A Summer Dissolved Oxygen Level Early Warning Method for Fishponds Based on Smart Grid Forecast Data

Abstract

This invention relates to a method for forecasting dissolved oxygen levels, specifically a method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data. The method includes the following steps: S1, constructing a high-resolution grid map of the fishpond area; S2, acquiring air pressure data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generating air pressure datasets for each grid unit through spatial interpolation; S3, acquiring dissolved oxygen data from multiple water quality monitoring stations located in the same coordinate system as the grid map; S4, acquiring forecast meteorological data for the target area, the forecast meteorological data including at least forecast air temperature and forecast air pressure, and spatially matching the forecast meteorological data with the grid map; S5, inputting the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of each grid unit into a pre-established dissolved oxygen warning model to calculate the forecast dissolved oxygen dataset for each grid unit.

Description

Technical Field

[0001] This invention relates to a method for forecasting dissolved oxygen levels, specifically a method for early warning of dissolved oxygen levels in fishponds during the summer based on intelligent grid forecast data. Background Technology

[0002] Summer is the riskiest season for aquaculture production. Prolonged high temperatures, frequent air pressure fluctuations, and frequent severe convective weather such as thunderstorms directly lead to drastic changes in dissolved oxygen levels in aquaculture ponds. Dissolved oxygen is a crucial indicator for maintaining normal respiration and growth in fish. When dissolved oxygen levels drop below a certain threshold, fish will surface to breathe, potentially leading to mass mortality and significant economic losses for fish farmers. Existing technologies, such as patent document ZL202510648084.8, disclose a method for predicting freshwater fishpond water temperature. This method can construct a fishpond water temperature forecasting model based on intelligent grid forecasting data, achieving gridded water temperature prediction and reducing errors caused by excessively large grids or neglect of local differences in traditional forecasting methods. However, this patent and most existing technologies primarily focus on water temperature prediction and do not systematically address the issue of spatially refined early warning of dissolved oxygen levels. Regarding dissolved oxygen level early warning, existing technologies generally employ two methods. The first method involves installing a single-point water quality monitoring probe in the fishpond to read the dissolved oxygen level at that point in real time, triggering an alarm when the value falls below a set threshold. This method only reflects the dissolved oxygen level at the probe's location. However, a fishpond can range from several acres to tens of acres, and dissolved oxygen levels vary significantly across different areas due to differences in water depth, aquatic plant distribution, feeding locations, aerator layout, and inlet / outlet points. The value measured by a single-point probe cannot represent the entire fishpond, often resulting in a situation where dissolved oxygen is adequate at the probe's location while fish in another corner are already experiencing oxygen deficiency and surfacing for air. The second method relies on farmers' experience or simple weather rules, such as turning on aerators in advance during hot, humid weather or low-pressure conditions. This method depends on personal experience, lacks quantitative data, and is prone to misjudgment or omission. The aforementioned existing technologies generally suffer from the following problems: First, insufficient spatial resolution, failing to identify spatial differences in dissolved oxygen levels within the fishpond, leading to the omission of oxygen-deficient areas. Second, the system relies on reactive or empirical predictions, lacking scientific early warning capabilities. By the time the probes alarm or fish are observed surfacing, losses are often already incurred or irreversible. Third, it is overly dependent on water quality monitoring probes, which require regular cleaning and calibration and are prone to malfunction or data drift. A failure in these probes paralyzes the entire early warning system. Fourth, traditional modeling methods for the relationship between meteorological factors and dissolved oxygen levels are too simplistic, mostly employing linear regression or empirical formulas. These methods fail to accurately depict the complex nonlinear relationships between various meteorological elements such as summer temperature, air pressure, sunlight, and wind speed and dissolved oxygen levels, resulting in significant prediction errors. Summary of the Invention

[0003] The technical problem solved by this invention is to provide a method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data, which can take into account spatial refinement, temporal predictability, high stability and high accuracy of prediction.

[0004] The basic solution provided by this invention is a method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data, comprising the following steps: S1. Construct a high-resolution grid map of the fishpond area, wherein the grid map covers the predicted target area and contains several grid cells; S2. Obtain air pressure data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate air pressure datasets for each grid unit through spatial interpolation; S3. Obtain dissolved oxygen data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate dissolved oxygen datasets for each grid unit using spatial interpolation. S4. Obtain the forecast meteorological data of the target area, the forecast meteorological data including at least forecast temperature and forecast air pressure, and spatially match the forecast meteorological data with the grid map to obtain the forecast meteorological dataset of each grid unit; S5. Input the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of the grid cell into the pre-established dissolved oxygen early warning model to calculate the forecast dissolved oxygen dataset of each grid cell. S6. Generate a fishpond dissolved oxygen level early warning distribution map based on the predicted dissolved oxygen level dataset.

[0005] Furthermore, S1 includes the following steps: S11. Collect the boundary coordinates of the fishpond to form a closed polygon; S12. Calculate the geometric center of the polygon based on the boundary coordinate points, and use it as the center point for mesh construction; S13. Using the center point as the origin, generate initial grid cells according to a preset grid size, so that the center point of the initial grid cells coincides with the origin; S14. Taking the initial grid cell as the center, iteratively expand outwards in all directions according to the grid size until the generated grid set completely covers the entire area within the boundary of the fishpond. S15. Delete grid cells that are completely outside the boundary of the fishpond, retain grid cells that intersect with the water area of ​​the fishpond, and assign a unique identifier and center point coordinates to each retained grid cell to obtain the high-resolution grid map.

[0006] Furthermore, S2 includes the following steps: S21. Identify multiple automatic weather stations located in the same coordinate system as the grid map; S22. Calculate the standard deviation of the distances between each automatic weather station and all boundary points on the fishpond boundary, and select stations that meet the preset spatial representativeness criteria based on the standard deviations:

[0007] in, This represents the standard deviation of the distances from the j-th automatic weather station to all fishpond boundary points. This represents the distance from the j-th station to the i-th boundary point. This represents the arithmetic mean of the distances from the j-th station to all M boundary points, where M is the total number of boundary points in the fishpond. S23. Based on the coordinates of the automatic weather stations and the standard deviation of their corresponding distances, generate the weather station coordinate values ​​for interpolation.

[0008] Furthermore, S2 also includes the following steps: S24. Determine the air pressure data of multiple automatic weather stations within a preset time range; S25. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S26. Calculate the distance between each target point and each automatic weather station; S27. Using the inverse distance weighted interpolation algorithm, the air pressure data of each automatic meteorological station is interpolated to each target point according to the distance to form the air pressure value of each target point. S28. Integrate the air pressure values ​​of all target points to form an air pressure dataset for a grid map; The inverse distance weighted interpolation algorithm calculates the air pressure value at the target point using the following formula:

[0009] in, This represents the air pressure value at the i-th target point. This represents the air pressure value at the j-th automatic weather station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th meteorological station, and α be a power parameter.

