Water bloom prediction method and device for backwater area

Abstract

The invention provides a water return area water bloom prediction method and device, and relates to the technical field of environmental science. The method comprises the following steps: acquiring geographic hydrological data and environment monitoring data of a target hydrological environment, and constructing a three-dimensional hydrodynamic water temperature model of a backwater area according to the geographic hydrological data and the environment monitoring data; constructing a three-dimensional eutrophication model of the backwater area based on the water temperature, the water depth, the horizontal flow velocity and the wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area; constructing a BP neural network model based on the backwater area three-dimensional eutrophication model, and training the BP neural network model; and obtaining a water bloom risk prediction result of the water return area of the target hydrological environment by using the neural network model. According to the method, the backwater area three-dimensional hydrodynamic water temperature model is constructed, data support is provided for chlorophyll a calculation of the backwater area eutrophication model, a large number of high-frequency detection devices do not need to be added, the model operation time is greatly shortened, and the model prediction efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of environmental science and technology, and in particular to a method and device for predicting water bloom in a backwater area. Background Art

[0002] Water resources management and environmental science are vital fields in modern society, especially in the construction of reservoirs and dams, which have greatly promoted the rational use and allocation of water resources. Reservoirs and dams not only solve the basic needs of human beings for water, such as water supply, irrigation, power generation, flood control and shipping, but also significantly improve the efficiency of water resource utilization and alleviate the problems of uneven distribution and shortage of water resources. In addition, these water conservancy facilities have also brought significant social and economic benefits, but they are accompanied by water pollution and water quality safety problems, which are becoming increasingly serious with climate change and increased human activities.

[0003] Especially during the process of reservoir storage and water transfer, reservoirs and the backwater areas closely connected to them face complex water quality challenges. Due to the rise in water levels and the slowdown in water flow, the diffusion capacity of water bodies in these areas decreases and the retention time of pollutants is prolonged, which often leads to problems such as eutrophication and algal blooms. These problems not only affect water quality, but may also endanger drinking water safety and public health, making the early warning and prediction of algal blooms an important task to ensure water quality safety.

[0004] At present, algal bloom prediction mainly relies on two types of models: process mechanism models and data-driven models. Process mechanism models simulate the growth environment of phytoplankton by analyzing the interaction between phytoplankton and environmental factors, while data-driven models rely on statistical analysis of a large number of environmental variables to predict the growth of phytoplankton. However, these models have encountered many challenges in their specific application in backwater areas. Due to the special hydrological and hydrodynamic conditions of backwater areas, general river and lake models are often not applicable, and special algal bloom prediction models need to be developed.

[0005] The main drawback of existing technologies is that process mechanism models rely on a large amount of high-frequency monitoring data to ensure the accuracy of the simulation, but this is often limited by economic and technical conditions in actual operations. In addition, the calculation process of such models is complex and time-consuming, which affects the timeliness of the prediction. Although data-driven models can improve prediction efficiency, they require a large amount of high-quality and diverse data as support in the early stage, which is difficult to achieve when data collection is limited. These limitations highlight the shortcomings of existing algal bloom prediction technologies in dealing with algal bloom problems in backwater areas, especially in the challenges of responding quickly to environmental changes and providing accurate early warnings. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for predicting water bloom in a backwater area. The method for predicting water bloom in a backwater area can achieve efficient and accurate prediction of water bloom in a backwater area of ​​a reservoir under limited data conditions.

[0007] In order to achieve the above-mentioned purpose of the present invention, the following technical solutions are particularly adopted:

[0008] In a first aspect, the present invention provides a method for predicting algal bloom in a backwater area, comprising:

[0009] Acquire geographic hydrological data and environmental monitoring data of the target hydrological environment, and construct a three-dimensional hydrodynamic and water temperature model of the backwater area based on the geographic hydrological data and the environmental monitoring data;

[0010] Based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area, a three-dimensional eutrophication model of the backwater area is constructed;

[0011] Based on the three-dimensional eutrophication model of the backwater area, a BP neural network model is constructed, and the BP neural network model is trained;

[0012] The trained neural network model is used to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment.

[0013] In an optional embodiment, the acquiring of geographic hydrological data and environmental monitoring data of the target hydrological environment, and constructing a three-dimensional hydrodynamic and water temperature model of the backwater area according to the geographic hydrological data and the environmental monitoring data, comprises:

[0014] Collecting the geographical hydrological data and the environmental monitoring data of the target hydrological environment, and constructing a one-dimensional hydrological model of the inflowing river, and calculating the riverbed roughness data in the target hydrological environment;

[0015] Based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, a one-dimensional hydrodynamic model of the inflow river is constructed, and the inflow flow data and the upstream flow data are calculated;

[0016] Based on the inflow data and the measured water level data of the reservoir, a two-dimensional hydrodynamic model of the reservoir is constructed, and the downstream water level data of the inflow river is calculated;

[0017] The upstream flow data and the measured water level data of the reservoir are used to construct a three-dimensional hydrodynamic and water temperature model of the backwater area.

[0018] In an optional implementation, the collecting of the geographic hydrological data and the environmental monitoring data of the target hydrological environment, constructing a one-dimensional hydrological model of the incoming river, and calculating the riverbed roughness data in the target hydrological environment include:

[0019] Collecting the geo-hydrological data and the environmental monitoring data in the target hydrological environment; wherein the geo-hydrological data include the surface type of the river entering the reservoir, the reservoir topographic map and the land use remote sensing data; the environmental monitoring data include rainfall data and evaporation data;

[0020] Dividing the inflowing river into a plurality of catchment areas according to the geographical hydrological data and the environmental monitoring data;

[0021] A one-dimensional hydrological model of the inflow river based on the NAM hydrological model is constructed, and each catchment area of ​​the inflow river is calibrated using the one-dimensional hydrological model of the inflow river to obtain the riverbed roughness data.

[0022] In an optional embodiment, based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, a one-dimensional hydrodynamic model of the inflow river is constructed, and the inflow flow data and the upstream flow data are calculated, including:

[0023] Obtaining a river system map and a topological relationship map of the river entering the reservoir, and performing generalization processing on the river channel of the river entering the reservoir and the reservoir;

[0024] Based on the river channel of the inflow river and the reservoir after generalization, the external boundary conditions and internal boundary conditions of the model are determined; wherein the external boundary conditions include the upstream water boundary of the river channel and the downstream water level boundary of each river channel of the inflow river; the internal boundary conditions include the setting of hydraulic structures on the river channel;

[0025] Based on the external boundary conditions and the internal boundary conditions, the MIKE 11HD hydrodynamic calculation model is coupled with the one-dimensional hydrological model of the inflow river to construct a one-dimensional hydrodynamic model of the inflow river, and the upstream flow data and the inflow flow data of the reservoir are calculated.

[0026] In an optional embodiment, the calculation expression of the MIKE 11HD hydrodynamic calculation model is: Among them, x and t represent the coordinates of the calculation point in space and time respectively; A represents the cross-sectional area of ​​the water flow; Q represents the flow rate; h represents the water level; q represents the lateral inflow rate; C represents the Xie Cai coefficient; R represents the hydraulic radius; α represents the momentum correction coefficient; and g represents the acceleration of gravity.

[0027] In an optional implementation, the step of constructing a two-dimensional hydrodynamic model of the reservoir based on the inflow flow data and the measured water level data of the reservoir, and calculating the downstream water level data of the inflow river, comprises:

[0028] Using the reservoir topographic map, two-dimensionally grid the reservoir;

[0029] Based on the reservoir topographic map processed by two-dimensional gridding, the land boundary conditions and water boundary conditions of the model are determined; wherein the land boundary conditions are obtained by the inflow flow data of the inflow river based on the one-dimensional hydrodynamic model of the inflow river; the water boundary conditions are determined by the inflow flow process of the reservoir at the beginning of the century, the measured water level and the reservoir capacity curve;

[0030] A two-dimensional hydrodynamic model of the reservoir is constructed, characteristic points are selected in the reservoir, and the measured water level data are used to calibrate the constructed two-dimensional hydrodynamic model of the reservoir to obtain the downstream water level data of the river entering the reservoir.

[0031] In an optional embodiment, the three-dimensional hydrodynamic and water temperature model of the backwater area is constructed by using the upstream flow data and the measured water level data of the reservoir, including:

[0032] The backwater area of ​​the target hydrological environment is subjected to three-dimensional mesh processing; wherein the three-dimensional mesh processing includes horizontal triangular meshes and quadrilateral meshes, as well as vertical meshes; the middle river channel of the river entering the reservoir adopts the quadrilateral mesh; the area other than the middle river channel of the river channel adopts the triangular mesh; the vertical mesh adopts a sigma and Z-Level mixed layering form;

[0033] Based on the backwater area processed by three-dimensional gridding, a three-dimensional hydrodynamic model of the backwater area is constructed;

[0034] Determine the upstream boundary condition and the downstream boundary condition of the backwater zone; wherein the upstream boundary condition uses the upstream flow data calculated by the one-dimensional hydrodynamic model of the inflow river as an input condition; and uses the intersection boundary of the backwater zone and the reservoir as the downstream boundary condition;

[0035] Using the three-dimensional hydrodynamic model of the backwater zone, based on the upstream boundary conditions and the downstream boundary conditions, the water depth of the backwater zone and the horizontal flow velocity of the backwater zone are calculated;

[0036] Based on the three-dimensional hydrodynamic model of the backwater area and the water temperature model, a three-dimensional hydrodynamic and water temperature model of the backwater area is constructed.

[0037] In an optional embodiment, in the three-dimensional hydrodynamic water temperature model of the backwater zone, the transmission equation of the water temperature model is: in, represents the turbulent diffusion coefficient in the vertical direction; Represents the source term generated by heat exchange with the atmosphere; F trepresents the horizontal diffusion term; T represents temperature; u, v, and w represent the velocity components of the water flow in the x, y, and z directions, respectively;

[0038] The horizontal diffusion term F t The equation is: in, represents the turbulent diffusion coefficient in the horizontal direction;

[0039] Among them, the boundary condition of the surface temperature is: when Z = η, in, represents the turbulent diffusion coefficient in the vertical direction; z represents the position coordinate in the vertical direction; η represents the water surface position; Represents the temperature gradient in the vertical direction; Q n represents the net heat flux, which is a measure of the heat exchange between the water body and the atmosphere; ρ0 represents the density of water; c p represents the specific heat capacity of water;

[0040] The bottom boundary condition is: when z = -d,

[0041] The heat exchange between water and air is calculated based on the following physical process: Q n =Q v +Q c +βq sr,net +Q lr,net ; Among them, Q n represents the net heat flux on the surface; c p represents the specific heat capacity of water; q v represents latent heat flux; q c represents the sensible heat flux; q sr,net represents the net shortwave radiation; q lr,net represents the net longwave radiation; Represents the amount of heat received or lost by a unit mass of water.

[0042] In an optional embodiment, the three-dimensional eutrophication model of the backwater area is constructed based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area, including:

[0043] Constructing the three-dimensional eutrophication model of the backwater area based on a preset eutrophication template;

[0044] Determine the boundary conditions and model forces of the three-dimensional eutrophication model of the backwater area; the boundary conditions include the boundary conditions of the upstream river channel and the boundary conditions of the downstream reservoir junction;

[0045] Based on the boundary conditions, the water temperature, water depth, horizontal flow velocity and wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area, and the light intensity obtained by the meteorological station are used as model forces and calibrated through the three-dimensional eutrophication model of the backwater area.

