Accurate calculation method of total phosphorus flux in lake

Abstract

The application discloses a kind of lake total phosphorus fluxes accurate accounting method, the application relates to water quality monitoring technical field, constructs three-dimensional water power and water quality coupling model and divides effective and invalid flux, combined with dynamic targeting monitoring, two-way LSTM inversion and closed loop check, realize the accurate adaptation accounting of lake total phosphorus fluxes, solve the scene adaptation defect of traditional river method, the advantages of the application are that: by constructing lake three-dimensional water power and water quality coupling numerical model and carrying out total phosphorus whole path tracking, effective flux and invalid flux are accurately divided, dynamic targeting monitoring system is replaced by traditional fixed point arrangement mode simultaneously, effectively solve the core defects that existing river type accounting method cannot adapt to lake complex flow field, it is difficult to distinguish lake internal circulation invalid flux, monitoring data is not representative, avoid flux repeated calculation and mistake problem, improve the matching degree of monitoring data on lake total phosphorus migration and transformation law.

Description

Technical Field

[0001] This invention relates to the field of water quality monitoring technology, specifically a method for accurately calculating the total phosphorus flux in lakes. Background Technology

[0002] Total phosphorus is a core control indicator for characterizing the quality of lake water environment and assessing the degree of lake eutrophication. Its input, migration, transformation and output processes in lake water directly determine the nutrient status of lake water and the health level of aquatic ecosystem.

[0003] The applicant discovered through a search that a Chinese patent, "A Method and Device for Real-Time Measurement of Total Phosphorus Flux in Water," with publication number "CN120369908A," discloses a method and device for real-time measurement of total phosphorus flux in water. This patent primarily involves deploying radar current meters and multi-parameter water quality sensor arrays at river monitoring sections to acquire real-time flow data and conventional water quality index data. It then constructs a multi-parameter coupled Long Short-Term Memory (LSTM) network inversion model to invert total phosphorus concentration. Finally, it combines this with flux calculation formulas to calculate the total phosphorus flux in the river in real time and uploads the data to a cloud platform. However, this approach has certain drawbacks. First, the underlying logic of this solution is entirely built around the scenario of unidirectional steady flow in rivers, making it unsuitable for lakes. The complex flow field coupled with wind-driven flow and throughput flow makes it difficult to distinguish the ineffective fluxes in the lake's internal circulation. The monitoring mode with fixed cross-sections at equal intervals cannot match the spatial heterogeneity of total phosphorus distribution in lakes, which easily leads to problems such as duplicate and incorrect flux calculations. Secondly, the inversion model of this scheme only couples conventional water quality indicators and does not include specific characteristic parameters that affect the changes in total phosphorus in lakes. Its generalization ability in the lake scenario is weak, and it only calculates the transit flux of a single cross-section, resulting in incomplete coverage of the calculation dimensions and a lack of a closed-loop verification link adapted to the full range of lake fluxes. The accuracy and comprehensiveness of the calculation results are difficult to guarantee. Therefore, we propose a precise calculation method for total phosphorus flux in lakes. Summary of the Invention

[0004] The purpose of this invention is to provide a precise method for calculating the total phosphorus flux of lakes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for accurate calculation of total phosphorus flux in lakes, the calculation method comprising the following steps:

[0006] Step 1: Collect basic data of the target lake over the past 5-10 years and conduct on-site surveys. Construct a three-dimensional coupled numerical model of hydrodynamics and water quality. After calibration and verification, simulate the entire trajectory of lake flow field and total phosphorus migration and transformation. Divide the effective flux that directly participates in the total phosphorus mass balance of the entire lake into ineffective fluxes that only circulate locally within the lake. Clarify the accounting boundaries and build the basic accounting framework.

[0007] Step 2: Based on the total phosphorus full-pathway tracking results, construct a dynamic targeted monitoring system. Set up fixed monitoring points at key nodes of total phosphorus input, sensitive areas of migration and transformation, and flux accounting control sections. Set up mobile monitoring points in areas with drastic changes in total phosphorus flux. Perform dynamic monitoring and collect laboratory measured calibration data. After standardizing the multi-source data, form a standardized accounting basic database.

[0008] Step 3: Construct the feature set and supervised learning dataset for the total phosphorus concentration inversion model based on standardized monitoring data, complete the data normalization process, and divide the training set, validation set, and test set in a 7:2:1 ratio;

[0009] Step 4: Construct a bidirectional LSTM inversion model that couples lake hydrodynamics, water quality and endogenous processes. Use a partitioned dataset to complete model training and validation, ensuring that the relative inversion error does not exceed 10%, and complete model finalization and calibration.

[0010] Step 5: Based on the defined accounting boundaries, hydrological data and total phosphorus concentration inversion results, conduct full-caliber total phosphorus flux item accounting, simultaneously complete the automatic elimination of ineffective fluxes, and form an effective total phosphorus flux item accounting list;

[0011] Step 6: Based on the law of conservation of total phosphorus in lakes, construct a flux balance closed-loop verification mechanism, set an allowable relative error threshold of 15%, verify the calculation results, and initiate a deviation correction process for results exceeding the threshold to complete accurate calculation.

[0012] Step 7: Output the final calculation results and complete the data archiving, establish a continuous iterative optimization mechanism for the method, and achieve broad adaptability of the method to different lake scenarios.

[0013] As a further aspect of the present invention: In step one, the basic data includes hydrological sequence data, water quality monitoring data, meteorological data, underwater topographic data, and sediment distribution and physicochemical property data. The hydrological sequence data includes water level, flow rate, and three-dimensional flow velocity data. The water quality monitoring data includes the five conventional parameters, total phosphorus, and chlorophyll. Meteorological data, including wind speed, wind direction, rainfall, and evaporation, are collected. Simultaneously, on-site surveys are conducted to verify basic information such as lake boundaries, inflow and outflow channels, and sediment zoning. Based on the above data, a three-dimensional hydrodynamic and water quality coupled numerical model of the target lake is constructed. On-site measured data are used to calibrate and verify the model parameters to ensure that the model simulation accuracy meets the requirements of the "Technical Specification for Numerical Simulation of Water Environment".

[0014] The calibrated model simulates the three-dimensional flow field distribution, water circulation path, and total phosphorus migration and transformation trajectory of lakes under different hydrological conditions (high water season, normal water season, and low water season) and meteorological conditions. Based on the simulation results, the total phosphorus flux of lakes is divided into effective flux and ineffective flux. Effective flux is the flux that directly participates in the total phosphorus mass balance of lakes and affects the overall water quality of lakes, including external input flux into the lake, internal release flux from sediments, atmospheric deposition flux, outflow flux from the lake, and biological fixation and removal flux. Ineffective flux is the circulating flux that only circulates within a single hydrological zone in the lake, does not cross the boundary of the hydrological zone, and does not participate in the total phosphorus mass balance of the whole lake. The accounting boundaries, spatial range, and time scale of each effective flux are determined, and the basic framework for the accounting of total phosphorus flux in lakes is established.

