Methods and devices for predicting the dynamics of cyanobacterial blooms in lakes
By combining a physical state description model and a dual-ensemble Kalman filter method, and using a neural network model to correct errors, the problem of low prediction accuracy for cyanobacterial blooms was solved, and high-precision prediction of lake cyanobacterial blooms was achieved.
Patent Information
- Application Number
- CN202210630132.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-06
- Publication Date
- 2025-12-02
- Estimated Expiration
- 2042-06-06
AI Technical Summary
Existing technologies for predicting cyanobacterial blooms have low accuracy, large errors in physical process-driven modeling, and a lack of interpretability in data-driven modeling, making it impossible to effectively correct random parameters.
By combining a physical state description model with a dual-ensemble Kalman filter method, and by correcting the randomness parameters and using a neural network model to correct the error, a dynamic prediction framework for cyanobacterial blooms in lakes is constructed.
It has achieved accurate prediction of cyanobacterial blooms, reduced prediction errors, and improved prediction accuracy.
Smart Images
Figure CN114819407B_ABST
Abstract
Description
Technical Field
[0001] This document relates to the field of computer technology, and in particular to a method and device for dynamic prediction of cyanobacterial blooms in lakes. Background Technology
[0002] In recent years, due to the rapid development of modern chemical and agricultural production, the eutrophication of global aquatic systems has become increasingly severe. The explosive proliferation of cyanobacteria in lakes, forming algal blooms, seriously affects human production, daily life, and economic development, and undermines the sustainable development of aquatic ecosystems. Algal bloom control has become a pressing global challenge, and in-depth research and effective prediction of cyanobacterial bloom processes have significant practical and socio-economic implications.
[0003] Currently, the modeling methods for predicting cyanobacterial blooms mainly include two types: physical process-driven modeling and data-driven modeling. However, both suffer from low prediction accuracy and inaccurate prediction results. Furthermore, they cannot effectively correct for the uncertain random parameters present in the physical models.
[0004] Physical process-driven modeling starts from the growth mechanism of cyanobacteria and predicts cyanobacterial growth by establishing a physical model between environmental variables and random parameters. Because the physical processes of cyanobacterial blooms are complex and greatly influenced by environmental factors, establishing an accurate model describing the cyanobacterial growth process is very difficult, leading to large prediction errors. Data-driven modeling collects large amounts of observational data and models the relationship between input and output data, often using neural network models and regression models. Data-driven models neglect the mechanistic analysis of cyanobacterial growth, require large amounts of observational data as training samples, have poor interpretability of prediction results, and have limitations in handling time series problems. Summary of the Invention
[0005] The purpose of this invention is to provide a method and apparatus for dynamic prediction of cyanobacterial blooms in lakes, aiming to solve the above-mentioned problems in the prior art.
[0006] This invention provides a method for dynamic prediction of cyanobacterial blooms in lakes, comprising:
[0007] A physical state description model is established based on the basic process of phytoplankton growth, and the parameters in the physical state description model are divided into deterministic parameters and random parameters.
[0008] The parameters and the state of the physical state description model are updated by using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model to obtain prediction results.
[0009] The residual value between the predicted result and the actual observation result is calculated. A neural network model is established based on the residual value. The error of the physical state description model is corrected by the neural network model to predict the cyanobacterial bloom in the lake.
[0010] This invention provides a dynamic prediction device for cyanobacterial blooms in lakes, comprising:
[0011] A partitioning module is established to build a physical state description model based on the basic process of phytoplankton growth, and to divide the parameters in the physical state description model into deterministic parameters and random parameters.
[0012] The prediction update module is used to update the parameters and the state of the physical state description model using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model to obtain prediction results.
[0013] The error prediction module is used to calculate the residual value between the prediction result and the actual observation result, establish a neural network model based on the residual value, and correct the error of the physical state description model through the neural network model to predict the cyanobacterial bloom in the lake.
[0014] This invention employs dual-ensemble Kalman filtering and neural network computer technology, combined with a physics-driven model and a data-driven model, to construct a dynamic prediction framework for lake cyanobacterial blooms. This achieves accurate prediction of lake cyanobacterial blooms, effectively reducing prediction errors and improving prediction accuracy. The dual-ensemble Kalman filtering method simultaneously updates the system state and random parameter states in the model, correcting the random parameters while predicting cyanobacterial blooms. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of the dynamic prediction method for cyanobacterial blooms in lakes according to an embodiment of the present invention.