[0046] In an optional embodiment, constructing a BP neural network model based on the three-dimensional eutrophication model of the backwater area and training the BP neural network model include:

[0047] Based on the downstream water level, upstream flow, temperature, light, upstream water quality and downstream water quality, different scenario plans are constructed, and based on the different scenario plans, the chlorophyll a concentration is calculated through the three-dimensional eutrophication model of the backwater area to form a training set;

[0048] Construct BP neural network model;

[0049] The BP neural network model is trained using the parameters in the training set as the model input and the chlorophyll a concentration to be fitted as the model output to obtain the trained BP neural network model.

[0050] In an optional implementation, the training of the BP neural network model further includes:

[0051] Using principal component analysis to reduce the dimension of parameters in the training set to obtain model input items;

[0052] The model input items are standardized to obtain the model input items after the standardization process; wherein the expression of the standardization process is: Wherein, x represents the original model input; xˋ represents the model input after standardization; μ represents the data mean; σ represents the data standard deviation;

[0053] Based on the cross-validation strategy, the model input items after the standardization process are divided into k folds, wherein the k-1 fold is used as the training data set, and the 1 fold samples are combined with the NSE, RMSE, and MAE evaluation indicators as the validation data set;

[0054] Execute the following training process k times: train the BP neural network model based on the training data set, and verify it through the verification data set to obtain a verification result;

[0055] The k-times verification results are taken as the training scores of the BP neural network model, and the BP neural network model is adjusted according to the training scores.

[0056] In an optional implementation, the constructing of the BP neural network model further includes:

[0057] The number of hidden layers of the BP neural network model is set to a single hidden layer; wherein the number of neurons in the single hidden layer is set to 1 / 3 to 2 times the number of neurons in the input layer;

[0058] Use the NSE indicator to evaluate and obtain the evaluation results;

[0059] If the evaluation result is that the NSE value reaches the preset threshold, it is determined that the performance meets expectations;

[0060] If the evaluation result is that the NSE value does not reach the preset threshold, it is determined that the performance does not meet expectations, and the number of hidden layers is set to double hidden layers; wherein, in the double hidden layers, the number of neurons in the first layer is set to 1 / 3 to 2 times the number of neurons in the input layer; the number of neurons in the second layer is set to 2 / 3 times the number of neurons in the input layer.

[0061] In an optional implementation, the constructing of the BP neural network model further includes:

[0062] Based on the Xavier method, the BP neural network model is weight initialized; wherein the expression of the weight initialization is: Among them, ω represents the weight matrix; U represents uniform distribution; n in Represents the number of neurons in the input layer of the neural network; n out represents the number of neurons in the output layer of the neural network; μ is determined by the type of activation function selected; the activation function is any one of Tanh, Sigmoid and Softsign;

[0063] If the activation function is Tanh, μ=6, if the activation function is Sigmoid, μ=96, if the activation function is Relu, μ=12;

[0064] Using different activation functions to train the BP neural network model, and evaluating it through performance evaluation indicators to obtain evaluation results of each activation function;

[0065] Wherein, the activation function Tanh is: The activation function Sigmoid is: The activation function Softsign is: Where x represents the sum of the input signal, including the weighted sum of the previous layer plus the bias term;

[0066] Based on the evaluation result, the activation function used in the BP neural network model is determined.

[0067] In an optional implementation, the constructing of the BP neural network model further includes:

[0068] Based on the chlorophyll a concentration, a time series model is established; wherein the expression of the time series model is: Among them, xt represents the actual observation value at time point T in the time series; represents the autoregressive coefficient; ε t represents the error term at time point t; ε t-1 , ε t-2 , ε t-q represents the first q error terms in the time series;

[0069] Taking the lag time as an input variable, the chlorophyll a concentration data under the lag time is calculated based on the time series model and used as a time training set;

[0070] Based on the time training set, the BP neural network model is trained and evaluated through performance evaluation indicators to obtain evaluation results corresponding to different lag times, and the BP neural network model is adjusted according to the evaluation results corresponding to different lag times.

[0071] In an optional embodiment, the algal bloom risk prediction result includes: if the chlorophyll a concentration C of the backwater zone of the target hydrological environment meets C≤10, the algal bloom risk prediction result is: the algal bloom degree is level 1; if the chlorophyll a concentration C of the backwater zone of the target hydrological environment meets 10<C≤15, the algal bloom risk prediction result is: the algal bloom degree is level 2; if the chlorophyll a concentration C of the backwater zone of the target hydrological environment meets 15<C≤50, the algal bloom risk prediction result is: the algal bloom degree is level 3; if the chlorophyll a concentration C of the backwater zone of the target hydrological environment meets 50<C≤100, the algal bloom risk prediction result is: the algal bloom degree is level 4; if the chlorophyll a concentration C of the backwater zone of the target hydrological environment meets C>100, the algal bloom risk prediction result is: the algal bloom degree is level 5.

[0072] In a second aspect, the present invention further provides a backwater area algal bloom prediction device, comprising:

[0073] An environmental data module, used to obtain geographical hydrological data and environmental monitoring data of the target hydrological environment, and to construct a three-dimensional hydrodynamic and water temperature model of the backwater area based on the geographical hydrological data and the environmental monitoring data;

[0074] The eutrophication module is also used to construct a three-dimensional eutrophication model of the backwater area based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area;

[0075] The neural network module is also used to construct a BP neural network model based on the three-dimensional eutrophication model of the backwater area, and train the BP neural network model;

[0076] The risk prediction module is used to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment by using the trained neural network model.

[0077] The backwater area bloom prediction method provided in this application provides data support for the chlorophyll a calculation of the backwater area eutrophication model by constructing a three-dimensional hydrodynamic and water temperature model of the backwater area, without adding a large number of high-frequency detection equipment. The neural network big data model based on the mechanism model data training realizes the reconstruction of the chlorophyll a prediction calculation mechanism, and under the premise of ensuring the prediction accuracy (the Nash efficiency coefficient of the chlorophyll a prediction result 3 days in advance reaches 0.972), the model operation time is greatly reduced and the model prediction efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present application and should not be regarded as limiting the scope of protection of the present application. For ordinary technicians in this field, other related drawings can also be obtained based on these drawings without creative work.

[0079] Figure 1 It is a schematic flow chart of Example 1 of the method for predicting water bloom in backwater area of ​​the present invention;

[0080] Figure 2 This is a schematic flow chart of the refinement of step S100 in Example 2 of the method for predicting water bloom in backwater areas of the present invention;

[0081] Figure 3 This is a schematic flow chart of the refinement of step S200 in Example 3 of the method for predicting water bloom in backwater areas of the present invention;

[0082] Figure 4 It is a schematic flow chart of the refinement of step S300 in Example 4 of the method for predicting water bloom in backwater area of ​​the present invention;

[0083] Figure 5 Schematic diagram of module connection of a device for predicting water bloom in a backwater area in an embodiment of the present invention. DETAILED DESCRIPTION

[0084] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0085] The components of the embodiments of the present application generally described and shown in the drawings herein may be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0086] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present application, are intended only to indicate specific features, numbers, steps, operations, elements, components, or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations of the foregoing items.

[0087] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0088] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present application.

[0089] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0090] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0091] Embodiment 1:

[0092] refer to Figure 1 In Example 1 of the present application, a method for predicting water bloom in a backwater area is provided, comprising:

[0093] Step S100, obtaining geographical hydrological data and environmental monitoring data of the target hydrological environment, and constructing a three-dimensional hydrodynamic and water temperature model of the backwater area according to the geographical hydrological data and the environmental monitoring data;

[0094] The target hydrological environment is the area where algal bloom prediction is required. The area may have certain hydrological and environmental characteristics, for example, it may include natural or artificial inflow rivers, river channels, reservoirs and other topographic and environmental characteristics. In addition, it may also target wetlands, estuaries and coastal waters, lakes and reservoirs, etc.

[0095] Each type of hydrological environment has its own unique hydrodynamic characteristics and environmental conditions, which determine the risk and management strategies of eutrophication. Therefore, the algal bloom prediction method needs to take these environmental characteristics into account and be adjusted and optimized according to the specific hydrological environment. Selecting the appropriate target hydrological environment is crucial to ensure the effectiveness and applicability of the prediction model.

[0096] As mentioned above, this step involves collecting geohydrological data and environmental monitoring data of the target hydrological environment.

[0097] Geographical hydrological data may include the geographical location of the water body, drainage area, river direction, riverbed topography, etc., while environmental monitoring data covers water temperature, flow rate, water level, rainfall, etc.

[0098] These data are used to construct a three-dimensional hydrodynamic water temperature model that can describe the physical and thermal behavior of water bodies, such as the direction and speed of water flow, and the distribution and changes of water temperature.

[0099] This model can accurately simulate the dynamic and thermal characteristics of specific water bodies, help understand complex water flow and temperature changes, and provide a basis for subsequent algal bloom predictions.

[0100] Step S200, constructing a three-dimensional eutrophication model of the backwater area based on the water temperature, water depth, horizontal flow velocity and wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area;

[0101] As mentioned above, based on the hydrodynamic water temperature model, this step further constructs a three-dimensional eutrophication model of the backwater area. This model uses the water temperature, water depth, horizontal flow velocity and wind speed data obtained in the previous step to simulate the movement of nutrients such as nitrogen and phosphorus and the dynamics of phytoplankton growth, thereby predicting the risk of eutrophication and algal blooms.

[0102] By combining with hydrodynamic and water temperature models, eutrophication models can more accurately simulate the behavior of nutrients and phytoplankton under complex hydrological conditions and increase the accuracy of predictions.

[0103] Ecological modeling software such as AQUATOX or EcoLab can be used, combined with the output of the hydrodynamic model, to simulate nutrient cycling and biological growth processes.

[0104] Step S300, constructing a BP neural network model based on the three-dimensional eutrophication model of the backwater area, and training the BP neural network model;

[0105] As mentioned above, based on the output of the three-dimensional eutrophication model of the backwater area, this step constructs a BP neural network model for training. This model will use the data provided by the eutrophication model as input, such as chlorophyll a concentration, nutrient salt level, etc., to learn the relationship between these parameters and the occurrence of algal blooms.

[0106] Neural networks can capture complex nonlinear relationships and improve the accuracy and efficiency of predicting algal bloom risks. Specifically, machine learning libraries such as TensorFlow or PyTorch can be used to train and verify the network by setting up a multi-layer perceptron (MLP) and selecting appropriate activation functions, loss functions, and optimizers.

[0107] Step S400, using the trained neural network model to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment.

[0108] As mentioned above, the trained neural network model is used to predict the risk of algal bloom in the backwater area of ​​the target hydrological environment. The model will output the possibility and severity of algal bloom to help decision makers take appropriate management measures. This step directly provides feasible prediction results, which helps to respond to potential environmental risks in a timely manner and optimize resource allocation and protection measures.

[0109] Trained neural network models can be used as inputs for real-time or newly collected hydrological and water quality data to obtain forecast results, which may also include the development of warning levels and recommended response strategies.

[0110] The steps in the above method all ensure the scientificity and practicality of the entire algal bloom prediction method through precise models and advanced technical means, providing an efficient technical support tool for water resources management.