[0015] As a further aspect of the present invention: In step two, based on the defined total phosphorus flux accounting boundary and path tracing results, a dynamic targeted monitoring system for total phosphorus in the lake is constructed. Fixed monitoring points are set up at total phosphorus input nodes (river mouths), migration and transformation sensitive areas (high-risk areas for endogenous release from sediments, algal aggregation areas, and coastal buffer zones), and flux accounting control sections (lake outlets and lake hydrological zoning interfaces). Simultaneously, based on the flow field simulation results from step one, mobile monitoring points are set up in areas with drastic changes in total phosphorus flux. The monitoring parameters at the fixed monitoring points include water level, three-dimensional flow velocity, water temperature, pH, dissolved oxygen, turbidity, conductivity, and chlorophyll. For total phosphorus, the monitoring parameters at mobile monitoring points remain consistent with those at fixed monitoring points;

[0016] The monitoring frequency is adjusted according to the hydrological period: During the normal water period, fixed points complete 3 samplings per normal water period, and mobile points complete 2 full-area surveys per normal water period; During the wet water period, fixed points complete 4 routine samplings, with 1 additional intensive sampling after heavy rainfall and during algal blooms; During the wet water period, mobile points complete 3 full-area surveys, with 1 additional full-area survey after heavy rainfall and during algal blooms; During the dry water period, fixed points complete 3 samplings, and mobile points complete 1 full-area survey.

[0017] All multi-source data acquired through monitoring are standardized, outliers are removed using the Laida criterion, and missing data are filled in using linear interpolation, forming a standardized and unified basic database for total phosphorus flux accounting. This ensures that the time scale, spatial coordinates, and accuracy of all data remain consistent, resulting in standardized monitoring data.

[0018] As a further aspect of the present invention: In step three, based on the standardized monitoring data obtained in step two, a feature set and a supervised learning dataset for the total phosphorus concentration inversion model adapted to the characteristics of lake water are constructed. First, the model input feature set is determined, divided into two categories: basic features and lake-specific features. The basic features include water level, water temperature, pH, dissolved oxygen, turbidity, conductivity, and three-dimensional flow velocity. The lake-specific features include chlorophyll. Concentration, wind speed, daily variation of water level, and vertical temperature gradient of water body; the model output label is the measured value of total phosphorus concentration detected by the national standard method in the laboratory.

[0019] All input feature data were standardized using Min-Max normalization to eliminate the influence of different units on model accuracy. The standardized time series data were converted into a supervised learning dataset using the sliding window method. The sliding window time step T was set to feature data from the sampling nodes of the past three hydrological periods. The duration of a single time step matched the sampling nodes of each hydrological period. The single time step was 1 month during the normal and dry seasons and 15 days during the wet season. The model output was the measured value of total phosphorus concentration at the current time step. The dataset was randomly divided into training, validation, and test sets in a ratio of 7:2:1. The training set was used for model parameter fitting, the validation set was used for model hyperparameter tuning, and the test set was used for final model accuracy verification.

[0020] As a further aspect of the present invention: In step four, a long short-term memory network inversion model coupling lake hydrodynamics, water quality and endogenous processes is constructed based on a supervised learning dataset. The model consists of an input layer, a bidirectional LSTM layer, an attention mechanism layer, a fully connected layer and an output layer stacked in sequence. The bidirectional LSTM layer is used to simultaneously capture the forward and backward time series dependencies of total phosphorus concentration changes, and the attention mechanism layer is used to automatically assign weights to lake-specific features, strengthen the learning of core features that have a significant impact on total phosphorus concentration changes, and solve the problem of poor generalization ability of traditional LSTM models under complex lake conditions.

[0021] The core calculation process of the model is as follows:

[0022] The input layer feeds the standardized feature data from step three into the bidirectional LSTM layer. The calculation formulas for the forget gate, input gate, cell state update, and output gate of the bidirectional LSTM unit are as follows:

[0023] Forgotten Gate: ;

[0024] Input Gate: ;

[0025] Candidate cell status: ;

[0026] Cell status update: ;

[0027] Output gate: ;

[0028] Hidden state output: ;

[0029] in, , , They are respectively The output values ​​of the forget gate, input gate, and output gate are constantly being forgotten. It is the Sigmoid activation function. , , , These are the weight matrices for the corresponding gating. for The hidden state at all times for Input feature data at time step , , , These are the bias terms for the corresponding gating. for The state of candidate cells at any given time. , They are respectively time, Cellular state at any given moment The hyperbolic tangent activation function is used. This is the element-wise multiplication operator;

[0030] The output data of the bidirectional LSTM layer is fed into the attention mechanism layer, and then processed by the weight allocation formula. Calculate the attention weights for each feature. For the first Attention weights for each feature For the first The importance score of each feature The total number of input features;

[0031] The output data of the attention mechanism layer is fed into the fully connected layer, mapped to the final total phosphorus concentration inversion value, and output by the output layer.

[0032] The model was trained using the training set partitioned in step three. The Adam optimizer was selected, and the learning rate was set to 0.001. Dropout layer and L2 regularization were used to prevent overfitting. The mean squared error (MSE) was used as the model loss function. The validation set loss was monitored, and training was stopped when the validation set loss did not decrease after 15 consecutive training rounds. The accuracy of the trained model was verified using the test set to ensure that the relative error between the model inversion value and the laboratory measured value did not exceed 10%. The model was finalized and calibrated, and high-precision total phosphorus concentration data were obtained.

[0033] As a further aspect of the present invention: In step five, based on the effective flux calculation boundary defined in step one, the standardized monitoring data obtained in step two, and the high-precision total phosphorus concentration data obtained in step four, a sub-item calculation of the total phosphorus flux of the lake is carried out, while the automatic removal of invalid fluxes is completed. First, for each type of effective flux defined in step one, the corresponding specific calculation formula is used for sub-item calculation, including:

[0034] 01. Formula for calculating the external input flux into the lake:

[0035] ;

[0036] In the formula, The total flux of external inputs into the lake during the accounting period. This represents the total number of rivers flowing into the lake. This represents the total number of time steps within the accounting period. for Time of the first Total phosphorus concentration inversion values ​​of each river section flowing into the lake for Time of the first The average flow velocity of each cross-section of the river flowing into the lake for Time of the first The cross-sectional area of ​​each river flowing into the lake. The duration of a single time step;

[0037] 02. Formula for calculating the outflow flux from the lake:

[0038] ;

[0039] In the formula, The total outflow flux from the lake during the accounting period. This represents the total number of control sections for lake outflow. for Time of the first Total phosphorus concentration inversion values ​​of each river section flowing out of the lake for Time of the first The average flow velocity of each cross-section of the river flowing out of the lake for Time of the first The cross-sectional area of ​​the river flowing out of the lake;

[0040] 03. Formula for calculating endogenous release flux from sediments:

[0041] ;

[0042] In the formula, This represents the total endogenous flux released from sediments during the accounting period. The total number of lake sediment zones. for Time of the first Phosphorus release rate per unit area of ​​each sediment zone For the first The area of ​​each sedimentary zone;