[0010] Furthermore, S3 includes the following steps: S31. Identify multiple water quality monitoring stations located in the same coordinate system as the grid map; S32. Calculate the standard deviation of the distance between each water quality monitoring station and the boundary point on the fishpond boundary, and select stations that meet the preset conditions for spatial representativeness based on the standard deviation. S33. Obtain dissolved oxygen data from multiple water quality monitoring stations within a preset time range; S34. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S35. Calculate the distance between each target point and each water quality monitoring station; S36. Using the inverse distance weighted interpolation algorithm, the dissolved oxygen data of each water quality monitoring station is interpolated to each target point according to the distance to form the dissolved oxygen value of each target point. S37. Integrate the dissolved oxygen values ​​of all target points to form the dissolved oxygen dataset of the grid map; wherein, the inverse distance weighted interpolation algorithm uses the following formula to calculate the dissolved oxygen value of the target points:

[0011] in, This represents the dissolved oxygen level at the i-th target location. This represents the dissolved oxygen value at the j-th water quality monitoring station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th water quality monitoring station, and let α be a power parameter.

[0012] Furthermore, S4 includes the following steps: S41. Connect to the meteorological intelligent grid forecasting system and obtain high-resolution forecast products covering the target forecast area; S42. Extract forecast temperature and forecast pressure data within the forecast time range from the high-resolution forecast product, wherein the forecast temperature includes at least the forecast maximum temperature; S43. Match the predicted temperature and predicted air pressure data with the spatial coordinates of the grid map; S44. Using a spatial interpolation method, the forecast temperature and forecast air pressure data are mapped to each grid cell in the grid map to obtain the forecast temperature and forecast air pressure values ​​for each grid cell. S45. Integrate the forecast temperature and forecast air pressure values ​​of all grid cells to form the forecast meteorological dataset of the grid map.

[0013] Furthermore, S5 includes the following steps: S51. Integrate the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of each grid unit to form a meteorological-water quality input feature set for each grid unit; S52. Input the meteorological-water quality input feature set into the pre-established dissolved oxygen level early warning model; S53. The dissolved oxygen level early warning model is constructed based on the random forest support vector regression algorithm, using the input features of each grid cell as independent variables and calculating the predicted dissolved oxygen level of each grid cell as dependent variables. S54. Integrate the predicted dissolved oxygen values ​​of all grid cells to form the predicted dissolved oxygen dataset of the grid map.

[0014] Furthermore, S6 includes the following steps: S61. Import the predicted dissolved oxygen dataset into a geographic information system tool; S62. Based on the risk assessment standard for dissolved oxygen in aquaculture, the predicted dissolved oxygen value is divided into multiple risk level intervals, and a corresponding visual label is assigned to each risk level interval. S63. Using GIS tools, the forecast dissolved oxygen value of each grid unit is mapped to the corresponding visual identifier according to its risk level range, and a distributed grid map of dissolved oxygen warning for the fishpond area is generated. S64. Output the dissolved oxygen level early warning grid distribution map.

[0015] Furthermore, the input features of the pre-established dissolved oxygen level early warning model in S5 are optimized, including the following steps: S55. Using the current forecast date as the time base, extract historical data within a preset time window as an evaluation sample set. Calculate the contribution of each candidate environmental factor to the change in dissolved oxygen on the evaluation sample set and sort them from largest to smallest contribution. The candidate environmental factors include at least multiple lag time data of air pressure, dissolved oxygen, temperature, and air pressure. S56. Based on the contribution ranking results, select the top N factors in terms of contribution from all candidate environmental factors as valid input features for the current forecast day, where N is a preset positive integer. S57. For each factor in the effective input features, search for the optimal lag period that minimizes the prediction error within its preset lag period search range, and use the historical data corresponding to the optimal lag period as the actual input value of the factor on the current forecast day. S58. Incorporate the latest observation data after the current forecast date into the historical database for updating the evaluation sample set for the next forecast date, thereby achieving daily dynamic adjustment of input features; The contribution is quantitatively evaluated based on the change in node purity in the random forest algorithm. The larger the change in node purity, the greater the impact of the factor on the prediction error and the higher the contribution. The lag time has a search range of one to three days.

[0016] The principle and advantages of this invention are as follows: The principle of the fishpond dissolved oxygen level early warning method proposed in this application can be understood as follows: First, the fishpond within the prediction area needs to be divided into small grids, each grid being an independent calculation unit. During division, a GPS is used to walk around the fishpond boundary, collecting the coordinates of the boundary points. These boundary points are connected to form an irregular polygon. Then, the geometric center of this polygon is calculated. Using this geometric center as the starting point, a square grid is drawn with a set side length, ensuring that the center of this grid coincides with the geometric center of the fishpond. Next, starting from this central grid, the grid is expanded outwards in the four directions (east, west, south, and north) one grid at a time with the same side length, until the area formed by all the grids completely covers the water surface within the fishpond boundary. Grids that fall entirely on the fishpond bank are deleted, retaining only those grids that intersect with the water surface. Each retained grid has its own number and center point coordinates, thus obtaining a high-resolution grid map of the fishpond. This grid map can divide a fishpond into dozens or even hundreds of small areas, each of which will subsequently receive a dissolved oxygen level prediction value.

[0017] Next, the data to be input into this grid model needs to be prepared. The data is divided into two categories: real-time data and forecast data. Real-time data includes air pressure and dissolved oxygen levels. Automatic weather stations record air pressure values, but there may only be a few weather stations around a fishpond, such as the weather monitoring station in Donghu Town, Fushun County. The air pressure values ​​of these few stations cannot directly represent the air pressure at each grid location within the fishpond because air pressure varies slightly spatially. This application uses an interpolation method to solve this problem. First, multiple automatic weather stations around the fishpond are selected, and the standard deviation of the distance from each station to all points on the fishpond boundary is calculated. The standard deviation reflects the spatial representativeness of the station; a larger standard deviation indicates that the station is relatively centrally located or relatively off-center from the edge of the fishpond, and can better reflect the gradient changes in air pressure around the fishpond. After selecting suitable stations, the actual air pressure data of these stations for the previous day is obtained. Then, each grid center point in the fishpond grid map is used as a target point, and the distance from each target point to each weather station is calculated. Using an inverse distance weighted interpolation algorithm, the closer a target point is to a monitoring station, the greater its air pressure value is assigned a weight. The air pressure values ​​from all stations are then weighted and averaged to obtain the air pressure value for that target point. This same calculation is performed on all target points, resulting in a previous day's actual air pressure value for each grid cell, forming the air pressure dataset for the entire fishpond grid. The same method is used for dissolved oxygen data. While water quality monitoring probes, such as fishpond water monitoring systems, may be installed inside or near the fishpond, the number of these probes is limited, and only a few points may have measured dissolved oxygen data. This application obtains the previous day's measured dissolved oxygen values ​​from these water quality monitoring stations. Similarly, representative stations are selected by calculating the standard deviation of the distance from the station to the fishpond boundary. Then, using the center point of each grid as the target point, the inverse distance weighted interpolation method is used to interpolate the dissolved oxygen values ​​of the stations onto each grid cell, ensuring that each grid cell receives a previous day's actual dissolved oxygen value.