[0111] In summary, the backwater area bloom prediction method provided in this embodiment provides data support for the chlorophyll a calculation of the backwater area eutrophication model by constructing a three-dimensional hydrodynamic water temperature model of the backwater area, without adding a large number of high-frequency detection equipment. The neural network big data model based on the mechanism model data training realizes the reconstruction of the chlorophyll a prediction calculation mechanism, and under the premise of ensuring the prediction accuracy (the Nash efficiency coefficient of the chlorophyll a prediction result 3 days in advance reaches 0.972), the model operation time is greatly reduced, and the model prediction efficiency is improved.

[0112] Embodiment 2:

[0113] refer to Figure 2Based on the above-mentioned embodiment 1, a method for predicting water bloom in a backwater area is provided in embodiment 2 of the present application. The step S100 obtains geographic hydrological data and environmental monitoring data of a target hydrological environment, and constructs a three-dimensional hydrodynamic and water temperature model of the backwater area according to the geographic hydrological data and the environmental monitoring data, including:

[0114] Step S110, collecting the geographical hydrological data and the environmental monitoring data of the target hydrological environment, constructing a one-dimensional hydrological model of the inflowing river, and calculating the riverbed roughness data in the target hydrological environment;

[0115] The above steps are the basis for the construction of the entire model, which involves collecting relevant data on the target hydrological environment, including topography, land use, hydrological cycle data (such as rainfall, evaporation), etc. These data are used to build a one-dimensional hydrological model, which is mainly used to simulate the flow and water quality changes of the river entering the reservoir. Through this step, basic data such as the flow conditions and riverbed roughness of the river can be obtained, which are crucial for understanding the dynamics of water flow and the establishment of subsequent models. This method can reflect the natural and artificial characteristics of the river in detail, which helps to accurately simulate and predict the behavior of water bodies.

[0116] Hydrological data can be collected by processing geographic information using GIS software, combined with hydrological measurement tools and techniques such as current meters. Hydrological models can be constructed and simulated using software such as HEC-HMS or SWAT.

[0117] As mentioned above, in the hydrodynamic model, riverbed roughness data is a key parameter used to describe the resistance or friction of the riverbed surface to the water flow. This parameter is an important factor in evaluating water flow velocity and water flow behavior because it directly affects the dynamic characteristics of the water flow, including flow velocity and flow direction. The riverbed roughness reflects the influence of riverbed materials (such as sand, gravel, silt, vegetation, etc.) and riverbed surface morphology (such as smoothness, roughness, presence or absence of slope and curvature, etc.) on the water flow. The higher the roughness of the riverbed, the greater the friction resistance to the water flow and the slower the flow rate; conversely, the lower the roughness, the faster the water flow.

[0118] Step S120, based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, construct a one-dimensional hydrodynamic model of the inflow river, and calculate the inflow flow data and the upstream flow data;

[0119] In the above, the riverbed roughness data obtained from the first step and the outflow data of the upstream reservoir are used to construct a one-dimensional hydrodynamic model of the inflow river. This model is mainly used to simulate the physical characteristics of the river such as flow velocity and flow, so as to obtain more detailed inflow data and upstream flow data, providing a basis for the hydrodynamic behavior of the reservoir.

[0120] This model provides an accurate description of the physical behavior of water flow, helping to predict flow changes and potential water quality problems in backwater areas.

[0121] Water flow simulation can be performed using hydrodynamic simulation software such as MIKE 11 or HEC-RAS, which can accurately calculate various physical parameters of water flow.

[0122] Step S130, constructing a two-dimensional hydrodynamic model of the reservoir based on the inflow flow data and the measured water level data of the reservoir, and calculating the downstream water level data of the inflow river;

[0123] As mentioned above, this step uses the inflow and measured water level data to construct a two-dimensional reservoir hydrodynamic model to simulate the flow and water level behavior of the reservoir water body. The model can provide the water level changes and water flow status of the reservoir, which is crucial for managing reservoir operations and predicting water quality changes in the reservoir area.

[0124] The two-dimensional model provides more detailed spatial variation information than the one-dimensional model, which helps to more fully understand the dynamic changes of the water body. Specifically, it can be implemented using MIKE 21 or other two-dimensional hydrodynamic simulation software, which can handle complex water body boundaries and internal flow problems.

[0125] Step S140, using the upstream flow data and the measured water level data of the reservoir, construct a three-dimensional hydrodynamic and water temperature model of the backwater area.

[0126] In the above, a three-dimensional hydrodynamic and water temperature model is constructed by combining upstream flow data and measured water level data of the reservoir to simulate the three-dimensional distribution of water flow, water level and water temperature in the backwater area. This model can provide detailed three-dimensional dynamics of the water body in the backwater area, including water temperature distribution, and provide necessary basic data for the subsequent eutrophication model.

[0127] Three-dimensional models can provide the most comprehensive description of water body dynamics, especially in complex backwater areas, and help to gain a deeper understanding of various physical, chemical and biological processes in the water body.

[0128] Specifically, three-dimensional models usually require advanced computing power and software support, such as MIKE 3D or DELFT3D, which can handle complex three-dimensional hydrodynamic and water temperature problems.

[0129] The method provided in this embodiment constructs a hydrodynamic model from one dimension to three dimensions by integrating geohydrological and environmental monitoring data, thereby providing a systematic method to predict and manage the risk of algal bloom in the backwater area. First, by collecting geohydrological data and environmental monitoring data and constructing a one-dimensional hydrological model of the inflowing river, this step uses existing data sources for decision support, optimizes resource use, and reduces the demand for new monitoring facilities. Then, the one-dimensional hydrodynamic model constructed using riverbed roughness and inflow data accurately simulates water flow dynamics, provides key water flow behavior information, and provides an accurate basis for water quality management. Further, combined with inflow and measured water level data, the two-dimensional hydrodynamic model can comprehensively analyze the hydrodynamic and water temperature distribution in the reservoir and improve the water management strategy. Finally, the three-dimensional hydrodynamic water temperature model integrates environmental factors, improves the response ability to complex environmental changes, and provides necessary input parameters for the eutrophication model. On the whole, this series of steps not only enhances the scientific nature and predictive ability of water body management, but also strengthens the prediction and response capabilities of potential algal bloom events in the backwater area, providing an effective tool for sustainable water resources management.

[0130] In summary, the above method can provide high-frequency data input such as flow, water level, water temperature, and water quality for the three-dimensional eutrophication model of the backwater area based on the data of existing meteorological stations, water quality monitoring stations, hydrological stations, etc. around the backwater area, by constructing a coupling hydrological eutrophication model of the inflow river (one-dimensional), the reservoir (two-dimensional), and the backwater area (three-dimensional), thereby driving the prediction of algal bloom in the backwater area without the need for additional high-frequency monitoring equipment.

[0131] Furthermore, the step S110 collects the geographic hydrological data and the environmental monitoring data of the target hydrological environment, constructs a one-dimensional hydrological model of the inflowing river, and calculates the riverbed roughness data in the target hydrological environment, including:

[0132] Step S111, collecting the geo-hydrological data and the environmental monitoring data in the target hydrological environment; wherein the geo-hydrological data include the surface type of the river entering the reservoir, the reservoir topographic map and the land use remote sensing data; the environmental monitoring data include rainfall data and evaporation data;

[0133] The above steps involve collecting basic data of the target hydrological environment, including geohydrological data and environmental monitoring data. Geohydrological data usually include river surface types, reservoir topographic maps, land use data, etc. These data help depict the physical framework and geographical characteristics of the entire water body. Environmental monitoring data such as rainfall and evaporation provide key dynamic information in the hydrological cycle.

[0134] Through this step, comprehensive data sets are obtained, which are the basis for building hydrological models and predicting the behavior of water bodies, thus ensuring the accuracy and reliability of the model because it is built based on realistic and detailed data inputs.

[0135] Specifically, this information can be collected through satellite remote sensing, ground measurements, and weather station data. GIS tools and remote sensing techniques can be used to analyze topographic maps and land use, while standard meteorological instruments are used to collect rainfall and evaporation data.

[0136] Step S112, dividing the inflow river into a plurality of catchment areas according to the geographic hydrological data and the environmental monitoring data;

[0137] As mentioned above, using the collected geo-hydrological data, the rivers entering the reservoir are divided into multiple catchment areas according to natural or artificial characteristics. Such a division helps to understand and simulate the behavior and characteristics of water flow in different areas in a more detailed way. Each catchment area can be evaluated and simulated separately, making the hydrological model more high-resolution and the prediction more targeted. This zoning method allows for detailed analysis of the hydrological dynamics of a specific area, helping to identify possible risk points and key intervention areas.

[0138] Specifically, hydrological modeling software such as HEC-HMS or SWAT can be used to perform hydrological zoning and model the flow and confluence of each area.

[0139] Step S113, constructing a one-dimensional hydrological model of the inflow river based on the NAM hydrological model, and calibrating each catchment area of ​​the inflow river through the one-dimensional hydrological model of the inflow river to obtain the riverbed roughness data.

[0140] In this step, the NAM (Nedbor-Afstromnings Model) hydrological model is used, which is a popular rainfall-runoff model used to simulate the hydrological behavior of rivers entering the reservoir. The model uses the partitioned data to predict the water flow and water quality conditions of each catchment area, so that detailed hydrological parameters such as flow, water level and potential water quality problems can be obtained for each catchment area.

[0141] The NAM model can take into account various hydrological factors and land use impacts, and provide accurate hydrological response predictions, especially under extreme climate events. The collected rainfall data and topographic data are input into the NAM model, and the riverbed roughness and water flow behavior are calculated through the model.

[0142] It should be noted that the NAM hydrological model is a lumped deterministic conceptual model used to simulate the rainfall runoff process in a watershed. It divides soil moisture into four parts: snow storage, surface storage, shallow or root zone storage, and groundwater storage (such as Figure 2 As shown in Figure 2, continuous calculations are performed to simulate the corresponding hydrological processes in the basin.

[0143] The meteorological input data of the NAM model include: rainfall, evaporation capacity and temperature (if the snowmelt model is used), and the output mainly includes: runoff and other hydrological elements such as soil moisture and groundwater recharge. The main parameters of the NAM model include the maximum water content of the surface aquifer Umax, the maximum water content of the root zone aquifer Lmax, the overland flow confluence coefficient CQOF, the soil interflow confluence time CKIF, the overland flow confluence time CK12, the overland flow abortion threshold TOF, the soil interflow abortion threshold TIF, the root zone groundwater recharge threshold TG and the base flow confluence time CKBF. The initial conditions of the model include the relative soil moisture content of the surface aquifer and the root zone aquifer at the beginning, as well as the initial values ​​of overland flow, soil interflow and base flow. The simulation results of the NAM model include surface runoff, soil interflow and base flow processes.

[0144] In this embodiment, firstly, for the target hydrological environment, the reservoir and surrounding topographic maps, land use remote sensing data, rainfall data from rain gauges, evaporation data, etc. in the geographical and hydrological data are collected; then, the rivers entering the reservoir are divided into multiple catchment areas according to the existing topographic data, land use type and water system distribution; finally, the NAM model can be used to set and adjust the hydrological parameter values ​​(including Umax, Lmax, CQOF, CK12, etc.) of each river entering the reservoir, and model calibration will be carried out later.