[0017] Figure 2 This is a schematic diagram illustrating the prediction of cyanobacterial blooms using the dual-ensemble Kalman filtering method according to an embodiment of the present invention.
[0018] Figure 3This is a schematic diagram illustrating the correction results of the bi-ensemble Kalman filtering method for random parameters in the model according to an embodiment of the present invention;
[0019] Figure 4 This is a schematic diagram of the residual between the predicted cyanobacterial bloom results and the actual observation results using the dual-ensemble Kalman filtering method according to an embodiment of the present invention;
[0020] Figure 5 This is a schematic diagram of the network structure of the nonlinear autoregressive model, i.e., the NARX neural network, according to an embodiment of the present invention.
[0021] Figure 6 This is a schematic diagram of the prediction results of cyanobacterial blooms by combining a nonlinear autoregressive model, namely the NARX neural network, according to an embodiment of the present invention.
[0022] Figure 7 This is a schematic diagram of the correction result of the nonlinear autoregressive model, namely the NARX neural network, on the residuals in an embodiment of the present invention.
[0023] Figure 8 This is a schematic diagram illustrating the training performance of the nonlinear autoregressive model, i.e., the NARX neural network, in different rounds according to an embodiment of the present invention.
[0024] Figure 9 This is a schematic diagram of a dynamic prediction device for lake cyanobacterial blooms according to an embodiment of the present invention. Detailed Implementation
[0025] To more accurately predict the growth trend of cyanobacteria in lakes, this invention proposes a dynamic prediction method for cyanobacterial blooms in lakes. This method combines physical process-driven modeling and data-driven modeling, utilizing dual-ensemble Kalman filtering and neural network computer technology to dynamically predict cyanobacterial blooms in lakes.
[0026] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.
[0027] Method Implementation Examples
[0028] According to embodiments of the present invention, a method for dynamic prediction of cyanobacterial blooms in lakes is provided. Figure 1 This is a flowchart of the dynamic prediction method for lake cyanobacterial blooms according to an embodiment of the present invention, such as... Figure 1As shown, the lake cyanobacterial bloom dynamic prediction method according to an embodiment of the present invention specifically includes:
[0029] Step 101: Establish a physical state description model based on the basic process of phytoplankton growth, and divide the parameters in the physical state description model into deterministic parameters and random parameters;
[0030] In step 101, based on the three main physical processes of phytoplankton growth, a physical state description model of cyanobacteria, i.e., a dynamic growth model, is established based on Formula 1:
[0031]
[0032] Where C represents the concentration of cyanobacteria in the lake, t represents time, and μ net This indicates the net growth rate of cyanobacteria in the lake. This represents the overall natural growth rate. ω represents the total natural loss rate, and ω0 represents the scouring loss rate.
[0033] Specifically:
[0034] Assume the total natural growth rate of cyanobacteria is μ compared to the maximum growth rate under ideal conditions. m The temperature model is φ Temp The nutrient model in the lake is φ Nutrient The light intensity model is φ Light Relatedly, according to Formula 2, the total natural growth rate can be described as follows:
[0035]
[0036] According to formula 3, the temperature model φ Temp The temperature coefficient θ of the growth rate μ , Ambient temperature T, Standard reference temperature T ref Described as:
[0037]
[0038] Among them, the temperature coefficient θ of the growth rate μ The parameter is random.
[0039] According to Formula 4, the nutrient model φ Nutrient Described as:
[0040]
[0041] Where N and P represent the nitrogen and phosphorus content in the lake, respectively, and K N K P These represent the half-saturation coefficients of nitrogen and phosphorus in the lake, respectively.
[0042] The light intensity model φ is determined according to formula 5. Light Modeling a saturated model using the Monod formula:
[0043]
[0044] Where I represents light intensity, K I The half-saturation coefficient of light;
[0045] Based on the sum of losses, the total natural loss rate is described using Formula 6.
