A method for establishing regional nutrient benchmarks for lakes and reservoirs with high sediment content

Abstract

The invention discloses a method for formulating a regional nutrient salt benchmark for lakes and reservoirs with high sediment content, comprising the following steps: S1, collecting monitoring data related to nutrient salt, total suspended particulate matter and target indicators of regional lakes and reservoirs, and constructing a regional lake and reservoir nutrient salt benchmark formulation database for establishing a quantitative response relationship between nutrient salt and target indicators; S2, selecting specific target indicator compliance values ​​and regression quantiles according to the risk preference or expert experience of decision makers; S3, using a quantile regression neural network model to establish a nonlinear pressure-response relationship between the target indicator and total nitrogen, total phosphorus and total suspended particulate matter, and using a Bayesian optimization method to optimize hyperparameters; S4, for a specific lake and reservoir, according to the average total suspended particle concentration of the lake and reservoir, providing a joint nutrient salt benchmark of total nitrogen and total phosphorus under specific target indicator compliance values ​​and regression quantile conditions.

Description

Technical Field

[0001] The invention belongs to the technical field of water quality monitoring and water body management, and specifically relates to a method for formulating a regional nutrient salt benchmark for lakes and reservoirs with high sediment content. Background Art

[0002] There are currently many methods for establishing nutrient benchmarks for lakes and reservoirs, usually statistical analysis, model inference, and pressure-response model methods. Among them, the pressure-response model method infers the nutrient benchmark by constructing the relationship between the pressure variable and the response variable. This method has strong applicability when dealing with lakes and reservoirs that are severely affected by human activities. It can construct a suitable mathematical model to describe the relationship between the pressure variable and the response variable based on the actual conditions such as the characteristics of the study area. Therefore, the pressure-response model method is of great significance in formulating nutrient benchmarks for the special water environment of lakes and reservoirs with high sediment content.

[0003] In the pressure-response model method, commonly used mathematical models include linear regression models and Bayesian hierarchical regression models. The linear regression model is mainly used to reveal the linear relationship between the pressure variable and the response variable, while the Bayesian hierarchical regression model can explore possible nonlinear relationships. However, in lakes and reservoirs with high sediment content, due to the particularity of the water environment, the relationship between the pressure variable and the response variable often presents complex nonlinear characteristics, which limits the application of the above two classic models. In addition, the traditional pressure-response model method usually calculates the baseline values ​​of total nitrogen and total phosphorus separately, but ignores the joint effect between different pressure variables. At the same time, previous methods mostly use the mean as the evaluation standard for water quality compliance, and this processing method cannot effectively avoid the impact of extreme values ​​on the results. Therefore, it is urgent to adopt a method that can handle nonlinear nutrient data and use quantiles as water quality evaluation standards to construct a joint nutrient benchmark suitable for lakes and reservoirs with high sediment content. Summary of the invention

[0004] To solve the above problems, the present invention proposes a method for formulating regional nutrient salt benchmarks for lakes and reservoirs with high sediment content. This method formulates quantile evaluation standards for the special water environment of lakes and reservoirs with high sediment content, and can jointly establish nutrient salt benchmarks, which can provide strong technical support for water quality evaluation and ecological management of lakes and reservoirs with high sediment content.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content includes the following steps:

[0007] S1. Collect monitoring data related to nutrients, total suspended particulate matter and target indicators of regional lakes and reservoirs, and build a regional lake and reservoir nutrient benchmark database to establish a quantitative response relationship between nutrients and target indicators;

[0008] S2. Select specific target indicator values ​​and regression quantiles based on the decision maker’s risk preference or expert experience;

[0009] S3. The quantile regression neural network model was used to establish the nonlinear pressure-response relationship between the target index and total nitrogen, total phosphorus and total suspended particulate matter, and the Bayesian optimization method was used for hyperparameter optimization;

[0010] The specific process of constructing the quantile regression neural network model in step S3 is:

[0011] S31. Divide the data set of the regional lake nutrient benchmark database into a training set, a validation set and a test set;

[0012] S32. Design the neural network architecture, including input layer, hidden layer and output layer, according to the actual situation and data scale;