[0145] In summary, in this embodiment, based on the elevation data of the inflowing rivers, river channel distribution, and land use type, the inflowing rivers are divided into zones, and combined with rainfall evaporation and other data, the MIKE11 NAM model is used to simulate the rainfall runoff process of each inflowing river. Subsequently, it is coupled with the one-dimensional hydrodynamic model of the inflowing rivers to calculate the runoff process of each inflowing river and provide source or side inflow data for the inflowing rivers.

[0146] Furthermore, the step S120 constructs a one-dimensional hydrodynamic model of the inflow river based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, and calculates the inflow flow data and the upstream flow data, including:

[0147] Step S121, obtaining a river system diagram and a topological relationship diagram of the river entering the reservoir, and performing generalization processing on the river channel of the river entering the reservoir and the reservoir;

[0148] The above steps are intended to simplify and generalize the geographic and hydrodynamic characteristics of the incoming rivers and their associated reservoirs. Generalization involves simplifying the complex river system into a format that is easier to simulate and analyze, usually including simplification of the river's length, width, depth, and flow direction.

[0149] Through generalization processing, a simplified river channel and reservoir model can be obtained, which highlights the main paths and key characteristics of the water flow and facilitates subsequent hydrodynamic analysis.

[0150] Simplify complex natural water systems, reduce model complexity, and improve computational efficiency while maintaining sufficient accuracy to predict key hydrological dynamics.

[0151] Specifically, GIS tools and hydrodynamic simulation software (such as MIKE 11 or HEC-RAS) can be used to generalize the river. These tools can help the modeler determine the key parts of the model and set the corresponding simulation parameters.

[0152] Step S122, based on the river channel of the inflow river and the reservoir after generalization processing, determine the external boundary conditions and internal boundary conditions of the model; wherein the external boundary conditions include the upstream water boundary of the river channel and the downstream water level boundary of each river channel of the inflow river; the internal boundary conditions include the setting of hydraulic structures on the river channel;

[0153] As mentioned above, setting the correct boundary conditions in the hydrodynamic model is the key to simulating water flow behavior. External boundary conditions usually involve the inlet and outlet of the river, such as upstream water and downstream water level; internal boundary conditions may include the setting of hydraulic structures such as dams and sluices. After setting these boundary conditions, the movement of water in the river system can be simulated more accurately, including flow rate, water level changes, etc. In this step, ensure that the model can accurately reflect the water flow behavior in the real world, especially when there is intervention from hydraulic structures.

[0154] Boundary conditions can be set and adjusted using MIKE 11HD or similar hydrodynamic calculation software. These software allow detailed configuration of various physical and technical parameters of the model.

[0155] Step S123, based on the external boundary conditions and the internal boundary conditions, the MIKE 11HD hydrodynamic calculation model is coupled with the one-dimensional hydrological model of the inflow river to construct a one-dimensional hydrodynamic model of the inflow river, and the upstream flow data and the inflow flow data of the reservoir are calculated.

[0156] As mentioned above, the MIKE 11HD model is used to further accurately simulate the hydrodynamic behavior of the river. By coupling with the one-dimensional hydrological model constructed earlier, the model can more comprehensively predict the flow, water level and related water quality changes in the river. The coupled model provides detailed predictions of water flow, water level and possible water quality changes, providing key data for algal bloom risk prediction. The model's predictive power and accuracy are improved, and it can handle both hydrological and hydrodynamic factors at the same time, which is particularly important for managing and predicting the impact of reservoirs on the downstream hydrological environment.

[0157] Hydrological data and hydrodynamic parameters can be integrated in MIKE 11HD software to carry out detailed hydrodynamic simulation and analysis.

[0158] Furthermore, the calculation expression of the MIKE 11HD hydrodynamic calculation model is:

[0159]

[0160] Among them, x and t represent the coordinates of the calculation point in space and time respectively; A represents the cross-sectional area of ​​the water flow; Q represents the flow rate; h represents the water level; q represents the lateral inflow rate; C represents the Xie Cai coefficient; R represents the hydraulic radius; α represents the momentum correction coefficient; and g represents the acceleration of gravity.

[0161] It should be noted that the MIKE11 HD hydrodynamic calculation model used in this embodiment is based on the vertically integrated mass and momentum conservation equations, namely the one-dimensional unsteady flow Saint-Venant equations, to simulate the flow state of a river or estuary.

[0162] It should be noted that the one-dimensional hydrodynamic model of the river channel adopts MIKE 11HD to construct the hydrodynamic model of the inflowing river, and is coupled with the NAM hydrological model of the inflowing river for joint calibration. On the one hand, it provides the inflow flow data for the construction of the two-dimensional hydrodynamic model of the reservoir, and on the other hand, it provides the upstream boundary flow data for the three-dimensional hydrodynamic model of the backwater area of ​​the inflowing river.

[0163] For example, first collect river system maps and their topological relationship maps to generalize rivers and reservoirs; then set internal and external boundaries. The external boundary conditions include the upstream water boundary of the river and the downstream water level boundary of each river entering the reservoir; the internal boundary conditions include the setting of hydraulic structures on the river; finally, select the measured flow data of important hydrological stations in the reservoir basin and the reservoir inflow flow as calibration targets, and couple the NAM model to carry out model calibration.

[0164] Furthermore, the step S130, based on the inflow data and the measured water level data of the reservoir, constructs a two-dimensional hydrodynamic model of the reservoir, and calculates the downstream water level data of the inflow river, including:

[0165] Step S131, using the reservoir topographic map, performing two-dimensional grid processing on the reservoir;

[0166] In this embodiment, based on the simulation calculation results of the one-dimensional hydrological and hydrodynamic model of the inflowing river, a two-dimensional hydrodynamic model of the reservoir is constructed to provide downstream boundary water level data for the three-dimensional hydrodynamic model of the backwater area of ​​the inflowing river.

[0167] The above steps involve using a topographic map of the reservoir to create a 2D gridded model. 2D gridding is a standard technique for simulating water flow and related dynamic processes, which converts complex water topography into a more computationally tractable grid structure. A detailed grid model is generated that accurately describes the physical form and hydrodynamic behavior of the reservoir in 2D space.

[0168] Two-dimensional gridding enables the model to capture details such as flow velocity, flow direction and water quality changes within the water body, thereby improving the spatial resolution and prediction accuracy of the model.

[0169] GIS software and hydrodynamic simulation software (such as MIKE 21 or Hec-RAS2D) can be used to grid the terrain map and set the appropriate grid size to balance computational efficiency and accuracy.

[0170] Step S132, based on the reservoir topographic map processed by two-dimensional gridding, determine the land boundary conditions and water boundary conditions of the model; wherein the land boundary conditions are obtained by the inflow data of the inflow river based on the one-dimensional hydrodynamic model of the inflow river; the water boundary conditions are determined by the inflow process of the reservoir at the beginning of the century, the measured water level and the reservoir capacity curve;

[0171] As mentioned above, setting appropriate boundary conditions is the key to building an accurate hydrodynamic model. Land boundary conditions involve the boundary where the water body contacts the land, such as the shape and characteristics of the coastline; water boundary conditions deal with the flow characteristics inside the water body, such as inlet flow and outlet water level.

[0172] This step can ensure that the model boundaries reflect the actual situation and provide correct input and output for water flow simulation. Appropriate boundary condition setting can ensure the reliability and accuracy of model output, which has direct practical significance for water quality and algal bloom management. Boundary conditions can be set by integrating flow data of the incoming river and water level data of the reservoir, and these conditions can be applied using hydrodynamic simulation software.

[0173] Step S133, constructing a two-dimensional hydrodynamic model of the reservoir, selecting characteristic points in the reservoir, using the measured water level data, calibrating based on the constructed two-dimensional hydrodynamic model of the reservoir, and obtaining the downstream water level data of the river entering the reservoir.

[0174] As mentioned above, after completing the gridding process and setting the boundary conditions, a two-dimensional hydrodynamic model of the reservoir is constructed, and the model is calibrated using the measured water level data. Calibration is the process of adjusting the model parameters to match the measured data to ensure that the model can accurately predict future hydrological conditions. Through the calibration process, the parameters of the model are adjusted to ensure that its output matches the actual observations, thereby improving the model's predictive reliability. The calibrated model can more accurately reflect the actual hydrodynamic conditions and provide accurate basic data for algal bloom prediction, especially under changing climate and water conditions.

[0175] The model can be calibrated using hydrodynamic simulation software (such as MIKE 21HD or other similar tools) and input with measured water levels and other hydrological data. The model parameters are adjusted using an optimization algorithm to minimize the difference between the model output and the observed data.

[0176] Furthermore, the step S140, using the upstream flow data and the measured water level data of the reservoir, constructs a three-dimensional hydrodynamic and water temperature model of the backwater area, including:

[0177] Step S141, performing three-dimensional mesh processing on the backwater area of ​​the target hydrological environment; wherein the three-dimensional mesh processing includes horizontal triangular meshes and quadrilateral meshes, as well as vertical meshes; the middle river channel of the river entering the reservoir adopts the quadrilateral mesh; the area outside the middle river channel of the river adopts the triangular mesh; the vertical mesh adopts a sigma and Z-Level mixed layering form;

[0178] The above steps are the starting point for building a three-dimensional hydrodynamic model. First, a three-dimensional hydrodynamic model of the backwater area can be constructed, and then a water temperature model can be constructed based on the three-dimensional hydrodynamic model of the backwater area, that is, a three-dimensional hydrodynamic and water temperature model of the backwater area. In this step, it is first necessary to convert the physical space of the backwater area into a three-dimensional grid. Such a gridding process allows the model to describe the flow and temperature distribution of the water body in detail in space. This produces a three-dimensional space model containing horizontal triangular and quadrilateral grids, as well as vertical (Z-Level and sigma layer) grids.

[0179] Three-dimensional gridding provides a high-resolution representation of complex hydrological environments and can capture detailed changes in water dynamics, such as flow velocity, water temperature gradient, etc.

[0180] Meshing can be done using 3D hydrodynamic simulation software such as MIKE 3D or DELFT3D. These software can automatically or manually create a suitable mesh structure based on the physical terrain and hydrological conditions.

[0181] Step S142, constructing a three-dimensional hydrodynamic model of the backwater area based on the backwater area processed by three-dimensional meshing;

[0182] Step S143, determining the upstream boundary conditions and downstream boundary conditions of the backwater area; wherein the upstream boundary condition uses the upstream flow data calculated by the one-dimensional hydrodynamic model of the inflow river as an input condition; and uses the intersection boundary of the backwater area and the reservoir as the downstream boundary condition;

[0183] To ensure the model runs accurately, accurate upstream and downstream boundary conditions need to be set. These conditions simulate the behavior of water entering and leaving the simulation area based on actual hydrological data. This ensures that the model's water flow data during simulation is consistent with the actual situation, especially flow and water level.

[0184] The correct setting of boundary conditions is the key to ensure the accuracy of simulation results, especially when simulating large-scale hydrological dynamics. The upstream boundary conditions are set using the flow data obtained from the one-dimensional hydrodynamic model of the inflowing river, and the intersection with the reservoir is used as the downstream boundary condition, which can be implemented through hydrodynamic simulation software.

