Water Resources Management and Aquaculture Data Analysis Methods and Systems

Abstract

This invention relates to the field of water resource management and aquaculture data analysis technology, and provides a method and system for water resource management and aquaculture data analysis. The system includes real-time collection of key water quality data from different geographical locations through a sensor network deployed in water resource areas, covering multiple core indicators affecting water quality. Based on historical statistical distribution analysis of ammonia nitrogen concentration, aquaculture areas are identified and their geographical boundaries are defined. Water flow direction fields are generated through water flow vector data processing to identify the upstream and downstream locations of each aquaculture area. After rasterizing the geographical boundaries, the spatial transport process of relevant water quality indicators is simulated. Simultaneously, a trained neural network is used to predict changes in dissolved oxygen concentration. Finally, various prediction results are integrated to generate a report, achieving refined management and accurate trend prediction of water resources in aquaculture scenarios.

Description

Technical Field

[0001] This invention relates to the field of water resource management and aquaculture data analysis technology, and more specifically, to methods and systems for water resource management and aquaculture data analysis. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] Water resources are the core foundation for the development of the aquaculture industry. The dynamic changes in water quality parameters are directly related to the survival status and aquaculture benefits of farmed organisms, and also affect the ecological balance of regional water resources. Therefore, accurate monitoring, analysis and prediction of water quality data in water resource areas have become the key to ensuring the sustainable development of the aquaculture industry and the rational use of water resources.

[0004] In the prior art, Chinese Patent No. CN108268569A discloses a water resources monitoring data acquisition and analysis system and method based on big data technology. It collects water quantity and water quality data through the telemetry terminal of the water resources monitoring station. After being received by the distributed message processing module, the data is parsed, sorted, transformed and stored by the distributed data processing module. The data is then processed by the real-time display, machine learning and graph analysis of the streaming data application module. Finally, the results are presented through the visualization module, aiming to achieve efficient acquisition, storage and application of water resources data.

[0005] However, this patent has obvious technical limitations, making it difficult to meet the specific needs of aquaculture scenarios: First, the data processing of this technology focuses on data parsing, classification and storage, and basic statistical analysis. Although it mentions applications such as machine learning, it does not consider the spatial transport characteristics of water quality parameters under the action of water flow, resulting in the analysis of water quality parameters remaining at the static data level and failing to reflect their dynamic spatial distribution patterns. Second, the technology lacks synergy in processing multiple types of water quality parameters and does not construct a multi-parameter coupled prediction model. It only outputs results through single-dimensional data analysis or simple integration, resulting in prediction results that cannot accurately reflect the comprehensive trend of future water quality changes.

[0006] Therefore, there is an urgent need for a water resource management and aquaculture data analysis method and system to achieve accurate identification, dynamic simulation and trend prediction of water quality in aquaculture areas. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for water resource management and aquaculture data analysis to improve the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows:

[0008] Firstly, this application provides a method for water resource management and aquaculture data analysis, including:

[0009] By deploying a sensor network in water resource areas, target data from multiple geographical locations are collected in real time. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data.

[0010] Historical statistical distribution analysis is performed on the ammonia nitrogen concentration data continuously monitored at each geographical location to generate a dynamic threshold. Adjacent geographical locations where the ammonia nitrogen concentration is consistently higher than the dynamic threshold are connected into continuous areas, marked as aquaculture areas, and the geographical boundary data of the aquaculture areas are output.

[0011] Extract water flow vector data for each geographic location, including flow velocity and flow direction angle. Use inverse distance weighted interpolation to spatially interpolate the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data contains a composite vector of water flow direction angle and flow velocity for each geographic location.

[0012] Based on the water flow direction angle in the water flow direction field data, the upstream and downstream geographical locations of each aquaculture area are identified according to the upstream and downstream water flow directions, respectively.

[0013] The geographic boundary data of each aquaculture area is rasterized into a set of grid cells; the combined vector of the water flow direction angle and flow velocity corresponding to each grid cell is extracted from the water flow direction field data.

[0014] Using ammonia nitrogen concentration data as the pollutant concentration field, combined with the flow velocity synthesis vector, the finite difference numerical calculation of the advection-diffusion equation is performed on the rasterized grid to simulate the spatial transport process of ammonia nitrogen concentration between grid cells in a specified time period in the future, and output the spatial distribution prediction sequence of ammonia nitrogen concentration in each aquaculture area.

[0015] Dissolved oxygen concentration data and water temperature data in each aquaculture area are used as input features and fed into a recurrent neural network trained on historical data to generate future time series predictions of the average dissolved oxygen concentration in that aquaculture area.

[0016] By integrating the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area, a prediction sequence report of future water quality parameters is generated.

[0017] Furthermore, after collecting target data from multiple geographic locations, it also includes:

[0018] For each target data point, the historical mean and standard deviation are calculated using a sliding window of a preset time length. The current collected value is compared with the historical mean. If the difference exceeds a preset multiple of the standard deviation, it is identified as an isolated outlier and removed. At the same time, for data missing periods caused by communication interruption, the sensor is used to perform linear interpolation to fill in the missing data at adjacent normal collection times before and after the missing period.

[0019] Furthermore, the step of generating a dynamic threshold specifically includes:

[0020] Based on the ammonia nitrogen concentration data continuously monitored at each geographical location, a sliding time window is used to extract historical concentration data within the window; the median of all historical concentration data within the window is taken as the baseline concentration for the window corresponding to that geographical location; the difference between each historical concentration data within the window and the baseline concentration is calculated, and the maximum value among all differences is taken as the concentration fluctuation range of that window.

[0021] Acquire the synchronous concentration data of all neighboring monitoring points within a preset radius of the geographic location in the sensor network, take the median of the synchronous concentration data of all neighboring monitoring points as the neighboring concentration benchmark, and calculate the difference between the neighboring concentration benchmark and the baseline concentration benchmark as the concentration deviation.

[0022] The adjustment amount is obtained by adding the concentration fluctuation amplitude and the concentration deviation; the dynamic threshold that changes dynamically with time and spatial environment is obtained by adding the baseline concentration benchmark and the adjustment amount.

[0023] Furthermore, the step of connecting adjacent geographical locations where ammonia nitrogen concentrations consistently exceed a dynamic threshold into a continuous region specifically includes:

[0024] Geographic locations where ammonia nitrogen concentration exceeds a dynamic threshold within a preset time period are marked as the initial point set;

[0025] Perform Delaunay triangulation on the initial point set to construct a triangular network.

[0026] A region growing algorithm is performed on the common edge cells of the triangular network to form a closed continuous region and determine its geographical boundary.