[0043] 04. Formula for calculating atmospheric deposition flux:

[0044] ;

[0045] In the formula, This represents the total atmospheric deposition flux during the accounting period. for Total phosphorus concentration in rainfall at any given time for Rainfall at any given moment The surface area of ​​the lake. for Atmospheric dry deposition flux per unit area at any given time;

[0046] 05. Formula for calculating biological fixation and removal flux:

[0047] ;

[0048] In the formula, This represents the total flux of biological fixation and removal within the accounting period. This is the net increase in lake algae during the accounting period. This is the phosphorus content coefficient of algae. To calculate the total mass of aquatic plant harvesting and fish catch during the accounting period, The phosphorus content coefficient for aquatic organisms;

[0049] After the sub-item calculations are completed, based on the lake flow field simulation results from step one, the invalid flux removal module automatically filters flux data that only circulates locally within the lake. The specific rule is as follows: for flux between any two monitoring sections within the lake, if the start and end points of the flux are both located within the same lake hydrological zone defined in step one and do not cross the calculation boundary, it is determined to be an invalid flux and removed from the total calculation results to avoid duplicate calculations of fluxes, thus forming a sub-item calculation list of the total effective phosphorus flux of the lake within the calculation period.

[0050] As a further aspect of the present invention: In step six, based on the law of conservation of total phosphorus mass in lakes, a closed-loop verification mechanism for total phosphorus flux balance is constructed to verify and correct the deviation of the full-caliber flux calculation results obtained in step five. First, the total phosphorus mass balance equation for lakes is established:

[0051] ;

[0052] In the formula, To calculate the change in total phosphorus storage in the lake during the accounting period, the change in total phosphorus storage in the lake is calculated using measured values. The calculation formula is as follows:

[0053] ;

[0054] In the formula, , These are the measured average total phosphorus concentrations of the lake water at the end and beginning of the accounting period, respectively. , These represent the lake's water volume at the end and beginning of the accounting period, respectively.

[0055] The allowable relative error threshold for flux balance verification is set at 15%, that is, when When the error is ≤15%, the calculation result is deemed to have passed the verification. When the error exceeds the allowable threshold, the deviation correction process is automatically initiated. First, the entire calculation process is traced back to check for outliers, model inversion deviations, and incorrect values ​​of calculation formula parameters. For model inversion deviations, the inversion model in step four is incrementally trained using newly added laboratory measured data to optimize model parameters and improve inversion accuracy. For formula parameter value issues, the calculation parameters are recalibrated based on measured data to correct the sub-item calculation results. The verification process is repeated until the calculation result meets the flux balance error threshold requirement, thus completing the accurate calculation of total phosphorus flux in the lake.

[0056] As a further aspect of the present invention: In step seven, after verification and correction are completed, the final accurate calculation result of total phosphorus flux in lakes is output, including a list of effective total phosphorus flux items for the entire lake within the calculation period, the contribution ratio of total phosphorus input, the spatiotemporal variation characteristics analysis of total phosphorus flux, and a report on the identification of key control nodes and priority control objects for total phosphorus input in lakes. At the same time, all calculation data, model parameters, and monitoring data are uniformly archived and stored in the total phosphorus flux database of lakes.

[0057] A continuous iterative optimization mechanism for the method is established. New monitoring data and laboratory measured data are collected quarterly to incrementally train and optimize the total phosphorus concentration inversion model in step four. Each year, based on changes in lake hydrological conditions and topography, the lake hydrodynamic-water quality coupling model and total phosphorus full-path tracking results from step one are updated. The monitoring point layout scheme and calculation parameters are optimized simultaneously to ensure that the calculation accuracy of the method always meets the requirements. At the same time, the core parameters of the model input feature set, the layout density and frequency of monitoring points, and the dimensions and formula parameters of flux calculation can be adjusted according to the type of target lake (shallow lake / deep lake / urban landscape lake / plateau lake), hydrological situation, water quality characteristics, and water morphology to achieve broad adaptability of the method to different lake scenarios.

[0058] Meanwhile, based on the flow field simulation results of step one, mobile monitoring points were deployed in areas with drastic changes in total phosphorus flux where the spatial gradient of total phosphorus concentration was greater than 0.05 mg / L / km and the daily variation frequency of flow velocity direction was greater than 3 times.

[0059] Monitoring frequency is adjusted according to the hydrological period phases: during the normal water period, fixed monitoring points complete 3 samplings per normal water period, and mobile monitoring points complete 2 full-area surveys per normal water period; during the high water period, fixed monitoring points complete 4 routine samplings, and during heavy rainfall with a daily rainfall exceeding 50mm, chlorophyll... After an algal bloom with a concentration greater than 20 μg / L, one additional intensive sampling is conducted; during the high-water season, mobile sampling points complete three full-area surveys, and after heavy rainfall or algal bloom events, one additional full-area survey is conducted; during the low-water season, fixed sampling points complete three samplings, and mobile sampling points complete one full-area survey.

[0060] Parallel water samples were collected and sent to the laboratory during each monitoring cycle. The total phosphorus concentration was detected using GB11893-89 "Determination of Total Phosphorus in Water - Ammonium Molybdate Spectrophotometric Method" to obtain the measured calibration dataset.

[0061] As a further aspect of this invention: hydrological zoning refers to lake sub-regions with relatively independent hydrodynamic circulation characteristics, divided based on lake flow field, hydrodynamic characteristics, and water quality differences; migration and transformation sensitive areas refer to water areas where the total phosphorus concentration and phosphorus release / conversion rate are significantly higher than the average level of the entire lake, significantly affecting changes in total phosphorus flux; flux accounting control sections refer to cross-sections with stable hydrological characteristics used to define the total phosphorus input, output, and regional flux transfer of the lake; total phosphorus flux refers to the set of total phosphorus fluxes covering all dimensions of lake external input, internal release, atmospheric deposition, outflow, and biological fixation and removal; incremental training refers to small-batch parameter optimization training based on newly added measured data on the basis of the established model, without changing the main structure of the model.

[0062] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0063] 1. This invention constructs a three-dimensional hydrodynamic and water quality coupled numerical model of lakes and conducts full-path tracking of total phosphorus, accurately distinguishing between effective and ineffective fluxes. Simultaneously, it employs a dynamic targeted monitoring system to replace the traditional fixed-point deployment method, effectively solving the core defects of existing river-type accounting methods that cannot adapt to the complex flow field of lakes, are difficult to distinguish ineffective fluxes in the lake's internal circulation, and suffer from insufficient representativeness of monitoring data. It avoids problems of duplicate and incorrect flux calculations, improves the matching degree of monitoring data with the migration and transformation patterns of total phosphorus in lakes, and can more comprehensively capture the spatiotemporal variation characteristics of total phosphorus under different hydrological and ecological conditions, providing a reliable data foundation for subsequent accounting. This significantly improves the rationality and data validity of total phosphorus flux accounting in lakes, making the accounting results closer to the actual phosphorus migration and transformation situation in lakes, and providing more realistic data support for lake pollution control.