[0185] Step S144, using the three-dimensional hydrodynamic model of the backwater area, based on the upstream boundary condition and the downstream boundary condition, calculate the water depth of the backwater area and the horizontal flow velocity of the backwater area;

[0186] As mentioned above, based on the constructed three-dimensional grid and the set boundary conditions, the model calculates the water depth and horizontal flow velocity in the backwater area, and obtains detailed water depth and flow velocity data, which have a direct impact on understanding and predicting the formation of algal blooms. This step provides comprehensive information about the dynamics of the water body, which helps to identify areas and conditions that may cause algal blooms.

[0187] These calculations can be performed using three-dimensional hydrodynamic simulation software, which runs the dynamic equations based on actual physical and hydrological data to produce results.

[0188] Step S145, constructing the three-dimensional hydrodynamic and water temperature model of the backwater area based on the three-dimensional hydrodynamic model of the backwater area and the water temperature model.

[0189] As mentioned above, the final step is to integrate the water temperature factor into the three-dimensional hydrodynamic model to create a comprehensive hydrodynamic water temperature model that can simulate the flow and temperature changes of the water body at the same time. Generate a comprehensive model that can simulate the dynamics of water flow and water temperature to provide the necessary environmental data for subsequent algal bloom risk prediction. Water temperature is one of the key factors affecting the occurrence of algal blooms. This model can accurately predict water temperature changes and help improve the accuracy of algal bloom prediction. Specifically, water temperature change equations and related thermal parameters can be added to the three-dimensional hydrodynamic model, and simulation software such as MIKE 3D or other similar tools can be used to perform these calculations.

[0190] For example, in this embodiment, MIKE 3FM is used to build a three-dimensional hydrodynamic and water quality coupling model for the backwater area. First, the MIKE 3HD module is used to establish a three-dimensional hydrodynamic and water temperature coupling model for the backwater area, and on this basis, the three-dimensional ECOLab module is coupled to establish a three-dimensional eutrophication model. Using the three-dimensional hydrodynamic and water temperature coupling model of the backwater area, the upstream flow data, downstream water level data, and meteorological station, hydrological station and water quality monitoring data of the backwater area are processed to provide spatiotemporal distribution data input for the construction of the eutrophication model of the backwater area, and the water quality and chlorophyll a concentration of the backwater area are output through model simulation calculation.

[0191] The MIKE 3FM model is based on the shallow water and Boussinesq assumptions to solve the Navier-Stokes equations for incompressible fluids. The horizontal grid in the model is divided into unstructured grids, and the vertical grid is divided into coordinates.

[0192] For the construction of the three-dimensional hydrodynamic model of the backwater area, first, a three-dimensional backwater area grid is set up, and triangles and quadrilaterals are used for gridding. The river channel part of the middle part of the river channel adopts quadrilateral grid, and other areas are constructed with triangular grids to meet the needs of grid depiction of actual terrain; the vertical grid adopts sigma and Z-Level mixed layering form.

[0193] Then, key parameters such as roughness type and horizontal eddy viscosity coefficient are set, and the wind speed and direction data of the meteorological station are used as the wind field conditions of the model;

[0194] Finally, the backwater area uses the water from the upstream river as the upstream open boundary, and the downstream is the intersection boundary of the backwater area and the Danjiangkou Reservoir as the downstream open boundary; the upstream open boundary uses the inflow condition calculated by the one-dimensional hydrodynamic model of the inflow river as the input condition, and the downstream open boundary uses the measured water level of the reservoir as the input condition. The water depth, horizontal flow velocity and other data of the backwater area are calculated through the model.

[0195] After the three-dimensional hydrodynamic model of the backwater area is constructed, the three-dimensional hydrodynamic and water-temperature model of the backwater area is constructed based on the three-dimensional hydrodynamic model of the backwater area and the water temperature model.

[0196] Furthermore, in the three-dimensional hydrodynamic water temperature model of the backwater area, the transmission equation of the water temperature model is: in, represents the turbulent diffusion coefficient in the vertical direction; Represents the source term generated by heat exchange with the atmosphere; F t represents the horizontal diffusion term; T represents temperature; u, v, and w represent the velocity components of the water flow in the x, y, and z directions, respectively; the horizontal diffusion term F t The equation is:

[0197] in, represents the turbulent diffusion coefficient in the horizontal direction;

[0198] It should be noted that the turbulent diffusion coefficients in the horizontal and vertical directions are constant values ​​or are determined in proportion by the eddy viscosity coefficient.

[0199] In this embodiment, a water temperature model is introduced based on a three-dimensional hydrodynamic model, and a complex water temperature transmission equation is set to simulate the spatial distribution of water temperature and its temporal changes. The equation includes factors such as convection, diffusion, and heat exchange between the water body and the atmosphere, and comprehensively simulates the dynamic changes of water body temperature.

[0200] The constructed model can describe in detail the temperature distribution of water bodies in three-dimensional space and its changes over time, taking into account a variety of thermodynamic processes. This model allows accurate prediction of ecological and chemical changes caused by changes in water temperature, especially the role of water temperature in the occurrence of algal blooms. It is essential for managing water health and predicting algal bloom risks.

[0201] Regarding boundary conditions:

[0202] (1) The boundary condition of the surface temperature is: when Z = η, in, represents the turbulent diffusion coefficient in the vertical direction; z represents the position coordinate in the vertical direction; η represents the water surface position; Represents the vertical temperature gradient; Q n represents the net heat flux, which is a measure of the heat exchange between the water body and the atmosphere; ρ0 represents the density of water (usually about 1000 kg / m 3 );c p represents the specific heat capacity of water [given as 4217 J / (kg·K)].

[0203] The bottom boundary condition is: when z = -d,

[0204] The heat exchange at the water-air surface is calculated based on the following physical processes:

[0205] Q n =q v +q c +βq sr,net +q lr,net ;(Formula 7) Among them, Q n represents the net heat flux on the surface (the sum of all heat flux components); c p =4217 J / (kg°k) represents the specific heat of water; q v represents latent heat flux (latent heat flux, associated with the evaporation or condensation of water); q c represents the sensible heat flux (related to direct heat conduction between the water body and the atmosphere); q sr,net represents the net shortwave radiation (net solar radiation input); q lr,net represents the net longwave radiation (infrared radiation exchange between water bodies and the atmosphere); Represents the amount of heat received or lost by a unit mass of water (used in temperature change calculations).

[0206] Together, these formulas form the basis for thermodynamic boundary conditions in water temperature models, which accurately describe how water bodies exchange heat with their surroundings through different physical processes. These models allow accurate predictions of changes in water temperature under different environmental conditions, which is critical for ecological models, water quality management, and environmental monitoring.

[0207] As mentioned above, the model can be constructed by using the atmospheric temperature and relative humidity measured at the meteorological station as input conditions respectively; the upstream and downstream water temperature boundaries use the water temperature measured data of the water quality monitoring station as input conditions, and the water temperatures of different layers are estimated through the three-dimensional hydrodynamic water temperature model of the backwater area.

[0208] Embodiment 3:

[0209] refer to Figure 3 Based on the above-mentioned embodiment 1, a method for predicting water bloom in a backwater area is provided in embodiment 3 of the present application. The step S200 constructs a three-dimensional eutrophication model of the backwater area based on the water temperature, water depth, horizontal flow velocity and wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area, including:

[0210] Step S210, constructing a three-dimensional eutrophication model of the backwater area based on a preset eutrophication template;

[0211] The above-mentioned three-dimensional eutrophication model of the backwater area is modified and constructed based on the predefined eutrophication template in ECO Lab.

[0212] It should be noted that the eutrophication template is a system composed of a series of differential equations, which are composed of differential equations describing the change process of 12 state variables of phytoplankton, chlorophyll a, zooplankton, detritus, dissolved oxygen and inorganic nutrients in the water body, as shown in Table 4.3. The entire model is calculated based on the circulation and conversion of carbon between different nutrient pools (phytoplankton, zooplankton, detritus, sediments), and the amount of carbon is assumed to be sufficient during the calculation process.

[0213] Table 1. List of state variables of the three-dimensional eutrophication model of the backwater area and the sediment system template

[0214] Chinese name Model Abbreviation Description in the model unit Carbon content of phytoplankton PC Phytoplankton C <![CDATA[gC / m 3 ]]> Nitrogen content of phytoplankton PN Phytoplankton N <![CDATA[gN / m 3 ]]> Phosphorus content of phytoplankton PP Phytoplankton P <![CDATA[gPm 3 ]]> Chlorophyll a CH Chlorophyll-a <![CDATA[g Chl / m 3 ]]> Carbon content of zooplankton ZC Zooplankton C <![CDATA[gC / m 3 ]]> Carbon content in the debris DC Detritus C <![CDATA[gC / m 3 ]]> Nitrogen content in debris DN Detritus N <![CDATA[gN / m 3 ]]> Phosphorus content in debris DP Detritus P <![CDATA[gP / m 3 ]]> Ammonia nitrogen <![CDATA[NH3]]> Ammonia N <![CDATA[gN / m 3 ]]> Nitrate <![CDATA[NO3]]> Nitrate N <![CDATA[gN / m 3 ]]> Inorganic phosphorus IP Inorganic phosphorous <![CDATA[gP / m 3 ]]> Dissolved oxygen DO Dissolved oxygen <![CDATA[g DO / m 3 ]]>

[0215] The life activities of phytoplankton in water bodies mainly consist of phytoplankton taking in nutrients to maintain its own growth, phytoplankton consuming nutrients in its body for metabolism, phytoplankton photosynthesis to produce oxygen, phytoplankton being preyed on by zooplankton, and phytoplankton settling and dying. In the model, these processes are represented by growth, predation, sinking and death.

[0216] Nutrients, light, and temperature all affect the growth process of phytoplankton, and the effects of these environmental factors are reflected in the equations through limiting functions. Phytoplankton is represented by four state variables: phytoplankton carbon, phytoplankton nitrogen, phytoplankton phosphorus, and phytoplankton chlorophyll a. The nutrient pools (carbon, nitrogen, and phosphorus) within the phytoplankton cells in the model are state variables because their uptake dynamics are separated from the uptake dynamics of phytoplankton carbon, resulting in nitrogen-carbon ratios and phosphorus-carbon ratios that vary over time. For carbon-based nutrient pools within phytoplankton cells, their source and sink terms are proportional to the corresponding carbon amounts.

[0217] In the above steps, a eutrophication model is constructed using data (water temperature, water depth, water velocity and wind speed) obtained from a three-dimensional hydrodynamic water temperature model. This model uses a preset eutrophication template that defines how to evaluate the behavior of nutrients (such as nitrogen and phosphorus) and their impact on the occurrence of algal blooms based on hydrological and meteorological data. This creates a model that can simulate and predict the dynamics of nutrients and the potential occurrence of algal blooms in a specific water body.

[0218] Specifically, ecological modeling software or customized numerical simulation programs can be used in conjunction with geographic information systems (GIS) to process spatial data and run these complex ecological models.

[0219] Step S220, determining the boundary conditions and model forces of the three-dimensional eutrophication model of the backwater area; the boundary conditions include the boundary conditions of the upstream river channel and the boundary conditions of the downstream reservoir junction;

[0220] As mentioned above, the open boundary conditions of the three-dimensional eutrophication model of the backwater area include two boundary conditions at the confluence of the upstream river channel and the downstream reservoir. The water quality conditions of the two boundaries are both based on the measured water quality data of the water quality monitoring station. The water quality monitoring indicators include total nitrogen, total phosphorus, ammonia nitrogen, and dissolved oxygen.