[0018] The methods for acquiring forecast data differ. This application connects to a meteorological intelligent grid operational data platform, which provides high-resolution numerical weather forecast products. The forecast maximum temperature and forecast air pressure for the next 24 hours are extracted from this platform; these two elements are important meteorological factors affecting dissolved oxygen changes in water bodies. Since the meteorological forecast product itself is also gridded, but its grid resolution may be coarser than that of the fishpond grid, spatial matching between the forecast data and the fishpond grid map is necessary. Using bilinear interpolation or spline interpolation, the maximum temperature and air pressure values ​​of the forecast field are accurately mapped to each grid cell of the fishpond, thus providing each grid with a forecast maximum temperature and air pressure for the current day.

[0019] With the data for four environmental factors for each grid—the previous day's actual air pressure, the previous day's actual dissolved oxygen level, the current day's forecast maximum temperature, and the current day's forecast air pressure—a model is needed to calculate the predicted dissolved oxygen level for the current day. This application employs the Random Forest Support Vector Regression algorithm, abbreviated as RF-SVR. The training process for this model is based on historical data. Data from the meteorological monitoring station and fishpond water monitoring system in Donghu Town, Fushun County, from 2021 to 2024 were collected. Using the current day's forecast dissolved oxygen level as the dependent variable, and considering the lag in meteorological and water body propagation, meteorological and water body factors, including average air temperature, maximum air temperature, average water temperature, maximum water temperature, precipitation, water vapor pressure, air pressure, sunshine duration, relative humidity, wind speed, dissolved oxygen level, and turbidity, were selected as independent variables for three time periods: the current day, the previous day, and the two days prior. Pearson correlation analysis identified the four most important factors affecting dissolved oxygen levels as the previous day's actual air pressure, the previous day's actual dissolved oxygen level, the current day's forecast maximum temperature, and the current day's forecast air pressure. Then, using random forest regression (a nonlinear regression method), the Gini index impurity was calculated, and the smallest Gini index was selected as the splitting attribute. This created a binary recursive splitting tree in the training dataset, establishing a fishpond dissolved oxygen level early warning model based on these four environmental factors. The model parameters used default values ​​from the corresponding R language packages. Random forest assessed and filtered the importance of each feature, then input the filtered features into support vector regression for training. Support vector regression used the RBF kernel function to handle complex nonlinear relationships, and the model hyperparameters were automatically optimized using a Bayesian optimization framework. After training, the four environmental factor data for each grid cell were input into the model, and the model output the current day's forecast dissolved oxygen value for that grid cell. This process was repeated for all grid cells within the fishpond to obtain the entire fishpond grid's forecast dissolved oxygen level dataset. To verify the applicability of the model, a representative station in Tuqiao Town, Fushun County, was selected for dissolved oxygen level testing in 2025. The cross-validation results showed that the coefficient of determination R squared reached 0.89 and the root mean square error was as low as 0.21 mg / L, which was significantly better than traditional models such as stepwise regression and linear regression.

[0020] The final step is to visually display the prediction results. The predicted dissolved oxygen dataset is imported into a Geographic Information System (GIS). Based on common indicators of fish surfacing and fish kills in aquaculture ponds, dissolved oxygen levels are divided into several risk levels: greater than 5.0 mg / L is the safe zone, 4.0 to 5.0 mg / L is the warning zone, 2.0 to 4.0 mg / L is the danger zone, and less than 2.0 mg / L is the emergency zone. Each level is represented by a different color, such as green for safe, yellow for warning, orange for danger, and red for emergency. The GIS tool identifies the corresponding risk level and color for each grid based on the predicted dissolved oxygen value, fills the grid with color, and finally generates a fishpond dissolved oxygen warning grid distribution map. This map clearly shows which areas in the fishpond have low dissolved oxygen levels and which areas are normal. Aquaculture farmers can see red or orange grids and know in advance whether to turn on aerators or take other oxygenation measures.

[0021] Compared with the prior art, the method of this application has several outstanding advantages: First, traditional dissolved oxygen monitoring and early warning mainly rely on single-point water quality probes installed in fishponds. These probes can only measure the dissolved oxygen level at their specific location, while a fishpond is often large, and dissolved oxygen levels at different locations can vary significantly due to factors such as water depth, water flow, light, and algae distribution. The probe readings cannot represent the entire fishpond, and the risk of oxygen deficiency in other areas may be overlooked. By the time fish begin to surface or even die off in a certain part of the pond, the probe may not have yet triggered an alarm, leading to reactive measures and losses already incurred. This application constructs a high-resolution grid map, dividing the fishpond into many small grids. Each grid independently calculates a predicted dissolved oxygen level, identifying oxygen-deficient dead zones or high-risk areas as well as relatively safe areas, achieving precise spatial early warning. Fish farmers can accurately locate the areas requiring oxygenation based on the grid map, rather than blindly increasing oxygen throughout the entire pond, saving energy and enabling timely problem-solving.

[0022] Secondly, traditional methods often only provide alerts based on current measured values ​​and cannot predict future trends in dissolved oxygen levels. By the time fish farmers see dissolved oxygen levels have dropped to dangerous levels, the fish may have already begun to experience oxygen deprivation. This application utilizes intelligent meteorological grid forecasting products to predict dissolved oxygen trends and potential risk areas 24 hours in advance. This means shifting from reactive measures after discovering oxygen deficiency to knowing a day in advance which areas are likely to experience oxygen deprivation the next day, allowing for proactive preparations such as turning on aerators or reducing feed intake, thus eliminating risks before they occur.

[0023] Third, traditional methods rely excessively on the stable operation of water quality probes. Probes are prone to failure, electrodes require regular calibration, biofilms easily adhere to their surfaces leading to measurement drift, and probes are not inexpensive, deterring many small and medium-sized aquaculture farmers from installing them in large quantities. If a probe malfunctions, the entire early warning system collapses. The method in this application reduces the absolute dependence on a single probe or even just a few probes, because even with dissolved oxygen data from only a small number of water quality monitoring points, a high-density dissolved oxygen prediction field for the entire fishpond grid can be generated through spatial interpolation and machine learning models. Meteorological data comes from the meteorological department's operational platform, ensuring high stability, wide coverage, and minimal interruption. Therefore, the robustness and reliability of the entire system are improved.