[0064] 2. This invention constructs a bidirectional LSTM inversion model that couples lake-specific feature parameters with an attention mechanism, and establishes a closed-loop verification mechanism for full-caliber flux component calculation and total phosphorus mass conservation. This effectively solves the shortcomings of traditional inversion models in lake scenarios, such as weak generalization ability, incomplete flux calculation dimensions, and lack of error calibration. It improves the stability and accuracy of total phosphorus concentration inversion, and realizes complete calculation of fluxes from multiple dimensions, including external input, internal release, and atmospheric deposition. Through closed-loop verification, it can automatically identify and correct calculation deviations, reduce overall calculation errors, and make the total phosphorus flux calculation results of lakes more reliable and comprehensive. It can clearly reflect the source and destination of total phosphorus, and provide a more accurate and reliable technical basis for lake eutrophication prevention and control, pollution source tracing, and the formulation of treatment plans. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the method steps in an embodiment of the present invention;

[0066] Figure 2This is a schematic diagram of the overall process for accurately calculating the total phosphorus flux in lakes in an embodiment of the present invention;

[0067] Figure 3 This is a schematic diagram of the bidirectional LSTM inversion model in an embodiment of the present invention. Detailed Implementation

[0068] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0069] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0070] Please see the appendix Figure 1 -Appendix Figure 3 This invention provides an accurate method for calculating the total phosphorus flux in lakes. The method includes the following steps:

[0071] Step 1: Collect basic data of the target lake over the past 5-10 years and conduct on-site surveys. Construct a three-dimensional coupled numerical model of hydrodynamics and water quality. After calibration and verification, simulate the entire trajectory of lake flow field and total phosphorus migration and transformation. Divide the effective flux that directly participates in the total phosphorus mass balance of the entire lake into ineffective fluxes that only circulate locally within the lake. Clarify the accounting boundaries and build the basic accounting framework.

[0072] Step 2: Based on the total phosphorus full-pathway tracking results, construct a dynamic targeted monitoring system. Set up fixed monitoring points at key nodes of total phosphorus input, sensitive areas of migration and transformation, and flux accounting control sections. Set up mobile monitoring points in areas with drastic changes in total phosphorus flux. Implement phased dynamic monitoring frequencies covering the entire hydrological cycle of normal water period, high water period, and low water period. Collect laboratory measured calibration data and standardize the multi-source data to form a standardized accounting basic database.

[0073] Step 3: Construct the feature set and supervised learning dataset for the total phosphorus concentration inversion model based on standardized monitoring data, complete the data normalization process, and divide the training set, validation set, and test set in a 7:2:1 ratio;

[0074] Step 4: Construct a bidirectional LSTM inversion model that couples lake hydrodynamics, water quality and endogenous processes. Use a partitioned dataset to complete model training and validation, ensuring that the relative inversion error does not exceed 10%, and complete model finalization and calibration.

[0075] Step 5: Based on the defined accounting boundaries, hydrological data and total phosphorus concentration inversion results, conduct full-caliber total phosphorus flux item accounting, simultaneously complete the automatic elimination of ineffective fluxes, and form an effective total phosphorus flux item accounting list;

[0076] Step 6: Based on the law of conservation of total phosphorus in lakes, construct a flux balance closed-loop verification mechanism, set an allowable relative error threshold of 15%, verify the calculation results, and initiate a deviation correction process for results exceeding the threshold to complete accurate calculation.

[0077] Step 7: Output the final accurate calculation results and complete data archiving, establish a continuous iterative optimization mechanism for the method, and achieve broad adaptability of the method to different lake scenarios.

[0078] Example

[0079] This embodiment uses a shallow, eutrophic lake in Anhui Province as the target lake. This lake is a typical shallow lake in the Yangtze River basin, with a water area of ​​approximately 780 km². 2 With an average water depth of 2.9m, there are 7 main rivers flowing into the lake and 1 river flowing out of the lake. Eutrophication and cyanobacterial blooms are prominent problems. Cyanobacterial blooms mostly occur during the high water season from June to August, making the engineering need for accurate calculation of total phosphorus flux urgent.

[0080] The calculation period for this embodiment is from January 1, 2024 to December 31, 2024 (a complete hydrological year). The specific implementation process is as follows:

[0081] Step 1: Basic Data Collection and Model Building, Flux Partitioning

[0082] Basic data on the lake from 2014 to 2023 were collected for nearly 10 years, including:

[0083] Hydrological data series: daily water level, flow rate, and three-dimensional flow velocity data, sourced from publicly available monitoring data from the local hydrological bureau;

[0084] Water quality monitoring data: Monthly routine five parameters, total phosphorus, and chlorophyll from 8 national monitoring stations The data comes from publicly available data from the local Department of Ecology and Environment.

[0085] Meteorological data: Daily wind speed, wind direction, rainfall, and evaporation data for the lake basin are sourced from the National Meteorological Science Data Center;

[0086] Underwater topographic data: The data used was 1:10000 underwater topographic data measured by the local authority in 2023.

[0087] Sediment distribution and physicochemical properties data: The data were obtained from the 2023 lake sediment survey conducted by the local provincial environmental science research institute.

[0088] Simultaneously, on-site surveys of the lake were conducted to verify basic information on the lake boundaries, seven inflow channels, one outflow channel, and three sedimentary zones (estuary sedimentary zone, central lake sedimentary zone, and coastal sedimentary zone).

[0089] Based on the above data, a three-dimensional coupled hydrodynamic and water quality numerical model of the lake was constructed using MIKE3FM. The horizontal grid size of the model was set to 50m×50m, the vertical grid was divided into 3 layers, and the time step was set to 300s.

[0090] The model parameters were calibrated using field-measured water level, flow velocity, and total phosphorus concentration data from January to June 2024, and validated using measured data from July to December 2024.

[0091] The coefficient of determination R between the simulated and measured water level values ​​after calibration 2 The relative error between the simulated and measured values ​​of flow velocity is ≤10%, and the relative error between the simulated and measured values ​​of total phosphorus concentration is ≤15%, which meets the requirements of SL / T167-2021 "Technical Specification for Numerical Simulation of Water Environment".

[0092] The calibrated model simulated different hydrological conditions during the high-water season (June-August), normal-water season (March-May, September-November), and low-water season (December-February of the following year), as well as the three-dimensional flow field distribution, water circulation path, and total phosphorus migration and transformation trajectory of the lake under different wind speeds and directions. Based on the simulation results, the lake was divided into six hydrological zones (Northwest Estuary Zone, Southwest Estuary Zone, Lake Center Zone, Coastal Zone, North Lake Zone, and Outflow Control Zone).

[0093] Total phosphorus flux is divided into effective flux and ineffective flux. Effective flux includes external input flux into the lake, internal release flux from sediments, atmospheric deposition flux, outflow flux from the lake, and biological fixation and removal flux. Ineffective flux is defined as circulating flux that only circulates within a single hydrological zone, does not cross the boundary of the hydrological zone, and does not participate in the total phosphorus mass balance of the entire lake.