[0027] Furthermore, after performing a region growing algorithm on the common edge cells of the triangular network to form a closed continuous region, the process also includes:

[0028] Identify the grid cells that constitute the geographical boundaries of a continuous region;

[0029] Obtain the water flow direction angle for each boundary grid cell;

[0030] The boundary grid cells whose flow direction angle points into the interior of the continuous region are marked as inflow-type boundaries;

[0031] The boundary grid cells whose flow direction angle points to the outside of the continuous region are marked as outflow boundaries;

[0032] Spatial distribution data of inflow and outflow boundaries are used as supplementary attributes to geographic boundary data.

[0033] Furthermore, the steps of identifying upstream and downstream geographical locations separately include:

[0034] Based on the water flow direction angle corresponding to each geographical location in the water flow direction field data, determine the upstream and downstream water flow directions for each aquaculture area;

[0035] For each aquaculture area, spatial extension identification is performed along the opposite direction of water flow from its geographical boundary data, and geographical locations within the extension range but not included in the aquaculture area are marked as upstream geographical locations;

[0036] For each aquaculture area, spatial extension identification is performed along the direction of water flow along its geographical boundary data, and the geographical locations within the extension range are marked as downstream geographical locations.

[0037] Furthermore, the recurrent neural network is a long short-term memory network; the steps of training with historical data include:

[0038] The time series of dissolved oxygen concentration and water temperature synchronously monitored in each aquaculture area during historical periods were obtained as the training sample set.

[0039] The training sample set is divided into training subsequences according to time sequence. Each training subsequence contains observation data from multiple consecutive time steps. The water temperature data and historical dissolved oxygen concentration data in the training subsequences are used as input features, and the dissolved oxygen concentration data of the next time step is used as the prediction target. The long short-term memory network is trained under supervision until its prediction error converges to the preset range.

[0040] Furthermore, after generating the future water quality parameter prediction sequence report, it also includes:

[0041] The predicted spatial distribution sequence of ammonia nitrogen concentration and the predicted time series of dissolved oxygen concentration were normalized with their respective standard limits.

[0042] The normalized ammonia nitrogen concentration index and dissolved oxygen concentration index are weighted and fused to generate a comprehensive water quality index for the aquaculture area within a specified time period.

[0043] Based on the changing trend and numerical range of the comprehensive water quality index, a matching is performed in a preset reference table to output the assessment level of the aquaculture area within a specified time period.

[0044] Secondly, this application also provides a water resource management and aquaculture data analysis system, including:

[0045] The data acquisition module is used to collect target data from multiple geographical locations in real time through a sensor network deployed in the water resource area. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data.

[0046] The region identification module is used to perform historical statistical distribution analysis on the ammonia nitrogen concentration data continuously monitored for each geographical location, generate a dynamic threshold, connect adjacent geographical locations where the ammonia nitrogen concentration is continuously higher than the dynamic threshold into a continuous region, mark it as an aquaculture region, and output the geographical boundary data of the aquaculture region.

[0047] The flow field construction module is used to extract water flow vector data for each geographic location, including flow velocity magnitude and flow direction angle. It uses inverse distance weighted interpolation to spatially interpolate the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data contains a composite vector of water flow direction angle and flow velocity for each geographic location.

[0048] The flow direction analysis module is used to identify the upstream and downstream geographical locations of each aquaculture area based on the water flow direction angle in the water flow direction field data, according to the upstream and downstream water flow directions.

[0049] The grid generation module is used to rasterize the geographic boundary data of each aquaculture area into a set of grid cells; and to extract the composite vector of the water flow direction angle and velocity corresponding to each grid cell from the water flow direction field data.

[0050] The simulation module is used to perform finite difference numerical calculations of the advection-diffusion equation on a rasterized grid, using ammonia nitrogen concentration data as the pollutant concentration field and combining it with the flow velocity synthesis vector. It simulates the spatial transport process of ammonia nitrogen concentration between grid cells in a specified time period in the future and outputs a predicted sequence of ammonia nitrogen concentration spatial distribution for each aquaculture area.

[0051] The prediction module is used to take dissolved oxygen concentration data and water temperature data in each aquaculture area as input features, input them into a recurrent neural network trained with historical data, and generate future time series prediction values ​​of the average dissolved oxygen concentration in the aquaculture area.

[0052] The report generation module integrates the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area to generate a report on the prediction sequence of future water quality parameters.

[0053] The beneficial effects of this invention are as follows:

[0054] This invention utilizes a sensor network deployed in water resource areas to collect real-time target data such as dissolved oxygen concentration, ammonia nitrogen concentration, water flow rate, and water temperature at various geographical locations. First, dynamic thresholds are generated from the continuously monitored ammonia nitrogen concentration data at each location through historical statistical distribution analysis. Adjacent locations where ammonia nitrogen concentration consistently exceeds the threshold are connected and marked as aquaculture areas, and their geographical boundary data is output. Next, water flow vector data at each location is extracted, and inverse distance weighted interpolation is used to generate water flow direction field data covering all monitored locations, thereby distinguishing the upstream and downstream locations of each aquaculture area. Subsequently, the geographical boundary data of the aquaculture areas is converted into a set of grid cells, and water flow-related data corresponding to each grid cell are extracted. Using ammonia nitrogen concentration as the pollutant concentration field, combined with flow velocity data, advection-diffusion equation numerical calculations are performed to simulate the future spatial transport process of ammonia nitrogen concentration and output a distribution prediction sequence. Simultaneously, dissolved oxygen concentration and water temperature data within the aquaculture area are input into a recurrent neural network trained on historical data to generate a predicted future time series value for the average dissolved oxygen concentration. Finally, the two types of prediction results are integrated to generate a water quality parameter prediction report. By combining the characteristics of aquaculture scenarios to construct a regional identification mechanism, introducing spatial interpolation and water flow field analysis to simulate parameter transport, and adopting a multi-parameter collaborative prediction model, the accurate identification, dynamic simulation, and trend prediction of water quality in aquaculture areas are achieved. Attached Figure Description

[0055] Figure 1 A flowchart of a water resource management and aquaculture data analysis method provided by the present invention;

[0056] Figure 2 This is a schematic diagram of the triangular mesh in this invention.

[0057] Figure 3 This is a schematic diagram of the structure of a water resource management and aquaculture data analysis system provided by the present invention.

[0058] In the diagram: 1. Sensor network; 2. Geographic boundary of aquaculture area; 3. Triangulation network; 201. Data acquisition module; 202. Region identification module; 203. Flow field construction module; 204. Flow direction analysis module; 205. Grid generation module; 206. Simulation module; 207. Prediction module; 208. Report generation module. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0060] An embodiment of the present invention proposes a method for water resource management and aquaculture data analysis, comprising:

[0061] S101 collects target data from multiple geographical locations in real time through a sensor network 1 deployed in the water resource area. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data.