[0046]
[0047] Where, σ m θ represents the maximum natural loss rate under ideal conditions. σ The temperature coefficient representing the natural loss rate is a random parameter;
[0048] According to Equation 7, the erosion loss rate ω0 of cyanobacteria is described based on the volume of erosion Q and the total volume of the lake V:
[0049]
[0050] Based on Formula 8, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is determined as follows:
[0051]
[0052] Step 102: Update the parameters and the state of the physical state description model using the dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model to obtain the prediction results;
[0053] Step 102 specifically includes:
[0054] Let Θ = [P, N, T, Q, I] represent the input variables, and θ = [θ μ ,θ σ The physical state description model of cyanobacteria, represented by ], is a dynamic growth model, where the parameters are uncertain.
[0055]
[0056] In the dual-set Kalman filtering method, a new equation is introduced to describe the dynamic changes of the randomness parameter:
[0057]
[0058] Where k represents the discrete time step. This represents a small perturbation applied to the random parameters. The mean is 0 and the variance is σ. T Gaussian random variables;
[0059] The concentration of cyanobacteria in the lake will be obtained by direct measurement using instruments. Considered to have observation error The measured value of cyanobacteria concentration is recorded as:
[0060]
[0061] Among them, observation error With a mean of 0 and a variance of σ d Gaussian random variables;
[0062] At time k = 0, based on the initial value of the cyanobacteria concentration state... prior distribution and parameter initial values prior distribution Monte Carlo sampling is performed on the model state and model parameters respectively, and N samples are taken to obtain a particle set;
[0063] At times k = 1, 2, ...:
[0064] Estimate the uncertainty parameters and system state concentrations using formulas 12 and 13:
[0065]
[0066]
[0067] The observed values of cyanobacterial concentration were calculated using Formula 14:
[0068]
[0069] Calculate the Kalman gain of the uncertain parameters using Equations 15 and 16. And update the randomness parameter:
[0070]
[0071]
[0072] According to Formula 17, the pseudo-observation value is updated based on the cyanobacteria concentration:
[0073]
[0074] Calculate the Kalman gain of the concentration using Equations 18 and 19, and update the concentration state:
[0075]
[0076]
[0077] Step 103: Calculate the residual value between the predicted result and the actual observation result, establish a neural network model based on the residual value, and use the neural network model to correct the error of the physical state description model to predict the cyanobacterial bloom in the lake.
[0078] Step 103 specifically includes:
[0079] The difference between the predicted value of cyanobacteria concentration obtained from the prediction and the actual observation value of cyanobacteria concentration obtained from the instrument measurement is used as the residual value, and a neural network model driven by the residual value is established.
[0080] A neural network model is used to predict the residual values, and the residual prediction results are combined with the prediction results to achieve the prediction of cyanobacterial blooms.
[0081] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0082] like Figure 1 As shown, the specific steps are as follows:
[0083] S1. Establish a dynamic growth model of cyanobacteria based on the four main physical processes of phytoplankton growth:
[0084]
[0085] Where C represents the concentration of cyanobacteria in the lake, t represents time, and μ m θ represents the maximum natural growth rate. μ The temperature coefficient representing the growth rate, where T represents the ambient temperature. ref The standard reference temperature is represented by N, the nitrogen content of the lake is represented by P, and the phosphorus content of the lake is represented by K. N K represents the half-saturation coefficient of nitrogen in a lake. P K represents the half-saturation coefficient of nitrogen in the lake, I represents the light intensity, and K represents the light intensity. I σ represents the half-saturation coefficient of light. m θ represents the maximum natural loss rate under ideal conditions. σ The temperature coefficient represents the natural loss rate, Q represents the volume lost, and V represents the total volume of the lake.
[0086] S2. Let Θ = [P, N, T, Q, I] represent the input variables, and θ = [θ μ ,θ σ The growth model of cyanobacteria can be represented as follows: [[Indicated by] an uncertain random parameter]
[0087]
[0088] S3. Introduce a new equation to describe the dynamic changes of random parameters:
[0089]
[0090] in, The mean is 0 and the variance is σ. T σ is a Gaussian random variable. T The value is: 1×10 -4.5 θ μ The initial value of θ is set to 1.16. σ The initial value is set to 1.05.