[0013] S33. Determine the hyperparameters that need to be optimized and their value ranges, including learning rate, batch size, number of network layers, and number of neurons; adopt the Bayesian optimization method, build a proxy model to approximate the true objective function, and use the acquisition function to efficiently search in the hyperparameter space to find the best hyperparameter combination; during the optimization process, continuously use the validation set to evaluate the performance of the proxy model, and update the proxy model based on the evaluation results;

[0014] S34, using the training set to train the quantile regression neural network model, during the training process, using the validation set data to input the quantile regression neural network model, adjusting the hyperparameters of the quantile regression neural network model according to the performance on the validation set, and recording the hyperparameter combination at this time when the performance on the validation set reaches the best;

[0015] S35, applying the best hyperparameter combination obtained in step S34, using the test set to evaluate the performance of the trained quantile regression neural network model, and analyzing the influence of the total nitrogen, total phosphorus and total suspended particulate matter environmental parameters on the target indicators based on the prediction results of the quantile regression neural network model;

[0016] S4. For specific lakes and reservoirs, based on the average total suspended particle concentration of the lakes and reservoirs, provide specific target indicator compliance values ​​and a combined nutrient benchmark of total nitrogen and total phosphorus under regression quantile conditions;

[0017] The specific process of step S4 is:

[0018] S41, calculating the average total suspended particle concentration of lakes and reservoirs with high sediment content according to the target indicator standard value and regression quantile finally determined in step S2;

[0019] S42. Using the trained quantile regression neural network model, the total nitrogen and total phosphorus values ​​of a specific lake reservoir are found under given target index attainment values ​​and regression quantiles through an iterative search method;

[0020] S43. Generate contour maps of total nitrogen and total phosphorus concentration values ​​to obtain a joint nutrient benchmark of total nitrogen and total phosphorus, which is used to visually display the areas where target indicators reach target values ​​at different total nitrogen and total phosphorus levels in specific high sediment content lakes and reservoirs.

[0021] Preferably, the target indicator includes chlorophyll a concentration; in step S2, the target indicator reaching the standard value is a chlorophyll a concentration equal to 0.010 mg / L, and the regression quantile is 90%.

[0022] Preferably, the specific process of step S1 is:

[0023] S11. Collect the concentration data of nutrients in lakes and reservoirs, including total nitrogen and total phosphorus;

[0024] S12. Collect ecological response data related to nutrient concentration and use the ecological response data as target indicators;

[0025] S13. Collect total suspended particulate matter concentration in lakes and reservoirs and other relevant water quality data;

[0026] S14. Conduct quality control on monitoring data, including eliminating abnormal data, eliminating duplicate data, and resampling data;

[0027] S15. Process the missing values ​​of the monitoring data and use the Kalman filter method to fill in the missing values. By fusing the measured values ​​and state estimation, the accuracy of the state estimation is improved, the missing values ​​are filled in, and a regional lake and reservoir nutrient benchmark database is constructed.

[0028] Preferably, the elimination of abnormal data in step S14 is to eliminate non-numeric data, wherein non-numeric data includes characters and null values; the elimination of duplicate data is to eliminate duplicate data in the time series; and the data resampling is to unify the data frequency to the daily frequency by taking the median method.

[0029] Preferably, the specific process of step S32 is:

[0030] S321, determine the input layer, the input characteristics include total nitrogen, total phosphorus and total suspended particulate matter;

[0031] S322, designing hidden layers, the number of hidden layers is obtained through cross validation, and a hyperbolic tangent function is used as an activation function to process nonlinear relationships;

[0032] S323. Determine the output layer, the output features are chlorophyll a concentration and its quantile estimation, and set the quantile loss function in the output layer. Adjust the weights and bias of the quantile regression neural network model through the gradient descent optimization algorithm to minimize the error between the predicted value and the actual value of the quantile regression neural network model at a specific quantile.

[0033] Preferably, the specific steps of optimizing the hyperparameters using the Bayesian optimization method in step S33 are:

[0034] S331, define the parameter space and determine the hyperparameters that need to be optimized, the hyperparameters including the number of layers of the neural network, the number of neurons in each layer, and the learning rate;

[0035] S332, using Gaussian process as a proxy model for the true approximate objective function;

[0036] S333, selecting the next potential hyperparameter combination for evaluation by obtaining a function;

[0037] S334. Train a quantile regression neural network on the selected hyperparameter combination and evaluate the performance on the validation set.