[0062] Specifically, sensor network 1 deployed in the water resource area collects target data from multiple geographical locations in real time. Sensor network 1 adopts a distributed deployment strategy, and monitoring points are preset according to the topographic features, water flow distribution and potential aquaculture area distribution of the water resource area to ensure that the collection range can fully cover the area to be monitored. Each sensor node transmits data back in real time through a wireless communication module. The collected target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data and water temperature data. The above four types of data are core parameters that characterize water resource quality and hydrological characteristics. Dissolved oxygen concentration is directly related to the adaptability of aquaculture organisms to the living environment. Ammonia nitrogen concentration is a key indicator reflecting the degree of pollution from aquaculture activities. Water flow data determines the diffusion range and rate of pollutants, and water temperature affects the stability of various chemical indicators in the water body.

[0063] After collecting target data from multiple geographical locations, the raw data needs to be preprocessed to eliminate the interference of outliers and the impact of missing data on subsequent analysis. This preprocessing process includes two core steps: outlier removal and data missing data imputation. Outlier removal employs a sliding window-based statistical analysis method. For each target data point, a historical mean and standard deviation are calculated using a sliding window of a preset time length (the preset time length can be adjusted according to the hydrological dynamics of the water resource area, typically ranging from 30 minutes to 2 hours, ensuring both the reflection of data temporal changes and avoiding response lag caused by an excessively long window). The principle is to construct a normal data fluctuation range using historical data within the sliding window; the historical mean represents the central tendency of the data within the window, and the standard deviation represents the dispersion of the data. The current collected value is compared with the historical mean. If the difference exceeds a preset multiple of the standard deviation (usually 2-3 times, based on the 3σ principle to accurately screen outliers), it is identified as an isolated outlier and removed. The beneficial effect of this design is that it can effectively eliminate outliers caused by sensor malfunctions, sudden interference, etc., ensuring data reliability. Meanwhile, for the data gaps caused by communication interruptions, linear interpolation is used to fill in the gaps. Specifically, data from the sensor at adjacent normal acquisition times before and after the gaps are extracted. Based on the continuity of data time series changes, supplementary data for the gaps are generated through linear fitting. The principle of this method is to use the correlation between adjacent time data to make reasonable inferences about the missing data. The beneficial effect is that it can ensure the integrity of the data sequence, avoid deviations in subsequent analysis caused by missing data, and ensure the continuity and integrity of the data time series.

[0064] S102, perform historical statistical distribution analysis on the continuously monitored ammonia nitrogen concentration data for each geographical location, generate a dynamic threshold, connect adjacent geographical locations where the ammonia nitrogen concentration is consistently higher than the dynamic threshold to form a continuous region, mark it as an aquaculture area, and output the geographical boundary data of the aquaculture area; such as Figure 2 Geographical boundaries of aquaculture areas in China 2.

[0065] Specifically, the principle behind generating dynamic thresholds lies in the fact that ammonia nitrogen, as a major product of the decomposition of pollutants such as uneaten feed and excrement in aquaculture activities, exhibits a significantly higher concentration in aquaculture areas compared to natural water bodies. However, the ammonia nitrogen concentration in natural water bodies fluctuates due to spatiotemporal factors such as hydrological dynamics and seasonal changes. Using fixed thresholds can easily lead to identification biases; therefore, it is necessary to construct a threshold system that dynamically adapts to time and space. Thus, the generation of dynamic thresholds is achieved collaboratively through three key sub-steps:

[0066] The first sub-step involves calculating the baseline concentration and the amplitude of concentration fluctuations. Specifically, historical ammonia nitrogen concentration data for each geographical location is extracted through a sliding time window. The setting of this sliding time window needs to be reasonably selected in combination with the hydrological dynamic characteristics of the water resource area (such as water flow velocity, water exchange cycle, etc.), and is usually set to 12 to 72 hours. The principle is to capture the natural fluctuation pattern of ammonia nitrogen concentration at that point through historical data samples of a sufficient duration, while avoiding insufficient sample size and inaccurate characterization of fluctuation patterns due to a window duration that is too short, or failure to respond to timely changes in concentration due to a window duration that is too long. After acquiring historical concentration data within the window, the median of the data is selected as the baseline concentration for the window corresponding to that geographical location. Compared with the mean, the median has a stronger ability to resist extreme value interference and can effectively avoid the impact of abnormally high or low values ​​caused by accidental factors (such as single pollutant emissions or instantaneous sensor errors) on the baseline value, ensuring that the baseline value can truly reflect the normal level of ammonia nitrogen concentration at that location. Subsequently, the difference between each historical concentration data within the window and the baseline concentration is calculated, and the maximum value among all differences is taken as the concentration fluctuation range of that window. The core function of this indicator is to quantify the maximum fluctuation range of ammonia nitrogen concentration at that location under normal conditions, providing a time-dimensional fluctuation basis for the subsequent dynamic adjustment of the threshold.

[0067] The second sub-step involves calculating the adjacent concentration benchmark and concentration deviation. The principle behind this is that the distribution of ammonia nitrogen concentration has spatial correlation. The concentration level at a single geographical location needs to be calibrated in conjunction with the concentration characteristics of the surrounding area to avoid misjudgments caused by differences in local natural water background concentrations (e.g., if the natural background ammonia nitrogen concentration in some areas is already high, relying solely on a single-point benchmark could easily lead to misjudgment as an aquaculture area). Specifically, firstly, the concurrent concentration data of all adjacent monitoring points within a preset radius of the geographical location in sensor network 1 are acquired. The value of this preset radius needs to be determined based on the sensor deployment density and water diffusion characteristics, ensuring coverage of the surrounding area that is correlated with the hydrological environment of the target location. Then, the median of the concurrent concentration data from all adjacent monitoring points is taken as the adjacent concentration benchmark. Using the median also ensures the stability of the adjacent area concentration benchmark. Finally, the difference between the adjacent concentration benchmark and the baseline concentration benchmark is calculated as the concentration deviation. This deviation quantifies the concentration difference between the target location and the surrounding area, providing a basis for the spatial dimension calibration of the threshold.

[0068] The third sub-step involves synthesizing the dynamic threshold. This is achieved by adding the concentration fluctuation amplitude and concentration deviation obtained earlier to obtain an adjustment amount, and then adding the baseline concentration to the adjustment amount to finally obtain a dynamic threshold that changes dynamically with time and spatial environment. The principle behind this synthesis logic is that the dynamic threshold must simultaneously cover the self-fluctuation in the time dimension and the regional differences in the spatial dimension, ensuring that the threshold can adapt to the ammonia nitrogen concentration distribution characteristics under different spatiotemporal conditions.