[0094] Define the accounting boundaries, spatial scope, and time scale of each effective flux, and complete the basic framework for calculating total phosphorus flux in lakes.

[0095] Step 2: Construction of Dynamic Targeted Monitoring System and Data Acquisition

[0096] Based on the total phosphorus full-pathway tracking results from step one, a dynamic targeted monitoring system for total phosphorus in this lake was constructed:

[0097] Fixed monitoring point layout: A total of 22 fixed monitoring points were set up at the total phosphorus input nodes such as the inflow points of the 7 rivers flowing into the lake, 7 high-risk areas for sediment endogenous release, 2 cyanobacteria accumulation areas, 2 coastal buffer zones, 1 lake outlet, and 6 hydrological zoning interfaces (there are overlapping points in different locations, such as the same points at the hydrological zoning interfaces and the high-risk areas for sediment endogenous release).

[0098] Mobile monitoring point deployment: Based on the flow field simulation results in step one, six mobile monitoring points were deployed in areas with drastic changes in total phosphorus flux, such as the estuary confluence area and the cyanobacteria accumulation area, where the spatial gradient of total phosphorus concentration is greater than 0.05 mg / L / km and the daily variation frequency of flow velocity direction is greater than 3 times.

[0099] Monitoring parameters: For both fixed and mobile monitoring points, the monitoring parameters include water level, three-dimensional flow velocity, water temperature, pH, dissolved oxygen, turbidity, conductivity, and chlorophyll. Total phosphorus;

[0100] The endogenous source monitoring area corresponds to fixed points;

[0101] Monitoring frequency:

[0102] During the normal water season (March-May and September-November): fixed locations will complete 3 samplings per normal water season period, and mobile locations will complete 2 full-area surveys per normal water season period;

[0103] High water season (June-August): Four routine samplings were completed at fixed locations, including heavy rainfall exceeding 50mm on the same day, and chlorophyll content was measured. After an algal bloom with a concentration greater than 20 μg / L, an additional high-density sampling should be conducted.

[0104] The mobile monitoring stations completed three full-area surveys during the high-water season, and one additional full-area survey was conducted after heavy rainfall and algal bloom events.

[0105] During the dry season (December to February): 3 samplings were completed at fixed locations, and 1 full-area survey was completed at mobile locations;

[0106] Endogenous monitoring area: Surface water bottom samples are collected once during each hydrological period, and three parallel samples are collected at each location.

[0107] Sample testing and data standardization: Three parallel water samples were collected simultaneously in each monitoring cycle and sent to the laboratory. The total phosphorus concentration was detected using GB11893-89 "Determination of Total Phosphorus in Water - Ammonium Molybdate Spectrophotometric Method" to obtain the measured calibration dataset.

[0108] All multi-source data acquired through monitoring were standardized, outliers were removed using the Laida criterion, and missing data were filled in using linear interpolation to form a standardized and unified basic database for calculating total phosphorus flux in lakes. This ensured that the time scale, spatial coordinates, and accuracy of all data remained consistent, resulting in standardized monitoring data.

[0109] Step 3: Construction of the Inversion Model Feature Set and Dataset

[0110] Based on the standardized monitoring data obtained in step two, a feature set and a supervised learning dataset for the total phosphorus concentration inversion model adapted to the water characteristics of the lake are constructed:

[0111] Feature set determination: Basic features include water level, water temperature, pH, dissolved oxygen, turbidity, conductivity, and three-dimensional flow velocity;

[0112] Lake-specific characteristics include chlorophyll The model outputs 11 input features, including concentration, wind speed, daily water level variation, and vertical temperature gradient of the water body. The model output label is the measured value of total phosphorus concentration obtained by laboratory testing using the national standard method.

[0113] Data preprocessing: Min-Max normalization is performed on all input feature data to eliminate the influence of different units on model accuracy;

[0114] Construction of supervised learning dataset: The standardized time series data is converted into a supervised learning dataset using the sliding window method. The sliding window time step T is set to the feature data of the sampling nodes of the past three hydrological periods. The duration of a single time step is matched with the sampling nodes of each hydrological period. The single time step is 1 month during the normal water period and the dry water period, and 15 days during the wet water period. The model output is the total phosphorus concentration inversion value of the current sampling node.

[0115] Dataset partitioning: The dataset is randomly divided into a training set (for model parameter fitting), a validation set (for model hyperparameter tuning), and a test set (for final model accuracy verification) in a ratio of 7:2:1.

[0116] Step 4: Construction and Training of the Bidirectional LSTM Inversion Model

[0117] Based on the supervised learning dataset constructed in step three, a bidirectional LSTM inversion model coupling lake hydrodynamics, water quality, and endogenous processes is built. The model consists of an input layer, a bidirectional LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer stacked sequentially.

[0118] Input layer: The number of neurons is consistent with the dimension of the feature set, set to 11;

[0119] Bidirectional LSTM layer: Two layers are set, with 64 neurons in each layer, to simultaneously capture the forward and backward time-series dependencies of total phosphorus concentration changes. The calculation formulas for the forget gate, input gate, cell state update, and output gate of the bidirectional LSTM unit are as follows:

[0120] Forgotten Gate: ;

[0121] Input Gate: ;

[0122] Candidate cell status: ;

[0123] Cell status update: ;

[0124] Output gate: ;

[0125] Hidden state output: ;

[0126] in, , , They are respectively The output values ​​of the forget gate, input gate, and output gate are constantly being forgotten. It is the Sigmoid activation function. , , , These are the weight matrices for the corresponding gating. for The hidden state at all times for Input feature data at time step , , , These are the bias terms for the corresponding gating. for The state of candidate cells at any given time. , They are respectively time, Cellular state at any given moment The hyperbolic tangent activation function is used. This is an element-wise multiplication operation;

[0127] Attention Mechanism Layer: Used to automatically assign weights to lake-specific features, enhancing core feature learning. The output data of the bidirectional LSTM layer is fed into the attention mechanism layer, and weights are assigned using the formula... Calculate the attention weights for each feature. For the first Attention weights for each feature For the first The importance score of each feature In this embodiment, the total number of input features is... =11;

[0128] Fully connected layers: Two layers are set, with 32 and 16 neurons respectively;

[0129] Output layer: Set up 1 neuron to output the total phosphorus concentration inversion value.

[0130] The model was trained using the training set partitioned in step three. The Adam optimizer was selected, with a learning rate of 0.001, a Dropout layer inactivation rate of 0.2, and an L2 regularization coefficient of 0.0001. The mean squared error (MSE) was used as the model loss function. The validation set loss was monitored, and training was stopped when the validation set loss did not decrease for 15 consecutive training rounds. The final training rounds were 132.

[0131] The accuracy of the trained model was verified using a test set. The average relative error between the model inversion value and the laboratory measured value was 6.8%, and the maximum relative error was 9.4%, which met the requirement that the relative error should not exceed 10%. The model was finalized and calibrated, and high-precision total phosphorus concentration data of the lake were obtained.