[0038] S335, updating the proxy model using the new evaluation result;

[0039] S336. Repeat steps S333-S335 until a stopping condition is met, where the stopping condition is an iteration upper limit or performance convergence.

[0040] After adopting the above technical scheme, the present invention has the following beneficial effects: the regional nutrient salt benchmark formulation method of the present invention is oriented to lakes and reservoirs with high sediment content, and outputs the target index and its quantile estimation value by constructing a quantile regression neural network model, and uses the quantile target value as the water quality compliance evaluation standard, thereby avoiding the extreme value problem of using the mean target value as the water quality compliance evaluation standard. At the same time, as a neural network model, the quantile regression neural network performs better than the traditional machine learning model when processing complex nonlinear high-dimensional water quality data. When inferring the nutrient salt benchmark for a specific lake and reservoir with high sediment content, a set of total nitrogen and total phosphorus concentrations that meet specific conditions are generated, and a joint nutrient salt benchmark for a specific lake and reservoir can be established, thereby fully considering the joint relationship between total nitrogen and total phosphorus. Therefore, the present invention not only provides a more scientific water quality compliance evaluation standard, but also formulates a more targeted and accurate joint nutrient salt benchmark, providing effective support for lake and reservoir water environment management, and has significant advantages and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the present invention;

[0042] Figure 2 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. 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.

[0044] like Figure 1 and Figure 2 As shown, a method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content includes the following steps:

[0045] S1. Collect monitoring data related to nutrients, total suspended particulate matter and target indicators of regional lakes and reservoirs, and build a regional lake and reservoir nutrient benchmark database to establish a quantitative response relationship between nutrients and target indicators; the target indicators include chlorophyll a concentration;

[0046] The specific process of step S1 is:

[0047] S11. Collect the concentration data of nutrients in lakes and reservoirs, including total nitrogen and total phosphorus;

[0048] S12. Collect ecological response data related to nutrient concentration and use the ecological response data as target indicators;

[0049] S13. Collect total suspended particulate matter concentration in lakes and reservoirs and other relevant water quality data;

[0050] S14. Conduct quality control on monitoring data, including eliminating abnormal data, eliminating duplicate data, and resampling data;

[0051] In step S14, the method of eliminating abnormal data is to eliminate non-numeric data, wherein non-numeric data includes characters and null values; the method of eliminating duplicate data is to eliminate duplicate data in the time series; the method of data resampling is to unify the data frequency into daily frequency by taking the median method;

[0052] S15. Process the missing values ​​of the monitoring data, use the Kalman filter method to fill the missing values, improve the accuracy of the state estimation by fusing the measured values ​​and the state estimation, complete the missing value filling, and construct a regional lake and reservoir nutrient salt benchmark database;

[0053] S2. Select specific target indicator values ​​and regression quantiles based on the decision maker’s risk preference or expert experience;

[0054] In step S2, the target indicator standard value is chlorophyll a concentration equal to 0.010 mg / L, and the regression quantile is 90%;

[0055] S3. The quantile regression neural network model (QRNN model) was used to establish the nonlinear pressure-response relationship between the target index and total nitrogen, total phosphorus and total suspended particulate matter, and the Bayesian optimization method was used for hyperparameter optimization;

[0056] The specific process of constructing the quantile regression neural network model in step S3 is:

[0057] S31. Divide the data set of the regional lake nutrient benchmark database into a training set, a validation set and a test set;

[0058] S32. Design the neural network architecture, including input layer, hidden layer and output layer, according to the actual situation and data scale;

[0059] The specific process of step S32 is:

[0060] S321, determine the input layer, the input characteristics include total nitrogen, total phosphorus and total suspended particulate matter;

[0061] S322, designing hidden layers, the number of hidden layers is obtained through cross validation, and a hyperbolic tangent function is used as an activation function to process nonlinear relationships;