[0069] After generating the dynamic threshold, the process further involves connecting adjacent geographical locations where ammonia nitrogen concentrations consistently exceed the dynamic threshold into a continuous region, marking it as aquaculture areas, and outputting geographical boundary data. The core of this process lies in extracting the contours of aquaculture areas through spatial aggregation, ensuring that the identified areas possess spatial continuity and integrity, conforming to the spatial distribution characteristics of actual aquaculture activities. The specific implementation includes the following steps:

[0070] One aspect is the labeling of the initial point set. Geographical locations where the ammonia nitrogen concentration exceeds a dynamic threshold within a preset time period are marked as the initial point set. The value of this preset time period needs to be determined in conjunction with the continuous characteristics of aquaculture activities, and is usually between 24 and 72 hours. The principle is that the increase in ammonia nitrogen concentration caused by aquaculture activities is continuous, while the instantaneous increase in concentration (such as sudden pollution or sensor failure) usually does not exceed this duration. By filtering through the duration, abnormal points caused by non-aquaculture factors can be eliminated, ensuring the effectiveness of the initial point set. Its beneficial effect is to initially filter out interference points, reduce the computational load of subsequent spatial aggregation, and improve the accuracy of aquaculture area identification.

[0071] Secondly, the construction of the triangulation network 3 and the region growth are performed. First, the initial point set is subjected to Delaunay triangulation to construct the triangulation network 3. The principle of Delaunay triangulation is that it can maximize the minimum interior angle of the triangle, avoid generating narrow and long triangles, and ensure that the triangulation network 3 can reasonably represent the spatial distribution relationship of the initial point set, providing a reliable spatial topology structure for the association judgment of adjacent points. Then, the region growth algorithm is performed on the common edge units of the triangulation network 3 to form a closed continuous region. The core logic of the region growth algorithm is to take any point in the initial point set as the seed point, and include the triangulation units that share the same edge as the triangulation unit of the seed point and also belong to the initial point set into the growth region. This process is repeated until no unit that meets the conditions can be included, and finally a closed region is formed.

[0072] To further enrich the application value of geographic boundary data, after forming a closed continuous area, this step adds a boundary attribute supplementation step, which includes: First, identifying the grid cells constituting the geographic boundary of the continuous area, achieved through edge cell screening of triangular mesh 3, that is, selecting cells that are only adjacent to triangular cells within a single area as boundary grid cells; then obtaining the water flow direction angle of each boundary grid cell, which comes from the water flow direction field data in the subsequent S103 step. The principle is that the water flow direction directly affects the diffusion path and range of pollutants in the aquaculture area, and the water flow direction attribute of the boundary has important reference value for subsequent water quality prediction; next, marking the boundary grid cells with the flow direction angle pointing into the continuous area as inflow-type boundaries, and marking the boundary grid cells with the flow direction angle pointing out of the continuous area as outflow-type boundaries; finally, using the spatial distribution data of inflow-type boundaries and outflow-type boundaries as supplementary attributes of geographic boundary data. The beneficial effect of this supplementary step is that it enriches the information dimensions of the geographic boundary data, so that the output boundary data not only contains spatial contour information, but also has hydrodynamic attributes. This provides key boundary condition support for the subsequent upstream and downstream geographic location identification in step S104 and the ammonia nitrogen concentration transport simulation in step S106, and improves the synergy and practicality of the entire data analysis method.

[0073] S103, extract water flow vector data for each geographic location, including flow velocity and flow direction angle, and use inverse distance weighted interpolation to perform spatial interpolation on the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data includes a composite vector of water flow direction angle and flow velocity for each geographic location.

[0074] Specifically, the extraction of water flow vector data involves analyzing the signal amplitude representing water flow intensity in the raw data, matching it with the previously calibrated amplitude-velocity correspondence, and calculating the flow velocity. Simultaneously, the phase information representing the flow direction in the raw data is analyzed, and angle correction is performed using the geographic coordinates of the monitoring points to eliminate directional deviations during data acquisition and ensure that the flow direction angle matches the actual geographic direction. Water flow vector data, as a core parameter representing the state of water movement, directly determines the pollutant diffusion rate through its flow velocity, while the flow direction angle is a key basis for determining the pollutant migration path and upstream-downstream relationships. The subsequent spatial interpolation stage employs the inverse distance weighted interpolation method. Based on the principle of spatial correlation, the extracted water flow vector data is fully interpolated across the entire region. Specifically, first, the neighborhood monitoring range of the interpolation point is defined based on the sensor deployment density and the continuity of water movement; then, using the reciprocal of the spatial distance between the interpolation point and neighboring monitoring points as the base weight, a weighting coefficient k=1.2 is introduced to optimize the allocation logic, and the flow velocity and flow direction angle are weighted separately to complete the spatial interpolation of the vector data. The final result is a water flow direction field data covering all geographical locations. This data system integrates the water flow motion characteristics of the entire region. The water flow direction angle provides a direct basis for subsequent identification of upstream and downstream flow directions, while the velocity composite vector fully reflects the coupling characteristics of the strength and direction of water movement, avoiding the limitations of single-parameter representation. The specific calculation process is as follows:

[0075] Basic weights The formula is:

[0076] (1)

[0077] In the formula, For point With point Spatial distance between them; These are the weighting coefficients; For point Point Interpolation weights; points For interpolation points; points It is one of the monitoring points in the neighborhood.

[0078] For the point to be interpolated Its neighborhood monitoring point set is ,point Flow rate and flow direction angle Interpolation value is:

[0079] (2)

[0080] (3)

[0081] In the formula, For point The magnitude of the flow velocity; For point The angle of flow direction.

[0082] S104, based on the water flow direction angle in the water flow direction field data, identifies the upstream and downstream geographical locations of each aquaculture area according to the upstream and downstream water flow directions.

[0083] Specifically, firstly, based on the water flow direction angles corresponding to each geographical location in the water flow direction field data, the reverse and forward flow directions for each aquaculture area are determined. The principle is that the water flow direction angles in the water flow direction field data have achieved full-area coverage through spatial interpolation and accurately represent the actual water movement direction at each geographical location. Since the migration and diffusion of pollutants in aquaculture areas directly depend on water flow, the reverse flow direction corresponds to the source direction of pollutants, and the forward flow direction corresponds to the diffusion direction of pollutants. By defining these directions, a material migration correlation between the aquaculture area and the surrounding water bodies can be established. The specific implementation method is as follows: First, the water flow direction angles corresponding to all geographical locations within and around the aquaculture area are extracted. An azimuth coordinate system is established with due north as the reference, and each angle is converted into an absolute azimuth value. Then, the reverse flow direction (i.e., the azimuth range differing from the water flow direction angle by 180°±5°, reserving an error threshold to accommodate minor water flow disturbances) is calculated through angle inversion. The forward flow direction is directly calculated along the azimuth range corresponding to the water flow direction angle, ensuring the accuracy and anti-interference of the direction definition.