[0132] Step 5: Calculation of total phosphorus flux and removal of ineffective flux

[0133] Based on the effective flux accounting boundary defined in step one, the standardized monitoring data obtained in step two, and the high-precision total phosphorus concentration data obtained in step four, the total phosphorus flux of the lake in 2024 was calculated separately, and the invalid flux was automatically removed.

[0134] The five types of effective flux were calculated separately, and the specific calculation results are as follows:

[0135] Total external flux into the lake, total number of rivers flowing into the lake =7, total sampling time steps during the hydrological period of the accounting cycle =14, the duration of a single time step is matched with the sampling nodes of each hydrological period. After verification, the total external flux into the lake in 2024 was... 590.60t;

[0136] Total endogenous flux released from sediments; total number of sedimentary zones in a lake =3, according to calculations, the total endogenous release flux of sediments in the lake in 2024 was 3. 188.50t;

[0137] Total atmospheric deposition flux and lake surface area =780km 2 According to calculations, the total atmospheric deposition flux of the lake in 2024 was... 109.3t;

[0138] Total outflow flux from the lake; Total number of outflow control sections =1, according to calculations, the total outflow from the lake in 2024 was 1. 314.67t;

[0139] The total flux of biological fixation and removal in the lake in 2024 was calculated. : 555.85t.

[0140] After the sub-item calculations are completed, based on the lake flow field simulation results from step one, the invalid flux removal module automatically filters flux data that only circulates locally within the lake. For fluxes between any two monitoring sections within the lake, if the starting point and ending point of the flux are both located within the same hydrological zone defined in step one and do not cross the calculation boundary, they are determined to be invalid fluxes and removed from the overall calculation results.

[0141] In this embodiment, a total of 41.6t of invalid flux was removed, avoiding duplicate calculation of flux, and finally forming a breakdown of the total effective phosphorus flux of the lake in 2024.

[0142] Step Six: Flux Balance Closed-Loop Verification and Deviation Correction

[0143] Based on the law of conservation of total phosphorus mass in lakes, a closed-loop verification mechanism for total phosphorus flux balance is constructed to verify and correct the deviation of the full-caliber flux calculation results obtained in step five:

[0144] Establish the mass balance equation for total phosphorus in lakes: The calculated net flux (net flux) on the right side of the equation is: 590.60 + 188.50 + 109.3 - 314.67 - 555.85 = 17.88t.

[0145] Calculation of measured changes in total phosphorus storage in lake water: Calculation formula: In the formula, the measured average total phosphorus concentration of the lake water at the start of the accounting period (January 1, 2024) is... =0.232mg / L, water volume =9.28×10 8 m 3 The measured average total phosphorus concentration of the lake water at the end of the accounting period (December 31, 2024). =0.249mg / L, water volume =9.45×10 8 m 3 .

[0146] Calculations show that the change in measured total phosphorus reserves... =0.249×9.45×10 8 ×10 -6 -0.232×9.28×10 8 ×10 -6 =19.25t.

[0147] Error verification: The allowable relative error threshold for flux balance verification is set to 15%. The calculated relative error is... ×100%≈7.12%, which meets the error threshold requirement of ≤15%, so the calculation result is deemed to have passed the verification, and the accurate calculation of the total phosphorus flux of the lake in 2024 is completed.

[0148] Step 7: Result Output and Iteration Mechanism Establishment

[0149] After verification, the final accurate calculation results of the total phosphorus flux of the lake in 2024 will be output, including a list of the total effective total phosphorus flux, the contribution ratio of total phosphorus input from each source, the spatiotemporal variation characteristics analysis of total phosphorus flux, and a report on the identification of key control nodes and priority control objects of total phosphorus sources.

[0150] According to calculations, the contribution of external sources of phosphorus input to the lake from rivers flowing into the lake accounts for 66.50%, the contribution of internal sources of phosphorus input from sediments accounts for 21.21%, and the contribution of atmospheric deposition accounts for 12.3%. Among them, the seven main rivers flowing into the lake are key channels for priority control of total phosphorus pollution.

[0151] Simultaneously, all accounting data, model parameters, and monitoring data are uniformly archived and stored in the total phosphorus flux database of the lake, establishing a continuous iterative optimization mechanism for the method to complete this accounting work.

[0152] Comparative Example

[0153] This comparative example uses the patent application CN120369908A, "A method and device for real-time measurement of total phosphorus flux in water," to calculate the total phosphorus flux for the same target lake and the same accounting period (January 1, 2024 to December 31, 2024) as the example. The specific steps are as follows:

[0154] S1. Analyze the water quality data of the lake in previous years and determine that the conventional water quality indicators with the strongest correlation with total phosphorus concentration are water temperature, pH, dissolved oxygen, turbidity and conductivity. Set up a monitoring section at the lake outlet and deploy the real-time measurement equipment for total phosphorus flux in water as described in this patent.

[0155] S2. The radar current meter is set up at equal intervals on the vertical line from the water surface to the riverbed of the monitoring section. The multi-parameter water quality sensor array is installed with the water flow direction aligned. The radar current meter is used to obtain real-time flow data of the section, and the multi-parameter water quality sensor array is used to obtain conventional water quality index data. The monitoring frequency is once a day.

[0156] S3. Construct a multi-parameter coupled Long Short-Term Memory (LSTM) inversion model, employ the moving pane algorithm to eliminate sensor drift, and invert the total phosphorus concentration using the inversion model. The model input features are conventional water quality indicators and flow velocity data. The same laboratory measured data as in the example were used for model training and validation. The average relative error between the model inversion value and the measured value was 13.8%.

[0157] S4. Based on the monitoring data, the total phosphorus flux is calculated in real time according to the flux calculation. The total phosphorus flux of the lake in 2024 is calculated to be 347.5t. The data is then uploaded to the cloud platform.

[0158] Comparison of core parameters between the examples and the comparative examples

[0159] Table 1. Comparison of core parameters between embodiments of the present invention and comparative examples:

[0160]

[0161] Analysis of Results from Examples and Comparative Examples

[0162] Comparison of Total Phosphorus Concentration Inversion Accuracy

[0163] The average relative error of total phosphorus concentration retrieval in the embodiment was 6.8%, far lower than the 13.8% of the comparative example. The method of this invention significantly improves the accuracy of total phosphorus concentration retrieval by incorporating lake-specific feature parameters, introducing a bidirectional LSTM network and attention mechanism, and matching dynamic targeted monitoring data during hydrological periods. It solves the core defects of traditional single-section LSTM models in complex flow fields and spatially heterogeneous scenarios in lakes, resulting in weak generalization ability and low accuracy. The retrieval accuracy meets the requirements of engineering applications.

[0164] Comparison of reliability of flux accounting results

[0165] The example calculation showed that the total phosphorus input flux of the lake in 2024 was 888.4t. After mass balance verification, the relative error was only 7.12%, which is far below the allowable threshold of 15%. The calculation result is consistent with the actual total phosphorus income and expenditure of the lake.