[0091] S4. The values of the deterministic parameters in the model are shown in Table 1:
[0092] Table 1
[0093]
[0094]
[0095] S5. Continuous direct measurements were conducted for 1461 days on environmental parameters such as cyanobacteria concentration C, ambient temperature T, nitrogen content N, phosphorus content P, light intensity I, and loss volume Q in the lake, and the measurement data were obtained.
[0096] S6. The concentration of cyanobacteria in the lake obtained by direct measurement using instruments. Considered to have observation error The measured value of cyanobacteria concentration is recorded as:
[0097]
[0098] in, The value is: 1×10 -1 .
[0099] S7. When applying the dual-set Kalman filter method to perform Monte Carlo sampling on the model state and model parameters, the number of samples N in the particle set is 20.
[0100] S8. At times k = 1, 2, ...:
[0101] Estimate uncertainty parameters and system state concentrations:
[0102]
[0103]
[0104] Observed values of cyanobacteria concentration:
[0105]
[0106] Calculate the Kalman gain for uncertain parameters And update the uncertain parameters:
[0107]
[0108]
[0109] Update false observations using cyanobacteria concentration.
[0110]
[0111] Calculate the Kalman gain for the concentration and update the concentration state:
[0112]
[0113]
[0114] The relationship between the predictions and actual observed values of cyanobacterial blooms using the dual-ensemble Kalman filter method is as follows: Figure 2 As shown, the correction results for the random parameters are as follows: Figure 3 As shown.
[0115] S9. The residual between the predicted cyanobacteria concentration obtained by the dual-ensemble Kalman filtering method and the true cyanobacteria concentration obtained by instrument measurement is expressed as γ. RCM (t).
[0116] The predicted residual results are as follows Figure 4 As shown.
[0117] S10. A nonlinear autoregressive model (NARX) is selected as the neural network model. The linear autoregressive model (NARX) can be represented by the input variable Θ(t) and the residual γ. RCM (t) is represented as:
[0118] γ RCM (t)=f[Θ(tD Θ ),…,Θ(t-1),γ RCM (tD r ),…,γ RCM (t-1)]
[0119] Where f[·] represents the nonlinear feedforward residual neural network equation, D Θ and D γ It is a hyperparameter in the equation. D Θ This represents the maximum order of input delay, with a value of 7. (D) γ This indicates the maximum order of output delay, with a value of 7.
[0120] S11. The selected nonlinear autoregressive model (NARX) neural network structure is as follows: Figure 5 As shown, our network consists of a hidden layer with k neurons, whose input is the systematic error γ. RCM And the system input variable Θ. Where, This represents the weight and bias of each neuron in the hidden layer, (ω) k b2) represents the weight and bias of each neuron in the output layer.
[0121] In this case, the hidden layer k=50, and 70% of the residual data is selected as samples. 70% of the samples are used as the training set, 15% as the validation set, and 15% as the test set.
[0122] The prediction results of cyanobacterial blooms after correction by a nonlinear autoregressive model (NARX) neural network are as follows: Figure 6 As shown, the corrected residual results are as follows: Figure 7 As shown, Figure 8 As shown.
[0123] In summary, this invention innovatively proposes a novel dynamic prediction method for cyanobacterial blooms, combining physical process-driven modeling with data-driven modeling, and integrating traditional data assimilation techniques with modern neural network methods. This effectively reduces prediction errors and improves the accuracy of cyanobacterial bloom prediction. The dual-ensemble Kalman filter method simultaneously updates the system state and random parameter states in the model, correcting the random parameters while predicting cyanobacterial blooms.
[0124] Device Examples
[0125] According to an embodiment of the present invention, a dynamic prediction device for cyanobacterial blooms in lakes is provided. Figure 9 This is a schematic diagram of a dynamic prediction device for lake cyanobacterial blooms according to an embodiment of the present invention, as shown below. Figure 9 As shown, the lake cyanobacterial bloom dynamic prediction device according to an embodiment of the present invention specifically includes:
[0126] A partitioning module 90 is used to establish a physical state description model based on the basic process of phytoplankton growth, and to partition the parameters in the physical state description model into deterministic parameters and stochastic parameters; the partitioning module 90 is specifically used for:
[0127] Based on the three main physical processes of phytoplankton growth, a physical state description model of cyanobacteria, i.e., a dynamic growth model, is established based on Equation 1:
[0128]
[0129] Where C represents the concentration of cyanobacteria in the lake, t represents time, and μ net This indicates the net growth rate of cyanobacteria in the lake. This represents the overall natural growth rate. ω represents the total natural loss rate, and ω0 represents the scouring loss rate.