[0084] Secondly, for each aquaculture area, spatial extension identification is performed along its geographical boundary data in the opposite direction of water flow. Geographical locations within the extension range but not included in the aquaculture area are marked as upstream geographical locations. The principle is that upstream geographical locations are potential sources of pollutants in the aquaculture area. Extending along the opposite direction of water flow can accurately locate the possible source areas of pollutants, while excluding points within the aquaculture area to avoid misjudgment. The definition of the extension range needs to be dynamically adjusted in conjunction with the flow velocity synthesis vector generated by S103. The higher the flow velocity, the stronger the pollutant migration ability, and the larger the extension range value. During the extension process, the spatial topological relationship between each geographical location and the aquaculture area is determined. A grid cell collision detection algorithm is used to filter out geographical locations within the extension range that do not belong to the aquaculture area and mark them as upstream geographical locations. This accurately locates the source area of ​​pollution input, providing spatial guidance for subsequent tracing of pollution causes.

[0085] Finally, for each aquaculture area, spatial extension identification is performed along the downstream direction of its geographical boundary data, and geographical locations within the extension range are marked as downstream geographical locations. The principle is that downstream geographical locations are receiving areas for pollutant diffusion from the aquaculture area, and their water quality is directly affected by pollutant emissions from the aquaculture area. Extending along the water flow can completely cover the pollution impact range. The extension range is kept consistent with the upstream to ensure data symmetry. Simultaneously, the extension path is optimized by combining the outflow-type boundary distribution of the aquaculture area boundary (derived from the S102 boundary attribute supplementation section), prioritizing extension along the downstream direction corresponding to the outflow-type boundary to improve the targeting of downstream area identification. Its beneficial effect is to completely define the environmental impact range of aquaculture activities, providing a clear downstream simulation boundary for subsequent ammonia nitrogen concentration transport simulation.

[0086] S105, rasterize the geographic boundary data of each aquaculture area into a set of grid cells; extract the composite vector of the water flow direction angle and flow velocity corresponding to each grid cell from the water flow direction field data.

[0087] Specifically, an adaptive grid partitioning strategy is adopted. The side length of the grid cell is dynamically determined based on the spatial scale of the aquaculture area, the sensor deployment density, and the dynamic characteristics of water flow. It is typically set to 1 / 3 to 1 / 2 of the distance between adjacent sensors. This ensures that the grid cell can accurately fit the irregular geographical boundary contours of the aquaculture area while avoiding computational redundancy due to an overly fine grid or spatial accuracy loss due to an overly coarse grid. Rasterization transforms irregular geographical boundaries into a set of regular grid cells, achieving standardized discretization of spatial data and establishing a unified spatial computation coordinate system.

[0088] Then, the combined vector of the flow direction angle and velocity corresponding to each grid cell is extracted from the flow direction field data. A spatial coordinate mapping matching mechanism is used: first, the center point of each rasterized grid cell is used as a feature sampling point to extract its geographic coordinate information; then, a coordinate alignment algorithm is used to accurately match the coordinates of this sampling point with the global coordinate system of the flow direction field data, thereby extracting the combined vector of the flow direction angle and velocity corresponding to that coordinate position. If the center point coordinates of a grid cell do not have directly corresponding original flow direction field data, linear interpolation is used to complete the extracted data based on the flow parameters of 3 to 5 valid data points in the neighborhood of that center point. The principle behind this operation is that the transport process of ammonia nitrogen concentration in water is directly driven by the flow motion state at each spatial location. The pollutant migration pattern of each grid cell needs to be strongly correlated with its corresponding flow direction and velocity; therefore, precise flow dynamic parameters need to be bound to each grid cell.

[0089] S106 uses ammonia nitrogen concentration data as the pollutant concentration field and combines it with the velocity synthesis vector to perform finite difference numerical calculations of the advection-diffusion equation on a rasterized grid. It simulates the spatial transport process of ammonia nitrogen concentration between grid cells in a specified future time period and outputs a predicted sequence of ammonia nitrogen concentration spatial distribution for each aquaculture area.

[0090] Specifically, ammonia nitrogen concentration data is used as the pollutant concentration field. The principle behind this setup is that ammonia nitrogen is a characteristic pollutant generated by aquaculture activities, and its concentration distribution directly reflects the diffusion trend of aquaculture pollution. Using it as the core research object allows for a targeted focus on the impact of aquaculture activities on the water resource environment. The construction of the pollutant concentration field needs to be based on the grid cells after the S105 step. The historical and real-time ammonia nitrogen concentration data corresponding to each grid cell are spatially assigned to form an initial concentration distribution field that precisely matches the grid system, ensuring the spatial correlation and accuracy of subsequent transport simulations. Subsequently, the velocity synthesis vector of each grid cell extracted in the aforementioned steps is combined. This velocity synthesis vector includes the velocity magnitude and flow direction angle, and is the core dynamic parameter driving the transport of ammonia nitrogen concentration. The principle is that the migration of pollutants in water is mainly dominated by water flow. The velocity synthesis vector can completely characterize the dragging and carrying capacity of water flow for pollutants, providing key dynamic basis for the quantitative simulation of the transport process.

[0091] Based on the aforementioned fundamental data, finite difference numerical calculations of the advection-diffusion equations are performed on a rasterized grid. The principle is that the advection-diffusion equations are classical kinetic equations describing pollutant migration, accurately characterizing the advection migration of pollutants with water flow and the diffusion process driven by concentration gradients. The rasterized grid provides a discretized computational platform for the finite difference method. By discretizing the continuous spatial and temporal domains into grid cells and time steps, complex partial differential equations can be transformed into a solvable set of algebraic equations, facilitating engineering implementation and improving computational efficiency. In specific implementation, the advection term is discretized using an upwind scheme to avoid numerical dispersion and ensure the accuracy of pollutant migration direction calculations. The diffusion term is discretized using a central difference scheme to ensure the stability of the diffusion process calculation. At the same time, the time step needs to be adaptively adjusted according to the grid cell size and flow velocity to satisfy the CFL (Courant-Friedrichs-Lewy) stability condition and avoid oscillating distortion in the calculation results. The above numerical calculations simulate the spatial transport process of ammonia nitrogen concentration between grid cells within a specified future time period. The specified future time period is flexibly set according to actual management needs. The principle is that water resource management and aquaculture regulation require strategies tailored to anticipated needs at different time scales. The simulation process iteratively calculates the ammonia nitrogen concentration value of each grid cell at each time step, dynamically extrapolating the transport process through concentration transfer calculations between adjacent grid cells. The final output is a predicted spatial distribution sequence of ammonia nitrogen concentration for each aquaculture area, containing ammonia nitrogen concentration data for each time node and each grid cell within the specified future time period. The specific calculation process is as follows:

[0092] Set grid cells At time step ammonia nitrogen concentration is The time step is The space step size is (Assuming a uniform mesh). The velocity vector is in Thoughts and The directional components are:

[0093] (4)

[0094] (5)

[0095] In the formula, For the flow rate at The amount; The magnitude of the flow velocity; For the flow rate at The amount; The angle represents the direction of water flow.