[0062] S323, determining an output layer, wherein the output feature is chlorophyll a concentration and its quantile estimation, and setting a quantile loss function in the output layer, and adjusting the weight and bias of the quantile regression neural network model by a gradient descent optimization algorithm to minimize the error between the predicted value and the actual value of the quantile regression neural network model at a specific quantile;

[0063] S33. Determine the hyperparameters that need to be optimized and their value ranges, including learning rate, batch size, number of network layers, and number of neurons; adopt the Bayesian optimization method, build a proxy model to approximate the true objective function, and use the acquisition function to efficiently search in the hyperparameter space to find the best hyperparameter combination; during the optimization process, continuously use the validation set to evaluate the performance of the proxy model, and update the proxy model based on the evaluation results;

[0064] The specific steps of optimizing the hyperparameters using the Bayesian optimization method in step S33 are:

[0065] S331, define the parameter space and determine the hyperparameters that need to be optimized, the hyperparameters including the number of layers of the neural network, the number of neurons in each layer, and the learning rate;

[0066] S332, using Gaussian process as a proxy model for the true approximate objective function;

[0067] S333, selecting the next potential hyperparameter combination for evaluation by obtaining a function;

[0068] S334. Train a quantile regression neural network on the selected hyperparameter combination and evaluate the performance on the validation set.

[0069] S335, updating the proxy model using the new evaluation result;

[0070] S336, repeat steps S333-S335 until a stop condition is met, where the stop condition is an iteration upper limit or performance convergence;

[0071] S34, using the training set to train the quantile regression neural network model, during the training process, using the validation set data to input the quantile regression neural network model, adjusting the hyperparameters of the quantile regression neural network model according to the performance on the validation set, and recording the hyperparameter combination at this time when the performance on the validation set reaches the best;

[0072] S35, applying the best hyperparameter combination obtained in step S34, using the test set to evaluate the performance of the trained quantile regression neural network model, and analyzing the influence of the total nitrogen, total phosphorus and total suspended particulate matter environmental parameters on the target indicators based on the prediction results of the quantile regression neural network model;

[0073] S4. For specific lakes and reservoirs, the joint nutrient salt benchmark of total nitrogen (TN) and total phosphorus (TP) under specific target indicator achievement values ​​and regression quantile conditions is given according to the average total suspended particle concentration of the lakes and reservoirs;

[0074] The specific process of step S4 is:

[0075] S41, calculating the average total suspended particle concentration of lakes and reservoirs with high sediment content according to the target indicator standard value and regression quantile finally determined in step S2;

[0076] S42. Using the trained quantile regression neural network model, the total nitrogen and total phosphorus values ​​of a specific lake reservoir are found under given target index attainment values ​​and regression quantiles through an iterative search method;

[0077] S43. Generate contour maps of total nitrogen and total phosphorus concentration values ​​to obtain a joint nutrient benchmark of total nitrogen and total phosphorus, which is used to visually display the areas where target indicators reach target values ​​at different total nitrogen and total phosphorus levels in specific high sediment content lakes and reservoirs.