[0130] The partitioning module 90 is specifically used for:
[0131] Assume the total natural growth rate of cyanobacteria is μ compared to the maximum growth rate under ideal conditions. m The temperature model is φ Temp The nutrient model in the lake is φ Nutrient The light intensity model is φ Light Relatedly, according to Formula 2, the total natural growth rate can be described as follows:
[0132]
[0133] According to formula 3, the temperature model φ Temp The temperature coefficient θ of the growth rate μ , Ambient temperature T, Standard reference temperature T ref Described as:
[0134]
[0135] Among them, the temperature coefficient θ of the growth rate u The parameter is random.
[0136] According to Formula 4, the nutrient model φ Nutrient Described as:
[0137]
[0138] Where N and P represent the nitrogen and phosphorus content in the lake, respectively, and K N K P These represent the half-saturation coefficients of nitrogen and phosphorus in the lake, respectively.
[0139] The light intensity model φ is determined according to formula 5. Light Modeling a saturated model using the Monod formula:
[0140]
[0141] Where I represents light intensity, K I The half-saturation coefficient of light;
[0142] Based on the sum of losses, the total natural loss rate is described using Formula 6.
[0143]
[0144] Where, σ mθ represents the maximum natural loss rate under ideal conditions. σ The temperature coefficient representing the natural loss rate is a random parameter;
[0145] According to Equation 7, the erosion loss rate ω0 of cyanobacteria is described based on the volume of erosion Q and the total volume of the lake V:
[0146]
[0147] Based on Formula 8, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is determined as follows:
[0148]
[0149] The prediction update module 92 is used to update the parameters and the state of the physical state description model using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth through the physical state description model to obtain prediction results.
[0150] The update prediction module 92 is specifically used for:
[0151] Let Θ = [P, N, T, Q, I] represent the input variables, and θ = [θ μ ,θ σ The physical state description model of cyanobacteria, represented by ], is a dynamic growth model, where the parameters are uncertain.
[0152]
[0153] In the dual-set Kalman filtering method, a new equation is introduced to describe the dynamic changes of the randomness parameter:
[0154]
[0155] Where k represents the discrete time step. This represents a small perturbation applied to the random parameters. The mean is 0 and the variance is σ. T Gaussian random variables;
[0156] The concentration of cyanobacteria in the lake will be obtained by direct measurement using instruments. Considered to have observation error The measured value of cyanobacteria concentration is recorded as:
[0157]
[0158] Among them, observation error With a mean of 0 and a variance of σ d Gaussian random variables;
[0159] At time k = 0, based on the initial value of the cyanobacteria concentration state... prior distribution and parameter initial values prior distribution Monte Carlo sampling is performed on the model state and model parameters respectively, and N samples are taken to obtain a particle set;
[0160] At times k = 1, 2, ...:
[0161] Estimate the uncertainty parameters and system state concentrations using formulas 12 and 13:
[0162]
[0163]
[0164] The observed values of cyanobacterial concentration were calculated using Formula 14:
[0165]
[0166] Calculate the Kalman gain of the uncertain parameters using Equations 15 and 16. And update the randomness parameter:
[0167]
[0168]
[0169] According to Formula 17, the pseudo-observation value is updated based on the cyanobacteria concentration:
[0170]
[0171] Calculate the Kalman gain of the concentration using Equations 18 and 19, and update the concentration state:
[0172]
[0173]
[0174] Error prediction module 94 is used to calculate the residual value between the prediction result and the actual observation result, establish a neural network model based on the residual value, correct the error of the physical state description model through the neural network model, and predict the cyanobacterial bloom in the lake.
[0175] The error prediction module 94 is specifically used for:
[0176] The difference between the predicted value of cyanobacteria concentration obtained from the prediction and the actual observation value of cyanobacteria concentration obtained from the instrument measurement is used as the residual value, and a neural network model driven by the residual value is established.
[0177] A neural network model is used to predict the residual values, and the residual prediction results are combined with the prediction results to achieve the prediction of cyanobacterial blooms.