[0096] Horizontal term Upwind discretization is employed:

[0097] (6)

[0098] Among them, if ,but ;like ,but .

[0099] diffusion term Using central difference discretization:

[0100] (7)

[0101] In the formula, This is the diffusion coefficient (determined by the characteristics of the water body).

[0102] The time derivative is obtained using forward differencing:

[0103] (8)

[0104] Therefore, the complete discrete equation is:

[0105] (9)

[0106] S107, Dissolved oxygen concentration data and water temperature data in each aquaculture area are used as input features and fed into a recurrent neural network trained with historical data to generate a future time series prediction value of the average dissolved oxygen concentration in the aquaculture area.

[0107] Specifically, the core logic of this step aims to leverage the strong correlation between dissolved oxygen concentration and water temperature within the aquaculture area, using a recurrent neural network trained on historical data to achieve time-series prediction. The specific technical details and principles are as follows: Dissolved oxygen concentration is a key environmental parameter for the survival of aquatic organisms, and its changes are significantly affected by water temperature—increased water temperature reduces dissolved oxygen saturation. Furthermore, there is a clear temporal correlation between changes in water temperature and dissolved oxygen concentration. Therefore, using both as input features can fully uncover the core driving factors affecting dissolved oxygen concentration changes, providing sufficient data support for accurate prediction. The recurrent neural network used is specifically a Long Short-Term Memory (LSTM) network. This network, through its gating mechanism (input gate, forget gate, output gate), can effectively capture long-range dependencies in time-series data, solving the gradient vanishing or gradient explosion problems of traditional recurrent neural networks (RNNs). It adapts to the long-term temporal variation characteristics of water quality data, ensuring the prediction model's ability to capture long-term trends in dissolved oxygen concentration.

[0108] The model training process is crucial for ensuring prediction accuracy. The specific implementation method is as follows: The training sample set should be selected from the time series of dissolved oxygen concentration and water temperature monitored synchronously during historical periods in each aquaculture area. Synchronous monitoring can ensure strict alignment of the two types of data in the time series dimension, avoiding distortion of the correlation between features and targets caused by time deviation. The historical periods should cover different seasons, hydrological conditions and aquaculture cycles to ensure that the sample set has sufficient representativeness and can improve the generalization ability of the model. The division of training subsequences must adhere to the principle of temporal continuity. Each subsequence contains observation data from multiple consecutive time steps. The number of time steps should be set in conjunction with the periodicity of dissolved oxygen concentration changes (e.g., covering a 12-24 hour period) to ensure that the subsequences can fully characterize the short-term fluctuations in dissolved oxygen concentration. Water temperature data and historical dissolved oxygen concentration data from the subsequences are used as input features, and the dissolved oxygen concentration at the next time step is used as the prediction target. Supervised training is employed to allow the model to learn the mapping relationship between the input features and the prediction target. During training, model parameters are iteratively optimized until the prediction error converges to a preset range. Convergence to the preset range indicates that the model's prediction accuracy meets the requirements of practical applications, avoiding overfitting or underfitting and ensuring the model's prediction reliability in real-world scenarios. The LSTM-based computation process is as follows:

[0109] LSTM introduces cell state and gating mechanisms (including forget gates) Input gate Output gate It can selectively remember, forget, and output information, thereby effectively capturing the long-term temporal dependence between dissolved oxygen concentration and characteristics such as water temperature.

[0110] This involves using an input gate to control the degree to which the current input information is integrated into the memory cells. That is, based on the hidden state at the previous moment... and current input The sigmoid function outputs a vector between 0 and 1, determining which new information is stored. Its formula is:

[0111] (10)

[0112] In the formula, The input gate activation vector; It is the sigmoid function, that is ; This is the weight matrix of the input gate; For time step The hidden state vector; For time step The input feature vector; is the bias vector of the input gate.

[0113] The forgetting gate is used to control the degree of forgetting in memory cells from the previous moment. Its formula is:

[0114] (11)

[0115] In the formula, The forget gate activation vector; Here is the weight matrix for the forget gate; This is the bias vector for the forget gate.

[0116] The output gate controls the effect of the memory cell on the output of the current hidden state; that is, the output gate regulates the memory cell. How to affect the current hidden state This is used for the final prediction. Its formula is:

[0117] (12)

[0118] In the formula, The output gate activation vector; This is the weight matrix of the output gate; This is the bias vector for the output gate.

[0119] New candidate memory information is generated using candidate memory cells. Each candidate memory cell generates a vector between -1 and 1 using the tanh function, representing potential new memory content. Its expression formula is:

[0120] (13)

[0121] In the formula, Candidate memory cell vectors; It is the hyperbolic tangent function. ; The weight matrix for candidate memory cells; is the bias vector of the candidate memory cells.

[0122] Memory cells are updated by combining the forgetting gate and the input gate. Among them, memory cells... It is the core of long-term memory, controlling the retention of old memories through the forgetting gate and the addition of new memories through the input gate, thus achieving continuous information updating. Its formula is:

[0123] (14)

[0124] In the formula, For time step Memory cell vector; For time step Memory cell vector; This is element-wise multiplication (Hadamard product).

[0125] The current hidden state is generated based on the output gate and memory cells. This is the output of the LSTM, used for prediction at the current time step and passed to the next time step. It filters information from the memory cells through the output gate, ensuring the output is relevant to the current task. Its formula is:

[0126] (15)

[0127] Hidden state is achieved through a fully connected layer. Predicted values ​​converted to dissolved oxygen concentration This completes the time series prediction. Its expression is:

[0128] (16)

[0129] In the formula, This is the weight matrix for the final prediction; This is the bias vector for the final prediction.

[0130] S108 integrates the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area to generate a future water quality parameter prediction sequence report.

[0131] Specifically, the principle of the integrated operation lies in the fact that the spatial distribution prediction sequence of ammonia nitrogen concentration focuses on the spatial diffusion trend of aquaculture pollution, while the time series prediction value of dissolved oxygen concentration focuses on the temporal changes of core environmental parameters for the survival of aquaculture organisms. These two are key complementary parameters characterizing water quality in aquaculture areas, and integration can achieve a comprehensive spatiotemporal representation of water quality conditions. Technically, through a time-series alignment mechanism, the data at each time point in the spatial distribution prediction sequence of ammonia nitrogen concentration are precisely matched with the time series prediction value of dissolved oxygen concentration during the same period. This binds full-time-series water quality parameters to each grid unit of the aquaculture area. The resulting future water quality parameter prediction sequence report can fully present the spatiotemporal evolution characteristics of water quality parameters, providing a structured and panoramic data foundation for subsequent assessments and avoiding the one-sidedness of assessments caused by single-dimensional data.