[0166] The comparative calculation yielded a total phosphorus flux of 347.5 t, with a relative error of approximately 19.4% compared to the measured change in total phosphorus storage in the lake. This calculation result significantly deviates from the actual total phosphorus cycle process in the lake and cannot provide reliable data support for lake management. Meanwhile, this invention avoids the repeated calculation of intra-lake circulation fluxes by eliminating invalid fluxes, thus solving the core adaptation deficiency of traditional river-type methods in lake scenarios.

[0167] Comparison of comprehensiveness and practicality of accounting

[0168] The example demonstrates comprehensive flux accounting for all sources of total phosphorus entering the lake, including external input, internal sediment release, atmospheric deposition, outflow, and biological fixation and removal. It fully covers the entire chain of total phosphorus input, transformation, and output in lakes, accurately identifies key control points and the contribution of each source, and can directly provide comprehensive data support for total phosphorus control and eutrophication treatment plan formulation. In contrast, the comparative example only calculates the transit flux at the outflow section, does not cover the core sources and internal transformation processes of total phosphorus in lakes, cannot reflect the true balance of total phosphorus in lakes, and can only achieve flux monitoring at a single section, which cannot meet the actual engineering needs of lake pollution prevention and control.

[0169] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for accurately calculating the total phosphorus flux of a lake, characterized in that, The accounting method includes the following steps: Step 1: Collect basic data of the target lake over the past 5-10 years and conduct on-site surveys. Construct a three-dimensional coupled numerical model of hydrodynamics and water quality. After calibration and verification, simulate the entire trajectory of lake flow field and total phosphorus migration and transformation. Divide the effective flux that directly participates in the total phosphorus mass balance of the entire lake into ineffective fluxes that only circulate locally within the lake. Clarify the accounting boundaries and build the basic accounting framework. Step 2: Based on the total phosphorus full-pathway tracking results, construct a dynamic targeted monitoring system. Set up fixed monitoring points at key nodes of total phosphorus input, sensitive areas of migration and transformation, and flux accounting control sections. Set up mobile monitoring points in areas with drastic changes in total phosphorus flux. Perform dynamic monitoring and collect laboratory measured calibration data. After standardizing the multi-source data, form a standardized accounting basic database. Step 3: Construct the feature set and supervised learning dataset for the total phosphorus concentration inversion model based on standardized monitoring data, complete the data normalization process, and divide the training set, validation set, and test set in a 7:2:1 ratio; Step 4: Construct a bidirectional LSTM inversion model that couples lake hydrodynamics, water quality and endogenous processes. Use a partitioned dataset to complete model training and validation, ensuring that the relative inversion error does not exceed 10%, and complete model finalization and calibration. Step 5: Based on the defined accounting boundaries, hydrological data and total phosphorus concentration inversion results, conduct full-caliber total phosphorus flux item accounting, simultaneously complete the automatic elimination of ineffective fluxes, and form an effective total phosphorus flux item accounting list; Step 6: Based on the law of conservation of total phosphorus in lakes, construct a flux balance closed-loop verification mechanism, set an allowable relative error threshold of 15%, verify the calculation results, and initiate a deviation correction process for results exceeding the threshold to complete accurate calculation. Step 7: Output the final calculation results and complete data archiving, and establish a continuous iterative optimization mechanism for the method.

2. The method for accurately calculating the total phosphorus flux of a lake according to claim 1, characterized in that: In step one, the basic data includes hydrological sequence data, water quality monitoring data, meteorological data, underwater topographic data, and sediment distribution and physicochemical property data. The hydrological sequence data includes water level, flow rate, and three-dimensional flow velocity data. The water quality monitoring data includes the five conventional parameters, total phosphorus, and chlorophyll. Meteorological data, including wind speed, wind direction, rainfall, and evaporation, were collected. Simultaneously, on-site surveys were conducted to verify the lake boundaries, inflow and outflow channels, and sediment zoning information. Based on the above data, a three-dimensional hydrodynamic and water quality coupled numerical model of the target lake was constructed. The model parameters were calibrated and verified using field measured data. The calibrated model simulated the three-dimensional flow field distribution, water circulation path, and total phosphorus migration and transformation trajectory of the lake under different hydrological and meteorological conditions. Based on the simulation results, the total phosphorus flux of the lake was divided into effective flux and ineffective flux. Effective flux is the flux that directly participates in the total phosphorus mass balance of the lake and affects the overall water quality of the lake. Ineffective flux is the circulating flux that only circulates within a single hydrological zone within the lake, does not cross the boundary of the hydrological zone, and does not participate in the total phosphorus mass balance of the entire lake.

3. The method for accurately calculating the total phosphorus flux of a lake according to claim 2, characterized in that: In step two, based on the defined total phosphorus flux accounting boundary and path tracing results, a dynamic targeted monitoring system for total phosphorus in the lake is constructed. Fixed monitoring points are set up at total phosphorus input nodes, migration and transformation sensitive areas, and flux accounting control sections. Simultaneously, based on the flow field simulation results from step one, mobile monitoring points are set up in areas with drastic changes in total phosphorus flux. The monitoring parameters at the fixed monitoring points include water level, three-dimensional flow velocity, water temperature, pH, dissolved oxygen, turbidity, conductivity, and chlorophyll. For total phosphorus, the monitoring parameters at mobile monitoring points remain consistent with those at fixed monitoring points; The monitoring frequency follows the hydrological period phased adjustment rule: during the normal water period, fixed points will complete 3 samplings in each time period, and mobile points will complete 2 full-area surveys in each time period of the normal water period. During the high-water season, fixed sampling points completed four routine samplings, with one additional intensive sampling after heavy rainfall and during algal blooms. Mobile sampling points completed three full-area surveys during the high-water season, with one additional full-area survey after heavy rainfall and during algal blooms. During the low-water season, fixed sampling points completed three samplings, and mobile sampling points completed one full-area survey. Simultaneously, parallel water samples were collected at each sampling point and sent to the laboratory for total phosphorus concentration detection using the ammonium molybdate spectrophotometric method according to national standards, to obtain a measured calibration dataset. All multi-source data acquired through monitoring were standardized, outliers were removed using the Laida criterion, and missing data were filled in using linear interpolation to form a basic database for total phosphorus flux accounting, resulting in standardized monitoring data.

4. The method for accurately calculating the total phosphorus flux of a lake according to claim 3, characterized in that: In step three, based on the standardized monitoring data obtained in step two, a feature set and a supervised learning dataset for the total phosphorus concentration inversion model adapted to the characteristics of the lake water are constructed. First, the model input feature set is determined, divided into two categories: basic features and lake-specific features. Basic features include water level, water temperature, pH, dissolved oxygen, turbidity, conductivity, and three-dimensional flow velocity. Lake-specific features include chlorophyll. The model output labels are the total phosphorus concentration measured by the national standard method in the laboratory, including concentration, wind speed, daily water level variation, and vertical temperature gradient of the water body. All input feature data are normalized by Min-Max normalization. The normalized time series data are converted into a supervised learning dataset using the sliding window method. The sliding window time step T is set to the feature data of the sampling nodes of the past three hydrological periods. The duration of a single time step is matched with the sampling node of each hydrological period. The single time step is 1 month during the normal water period and the dry water period, and 15 days during the wet water period. The model output is the measured total phosphorus concentration of the current sampling node.