[0221] The model forces include water temperature, light intensity, water depth, horizontal flow velocity, vertical water layer thickness and wind speed, etc. Among them, water temperature, water depth, horizontal flow velocity and wind speed are built-in forces. Water temperature is obtained through the three-dimensional water temperature model of the backwater area; water depth, horizontal flow velocity and wind speed are read from the three-dimensional hydrodynamic model of the backwater area; light intensity is obtained from the surrounding meteorological stations.

[0222] Set parameters related to algae growth and bloom process. Some parameters will be adjusted during model calibration and verification.

[0223] Step S230, based on the boundary conditions, the water temperature, water depth, horizontal flow velocity and wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area, and the light intensity obtained by the meteorological station are used as model forces, and the three-dimensional eutrophication model of the backwater area is calibrated.

[0224] Calibration is the process of adjusting model parameters to ensure that the model output is consistent with actual observed data. This step uses actual observed water quality data to adjust the model to ensure its accuracy in future applications. The established three-dimensional eutrophication model of the backwater area was calibrated and verified using data from two water quality monitoring stations upstream and downstream of the backwater area.

[0225] In summary, the method of constructing a three-dimensional eutrophication model in this example provides a systematic solution for assessing and managing algal blooms in backwater areas. This method combines detailed hydrological data, ecological models, and field monitoring data to predict the occurrence of algal blooms through a scientific method and provide decision support for management.

[0226] Embodiment 4:

[0227] refer to Figure 4 Based on the above-mentioned embodiment 1, a method for predicting water bloom in a backwater area is provided in embodiment 4 of the present application. The step S300, based on the three-dimensional eutrophication model of the backwater area, constructs a BP neural network model, and trains the BP neural network model, including:

[0228] Step S310, constructing different scenario plans based on the downstream water level, upstream flow, temperature, light, upstream water quality and downstream water quality, and calculating the chlorophyll a concentration through the three-dimensional eutrophication model of the backwater area based on the different scenario plans to form a training set;

[0229] As mentioned above, this step involves using a three-dimensional eutrophication model to simulate chlorophyll a concentrations based on different environmental parameters such as downstream water level, upstream flow, air temperature, light, upstream water quality, and downstream water quality. These data, which reflect the possible development of algal blooms under various environmental conditions, are used to generate a training set for the neural network model, thereby generating a data set containing a variety of predictor variables and a response variable (chlorophyll a concentration), which will be used to train the neural network model.

[0230] By simulating the chlorophyll a concentration under different environmental conditions, the neural network model can learn the complex relationship between environmental factors and the development of algal blooms, thereby enhancing the model's generalization ability and prediction accuracy.

[0231] Specifically, environmental simulation software or a custom numerical simulation script can be used to run the eutrophication model and output a data set for training. Usually, the data needs to be preprocessed, such as normalization, to optimize the training effect of the neural network.

[0232] Step S320, constructing a BP neural network model;

[0233] After obtaining the training dataset, the next step is to build a BP (back propagation) neural network model, which is a widely used supervised learning network that can learn complex nonlinear relationships between input (environmental parameters) and output (chlorophyll a concentration).

[0234] Step S330, using the parameters in the training set as model input and the chlorophyll a concentration to be fitted as model output, the BP neural network model is trained to obtain the trained BP neural network model.

[0235] In this step, the constructed BP neural network model is trained using the data set generated in step S310. The training process involves adjusting network parameters (such as weights and biases) to minimize the difference between the predicted output and the actual data. The trained neural network model can accurately predict the concentration of chlorophyll a based on the input environmental parameters.

[0236] The training process adapts the model to specific data characteristics and patterns, improving the accuracy and reliability of predictions.

[0237] Model parameters can be iteratively adjusted during training by defining loss functions (such as mean squared error) and optimization algorithms (such as stochastic gradient descent). Training can be performed on high-performance computing resources to handle large amounts of data and complex network structures.

[0238] Furthermore, the step S330 of training the BP neural network model further includes:

[0239] Step S331, using principal component analysis method to reduce the dimension of the parameters in the training set to obtain model input items;

[0240] It should be noted that, since the data set has relatively low volatility, contains less implicit information, and contains more water quality parameters, using all of them as input items can easily lead to "dimensionality disaster" and greatly increase the model operation load. Therefore, in this embodiment, the data is preprocessed. Specifically, principal component analysis (PCA) can be used to reduce the data dimension and use it as the model input item. The principal component selection is based on the cumulative contribution rate being higher than 85%.

[0241] As mentioned above, before training the neural network, PCA is used to reduce the number of features in the data set, which helps to reduce model complexity and avoid overfitting. In this embodiment, principal components are extracted from the original data set to cumulatively explain at least 85% of the data variance as input to the neural network, thereby enabling a simplified data set that retains most of the information but has a lower dimension.

[0242] It simplifies the input of the model, speeds up the training, and may also improve the generalization ability of the model.

[0243] Specifically, the PCA algorithm can be used in the data preprocessing stage, which is usually implemented in Python through the PCA class of the scikit-learn library.

[0244] Step S332, standardize the model input items to obtain the model input items after standardization; wherein the expression of the standardization is:

[0245] Wherein, x represents the original model input; xˋ represents the model input after standardization; μ represents the data mean; σ represents the data standard deviation;

[0246] As mentioned above, the neural network model is sensitive to the input data features. In order to eliminate the order of magnitude differences of data of different dimensions, the input data is converted into a standard distribution form with a mean of 0 and a variance of 1 by standardization. In this way, each feature contributes equally to the model, preventing some features with large dimensions from dominating the learning process. The mean and standard deviation of each feature are calculated, and all data points are converted using these statistics. All features have the same scale, and the convergence speed of the model training may be faster.

[0247] This step is used to prevent the gradient from disappearing or exploding and improve the stability of the model.

[0248] Step S333, based on the cross-validation strategy, the model input items after the standardization process are divided into k folds, wherein the k-1 fold is used as the training data set, and the 1 fold samples are combined with the NSE, RMSE, and MAE evaluation indicators as the verification data set;

[0249] As mentioned above, the original data processed by PCA dimensionality reduction and standardization were used to train the BP neural network model of the backwater area, and the differences in the simulation results of the two were compared through cross-validation. The main evaluation indicators were: NSE, RMSE, MAE, MRE, and MRE_1. At the same time, in order to analyze the fluctuation of the cross-validation results, the standard error (Standard Error) of the NSE score was calculated, and the index was marked as Std(NSE).

[0250] The k-folders cross-validation strategy is adopted to realize the division of training set and validation set while ensuring that the sample data is fully utilized. The entire training set samples are divided into k folds, and training is performed on the training set consisting of k-1 fold samples. The model training is evaluated on the validation set consisting of 1 fold samples by combining NSE, RMSE, and MAE evaluation indicators. The model is executed k times in sequence, and the average of k times is taken as the final score of the model. This score is used as the basis for evaluating the parameter tuning results.

[0251] In this step, by dividing the data into multiple parts and using one part as the validation set and the rest as the training set alternately, the generalization ability of the model can be better evaluated. The data is divided into k subsets, and one subset is set aside each time as validation data, and the rest is used as training data. In this way, the performance of the model on multiple different data subsets can be obtained, reducing the model's dependence on a specific data distribution.

[0252] Constructing a validation data set reduces the risk of model overfitting and improves the reliability of the model.

[0253] You can use k-fold cross validation during model training, usually using the cross_val_score function of the scikit-learn library.

[0254] Step S334, executing the following training process k times: training the BP neural network model based on the training data set, and verifying it through the verification data set to obtain a verification result;

[0255] As mentioned above, by repeating training and validation, the performance of the model on different training and validation data sets can be evaluated, helping to identify the stability and reliability of the model. In each iteration, different training and validation data sets are selected to train and test the model. This ensures that each data point has the opportunity to serve as validation data, thereby more comprehensively evaluating the model.

[0256] After each training, the model is evaluated on the validation set to obtain performance indicators (such as accuracy, loss, etc.), which can be used to measure the generalization ability of the model. This provides a comprehensive view of the model's performance and helps avoid overfitting the model to a single data set.

[0257] Step S335, taking the k-times verification results as the training scores of the BP neural network model, and adjusting the BP neural network model according to the training scores.

[0258] As mentioned above, by collecting the results of multiple training and validation runs, we can get the average performance of the model on different subsets, which helps to evaluate the overall performance of the model and adjust the model parameters based on this information. The results of k validation runs are summarized, and the average score is usually used to evaluate the overall performance of the model. Based on the average score of k runs, the quality of the model parameters can be judged more accurately, so as to optimize the model.

[0259] This step makes the model adjustment more scientific and systematic, based on objective data rather than the results of a single experiment.

[0260] In the model training process, the network structure or learning parameters can be adjusted according to the average values ​​of evaluation indicators (such as accuracy, recall, F1 score, etc.), and these parameters can be optimized using the back propagation algorithm.

[0261] The above method in this embodiment can ensure that the BP neural network can not only learn the intrinsic pattern of the data, but also has good generalization ability and good prediction performance for unknown data. In actual operation, these steps usually need to be performed using a specialized machine learning and statistical software library to ensure computational efficiency and accuracy of the results.

[0262] Furthermore, the step S320, constructing the BP neural network model, also includes:

[0263] Step S3211, setting the number of hidden layers of the BP neural network model to a single hidden layer; wherein the number of neurons in the single hidden layer is set to 1 / 3 to 2 times the number of neurons in the input layer;

[0264] It should be noted that the structure of the hidden layer of the neural network model includes the number of hidden layers and the number of neurons in a single layer. Generally speaking, the more hidden neurons and hidden layers there are, the stronger its ability to mine and extract data features is. However, at the same time, the existence of the curse of dimensionality also increases the time and storage space required for model operation exponentially, and the complex hidden layer structure may cause the model to overfit, affecting the generalization ability of the model.

[0265] Theoretically, a single hidden layer neural network with a nonlinear activation function is sufficient to extract the nonlinear mapping relationship between Chl-a and the input environmental variables.

[0266] In this embodiment, the applicability of a single hidden layer neural network to a two-point model will be examined first. Based on empirical methods and considering the model computational load, the range of single hidden layer neurons is set between 1 / 3 and 2 times (i.e., 2-12) of the input layer neurons, and the model performance is verified and evaluated.

[0267] In this step, a single hidden layer neural network is first set up to simplify the structure of the model, which helps to quickly implement and initially observe the performance of the model. In the neural network structure, only one hidden layer is included, and the number of neurons can be adjusted according to the size of the input layer, thus creating a single hidden layer BP neural network model. Simplifying the model can reduce the consumption of computing resources and help avoid overfitting.

[0268] Step S3212, using the NSE indicator to perform evaluation and obtain an evaluation result;

[0269] As mentioned above, since NSE is more sensitive to parameter settings in practice, NSE is used as the main evaluation indicator. At the same time, considering the volatility of the model's own fitting effect, a polynomial regression curve (PolynomialRegression) is fitted to the NSE score value to present its main change trend.