[0024] Fourth, while existing studies on dissolved oxygen prediction based on meteorological factors have verified the influence of meteorological data on dissolved oxygen levels, most employ simple linear regression or stepwise regression models. These statistical models struggle to accurately characterize the complex nonlinear relationship between meteorological factors and dissolved oxygen, especially during hot summer weather when the interaction of factors such as temperature, air pressure, sunlight, and wind force results in a non-simple additive or subtractive effect on dissolved oxygen. This application employs a hybrid machine learning model combining random forest and support vector regression. Random forest can handle nonlinear relationships and feature interactions, while support vector regression uses kernel functions to map low-dimensional nonlinear problems to high-dimensional linear problems. The combination of these two models significantly improves forecast performance. Validated with actual data, the coefficient of determination reaches 0.89, and the root mean square error is only 0.21 mg / L. This accuracy is remarkably high in the field of dissolved oxygen forecasting and meets the accuracy requirements for early warnings in practical production.

[0025] Fifth, traditional methods typically produce predictions as a series of numbers or a curve, which is not intuitive enough. Fishery farmers may not be able to immediately determine which areas are at risk. This application ultimately uses GIS to generate a color-coded grid early warning map, with different colors indicating different risk levels. Fishery farmers can easily see where the risk is in the pond, even without specialized knowledge, leading to better decision support. Furthermore, this map can be printed out or stored on a mobile phone for convenient reference during pond inspections. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of an embodiment of the present invention. Detailed Implementation

[0027] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data includes the following steps: S1. Construct a high-resolution grid map of the fishpond area, wherein the grid map covers the predicted target area and contains several grid cells; S2. Obtain air pressure data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate air pressure datasets for each grid unit through spatial interpolation; S3. Obtain dissolved oxygen data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate dissolved oxygen datasets for each grid unit using spatial interpolation. S4. Obtain the forecast meteorological data of the target area, the forecast meteorological data including at least forecast temperature and forecast air pressure, and spatially match the forecast meteorological data with the grid map to obtain the forecast meteorological dataset of each grid unit; S5. Input the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of the grid cell into the pre-established dissolved oxygen early warning model to calculate the forecast dissolved oxygen dataset of each grid cell. S6. Generate a fishpond dissolved oxygen level early warning distribution map based on the predicted dissolved oxygen level dataset.

[0028] Specifically, taking a ten-acre fishpond as an example, the fish farmer used a handheld GPS device to collect eighty boundary coordinate points along the pond's edge, forming a closed polygon. The geometric center point of the polygon was calculated. Using the center point as the origin, a grid with a side length of twenty meters was created, generating the first square grid with its center coinciding with the origin. This grid was gradually expanded in the four directions (east, west, south, and north) until it covered the entire water surface. Grids completely on the bank were deleted, retaining those that intersected with the water surface, resulting in forty-eight valid grids, each with a unique number and center point coordinates. The fish farmer obtained the previous day's average air pressure values ​​from five automatic weather stations around the fishpond from the local meteorological bureau: station A was 1005.1 hPa, station B was 1004.7 hPa, station C was 1005.3 hPa, station D was 1004.9 hPa, and station E was 1005.0 hPa. Using the center points of 48 grids as target points, the previous day's air pressure value for each grid was obtained using inverse distance weighted interpolation. Grid 001 had an air pressure of 1005.0 hPa, grid 002 had 1004.9 hPa, and the values ​​for the remaining grids ranged from 1004.8 to 1005.2 hPa. The fish farmer installed two water quality monitoring probes in the fishpond to obtain the previous day's average dissolved oxygen value: probe 1 showed 4.8 mg / L, and probe 2 showed 5.2 mg / L. Using the center points of the 48 grids as target points, the previous day's dissolved oxygen value for each grid was obtained using inverse distance weighted interpolation. Grid 001 had 4.9 mg / L, and grid 002 had 5.0 mg / L. The fish farmer obtained the forecast maximum temperature of 34.5 degrees Celsius and the forecast air pressure of 1005.5 hPa from the intelligent meteorological grid forecast platform. The forecast data was spatially matched and bilinearly interpolated with the fishpond grid to obtain the forecast maximum temperature and pressure for each grid. Grid 001 had a forecast maximum temperature of 34.5 degrees Celsius and a forecast pressure of 1005.5 hPa, while grid 002 had a forecast maximum temperature of 34.6 degrees Celsius and a forecast pressure of 1005.4 hPa. The fish farmers input the previous day's pressure, dissolved oxygen level, forecast maximum temperature, and forecast pressure for each grid into a pre-trained random forest support vector regression model. The model output the forecast dissolved oxygen level for the current day for each grid: 5.3 mg / L for grid 001, 5.1 mg / L for grid 002, and 3.2 mg / L for grid 010. Aquaculture farmers import the predicted dissolved oxygen levels from 48 grids into GIS software. The data is categorized as follows: green for safe levels above 5.0 mg / L, yellow for warning levels between 4.0 and 5.0 mg / L, orange for dangerous levels between 2.0 and 4.0 mg / L, and red for emergency levels. This generates a distributed grid map for fishpond dissolved oxygen levels. Grid 010 on the map is displayed in orange, allowing farmers to turn on aerators in the corresponding area in advance. Traditional methods rely on two probes to monitor only two points, making it impossible to predict dissolved oxygen levels in other areas the following day. This method achieves full grid coverage and 24-hour advance warning.

[0029] In S5, fish farmers compile four data points for each grid cell: the previous day's air pressure, the previous day's dissolved oxygen level, the forecast maximum temperature for the current day, and the forecast air pressure for the current day, into an input vector. Then, a pre-trained dissolved oxygen level early warning model is invoked. This model is based on a random forest support vector regression algorithm, and the training data comes from historical observation data from three nearby fishponds over the past three years. After inputting these four values, the model outputs the forecast dissolved oxygen level for that grid cell. This process is repeated for each of the forty-eight grid cells, resulting in forty-eight forecast dissolved oxygen levels.

[0030] In S6, fish farmers import the predicted dissolved oxygen levels of these 48 grids into GIS software on their computers. The GIS software, based on commonly used dissolved oxygen risk standards in aquaculture, displays grids with predicted values ​​greater than 5 mg / L in green, 4-5 mg / L in yellow, 2-4 mg / L in orange, and less than 2 mg / L in red. This generates a fishpond dissolved oxygen warning distribution map, with each grid corresponding to a color. If fish farmers see several red grids near the northwest corner of the fishpond, they know that the dissolved oxygen level there will be very low the next day, requiring them to turn on the aerators in advance. The advantage of this method is that traditional methods can only rely on two probes to monitor two points, completely unaware of oxygen deficiency in other areas. This method, however, can predict the dissolved oxygen level of every grid in the entire fishpond, greatly improving spatial accuracy. Furthermore, traditional methods only show the current dissolved oxygen value and cannot predict the situation tomorrow; this method provides a warning 24 hours in advance, giving fish farmers ample preparation time.