5. The method for accurately calculating the total phosphorus flux of a lake according to claim 4, characterized in that: In step four, a long short-term memory network inversion model coupling lake hydrodynamics, water quality and endogenous processes is constructed based on the supervised learning dataset. The model consists of an input layer, a bidirectional LSTM layer, an attention mechanism layer, a fully connected layer and an output layer stacked in sequence. The calculation process of the model is as follows: The input layer feeds the standardized feature data from step three into the bidirectional LSTM layer. The calculation formulas for the forget gate, input gate, cell state update, and output gate of the bidirectional LSTM unit are as follows: Forgotten Gate: ; Input Gate: ; Candidate cell status: ; Cell status update: ; Output gate: ; Hidden state output: ; in, , , They are respectively The output values ​​of the forget gate, input gate, and output gate are constantly being forgotten. It is the Sigmoid activation function. , , , These are the weight matrices for the corresponding gating. for The hidden state at all times for Input feature data at time step , , , These are the bias terms for the corresponding gating. for The state of candidate cells at any given time. , They are respectively time, Cellular state at any given moment The hyperbolic tangent activation function is used. This is the element-wise multiplication operator; The output data of the bidirectional LSTM layer is fed into the attention mechanism layer, and then processed by the weight allocation formula. Calculate the attention weights for each feature. For the first Attention weights for each feature For the first The importance score of each feature The total number of input features; The output data of the attention mechanism layer is fed into the fully connected layer, mapped to the final total phosphorus concentration inversion value, and output by the output layer. The model was trained using the training set partitioned in step three. The Adam optimizer was selected, and the learning rate was set to 0.

001. Training was stopped when the loss on the validation set did not decrease after 15 consecutive training rounds, thus obtaining high-precision total phosphorus concentration data.

6. The method for accurately calculating the total phosphorus flux of a lake according to claim 5, characterized in that: In step five, based on the effective flux calculation boundaries defined in step one, the standardized monitoring data obtained in step two, and the high-precision total phosphorus concentration data obtained in step four, a sub-item calculation of the total phosphorus flux across the entire lake is carried out. Simultaneously, the automatic removal of invalid fluxes is completed. Firstly, for each type of effective flux defined in step one, the corresponding specific calculation formulas are used for sub-item calculation, including:

01. Formula for calculating the external input flux into the lake: ； In the formula, The total flux of external inputs into the lake during the accounting period. This represents the total number of rivers flowing into the lake. This represents the total number of time steps within the accounting period. for Time of the first Total phosphorus concentration inversion values ​​of each river section flowing into the lake for Time of the first The average flow velocity at each cross-section of the river flowing into the lake for Time of the first The cross-sectional area of ​​the rivers flowing into the lake. The duration of a single time step; 02. Formula for calculating the outflow flux from the lake: ； In the formula, The total outflow flux from the lake during the accounting period. This represents the total number of control sections for lake outflow. for Time of the first Total phosphorus concentration inversion values ​​of each river section flowing out of the lake for Time of the first The average flow velocity of each cross-section of the river flowing out of the lake for Time of the first The cross-sectional area of ​​the river flowing out of the lake; 03. Formula for calculating endogenous release flux from sediments: ； In the formula, This represents the total endogenous flux released from sediments during the accounting period. The total number of lake sediment zones. for Time of the first Phosphorus release rate per unit area of ​​each sediment zone For the first The area of ​​each sedimentary zone; 04. Formula for calculating atmospheric deposition flux: ； In the formula, This represents the total atmospheric deposition flux during the accounting period. for Total phosphorus concentration in rainfall at any time for Rainfall at any given moment The surface area of ​​the lake. for Atmospheric dry deposition flux per unit area at any given time; 05. Formula for calculating biological fixation and removal flux: ； In the formula, This represents the total flux of biological fixation and removal within the accounting period. This is the net increase in lake algae during the accounting period. This is the phosphorus content coefficient of algae. To calculate the total mass of aquatic plant harvesting and fish catch during the accounting period, The phosphorus content coefficient for aquatic organisms; After the sub-item calculations are completed, based on the lake flow field simulation results from step one, the invalid flux removal module automatically filters flux data that only circulates locally within the lake. The specific rule is as follows: for flux between any two monitoring sections within the lake, if the start and end points of the flux are both located within the same lake hydrological zone defined in step one and do not cross the calculation boundary, it is determined to be an invalid flux and removed from the total calculation results, forming a sub-item calculation list of the total effective phosphorus flux of the lake within the calculation period.

7. The method for accurately calculating the total phosphorus flux of a lake according to claim 6, characterized in that: In step six, based on the law of conservation of total phosphorus mass in lakes, a closed-loop verification mechanism for total phosphorus flux balance is constructed to verify and correct the deviation of the full-caliber flux calculation results obtained in step five. First, the total phosphorus mass balance equation for lakes is established: ； In the formula, To calculate the change in total phosphorus storage in the lake during the accounting period, the change in total phosphorus storage in the lake is calculated using measured values. The calculation formula is as follows: ； In the formula, , These are the measured average total phosphorus concentrations of the lake water at the end and beginning of the accounting period, respectively. , These represent the lake's water volume at the end and beginning of the accounting period, respectively. The allowable relative error threshold for flux balance verification is set at 15%, that is, when When the error is ≤15%, the calculation result is deemed to have passed the verification. When the error exceeds the allowable threshold, the deviation correction process is automatically initiated. First, the entire calculation process is traced back to check for abnormal data values, model inversion deviations, and incorrect values ​​of calculation formula parameters. For model inversion deviations, newly added laboratory measured data is used to incrementally train the inversion model in step four to optimize model parameters, improve inversion accuracy, and complete the accurate calculation of total phosphorus flux in lakes.

8. The method for accurately calculating the total phosphorus flux of a lake according to claim 7, characterized in that: Step seven, after verification and correction, outputs the final accurate calculation results of total phosphorus flux in lakes, including a list of effective total phosphorus flux items for the entire lake within the calculation period, the contribution ratio of total phosphorus input, spatiotemporal variation characteristics analysis of total phosphorus flux, and a report on the identification of key control nodes and priority control objects for total phosphorus input in lakes. At the same time, all calculation data, model parameters, and monitoring data are uniformly archived and stored in the total phosphorus flux database of lakes. Establish a continuous iterative optimization mechanism for the method. Collect newly added monitoring data and laboratory measured data every quarter to incrementally train and optimize the parameters of the total phosphorus concentration inversion model in step four. Every year, based on the changes in lake hydrological conditions and topography, update the lake hydrodynamic and water quality coupling model and the total phosphorus full path tracking results in step one, and simultaneously optimize the monitoring point layout plan and calculation parameters.

Images