[0178] The embodiments of the present invention are device embodiments corresponding to the above method embodiments. The specific operation of each module can be understood with reference to the description of the method embodiments, and will not be repeated here.
[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for dynamic prediction of cyanobacterial blooms in lakes, characterized in that, include: A physical state description model is established based on the basic process of phytoplankton growth, and the parameters in the physical state description model are divided into deterministic parameters and random parameters. The parameters and the state of the physical state description model are updated by using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model to obtain prediction results. The process involves calculating the residual between the predicted result and the actual observation result, establishing a neural network model based on the residual, and correcting the error of the physical state description model using the neural network model to predict cyanobacterial blooms in lakes. Specifically, this includes: using the difference between the predicted cyanobacterial concentration and the actual observed cyanobacterial concentration obtained from instrument measurements as the residual, establishing a neural network model driven by the residual; using the neural network model to predict the residual; and combining the residual prediction result with the predicted result to achieve prediction of cyanobacterial blooms. Specifically, the physical state description model established based on the basic growth process of phytoplankton includes: Based on the three main physical processes of phytoplankton growth, a physical state description model of cyanobacteria, i.e., a dynamic growth model, is established based on Equation 1: Formula 1: in, Indicates the concentration of cyanobacteria in the lake. Indicates time, This indicates the net growth rate of cyanobacteria in the lake. This represents the overall natural growth rate. This represents the total natural loss rate. Indicates the scouring loss rate; The physical state description model of cyanobacteria, i.e., the dynamic growth model, established based on the three main physical processes of phytoplankton growth, specifically includes: Assume the total natural growth rate of cyanobacteria is equal to the maximum growth rate under ideal conditions. Temperature model is Nutrient models in lakes are The light intensity model is Relatedly, according to Formula 2, the total natural growth rate can be described as follows: Formula 2: According to Formula 3, the temperature model Temperature coefficient of growth rate Environmental observation temperature Standard reference temperature Described as: Formula 3: Among them, the temperature coefficient of growth rate The parameter is random. According to Formula 4, the nutrient model Described as: Formula 4: in, , These represent the nitrogen and phosphorus content in the lake, respectively. , These represent the half-saturation coefficients of nitrogen and phosphorus in the lake, respectively. The light intensity model is determined according to formula 5. Modeling a saturated model using the Monod formula: Formula 5: in, Indicates light intensity. The half-saturation coefficient of light; Based on the sum of losses, the total natural loss rate is described using Formula 6. : Formula 6: in, This represents the maximum natural loss rate under ideal conditions. The temperature coefficient representing the natural loss rate is a random parameter; According to Formula 7, based on the volume of loss and the total volume of the lake Describes the scouring loss rate of cyanobacteria : Formula 7: Based on Formula 8, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is determined as follows: Formula 8.
2. The method according to claim 1, characterized in that, The parameters and the state of the physical state description model are updated using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model. The specific prediction results include: Will Represented as input variables, Represented as uncertain random parameters, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is denoted as: Formula 9; In the dual-set Kalman filtering method, a new equation is introduced to describe the dynamic changes of the randomness parameter: Formula 10; in, Represents the discrete time step. This represents a small perturbation applied to the random parameters. The mean is 0 and the variance is Gaussian random variables; The concentration of cyanobacteria in the lake will be obtained by direct measurement using instruments. Considered to have observation error The measured value of cyanobacteria concentration is recorded as: Formula 11; Among them, observation error With a mean of 0 and a variance of Gaussian random variables; At any moment At that time, based on the initial value of cyanobacteria concentration. prior distribution and parameter initial values prior distribution Monte Carlo sampling was performed on the model state and model parameters respectively, and the results were obtained. Each sample yields a particle set; At any moment hour: Estimate the uncertainty parameters and system state concentrations using formulas 12 and 13: Formula 12; Formula 13; The observed values of cyanobacterial concentration were calculated using Formula 14: Formula 14; Calculate the Kalman gain of the uncertain parameters using Equations 15 and 16. And update the randomness parameter: Formula 15; Formula 16; According to Formula 17, the pseudo-observation value is updated based on the cyanobacteria concentration: Formula 17; Calculate the Kalman gain of the concentration using Equations 18 and 19, and update the concentration state: Formula 18; Official 19.