[0132] After generating the report, the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration are normalized with their respective standard limits. The principle is that the original predicted values ​​of ammonia nitrogen and dissolved oxygen have different dimensions (the former is in units of concentration, and the latter is also in units of concentration but with significantly different standard limits), and direct comparison and fusion would result in weight imbalance. Technically, it is necessary to retrieve the corresponding industry or national standard water quality limits for both (such as the ammonia nitrogen standard limit and dissolved oxygen standard limit for freshwater aquaculture) as benchmarks, and use the extreme value normalization method to eliminate the dimensional differences, ensuring that after normalization, both indices are within the [0,1] range, and that the trend of the index changes is positively correlated with water quality (the higher the ammonia nitrogen concentration, the closer the index is to 1; the higher the dissolved oxygen concentration, the closer the index is to 1), thereby achieving comparability of different water quality parameters and providing a unified data scale for subsequent fusion. The normalization expression is:

[0133] (17)

[0134] (18)

[0135] In the formula, This is a predicted value for ammonia nitrogen concentration; The standard limit for ammonia nitrogen concentration (such as the national standard upper limit); This is the normalized ammonia nitrogen concentration index; the closer the value is to 1, the higher the concentration. This is a predicted value for dissolved oxygen concentration; The standard limit for dissolved oxygen concentration (such as ideal saturation value or minimum required value); This is the normalized dissolved oxygen concentration index; the closer the value is to 1, the higher the concentration.

[0136] The normalized ammonia nitrogen concentration index and dissolved oxygen concentration index are weighted and fused to generate a comprehensive water quality index for the aquaculture area within a specified time period. The principle is that the impact weights of ammonia nitrogen and dissolved oxygen on aquaculture water quality and aquatic life differ, requiring quantification of their weighting based on the characteristics of the aquaculture scenario. Weighting coefficients are determined using the analytic hierarchy process (e.g., ammonia nitrogen is weighted at 0.6 and dissolved oxygen at 0.4 in high-density aquaculture scenarios, aligning with core pollution control needs). The two normalized indices are then weighted and summed to obtain the comprehensive water quality index for the specified time period, achieving a comprehensive quantitative characterization of water quality, balancing pollution levels with environmental suitability, and avoiding the limitations of single-parameter assessment. The expression is:

[0137] (19)

[0138] (20)

[0139] In the formula, Weighting of ammonia nitrogen concentration; The weighting is based on the dissolved oxygen concentration. The value is a comprehensive water quality index; the closer it is to 1, the worse the water quality.

[0140] Based on the changing trends and numerical ranges of the comprehensive water quality index, a matching process is performed in a pre-set reference table to output the assessment level of the aquaculture area within a specified time period. The principle behind this level output is to transform abstract index data into an intuitive basis for management decisions through standardized level classification. The pre-set reference table classifies levels according to the comprehensive water quality index range (e.g., [0, 0.3] is excellent, [0.3, 0.6] is good, and [0.6, 1.0] is poor), while also relating it to the index's changing trends (e.g., if it continues to rise, the level is downgraded by one level). After matching, the assessment level is output, thereby reducing the data interpretation cost for management and providing a clear and implementable basis for aquaculture regulation (such as water exchange and feed reduction) and water resource protection decisions.

[0141] like Figure 3 As shown, based on the same inventive concept, this embodiment provides a water resource management and aquaculture data analysis system, including:

[0142] The data acquisition module is used to collect target data from multiple geographical locations in real time through a sensor network 1 deployed in the water resource area. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data.

[0143] The region identification module is used to perform historical statistical distribution analysis on the ammonia nitrogen concentration data continuously monitored for each geographical location, generate a dynamic threshold, connect adjacent geographical locations where the ammonia nitrogen concentration is continuously higher than the dynamic threshold into a continuous region, mark it as an aquaculture region, and output the geographical boundary data of the aquaculture region.

[0144] The flow field construction module is used to extract water flow vector data for each geographic location, including flow velocity magnitude and flow direction angle. It uses inverse distance weighted interpolation to spatially interpolate the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data contains a composite vector of water flow direction angle and flow velocity for each geographic location.

[0145] The flow direction analysis module is used to identify the upstream and downstream geographical locations of each aquaculture area based on the water flow direction angle in the water flow direction field data, according to the upstream and downstream water flow directions.

[0146] The grid generation module is used to rasterize the geographic boundary data of each aquaculture area into a set of grid cells; and to extract the composite vector of the water flow direction angle and velocity corresponding to each grid cell from the water flow direction field data.

[0147] The simulation module is used to perform finite difference numerical calculations of the advection-diffusion equation on a rasterized grid, using ammonia nitrogen concentration data as the pollutant concentration field and combining it with the flow velocity synthesis vector. It simulates the spatial transport process of ammonia nitrogen concentration between grid cells in a specified time period in the future and outputs a predicted sequence of ammonia nitrogen concentration spatial distribution for each aquaculture area.

[0148] The prediction module is used to take dissolved oxygen concentration data and water temperature data in each aquaculture area as input features, input them into a recurrent neural network trained with historical data, and generate future time series prediction values ​​of the average dissolved oxygen concentration in the aquaculture area.

[0149] The report generation module integrates the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area to generate a report on the prediction sequence of future water quality parameters.

[0150] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0151] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for water resource management and aquaculture data analysis, characterized in that, include: By deploying a sensor network in water resource areas, target data from multiple geographical locations are collected in real time. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data. Historical statistical distribution analysis is performed on the ammonia nitrogen concentration data continuously monitored at each geographical location to generate a dynamic threshold. Adjacent geographical locations where the ammonia nitrogen concentration is consistently higher than the dynamic threshold are connected as continuous regions, marked as aquaculture areas, and the geographical boundary data of the aquaculture areas are output. Extract water flow vector data for each geographic location, including flow velocity and flow direction angle. Use inverse distance weighted interpolation to spatially interpolate the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data contains a composite vector of water flow direction angle and flow velocity for each geographic location. Based on the water flow direction angle in the water flow direction field data, the upstream and downstream geographical locations of each aquaculture area are identified according to the upstream and downstream water flow directions, respectively. The geographic boundary data of each aquaculture area is rasterized into a set of grid cells; Extract the combined vector of the flow direction angle and flow velocity corresponding to each grid cell from the flow direction field data; Using the ammonia nitrogen concentration data as the pollutant concentration field, combined with the flow velocity synthesis vector, the finite difference numerical calculation of the advection-diffusion equation is performed on the rasterized grid to simulate the spatial transport process of ammonia nitrogen concentration between grid cells in a future specified time period, and output the spatial distribution prediction sequence of ammonia nitrogen concentration in each aquaculture area. Dissolved oxygen concentration data and water temperature data in each aquaculture area are used as input features and fed into a recurrent neural network trained on historical data to generate future time series predictions of the average dissolved oxygen concentration in that aquaculture area. By integrating the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area, a future water quality parameter prediction sequence report is generated.