[0078] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content, characterized in that: The following steps are involved: S1. Collect monitoring data related to nutrients, total suspended particulate matter and target indicators of regional lakes and reservoirs, and build a regional lake and reservoir nutrient benchmark database to establish a quantitative response relationship between nutrients and target indicators; S2. Select specific target indicator values ​​and regression quantiles based on the decision maker’s risk preference or expert experience; S3. The quantile regression neural network model was used to establish the nonlinear pressure-response relationship between the target index and total nitrogen, total phosphorus and total suspended particulate matter, and the Bayesian optimization method was used for hyperparameter optimization; The specific process of constructing the quantile regression neural network model in step S3 is: S31. Divide the data set of the regional lake nutrient benchmark database into a training set, a validation set and a test set; S32. Design the neural network architecture, including input layer, hidden layer and output layer, according to the actual situation and data scale; S33, determining the hyperparameters to be optimized and the value ranges of the hyperparameters, the hyperparameters including the learning rate, the batch size, the number of network layers, and the number of neurons; The Bayesian optimization method is used to build a proxy model to approximate the true objective function, and the acquisition function is used to efficiently search in the hyperparameter space to find the best hyperparameter combination. During the optimization process, the performance of the proxy model is continuously evaluated using the validation set, and the proxy model is updated based on the evaluation results. S34, using the training set to train the quantile regression neural network model, during the training process, using the validation set data to input the quantile regression neural network model, adjusting the hyperparameters of the quantile regression neural network model according to the performance on the validation set, and recording the hyperparameter combination at this time when the performance on the validation set reaches the best; S35, applying the best hyperparameter combination obtained in step S34, using the test set to evaluate the performance of the trained quantile regression neural network model, and analyzing the influence of the total nitrogen, total phosphorus and total suspended particulate matter environmental parameters on the target indicators based on the prediction results of the quantile regression neural network model; S4. For specific lakes and reservoirs, based on the average total suspended particle concentration of the lakes and reservoirs, provide specific target indicator compliance values ​​and a combined nutrient benchmark of total nitrogen and total phosphorus under regression quantile conditions; The specific process of step S4 is: S41, calculating the average total suspended particle concentration of lakes and reservoirs with high sediment content according to the target indicator standard value and regression quantile finally determined in step S2; S42. Using the trained quantile regression neural network model, the total nitrogen and total phosphorus values ​​of a specific lake reservoir are found under given target index attainment values ​​and regression quantiles through an iterative search method; S43. Generate contour maps of total nitrogen and total phosphorus concentration values ​​to obtain a joint nutrient benchmark of total nitrogen and total phosphorus, which is used to visually display the areas where target indicators reach target values ​​at different total nitrogen and total phosphorus levels in specific high sediment content lakes and reservoirs.

2. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content as claimed in claim 1, characterized in that: The target indicator includes chlorophyll a concentration; in step S2, the target indicator reaching the standard value is chlorophyll a concentration equal to 0.010 mg / L, and the regression quantile is 90%.

3. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content as claimed in claim 1, characterized in that: The specific process of step S1 is: S11. Collect the concentration data of nutrients in lakes and reservoirs, including total nitrogen and total phosphorus; S12. Collect ecological response data related to nutrient concentration and use the ecological response data as target indicators; S13. Collect total suspended particulate matter concentration in lakes and reservoirs and other relevant water quality data; S14. Conduct quality control on monitoring data, including eliminating abnormal data, eliminating duplicate data, and resampling data; S15. Process the missing values ​​of the monitoring data and use the Kalman filter method to fill in the missing values. By fusing the measured values ​​and state estimation, the accuracy of the state estimation is improved, the missing values ​​are filled in, and a regional lake and reservoir nutrient benchmark database is constructed.

4. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content as claimed in claim 3, characterized in that: The elimination of abnormal data in step S14 is to eliminate non-numeric data, wherein non-numeric data includes characters and null values; the elimination of duplicate data is to eliminate duplicate data in the time series; and the data resampling is to unify the data frequency to the daily frequency by taking the median method.

5. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content as claimed in claim 1, characterized in that: The specific process of step S32 is: S321, determine the input layer, the input characteristics include total nitrogen, total phosphorus and total suspended particulate matter; S322, designing hidden layers, the number of hidden layers is obtained through cross validation, and a hyperbolic tangent function is used as an activation function to process nonlinear relationships; S323. Determine the output layer, the output features are chlorophyll a concentration and its quantile estimation, and set the quantile loss function in the output layer. Adjust the weights and bias of the quantile regression neural network model through the gradient descent optimization algorithm to minimize the error between the predicted value and the actual value of the quantile regression neural network model at a specific quantile.

6. A method for establishing a regional nutrient salt benchmark for lakes and reservoirs with high sediment content as claimed in claim 1, characterized in that: The specific steps of optimizing the hyperparameters using the Bayesian optimization method in step S33 are: S331, define the parameter space and determine the hyperparameters that need to be optimized, the hyperparameters including the number of layers of the neural network, the number of neurons in each layer, and the learning rate; S332, using Gaussian process as a proxy model for the true approximate objective function; S333, selecting the next potential hyperparameter combination for evaluation by obtaining a function; S334. Train a quantile regression neural network on the selected hyperparameter combination and evaluate the performance on the validation set. S335, updating the proxy model using the new evaluation result; S336. Repeat steps S333-S335 until a stopping condition is met, where the stopping condition is an iteration upper limit or performance convergence.

Images