[0270] Step S3213, if the evaluation result is that the NSE value reaches a preset threshold, it is determined that the performance meets expectations;

[0271] Step S3214, if the evaluation result is that the NSE value does not reach the preset threshold, it is determined that the performance does not meet expectations, and the number of hidden layers is set to double hidden layers; wherein, in the double hidden layers, the number of neurons in the first layer is set to 1 / 3 to 2 times the number of neurons in the input layer; the number of neurons in the second layer is set to 2 / 3 times the number of neurons in the input layer.

[0272] As mentioned above, if the number of hidden neurons reaches the upper limit and is not enough to explain the correlation between variables, consider building a double hidden layer neural network. The double hidden layer neural network has stronger feature extraction capabilities than a single hidden layer, and can effectively reduce the number of hidden neurons required.

[0273] Similarly, according to the empirical method, the number of neurons in the first hidden layer is set to 2-12, and the number of neurons in the second hidden layer is set to 2 / 3 of the first layer (rounded up), making it slightly smaller than the first layer to improve the ability to summarize features. Finally, the optimization effect of the double hidden layer neurons is evaluated by calculating the model NSE score.

[0274] In this embodiment, the above method embodies an iterative and adaptive method to optimize the structure and parameters of the neural network model, ensuring that the model is neither too simple nor too complex, so as to achieve the best performance in a given application context. In this way, it can be ensured that the model has good generalization ability and prediction accuracy on a specific problem.

[0275] Furthermore, the step S320, constructing the BP neural network model, also includes:

[0276] Step S3221, based on the Xavier method, weight initialization is performed on the BP neural network model; wherein the expression of the weight initialization is: Among them, ω represents the weight matrix; U represents uniform distribution; n in Represents the number of neurons in the input layer of the neural network; n out Represents the number of neurons in the output layer of the neural network; μ is determined by the type of activation function selected; the activation function is any one of Tanh, Sigmoid and Softsign; if the activation function is Tanh, μ=6, if the activation function is Sigmoid, μ=96, if the activation function is Relu, μ=12;

[0277] It should be noted that for the neural network model, the random initialization strategy of each neuron weight helps to break the symmetry between each neuron node and improve the feature extraction ability of the neural network.

[0278] Therefore, when building the preset model, the parameter initialization strategy in this embodiment adopts the Xavier method. Correct weight initialization can help prevent the gradient vanishing or exploding problem in the early stage of training, and promote the network to converge faster and more effectively. The Xavier initialization method (such as the weight initialization expression) is adopted, which automatically adjusts the initial size of the weight according to the number of neurons in the input layer and the output layer. This makes the neural network have a more uniform activation and gradient distribution in the early stage of training, which helps to improve the training efficiency and stability of the network.

[0279] Step S3222, using different activation functions to train the BP neural network model, and evaluating it through performance evaluation indicators to obtain evaluation results of each activation function;

[0280] Wherein, the activation function Tanh is: The activation function Sigmoid is: The activation function Softsign is: (Formula 12) Where x represents the sum of the input signal, including the weighted sum of the previous layer plus the bias term;

[0281] Step S3223: Determine the activation function used in the BP neural network model based on the evaluation result.

[0282] The Xavier method sets the initial weight of each neuron to a reasonable standard deviation or boundary under normal distribution or uniform distribution in combination with the model scale, thus avoiding the problem of model training failure caused by gradient explosion or gradient disappearance as the number of network layers increases.

[0283] The gradient is a function of the derivative of the activation function, so the Xavier method is sensitive to the activation function. Its assumptions determine that it is more suitable for activation functions that are symmetric about 0, such as Tanh, Sigmoid, Softsign, etc. For the model preset Relu, the activation function has general adaptability. Among them, Softsign has similar function characteristics to Tanh, but its inverse calculation takes longer.

[0284] Wherein, the activation function Relu is: f(x)=max(0,x). (Formula 13)

[0285] As mentioned above, in this embodiment, Relu, Sigmoid, and Tanh are used as activation functions to train the neural network model, and the model scores are compared and analyzed. The evaluation index values ​​include: NSE, Std, RMSE, MAE, MRE, and MRE_1.

[0286] Furthermore, the construction of the BP neural network model also includes:

[0287] Step S3231, based on the chlorophyll a concentration, establish a time series model; wherein the expression of the time series model is: Among them, xt represents the actual observation value at time point t in the time series; represents the autoregressive coefficient; ε t represents the error term at time point t; ε t-1 , ε t-2 , ε t-q represents the first q error terms in the time series;

[0288] It should be noted that the study found that the data records obtained from long-term continuous observations of Chl-a concentration (chlorophyll a concentration) in water bodies have obvious time series characteristics, and are suitable for fitting and prediction using time series models. That is, there is a correlation between the current fluctuations in Chl-a and past Chl-a data. This hidden dependence between data will help improve the model's fitting and prediction performance. At the same time, in scenarios where short-term Chl-a predictions are required, past Chl-a monitoring data are often known. Therefore, the Chl-a observations that lag behind the model output results in observation time are added to the model as data input to examine changes in model performance, thereby further optimizing the model.

[0289] The above time series model: Autoregressive Moving Average (ARMA) model, its formula can include two parts: (1) The left side of the equation consists of the independent variable x t and its lagged time series x t-p The autoregressive (AR) part of the equation; (2) The right side of the equation consists of the error term, also known as the disturbance term ε t and its lagged time series ε t-q The moving average (MA) part of the composition.

[0290] That is, when applied to chlorophyll a simulation, the AR part shows the correlation between Chl-a and its own past concentration levels, and the MA part shows the correlation between Chl-a and various external environmental disturbances, that is, the various environmental variable inputs in the aforementioned modeling and their past data.

[0291] Therefore, in order to determine the time series characteristics of chlorophyll a within the range of two characteristic points t1 and h2 in the Shending River Basin, in this embodiment, the idea of ​​Box-Jenkins ARMA time series modeling is used to perform pattern recognition on the chlorophyll a time series.

[0292] Step S3232, using the lag time as an input variable, calculating the chlorophyll a concentration data under the lag time based on the time series model, and using it as a time training set;

[0293] Step S3233, based on the time training set, the BP neural network model is trained, and evaluated through performance evaluation indicators to obtain evaluation results corresponding to different lag times, and the BP neural network model is adjusted according to the evaluation results corresponding to different lag times.

[0294] As mentioned above, the autocorrelation coefficient (ACF) and partial correlation coefficient (PACF) of Chl-a compared with its own lagged data were calculated, and the mean of the scenario schemes was taken for analysis.

[0295] We selected 1-day time-lag data as model input to further train the model and analyze its model score. Secondly, considering that the results obtained by using autocorrelation coefficient and partial correlation coefficient graph for pattern recognition in time series analysis are often an approximate range and still need to be further verified, we continued to add 4-day and 7-day time-lag data to the model to calculate and analyze its model score.

[0296] Furthermore, the algal bloom risk prediction results include:

[0297] If the chlorophyll a concentration C in the backwater zone of the target hydrological environment meets C≤10, the algal bloom risk prediction result is: the algal bloom degree is level 1; if the chlorophyll a concentration C in the backwater zone of the target hydrological environment meets 10<C≤15, the algal bloom risk prediction result is: the algal bloom degree is level 2; if the chlorophyll a concentration C in the backwater zone of the target hydrological environment meets 15<C≤50, the algal bloom risk prediction result is: the algal bloom degree is level 3; if the chlorophyll a concentration C in the backwater zone of the target hydrological environment meets 50<C≤100, the algal bloom risk prediction result is: the algal bloom degree is level 4; if the chlorophyll a concentration C in the backwater zone of the target hydrological environment meets C>100, the algal bloom risk prediction result is: the algal bloom degree is level 5.

[0298] In this embodiment, the Nash efficiency coefficient of the chlorophyll a prediction BP neural network model for the backwater area based on the mechanism model data training can reach 0.972 for the chlorophyll a prediction result 3 days in advance. According to the prediction results of the chlorophyll a concentration in the backwater area, the degree of water bloom in the backwater area can be evaluated with reference to the following table, providing a basis for the subsequent formulation of water bloom prevention and control measures.

[0299] Table 2. Bloom degree classification based on chlorophyll a prediction results

[0300] Bloom severity level Chlorophyll a concentration C (μg / L) Bloom severity level Chlorophyll a concentration C (μg / L) Level 1 C＜10 Level 4 50＜C＜100 Level 2 10＜C＜15 Level 5 C＞100 Level 3 15＜C＜50 / /

[0301] It can be understood that the device of this embodiment corresponds to the backwater area water bloom prediction method of the above embodiment, and the options in the above embodiment are also applicable to this embodiment, so they will not be repeated here.

[0302] In addition, reference Figure 5The present application also provides a backwater area algal bloom prediction device, including: an environmental data module 10, used to obtain the geographic hydrological data and environmental monitoring data of the target hydrological environment, and construct a three-dimensional hydrodynamic and water temperature model of the backwater area based on the geographic hydrological data and the environmental monitoring data; a eutrophication module 20, which is also used to construct a three-dimensional eutrophication model of the backwater area based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic and water temperature model of the backwater area; a neural network module 30, which is also used to construct a BP neural network model based on the three-dimensional eutrophication model of the backwater area, and train the BP neural network model; a risk prediction module 40, which is used to use the trained neural network model to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment.

[0303] The present application also provides a computer-readable storage medium for storing the computer program used in the above-mentioned computer device. For example, the computer-readable storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0304] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flow diagram, and the combination of boxes in the structure diagram and / or the flow diagram, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0305] In addition, the functional modules or units in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0306] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0307] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for predicting algal bloom in a backwater area, characterized in that: include: Acquire geographic hydrological data and environmental monitoring data of the target hydrological environment, and construct a three-dimensional hydrodynamic and water temperature model of the backwater area based on the geographic hydrological data and the environmental monitoring data; Based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area, a three-dimensional eutrophication model of the backwater area is constructed; Based on the three-dimensional eutrophication model of the backwater area, a BP neural network model is constructed, and the BP neural network model is trained; The trained neural network model is used to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment.

2. The method for predicting water bloom in backwater area according to claim 1, characterized in that: The step of acquiring geographic hydrological data and environmental monitoring data of the target hydrological environment and constructing a three-dimensional hydrodynamic and water temperature model of the backwater area according to the geographic hydrological data and the environmental monitoring data includes: Collecting the geographical hydrological data and the environmental monitoring data of the target hydrological environment, and constructing a one-dimensional hydrological model of the inflowing river, and calculating the riverbed roughness data in the target hydrological environment; Based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, a one-dimensional hydrodynamic model of the inflow river is constructed, and the inflow flow data and the upstream flow data are calculated; Based on the inflow data and the measured water level data of the reservoir, a two-dimensional hydrodynamic model of the reservoir is constructed, and the downstream water level data of the inflow river is calculated; The upstream flow data and the measured water level data of the reservoir are used to construct a three-dimensional hydrodynamic and water temperature model of the backwater area.

3. The method for predicting water bloom in backwater area according to claim 2, characterized in that: The collecting of the geographical hydrological data and the environmental monitoring data of the target hydrological environment, and constructing a one-dimensional hydrological model of the inflowing river, and calculating the riverbed roughness data in the target hydrological environment include: Collecting the geo-hydrological data and the environmental monitoring data in the target hydrological environment; wherein the geo-hydrological data include the surface type of the river entering the reservoir, the reservoir topographic map and the land use remote sensing data; the environmental monitoring data include rainfall data and evaporation data; Dividing the inflowing river into a plurality of catchment areas according to the geographical hydrological data and the environmental monitoring data; A one-dimensional hydrological model of the inflow river based on the NAM hydrological model is constructed, and each catchment area of ​​the inflow river is calibrated using the one-dimensional hydrological model of the inflow river to obtain the riverbed roughness data.