[0031] S1 includes the following steps: S11. Collect the boundary coordinates of the fishpond to form a closed polygon; S12. Calculate the geometric center of the polygon based on the boundary coordinate points, and use it as the center point for mesh construction; S13. Using the center point as the origin, generate initial grid cells according to a preset grid size, so that the center point of the initial grid cells coincides with the origin; S14. Taking the initial grid cell as the center, iteratively expand outwards in all directions according to the grid size until the generated grid set completely covers the entire area within the boundary of the fishpond. S15. Delete grid cells that are completely outside the boundary of the fishpond, retain grid cells that intersect with the water area of ​​the fishpond, and assign a unique identifier and center point coordinates to each retained grid cell to obtain the high-resolution grid map.

[0032] Specifically, for example, fish farmers use high-precision mapping software on a tablet to select coordinates along the bank of their fishpond. The fishpond's shape resembles a twisted bean pod, not a regular rectangle or circle. The farmer selects a point every two meters, totaling 210 points, which are then connected sequentially to form a closed, irregular polygon. The farmer inputs the coordinates of these points into the computer, which automatically calculates the average of the coordinates of all vertices of the polygon to obtain its geometric center point. This center point is located southeast of the fishpond's surface. The farmer sets the grid side length to ten meters and draws a ten-meter by ten-meter square grid with the geometric center point as the origin, ensuring that the intersection of the two diagonals of this square falls precisely on the geometric center point. Then, starting from this central grid, a grid is expanded in four directions: east, west, south, and north, each grid with a side length of ten meters, with the newly generated grid adjacent to the central grid. The farmer checks whether the newly generated grid has covered all the fishpond's boundary points; if not, the grid continues to expand outwards in a ring until the outermost grid boundary extends beyond the fishpond's bank. Finally, the grids are trimmed using the pond boundary polygons. Grids completely on the shore are deleted, while those partially on the water and partially on the shore are retained. Only grids completely on the shore are deleted. This results in 120 valid grids, each with a number (e.g., 001 to 120) and the latitude and longitude coordinates of its center point. This completes the construction of a high-resolution grid map of an irregular fishpond. In existing technologies, many fish farmers manage the fishpond as a single unit or place only a probe in the center, ignoring the spatial differences within the pond. This method, through fine grid division, allows for the separate calculation and representation of different dissolved oxygen levels in different areas of the fishpond.

[0033] S2 includes the following steps: S21. Identify multiple automatic weather stations located in the same coordinate system as the grid map; S22. Calculate the standard deviation of the distances between each automatic weather station and all boundary points on the fishpond boundary, and select stations that meet the preset spatial representativeness criteria based on the standard deviations:

[0034] in, This represents the standard deviation of the distances from the j-th automatic weather station to all fishpond boundary points. This represents the distance from the j-th station to the i-th boundary point. This represents the arithmetic mean of the distances from the j-th station to all M boundary points, where M is the total number of boundary points in the fishpond. S23. Based on the coordinates of the automatic weather stations and the standard deviation of their corresponding distances, generate the weather station coordinate values ​​for interpolation.

[0035] Specifically, consider a fishpond with six automatic weather stations within a five-kilometer radius. The coordinates of these stations and the fishpond's grid map both use the CGCS2000 coordinate system to ensure accurate distance calculations. The fish farmer first obtains the coordinates of these six stations. Then, they calculate the distance from each station to every boundary point on the fishpond's boundary. There are 60 boundary points in total. For the first station, the distances to each of the 60 boundary points are calculated, resulting in 60 distance values. The average of these 60 values ​​is then calculated, followed by the sum of the squares of the differences between each distance and the average, and finally, the standard deviation. The standard deviations for the remaining five stations are calculated similarly. Let's assume the calculated results are: Station A standard deviation of 260 meters, Station B standard deviation of 180 meters, Station C standard deviation of 310 meters, Station D standard deviation of 95 meters, Station E standard deviation of 260 meters, and Station F standard deviation of 110 meters.

[0036] The fish farmer prioritized selecting stations with larger standard deviations. A larger standard deviation indicates a wider range of distance variation relative to the edge of the fishpond, suggesting a more suitable or less oblique location that better reflects spatial pressure variations around the fishpond. Therefore, the three stations with the largest standard deviations—station C (310 meters), station E (260 meters), and station A (230 meters)—were chosen for interpolation. The coordinates of these stations were recorded for subsequent pressure interpolation calculations. Traditional methods often simply select the closest stations to the fishpond or directly use meteorological station data from the administrative region where the fishpond is located. This may miss distant stations that significantly impact the fishpond's microclimate or introduce stations heavily influenced by local topography. This method, through standard deviation screening, improves the spatial representativeness of the selected stations.

[0037] S2 further includes the following steps: S24. Determine the air pressure data of multiple automatic weather stations within a preset time range; S25. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S26. Calculate the distance between each target point and each automatic weather station; S27. Using the inverse distance weighted interpolation algorithm, the air pressure data of each automatic meteorological station is interpolated to each target point according to the distance to form the air pressure value of each target point. S28. Integrate the air pressure values ​​of all target points to form an air pressure dataset for a grid map; The inverse distance weighted interpolation algorithm calculates the air pressure value at the target point using the following formula:

[0038] in, This represents the air pressure value at the i-th target point. This represents the air pressure value at the j-th automatic weather station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th meteorological station, and α be a power parameter.

[0039] Specifically, let the previous day's average air pressure values ​​for three automatic weather stations C, E, and A be: Station C 1005.3 hPa, Station E 1004.9 hPa, and Station A 1005.1 hPa. The fishpond grid map has 120 grid center points as target locations. Taking grid 001 as an example, input the air pressure values ​​of the three stations and the distances from the center point of grid 001 to the three stations: Station C 1200 meters, Station E 800 meters, and Station A 500 meters. With the power parameter α set to 2, an inverse distance weighted interpolation algorithm is used to output the air pressure value of grid 001 as 1005.0 hPa. The same operation is performed on all 120 grids, outputting the air pressure value for each grid to form an air pressure dataset. Traditional methods use the air pressure of a single station to represent the entire pond, resulting in significant errors. This method calculates the air pressure value independently for each grid, which better reflects the actual spatial distribution.

[0040] S3 includes the following steps: S31. Identify multiple water quality monitoring stations located in the same coordinate system as the grid map; S32. Calculate the standard deviation of the distance between each water quality monitoring station and the boundary point on the fishpond boundary, and select stations that meet the preset conditions for spatial representativeness based on the standard deviation. S33. Obtain dissolved oxygen data from multiple water quality monitoring stations within a preset time range; S34. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S35. Calculate the distance between each target point and each water quality monitoring station; S36. Using the inverse distance weighted interpolation algorithm, the dissolved oxygen data of each water quality monitoring station is interpolated to each target point according to the distance to form the dissolved oxygen value of each target point. S37. Integrate the dissolved oxygen values ​​of all target points to form the dissolved oxygen dataset of the grid map; wherein, the inverse distance weighted interpolation algorithm uses the following formula to calculate the dissolved oxygen value of the target points:

[0041] in, This represents the dissolved oxygen level at the i-th target location. This represents the dissolved oxygen value at the j-th water quality monitoring station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th water quality monitoring station, and let α be a power parameter.