2. The water resource management and aquaculture data analysis method according to claim 1, characterized in that, After collecting target data from multiple geographical locations, the process also includes: For each target data point, the historical mean and standard deviation are calculated using a sliding window of a preset time length. The current collected value is compared with the historical mean. If the difference exceeds a preset multiple of the standard deviation, it is identified as an isolated outlier and removed. Simultaneously, for data gaps caused by communication interruptions, the sensor is used to perform linear interpolation to fill in the gaps using data from adjacent normal collection times before and after the gap period.

3. The water resource management and aquaculture data analysis method according to claim 1, characterized in that, The step of generating the dynamic threshold specifically includes: Based on the ammonia nitrogen concentration data continuously monitored at each geographical location, a sliding time window is used to extract historical concentration data within the window; the median of all historical concentration data within the window is taken as the baseline concentration for the window corresponding to that geographical location; the difference between each historical concentration data within the window and the baseline concentration is calculated, and the maximum value among all differences is taken as the concentration fluctuation range of that window. Obtain the synchronous concentration data of all neighboring monitoring points within a preset radius of the geographic location in the sensor network, take the median of the synchronous concentration data of all neighboring monitoring points as the neighboring concentration benchmark; calculate the difference between the neighboring concentration benchmark and the baseline concentration benchmark as the concentration deviation; The adjustment amount is obtained by adding the concentration fluctuation amplitude and the concentration deviation; the dynamic threshold that changes dynamically with time and spatial environment is obtained by adding the baseline concentration benchmark and the adjustment amount.

4. The water resource management and aquaculture data analysis method according to claim 3, characterized in that, The step of connecting adjacent geographical locations where the ammonia nitrogen concentration is consistently higher than the dynamic threshold into a continuous region specifically includes: Geographic locations where the ammonia nitrogen concentration is higher than the dynamic threshold within a preset time period are marked as the initial point set; Perform Delaunay triangulation on the initial point set to construct a triangulated network. A region growing algorithm is performed on the common edge cells of the triangular network to form a closed continuous region and determine its geographical boundary.

5. The water resource management and aquaculture data analysis method according to claim 4, characterized in that, After performing a region growing algorithm on the common edge cells of the triangular network to form a closed continuous region, the method further includes: Identify the grid cells that constitute the geographical boundary of the continuous region; Obtain the water flow direction angle for each boundary grid cell; The boundary grid cells whose flow direction angle points into the interior of the continuous region are marked as inflow-type boundaries; The boundary grid cells whose flow direction angle points to the outside of the continuous region are marked as outflow-type boundaries; The spatial distribution data of the inflow and outflow boundaries are used as supplementary attributes of the geographic boundary data.

6. The method for water resource management and aquaculture data analysis according to claim 1, characterized in that, The steps of identifying the upstream and downstream geographical locations respectively specifically include: Based on the water flow direction angle corresponding to each geographical location in the water flow direction field data, the counter-current direction and the downstream direction corresponding to each aquaculture area are determined. For each aquaculture area, spatial extension identification is performed along the opposite direction of water flow from its geographical boundary data, and geographical locations within the extension range but not included in the aquaculture area are marked as upstream geographical locations; For each aquaculture area, spatial extension identification is performed along the direction of water flow along its geographical boundary data, and the geographical locations within the extension range are marked as downstream geographical locations.

7. The method for water resource management and aquaculture data analysis according to claim 1, characterized in that, The recurrent neural network is a long short-term memory network; The steps of training with historical data include: The time series of dissolved oxygen concentration and water temperature synchronously monitored in each aquaculture area during historical periods were obtained as the training sample set. The training sample set is divided into training subsequences according to time sequence. Each training subsequence contains observation data from multiple consecutive time steps. The water temperature data and historical dissolved oxygen concentration data in the training subsequences are used as input features, and the dissolved oxygen concentration data of the next time step is used as the prediction target. The long short-term memory network is trained under supervision until its prediction error converges to a preset range.

8. The method for water resource management and aquaculture data analysis according to claim 1, characterized in that, After generating the future water quality parameter prediction sequence report, the method also includes: The predicted spatial distribution sequence of ammonia nitrogen concentration and the predicted time series value of dissolved oxygen concentration are respectively normalized with their corresponding standard limits. The normalized ammonia nitrogen concentration index and dissolved oxygen concentration index are weighted and fused to generate a comprehensive water quality index for the aquaculture area within a specified time period. Based on the changing trend and numerical range of the comprehensive water quality index, a matching is performed in a preset reference table to output the assessment level of the aquaculture area within a specified time period.

9. A water resource management and aquaculture data analysis system, based on the water resource management and aquaculture data analysis method according to claim 1, characterized in that, include: The data acquisition module is used to collect target data from multiple geographical locations in real time through a sensor network deployed in the water resource area. The target data includes dissolved oxygen concentration data, ammonia nitrogen concentration data, water flow data, and water temperature data. The region identification module is used to perform historical statistical distribution analysis on the ammonia nitrogen concentration data continuously monitored for each geographical location, generate a dynamic threshold, connect adjacent geographical locations where the ammonia nitrogen concentration is continuously higher than the dynamic threshold into a continuous region, mark it as an aquaculture region, and output the geographical boundary data of the aquaculture region. The flow field construction module is used to extract water flow vector data for each geographic location, including flow velocity magnitude and flow direction angle. It uses inverse distance weighted interpolation to spatially interpolate the vector data to generate water flow direction field data covering all geographic locations. The water flow direction field data contains a composite vector of water flow direction angle and flow velocity for each geographic location. The flow direction analysis module is used to identify the upstream and downstream geographical locations of each aquaculture area based on the water flow direction angle in the water flow direction field data, according to the upstream and downstream water flow directions. The grid partitioning module is used to rasterize the geographic boundary data of each aquaculture area into a set of grid cells; Extract the combined vector of the flow direction angle and flow velocity corresponding to each grid cell from the flow direction field data; The simulation module is used to perform finite difference numerical calculations of the advection-diffusion equation on the rasterized grid, using the ammonia nitrogen concentration data as the pollutant concentration field and combining it with the flow velocity synthesis vector, to simulate the spatial transport process of ammonia nitrogen concentration between grid cells in a future specified time period, and output a predicted sequence of ammonia nitrogen concentration spatial distribution for each aquaculture area. The prediction module is used to take dissolved oxygen concentration data and water temperature data in each aquaculture area as input features, input them into a recurrent neural network trained with historical data, and generate future time series prediction values ​​of the average dissolved oxygen concentration in the aquaculture area. The report generation module is used to integrate the spatial distribution prediction sequence of ammonia nitrogen concentration and the time series prediction value of dissolved oxygen concentration in each aquaculture area to generate a future water quality parameter prediction sequence report.

Images