing module, calculates the state deviation value and feature sensitivity value, inputs the correlation model to obtain the optimal feedback parameters, and transmits the optimal feedback parameters to the feature collaborative processing module and the trajectory generation and control module. The feature collaborative processing module adjusts the feature extraction dimension of the spatiotemporal parsing layer and the generation rules of the feature derivation layer based on the optimal feedback parameters, and the trajectory generation control module adjusts the correlation and transfer coefficients in the spatiotemporal coupled trajectory generation algorithm based on the optimal feedback parameters. The model adaptation verification module of the integrated marine water environment analysis model is set up. This module receives the new water environment state probability distribution results output by the feature co-processing module after parameter adjustment, and determines the water environment state with the highest probability from the new water environment state probability distribution results as the predicted state. The predicted state is compared with the historical water environment state records. If the number of times the predicted state matches the historical records exceeds the preset adaptation threshold, it means that the integrated marine water environment analysis model has achieved dynamic adaptation to the current marine water environment state. The core control module, multi-sea area data access module, trajectory generation and control module, feature collaborative processing module, parameter self-optimization module and model adaptation and verification module are integrated to form an integrated marine water environment analysis model. Multi-sea area adaptation tests were conducted on the integrated analysis model of marine water environment. Monitoring data of multiple sea area elements under different seasons and climatic conditions were input to enable the integrated analysis model of marine water environment to accurately map the water environment state through parameter self-optimization in various scenarios, thus completing the final construction of the integrated analysis model of marine water environment.

9. A model building system for integrated analysis of marine water environment, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the model building method for integrated analysis of marine water environment as described in any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. A processor of the model building system for integrated analysis of marine water environment reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the model building system for integrated analysis of marine water environment to perform the model building method for integrated analysis of marine water environment as described in any one of claims 1 to 8.

Images