[0042] Specifically, assuming the fish farmer installs three water quality monitoring probes, the standard deviation of the distance from each probe to fifty fishpond boundary points is calculated: Probe 1 has a standard deviation of 45 meters, Probe 2 has a standard deviation of 68 meters, and Probe 3 has a standard deviation of 22 meters. The two probes with the largest standard deviations, Probe 2 and Probe 1, are selected as interpolation stations. The average dissolved oxygen (DO) values ​​from the previous day are obtained for both probes: 4.8 mg / L for Probe 2 and 5.2 mg / L for Probe 1. A fishpond grid map has ninety grid center points as target locations. Taking grid 001 as an example, the DDO values ​​from the two probes and the distances from the center point of grid 001 to Probe 2 (30 meters) and to Probe 1 (60 meters) are input. The power parameter α is set to 2, and an inverse distance weighted interpolation algorithm is used to output the DDO value of grid 001 as 4.88 mg / L. The same operation is performed on all ninety grids, outputting the DDO value for each grid, forming a dissolved oxygen dataset. Traditional methods rely solely on probe readings, assuming that dissolved oxygen levels are uniform throughout the pond. This method, however, uses interpolation to reflect the differences in dissolved oxygen levels within the fishpond.

[0043] S4 includes the following steps: S41. Connect to the meteorological intelligent grid forecasting system and obtain high-resolution forecast products covering the target forecast area; S42. Extract forecast temperature and forecast pressure data within the forecast time range from the high-resolution forecast product, wherein the forecast temperature includes at least the forecast maximum temperature; S43. Match the predicted temperature and predicted air pressure data with the spatial coordinates of the grid map; S44. Using a spatial interpolation method, the forecast temperature and forecast air pressure data are mapped to each grid cell in the grid map to obtain the forecast temperature and forecast air pressure values ​​for each grid cell. S45. Integrate the forecast temperature and forecast air pressure values ​​of all grid cells to form the forecast meteorological dataset of the grid map.

[0044] Specifically, fish farmers access the intelligent grid weather forecasting platform via a mobile app. This platform provides 24-hour forecasts with a resolution of 500 meters. For example, the fish farmer received a forecast high of 37.5 degrees Celsius and a forecast air pressure of 998.6 hPa for August 11th. The fishpond grid map consists of sixty grids, each with a side length of fifteen meters. The coordinates of the center points of the sixty grids are spatially matched with the forecast grid, and bilinear interpolation is used to obtain the forecast high temperature and air pressure values ​​for each grid. Grid 001 forecasts a high temperature of 37.3 degrees Celsius and a forecast air pressure of 998.5 hPa; grid 002 forecasts a high temperature of 37.4 degrees Celsius and a forecast air pressure of 998.6 hPa; and grid 060 forecasts a high temperature of 37.6 degrees Celsius and a forecast air pressure of 998.7 hPa. This forms a forecast meteorological dataset for the sixty grids. Traditional methods use single temperature values ​​at the city level, while this method uses high-resolution grid forecasts and interpolates them to each grid in the fishpond, improving the accuracy of local meteorological elements.

[0045] S5 includes the following steps: S51. Integrate the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of each grid unit to form a meteorological-water quality input feature set for each grid unit; S52. Input the meteorological-water quality input feature set into the pre-established dissolved oxygen level early warning model; S53. The dissolved oxygen level early warning model is constructed based on the random forest support vector regression algorithm, using the input features of each grid cell as independent variables and calculating the predicted dissolved oxygen level of each grid cell as dependent variables. S54. Integrate the predicted dissolved oxygen values ​​of all grid cells to form the predicted dissolved oxygen dataset of the grid map.

[0046] Specifically, the farmers have already obtained the previous day's air pressure, dissolved oxygen level, forecasted maximum temperature, and forecasted air pressure for each grid cell. Taking grid cell 001 as an example, four feature values ​​are input: previous day's air pressure 1001.2 hPa, previous day's dissolved oxygen level 5.1 mg / L, forecasted maximum temperature 34.5°C, and forecasted air pressure 1001.5 hPa. These four values ​​are then input into a pre-trained random forest support vector regression model, which outputs a forecasted dissolved oxygen level of 5.3 mg / L for that grid cell. For all sixty grid cells, their respective feature values ​​are input, and the model outputs sixty forecasted dissolved oxygen levels: 5.3 for grid cell 001, 5.1 for 002, 3.2 for 030, 4.0 for 045, and 5.5 for 060. This forms a dataset of forecasted dissolved oxygen levels for sixty grid cells. The root mean square error of simple linear regression models in the prior art is usually above 0.5 mg / L, while the root mean square error of this model after cross-validation is 0.21 mg / L, which significantly improves the prediction accuracy.

[0047] S6 includes the following steps: S61. Import the predicted dissolved oxygen dataset into a geographic information system tool; S62. Based on the risk assessment standard for dissolved oxygen in aquaculture, the predicted dissolved oxygen value is divided into multiple risk level intervals, and a corresponding visual label is assigned to each risk level interval. S63. Using GIS tools, the forecast dissolved oxygen value of each grid unit is mapped to the corresponding visual identifier according to its risk level range, and a distributed grid map of dissolved oxygen warning for the fishpond area is generated. S64. Output the dissolved oxygen level early warning grid distribution map.

[0048] Specifically, fish farmers saved the predicted dissolved oxygen data set for sixty grids as a CSV file and imported it into QGIS software. They then imported the pond boundary shape file and the center point layers of the sixty grids into QGIS. Risk levels were set according to aquaculture standards: a predicted dissolved oxygen level greater than 5.0 mg / L was considered safe and displayed in green; 4.0 to 5.0 was considered warning and displayed in yellow; 2.0 to 4.0 was considered dangerous and displayed in orange; and less than 2.0 was considered emergency and displayed in red. QGIS automatically assigned colors based on the predicted dissolved oxygen value for each grid. Grid 001 with a predicted dissolved oxygen level of 5.3 was displayed in green, grid 002 with 5.1 was displayed in green, grid 030 with 3.2 was displayed in orange, grid 045 with 4.0 was displayed in yellow, and grid 060 with 5.5 was displayed in green. After generating the Tyson polygon grid map, it was exported as a PNG image. Seeing the orange area in the northwest of the map, the fish farmers turned on the aerators in that area in advance, preventing the fish from surfacing due to excessively low dissolved oxygen levels the following morning. Traditional methods either wait until the fish surface before taking action or waste electricity by aerating the entire pond overnight. This method achieves precise early warning and precise aeration.