3. A dynamic prediction device for cyanobacterial blooms in lakes, characterized in that, include: A partitioning module is established to build a physical state description model based on the basic process of phytoplankton growth, and to divide the parameters in the physical state description model into deterministic parameters and random parameters. The prediction update module is used to update the parameters and the state of the physical state description model using a dual-ensemble Kalman filter method, thereby correcting the randomness parameters and predicting cyanobacteria growth using the physical state description model to obtain prediction results. An error prediction module is used to calculate the residual value between the prediction result and the actual observation result, establish a neural network model based on the residual value, correct the error of the physical state description model through the neural network model, and predict the cyanobacterial bloom in the lake. Specifically, the partitioning module is used for: Based on the three main physical processes of phytoplankton growth, a physical state description model of cyanobacteria, i.e., a dynamic growth model, is established based on Equation 1: Formula 1: in, Indicates the concentration of cyanobacteria in the lake. Indicates time, This indicates the net growth rate of cyanobacteria in the lake. This represents the overall natural growth rate. This represents the total natural loss rate. Indicates the scouring loss rate; The partitioning module is specifically used for: Assume the total natural growth rate of cyanobacteria is equal to the maximum growth rate under ideal conditions. Temperature model is Nutrient models in lakes are The light intensity model is Relatedly, according to Formula 2, the total natural growth rate can be described as follows: Formula 2: According to Formula 3, the temperature model Temperature coefficient of growth rate Environmental observation temperature Standard reference temperature Described as: Formula 3: Among them, the temperature coefficient of growth rate The parameter is random. According to Formula 4, the nutrient model Described as: Formula 4: in, , These represent the nitrogen and phosphorus content in the lake, respectively. , These represent the half-saturation coefficients of nitrogen and phosphorus in the lake, respectively. The light intensity model is determined according to formula 5. Modeling a saturated model using the Monod formula: Formula 5: in, Indicates light intensity. The half-saturation coefficient of light; Based on the sum of losses, the total natural loss rate is described using Formula 6. : Formula 6: in, This represents the maximum natural loss rate under ideal conditions. The temperature coefficient representing the natural loss rate is a random parameter; According to Formula 7, based on the volume of loss and the total volume of the lake Describes the scouring loss rate of cyanobacteria : Formula 7: Based on Formula 8, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is determined as follows: Formula 8: The error prediction module is specifically used for: The difference between the predicted value of cyanobacteria concentration obtained from the prediction and the actual observation value of cyanobacteria concentration obtained from the instrument measurement is used as the residual value, and a neural network model driven by the residual value is established. A neural network model is used to predict the residual values, and the residual prediction results are combined with the prediction results to achieve the prediction of cyanobacterial blooms.
4. The apparatus according to claim 3, characterized in that, The update prediction module is specifically used for: Will Represented as input variables, Represented as uncertain random parameters, the physical state description model of cyanobacteria, i.e., the dynamic growth model, is denoted as: Formula 9; In the dual-set Kalman filtering method, a new equation is introduced to describe the dynamic changes of the randomness parameter: Formula 10; in, Represents the discrete time step. This represents a small perturbation applied to the random parameters. The mean is 0 and the variance is Gaussian random variables; The concentration of cyanobacteria in the lake will be obtained by direct measurement using instruments. Considered to have observation error The measured value of cyanobacteria concentration is recorded as: Formula 11; Among them, observation error With a mean of 0 and a variance of Gaussian random variables; At any moment At that time, based on the initial value of cyanobacteria concentration. prior distribution and parameter initial values prior distribution Monte Carlo sampling was performed on the model state and model parameters respectively, and the results were obtained. Each sample yields a particle set; At any moment hour: Estimate the uncertainty parameters and system state concentrations using formulas 12 and 13: Formula 12; Formula 13; The observed values of cyanobacterial concentration were calculated using Formula 14: Formula 14; Calculate the Kalman gain of the uncertain parameters using Equations 15 and 16. And update the randomness parameter: Formula 15; Formula 16; According to Formula 17, the pseudo-observation value is updated based on the cyanobacteria concentration: Formula 17; Calculate the Kalman gain of the concentration using Equations 18 and 19, and update the concentration state: Formula 18; Official 19.