4. The method for predicting water bloom in backwater area according to claim 2, characterized in that: Based on the riverbed roughness data and the outflow flow data of the upstream reservoir of the inflow river in the target hydrological environment, a one-dimensional hydrodynamic model of the inflow river is constructed, and the inflow flow data and the upstream flow data are calculated, including: Obtaining a river system map and a topological relationship map of the river entering the reservoir, and performing generalization processing on the river channel of the river entering the reservoir and the reservoir; Based on the river channel of the inflow river and the reservoir after generalization, the external boundary conditions and internal boundary conditions of the model are determined; wherein the external boundary conditions include the upstream water boundary of the river channel and the downstream water level boundary of each river channel of the inflow river; the internal boundary conditions include the setting of hydraulic structures on the river channel; Based on the external boundary conditions and the internal boundary conditions, a MIKE 11HD hydrodynamic calculation model is used to couple the one-dimensional hydrological model of the inflow river to construct a one-dimensional hydrodynamic model of the inflow river, and the upstream flow data and the inflow flow data of the reservoir are calculated; The calculation expression of the MIKE 11HD hydrodynamic calculation model is: Among them, x and t represent the coordinates of the calculation point in space and time respectively; A represents the cross-sectional area of ​​the water flow; Q represents the flow rate; h represents the water level; q represents the lateral inflow rate; C represents the Xie Cai coefficient; R represents the hydraulic radius; α represents the momentum correction coefficient; and g represents the acceleration of gravity.

5. The method for predicting water bloom in backwater area according to claim 2, characterized in that: The step of constructing a two-dimensional hydrodynamic model of the reservoir based on the inflow flow data and the measured water level data of the reservoir, and calculating the downstream water level data of the inflow river, comprises: Using the reservoir topographic map, two-dimensionally grid the reservoir; Based on the reservoir topographic map processed by two-dimensional gridding, the land boundary conditions and water boundary conditions of the model are determined; wherein the land boundary conditions are obtained by the inflow flow data of the inflow river based on the one-dimensional hydrodynamic model of the inflow river; the water boundary conditions are determined by the inflow flow process of the reservoir at the beginning of the century, the measured water level and the reservoir capacity curve; A two-dimensional hydrodynamic model of the reservoir is constructed, characteristic points are selected in the reservoir, and the measured water level data are used to calibrate the constructed two-dimensional hydrodynamic model of the reservoir to obtain the downstream water level data of the river entering the reservoir.

6. The method for predicting water bloom in backwater area according to claim 2, characterized in that: The method of constructing a three-dimensional hydrodynamic and water temperature model of the backwater area by using the upstream flow data and the measured water level data of the reservoir includes: The backwater area of ​​the target hydrological environment is subjected to three-dimensional mesh processing; wherein the three-dimensional mesh processing includes horizontal triangular meshes and quadrilateral meshes, as well as vertical meshes; the middle river channel of the river entering the reservoir adopts the quadrilateral mesh; the area other than the middle river channel of the river channel adopts the triangular mesh; the vertical mesh adopts a sigma and Z-Level mixed layering form; Based on the backwater area processed by three-dimensional gridding, a three-dimensional hydrodynamic model of the backwater area is constructed; Determine the upstream boundary condition and the downstream boundary condition of the backwater zone; wherein the upstream boundary condition uses the upstream flow data calculated by the one-dimensional hydrodynamic model of the inflow river as an input condition; and uses the intersection boundary of the backwater zone and the reservoir as the downstream boundary condition; Using the three-dimensional hydrodynamic model of the backwater zone, based on the upstream boundary conditions and the downstream boundary conditions, the water depth of the backwater zone and the horizontal flow velocity of the backwater zone are calculated; Based on the three-dimensional hydrodynamic model of the backwater area and the water temperature model, construct the three-dimensional hydrodynamic and water temperature model of the backwater area; In the three-dimensional hydrodynamic water temperature model of the backwater zone, the transmission equation of the water temperature model is: in, represents the turbulent diffusion coefficient in the vertical direction; Represents the source term generated by heat exchange with the atmosphere; F t represents the horizontal diffusion term; T represents temperature; u, v, and w represent the velocity components of the water flow in the x, y, and z directions, respectively; The horizontal diffusion term F t The equation is: in, represents the turbulent diffusion coefficient in the horizontal direction; The boundary condition of the surface temperature is: When z = η, in, represents the turbulent diffusion coefficient in the vertical direction; z represents the position coordinate in the vertical direction; η represents the water surface position; Represents the vertical temperature gradient; Q n represents the net heat flux, which is a measure of the heat exchange between the water body and the atmosphere; ρ0 represents the density of water; c p represents the specific heat capacity of water; The bottom boundary condition is: When z = -d, The heat exchange at the water-air surface is calculated based on the following physical processes: Q n =q v +q c +βq sr,net +q l3,net ; Among them, Q n represents the net heat flux on the surface; c p represents the specific heat capacity of water; q v represents latent heat flux; q c represents the sensible heat flux; q sr,net represents the net shortwave radiation; q lr,net represents the net longwave radiation; Represents the amount of heat received or lost by a unit mass of water.

7. The method for predicting water bloom in backwater area according to claim 1, characterized in that: The three-dimensional eutrophication model of the backwater area is constructed based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area, including: Constructing the three-dimensional eutrophication model of the backwater area based on a preset eutrophication template; Determine the boundary conditions and model forces of the three-dimensional eutrophication model of the backwater area; the boundary conditions include the boundary conditions of the upstream river channel and the boundary conditions of the downstream reservoir junction; Based on the boundary conditions, the water temperature, water depth, horizontal flow velocity and wind speed obtained by the three-dimensional hydrodynamic water temperature model of the backwater area, and the light intensity obtained by the meteorological station are used as model forces and calibrated through the three-dimensional eutrophication model of the backwater area.

8. The method for predicting water bloom in backwater area according to claim 1, characterized in that: The method of constructing a BP neural network model based on the three-dimensional eutrophication model of the backwater area and training the BP neural network model includes: Based on the downstream water level, upstream flow, temperature, light, upstream water quality and downstream water quality, different scenario plans are constructed, and based on the different scenario plans, the chlorophyll a concentration is calculated through the three-dimensional eutrophication model of the backwater area to form a training set; Construct BP neural network model; The BP neural network model is trained by taking the parameters in the training set as the model input and the chlorophyll a concentration to be fitted as the model output to obtain the trained BP neural network model; The training of the BP neural network model further includes: Using principal component analysis to reduce the dimension of parameters in the training set to obtain model input items; The model input items are standardized to obtain the model input items after the standardization process; wherein the expression of the standardization process is: Wherein, x represents the original model input; x` represents the model input after standardization; μ represents the data mean; σ represents the data standard deviation; Based on the cross-validation strategy, the model input items after the standardization process are divided into k folds, wherein the k-1 fold is used as the training data set, and the 1 fold samples are combined with the NSE, RMSE, and MAE evaluation indicators as the validation data set; Execute the following training process k times: train the BP neural network model based on the training data set, and verify it through the verification data set to obtain a verification result; Taking the k-times verification results as the training scores of the BP neural network model, and adjusting the BP neural network model according to the training scores; The construction of the BP neural network model also includes: The number of hidden layers of the BP neural network model is set to a single hidden layer; wherein the number of neurons in the single hidden layer is set to 1 / 3 to 2 times the number of neurons in the input layer; Use the NSE indicator to evaluate and obtain the evaluation results; If the evaluation result is that the NSE value reaches the preset threshold, it is determined that the performance meets expectations; If the evaluation result is that the NSE value does not reach the preset threshold, it is determined that the performance does not meet expectations, and the number of hidden layers is set to double hidden layers; wherein, in the double hidden layers, the number of neurons in the first layer is set to 1 / 3 to 2 times the number of neurons in the input layer; the number of neurons in the second layer is set to 2 / 3 times the number of neurons in the input layer; The construction of the BP neural network model also includes: Based on the Xavier method, the BP neural network model is weight initialized; wherein the expression of the weight initialization is: Among them, ω represents the weight matrix; U represents uniform distribution; n in Represents the number of neurons in the input layer of the neural network; n out represents the number of neurons in the output layer of the neural network; μ is determined by the type of activation function selected; the activation function is any one of Tanh, Sigmoid and Softsign; If the activation function is Tanh, μ=6, if the activation function is Sigmoid, μ=96, if the activation function is Relu, μ=12; Using different activation functions to train the BP neural network model, and evaluating it through performance evaluation indicators to obtain evaluation results of each activation function; Wherein, the activation function Tanh is: The activation function Sigmoid is: The activation function Softsign is: Where x represents the sum of the input signal, including the weighted sum of the previous layer plus the bias term; Based on the evaluation result, determining the activation function used in the BP neural network model; The construction of the BP neural network model also includes: Based on the chlorophyll a concentration, a time series model is established; wherein the expression of the time series model is: Among them, xt represents the actual observation value at time point t in the time series; represents the autoregressive coefficient; ε t represents the error term at time point t; ε t-1 , ε t-2 , ε t-q represents the first q error terms in the time series; Taking the lag time as an input variable, the chlorophyll a concentration data under the lag time is calculated based on the time series model and used as a time training set; Based on the time training set, the BP neural network model is trained and evaluated through performance evaluation indicators to obtain evaluation results corresponding to different lag times, and the BP neural network model is adjusted according to the evaluation results corresponding to different lag times.

9. The method for predicting water bloom in backwater area according to claim 1, characterized in that: The algal bloom risk prediction results include: If the chlorophyll a concentration C composite C of the backwater area of ​​the target hydrological environment is ≤10, the algal bloom risk prediction result is: the algal bloom degree level is level 1; If the chlorophyll a concentration C in the backwater area of ​​the target hydrological environment meets 10<C≤15, the algal bloom risk prediction result is: the algal bloom degree level is level 2; If the chlorophyll a concentration C in the backwater area of ​​the target hydrological environment meets 15<C≤50, the algal bloom risk prediction result is: the algal bloom degree level is level 3; If the chlorophyll a concentration C of the backwater area of ​​the target hydrological environment meets 50<C≤100, the algal bloom risk prediction result is: the algal bloom degree level is level 4; If the chlorophyll a concentration C of the backwater area of ​​the target hydrological environment meets C>100, the algal bloom risk prediction result is: the algal bloom degree level is level 5.

10. A device for predicting water bloom in a backwater area, characterized in that: include: An environmental data module, used to obtain geographical hydrological data and environmental monitoring data of the target hydrological environment, and to construct a three-dimensional hydrodynamic and water temperature model of the backwater area based on the geographical hydrological data and the environmental monitoring data; The eutrophication module is also used to construct a three-dimensional eutrophication model of the backwater area based on the water temperature, water depth, horizontal flow velocity and wind speed obtained from the three-dimensional hydrodynamic water temperature model of the backwater area; The neural network module is also used to construct a BP neural network model based on the three-dimensional eutrophication model of the backwater area, and train the BP neural network model; The risk prediction module is used to obtain the algal bloom risk prediction result of the backwater area of ​​the target hydrological environment by using the trained neural network model.

Images