[0049] The input feature optimization of the dissolved oxygen level early warning model pre-established in S5 includes the following steps: S55. Using the current forecast date as the time base, extract historical data within a preset time window as an evaluation sample set. Calculate the contribution of each candidate environmental factor to the change in dissolved oxygen on the evaluation sample set and sort them from largest to smallest contribution. The candidate environmental factors include at least multiple lag time data of air pressure, dissolved oxygen, temperature, and air pressure. S56. Based on the contribution ranking results, select the top N factors in terms of contribution from all candidate environmental factors as valid input features for the current forecast day, where N is a preset positive integer. S57. For each factor in the effective input features, search for the optimal lag period that minimizes the prediction error within its preset lag period search range, and use the historical data corresponding to the optimal lag period as the actual input value of the factor on the current forecast day. S58. Incorporate the latest observation data after the current forecast date into the historical database for updating the evaluation sample set for the next forecast date, thereby achieving daily dynamic adjustment of input features; The contribution is quantitatively evaluated based on the change in node purity in the random forest algorithm. The larger the change in node purity, the greater the impact of the factor on the prediction error and the higher the contribution. The lag time has a search range of one to three days.

[0050] For example, in a fishpond, the farmer has been running a dissolved oxygen level early warning model continuously for a month. On July 20th, the system needs to make a forecast for July 21st. The system first extracts historical observation data from the most recent 30 days, from June 21st to July 20th, as the evaluation sample set.

[0051] This data includes the previous day's actual dissolved oxygen level, actual air pressure, average temperature, vapor pressure, and average wind speed for each grid within the fishpond, as well as the same data for the corresponding two and three days prior, totaling 15 candidate factors. The system inputs this 30-day data into a random forest algorithm. The algorithm calculates the sum of the reductions in Gini index impurity caused by each factor during decision tree splitting, using this sum to quantify the contribution of each factor to dissolved oxygen prediction. After calculation, the system outputs the contribution ranking results of the 15 factors. The first factor was the dissolved oxygen level from the previous day, with a total decrease in impurity of 152; the second factor was the forecast maximum temperature for the day, with a total decrease in impurity of 123; the third factor was the actual air pressure from the previous day, with a total decrease in impurity of 98; the fourth factor was the forecast air pressure for the day, with a total decrease in impurity of 87; the fifth factor was the water vapor pressure from the previous day, with a total decrease in impurity of 56; and the total decrease in impurity of the remaining factors was all below 40.

[0052] The system selects the top four factors based on preset rules: the previous day's dissolved oxygen level, the forecast maximum temperature for the current day, the previous day's actual air pressure, and the forecast air pressure for the current day, as valid input features for the July 21st forecast. Next, the system searches for the optimal lag period for each of these four valid factors. For the previous day's dissolved oxygen level, the system attempts to predict the dissolved oxygen level on July 20th using dissolved oxygen values ​​from July 19th, July 18th, and July 17th as input. Comparison shows that using data from July 19th (1 day lag) results in the smallest prediction error, so its optimal lag period is determined to be 1. For the previous day's actual air pressure, comparisons are made using air pressure values ​​from July 19th, July 18th, and July 17th. Using data from July 18th (2 days lag) results in the smallest error, so its optimal lag period is determined to be 2. For the forecast maximum temperature for the current day, the forecast value directly corresponds to the temperature on July 21st, so there is no lag issue, and the optimal lag period is 0. The forecast air pressure for that day is also 0. In this way, the system determines the optimal input data sources for the four factors used in the July 21 forecast: the actual dissolved oxygen value of the previous day, July 20; the actual air pressure values ​​of the previous two days, July 19; the forecast maximum temperature value for July 21; and the forecast air pressure value for July 21.

[0053] After the forecast was completed on July 21, the system updated the sample set. Data from June 21 was removed from the evaluation sample set, and actual observation data from July 21 was added, resulting in a new evaluation sample set covering 30 days from June 22 to July 21. The system then recalculated the contribution ranking of 15 factors in this new sample set. Because the weather on July 21 was cloudy and rainy, unlike the sunny and hot weather of the previous ten days, the contribution ranking changed. The new ranking showed that the previous day's water vapor pressure increased from a decrease in impurity of 56 to 93, moving from 5th to 3rd place, while the previous day's air pressure decreased to 82, moving to 4th place. The forecast maximum temperature remained 2nd, but the value dropped to 101, and the previous day's dissolved oxygen remained 1st, but the value dropped to 134. Based on this, the system adjusted the effective input features, determining the top 4 contributing factors in the new sample set as the effective inputs for the July 22 forecast: the previous day's dissolved oxygen, the forecast maximum temperature, the previous day's water vapor pressure, and the previous day's actual air pressure. Simultaneously, the optimal lag period was re-searched for these four factors. It was found that the lag period for water vapor pressure changed from 0 to 1 under cloudy and rainy weather, meaning that the water vapor pressure of the previous day had the best predictive effect on the dissolved oxygen level of the current day, while data with a lag of 2 or 3 days had larger errors. The system applied these updated factor combinations and optimal lag periods to the dissolved oxygen forecast for July 22. Taking grid 015 as an example, the system input the actual dissolved oxygen value of 5.0 mg / L for grid 015 on July 21, the predicted maximum temperature of 30.2 degrees Celsius for grid 015 on July 22, the actual water vapor pressure of grid 015 on July 21, and the actual air pressure of grid 015 on July 20, and the model output the predicted dissolved oxygen value of 4.8 mg / L for grid 015 on July 22. Without a dynamic filtering mechanism and using four fixed initial factors, the forecast error for cloudy and rainy weather on July 21 might have increased because the important factor of water vapor pressure was not included.

[0054] After adopting dynamic screening, the model can automatically adjust the combination of input factors according to recent weather changes, ensuring that the forecast is always based on the most effective environmental factors at present. Aquaculture farmers can view the daily changes in factor rankings after dynamic screening through the system interface, understanding whether the main factors affecting dissolved oxygen levels in fishponds are air temperature, air pressure, or water vapor pressure. When the system indicates that water vapor pressure has ranked high in contribution for several consecutive days, farmers know that the risk of dissolved oxygen decline is higher than usual under recent rainy weather, and they need to prepare for oxygenation in advance. Most existing dissolved oxygen prediction models use a fixed set of input factors, which do not change regardless of weather conditions; factor combinations suitable for sunny days may not be suitable for rainy days. This solution re-evaluates factor contribution and optimal lag time daily, enabling the model to adapt to different weather conditions and maintain a high level of forecast accuracy under all-weather conditions.

[0055] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data, characterized in that: Includes the following steps: S1. Construct a high-resolution grid map of the fishpond area, wherein the grid map covers the predicted target area and contains several grid cells; S2. Obtain air pressure data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate air pressure datasets for each grid unit through spatial interpolation; S3. Obtain dissolved oxygen data from multiple water quality monitoring stations located in the same coordinate system as the grid map, and generate dissolved oxygen datasets for each grid unit using spatial interpolation. S4. Obtain the forecast meteorological data of the target area, the forecast meteorological data including at least forecast temperature and forecast air pressure, and spatially match the forecast meteorological data with the grid map to obtain the forecast meteorological dataset of each grid unit; S5. Input the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of the grid cell into the pre-established dissolved oxygen early warning model to calculate the forecast dissolved oxygen dataset of each grid cell. S6. Generate a fishpond dissolved oxygen level early warning distribution map based on the predicted dissolved oxygen level dataset.

2. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 1, characterized in that: S1 includes the following steps: S11. Collect the boundary coordinates of the fishpond to form a closed polygon; S12. Calculate the geometric center of the polygon based on the boundary coordinate points, and use it as the center point for mesh construction; S13. Using the center point as the origin, generate initial grid cells according to a preset grid size, so that the center point of the initial grid cells coincides with the origin; S14. Taking the initial grid cell as the center, iteratively expand outwards in all directions according to the grid size until the generated grid set completely covers the entire area within the boundary of the fishpond. S15. Delete grid cells that are completely outside the boundary of the fishpond, retain grid cells that intersect with the water area of ​​the fishpond, and assign a unique identifier and center point coordinates to each retained grid cell to obtain the high-resolution grid map.

3. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 2, characterized in that: S2 includes the following steps: S21. Identify multiple automatic weather stations located in the same coordinate system as the grid map; S22. Calculate the standard deviation of the distances between each automatic weather station and all boundary points on the fishpond boundary, and select stations that meet the preset spatial representativeness criteria based on the standard deviations: in, This represents the standard deviation of the distances from the j-th automatic weather station to all fishpond boundary points. This represents the distance from the j-th station to the i-th boundary point. This represents the arithmetic mean of the distances from the j-th station to all M boundary points, where M is the total number of boundary points in the fishpond. S23. Based on the coordinates of the automatic weather stations and the standard deviation of their corresponding distances, generate the weather station coordinate values ​​for interpolation.

4. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 3, characterized in that: S2 further includes the following steps: S24. Determine the air pressure data of multiple automatic weather stations within a preset time range; S25. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S26. Calculate the distance between each target point and each automatic weather station; S27. Using the inverse distance weighted interpolation algorithm, the air pressure data of each automatic meteorological station is interpolated to each target point according to the distance to form the air pressure value of each target point. S28. Integrate the air pressure values ​​of all target points to form an air pressure dataset for a grid map; The inverse distance weighted interpolation algorithm calculates the air pressure value at the target point using the following formula: in, This represents the air pressure value at the i-th target point. This represents the air pressure value at the j-th automatic weather station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th meteorological station, and α be a power parameter.

5. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 4, characterized in that: S3 includes the following steps: S31. Identify multiple water quality monitoring stations located in the same coordinate system as the grid map; S32. Calculate the standard deviation of the distance between each water quality monitoring station and the boundary point on the fishpond boundary, and select stations that meet the preset conditions for spatial representativeness based on the standard deviation. S33. Obtain dissolved oxygen data from multiple water quality monitoring stations within a preset time range; S34. Determine multiple target points corresponding to the grid map, wherein the target points are the grid center points in the grid map; S35. Calculate the distance between each target point and each water quality monitoring station; S36. Using the inverse distance weighted interpolation algorithm, the dissolved oxygen data of each water quality monitoring station is interpolated to each target point according to the distance to form the dissolved oxygen value of each target point. S37. Integrate the dissolved oxygen values ​​of all target points to form the dissolved oxygen dataset of the grid map; wherein, the inverse distance weighted interpolation algorithm uses the following formula to calculate the dissolved oxygen value of the target points: in, This represents the dissolved oxygen level at the i-th target location. This represents the dissolved oxygen value at the j-th water quality monitoring station. These are the weighting coefficients. , Let α be the distance from the i-th target point to the j-th water quality monitoring station, and let α be a power parameter.

6. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 5, characterized in that: S4 includes the following steps: S41. Connect to the meteorological intelligent grid forecasting system and obtain high-resolution forecast products covering the target forecast area; S42. Extract forecast temperature and forecast pressure data within the forecast time range from the high-resolution forecast product, wherein the forecast temperature includes at least the forecast maximum temperature; S43. Match the predicted temperature and predicted air pressure data with the spatial coordinates of the grid map; S44. Using a spatial interpolation method, the forecast temperature and forecast air pressure data are mapped to each grid cell in the grid map to obtain the forecast temperature and forecast air pressure values ​​for each grid cell. S45. Integrate the forecast temperature and forecast air pressure values ​​of all grid cells to form the forecast meteorological dataset of the grid map.

7. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 6, characterized in that: S5 includes the following steps: S51. Integrate the air pressure dataset, dissolved oxygen dataset, and forecast meteorological dataset of each grid unit to form a meteorological-water quality input feature set for each grid unit; S52. Input the meteorological-water quality input feature set into the pre-established dissolved oxygen level early warning model; S53. The dissolved oxygen level early warning model is constructed based on the random forest support vector regression algorithm, using the input features of each grid cell as independent variables and calculating the predicted dissolved oxygen level of each grid cell as dependent variables. S54. Integrate the predicted dissolved oxygen values ​​of all grid cells to form the predicted dissolved oxygen dataset of the grid map.

8. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 7, characterized in that: S6 includes the following steps: S61. Import the predicted dissolved oxygen dataset into a geographic information system tool; S62. Based on the risk assessment standard for dissolved oxygen in aquaculture, the predicted dissolved oxygen value is divided into multiple risk level intervals, and a corresponding visual label is assigned to each risk level interval. S63. Using GIS tools, the forecast dissolved oxygen value of each grid unit is mapped to the corresponding visual identifier according to its risk level range, and a distributed grid map of dissolved oxygen warning for the fishpond area is generated. S64. Output the dissolved oxygen level early warning grid distribution map.

9. The method for early warning of dissolved oxygen levels in fishponds during summer based on intelligent grid forecast data according to claim 8, characterized in that: The input feature optimization of the dissolved oxygen level early warning model pre-established in S5 includes the following steps: S55. Using the current forecast date as the time base, extract historical data within a preset time window as an evaluation sample set. Calculate the contribution of each candidate environmental factor to the change in dissolved oxygen on the evaluation sample set and sort them from largest to smallest contribution. The candidate environmental factors include at least multiple lag time data of air pressure, dissolved oxygen, temperature, and air pressure. S56. Based on the contribution ranking results, select the top N factors in terms of contribution from all candidate environmental factors as valid input features for the current forecast day, where N is a preset positive integer. S57. For each factor in the effective input features, search for the optimal lag period that minimizes the prediction error within its preset lag period search range, and use the historical data corresponding to the optimal lag period as the actual input value of the factor on the current forecast day. S58. Incorporate the latest observation data after the current forecast date into the historical database for updating the evaluation sample set for the next forecast date, thereby achieving daily dynamic adjustment of input features; The contribution is quantitatively evaluated based on the change in node purity in the random forest algorithm. The larger the change in node purity, the greater the impact of the factor on the prediction error and the higher the contribution. The lag time has a search range of one to three days.

Images