Method, medium and system for determining depth-fixed stratification parameters of a multi-channel water sampler

Abstract

The present invention provides a method, medium and system for determining the depth-fixed stratification parameters of a multi-channel water sampler, belonging to the technical field of the use of multi-channel water samplers, including: using multiple groups of water sampling project data to establish a basic water body change model, and training a multi-layer perceptron neural network model with augmented data to obtain a water body change model. Then, based on the water body distribution data, a genetic algorithm is used to calculate the optimal sampling parameters. Finally, a regression neural network model is established, and the environmental parameters of the current water area to be measured are input to obtain the depth-fixed stratification parameters of the multi-channel water sampler. This method utilizes historical sampling data and realizes the refined modeling of the target water body and the intelligent optimization of sampling parameters through technologies such as data augmentation, neural networks, and genetic algorithms, providing effective support for multi-channel water sampling monitoring. It solves the technical problem that in the prior art, the determination of the depth-fixed stratification parameters of water samplers often relies on the experience of operators, and the representativeness and reliability of sampling data are insufficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of the use of multi-channel water samplers. Specifically, it relates to a method, medium, and system for determining depth and layer parameters of a multi-channel water sampler. Background Art

[0002] Water quality monitoring is an important part of environmental protection and resource management and is crucial for maintaining ecological balance and ensuring drinking water safety. Traditional water quality monitoring methods mainly include two links: on-site sampling and laboratory analysis. In the on-site sampling stage, sampling personnel need to manually determine the sampling point location, sampling depth, and sampling time according to the monitoring plan and take samples on-site; in the laboratory analysis stage, the collected water samples are transported back to the laboratory for testing and analysis of physical, chemical, and biological indicators. This sampling-transportation-analysis method has some key problems:

[0003] First, the sampling process relies on manual operation, which has certain subjectivity and randomness. The selection of sampling points, depth, and time depends on the experience judgment of sampling personnel, making it difficult to ensure representativeness of the entire water body and possibly leading to deviations in sampling results. At the same time, the sampling efficiency of manual operation is low, making it difficult to meet the needs of dynamic water quality monitoring.

[0004] Second, the processes of water sample collection, transportation, and preservation are prone to introducing pollution and measurement errors. Contact pollution during the sampling process, temperature changes during transportation, and different preservation conditions will all affect the properties of water samples, thus affecting the reliability of the final analysis results.

[0005] Third, laboratory analysis usually takes a cycle of 2 - 7 days, unable to provide timely water quality information and making it difficult to respond quickly to the water body situation. In some emergency situations, such as chemical leaks and industrial wastewater discharges, it is impossible to detect and take corresponding measures in time, easily causing environmental pollution accidents.

[0006] To solve the above problems, automated water quality sampling equipment such as multi-channel water samplers has been developed in recent years, such as an intelligent multi-channel water sample collection device disclosed in the utility model with the publication number CN202321995114.5. Such equipment can automatically collect water samples and is equipped with on-line sensors, enabling on-site real-time monitoring of water quality indicators. Compared with traditional sampling methods, automated water sampling equipment has the following advantages:

[0007] 1) Sampling points can be programmably set, with a wide coverage range, capable of monitoring the overall condition of the water body;

[0008] 2) The sampling time interval can be flexibly adjusted, suitable for the needs of dynamic water quality monitoring;

[0009] 3) On-site instant analysis, without the need for water sample transportation, capable of quickly detecting water quality anomalies;

[0010] 4) The data can be remotely transmitted, which is conducive to information management and early warning.

[0011] However, when using this device, the determination of the depth stratification parameters of the water sampler often relies on the experience of the operator, and the representativeness and reliability of the sampling data are insufficient. Summary of the Invention

[0012] In view of this, the present invention provides a method, medium and system for determining the depth stratification parameters of a multi-channel water sampler, which can solve the technical problem that the prior art often relies on the experience of the operator to determine the depth stratification parameters of the water sampler, and the representativeness and reliability of the sampling data are insufficient.

[0013] The present invention is implemented as follows:

[0014] The first aspect of the present invention provides a method for determining the depth stratification parameters of a multi-channel water sampler, which includes the following steps:

[0015] S10. Collect multiple groups of historical water sampling project data, where each group of water sampling project data is specifically the project data of the historical water sampling projects of a successful multi-channel water sampler in different regions of the same water body;

[0016] S20. For the environmental parameters and sampling analysis data in each group of project data, establish and fit the basic change model of the water body;

[0017] S30. Based on the basic change model, expand each group of project data to obtain corresponding multiple groups of expanded project data;

[0018] S40. Use each group of water sampling project data and its corresponding expanded project data as the training set, use the multiple groups of water sampling project data as the verification set, and train a multi-layer perceptron neural network model to obtain the water body change model;

[0019] S50. For each group of water sampling project data in the expanded project data, input the environmental parameters and sampling analysis parameters therein into the water body change model to obtain the water body distribution data;

[0020] S60. For each group of water sampling project data in the expanded project data, use the sampling parameters as the initial population, use the maximization of the sampling coverage rate and the minimization of the sampling error as the objective function, and use the obtained water body distribution to calculate the optimal sampling parameters by using the genetic algorithm;

[0021] S70. For all the augmented project data, establish a training data set, where the input of the training is the environmental parameters and the output of the training is the sampling parameters, and train a regression neural network to obtain the sampling parameter determination model;

[0022] S80. Obtain the environmental parameters of the current water area to be measured, input them into the sampling parameter determination model, and obtain the sampling parameters, which are used as the depth stratification parameters of the multi-channel water sampler for the current water area to be measured.

[0023] Among them, the step S10 specifically includes:

[0024] Step 101. Obtain the historical project data of the same water body in different areas where sampling has been successfully carried out using a multi-channel water sampler, including environmental parameters, sampling parameters, and sampling analysis data; among them, the environmental parameters include water depth, water quality, water temperature, and flow velocity, etc., the sampling parameters include sampling time interval, sampling depth, and sampling volume, and the sampling analysis data includes physical indicators, biological indicators, and chemical indicators.

[0025] Step 102. Organize and classify and store the collected historical project data to provide basic data for establishing a water body change model in the future.

[0026] Among them, the specific implementation manner of the step S20 is as follows:

[0027] Step 201. Select a suitable mathematical modeling method, such as regression analysis, time series analysis, etc., and establish a basic water body change model according to the environmental parameters and sampling analysis data collected in step S10.

[0028] Step 202. Perform parameter fitting and verification on the established water body change model to ensure that the model can accurately describe the internal relationship between the water body environmental parameters and the sampling analysis indicators.

[0029] Among them, the specific implementation manner of the step S30 is as follows:

[0030] Step 301. Select a suitable statistical method, such as Monte Carlo random simulation, Bootstrap sampling, etc., and expand each group of historical project data collected in step S10 to generate more diverse simulation data.

[0031] Step 302. Use the multiple groups of project data obtained after expansion as the input data for the subsequent steps.

[0032] Among them, the specific implementation manner of the step S40 is as follows:

[0033] Step 401. Use the original project data in step S10 as the training set and the data obtained by expansion in step S30 as the validation set to train a multi-layer perceptron (MLP) neural network model.

[0034] Step 402. By repeatedly adjusting the structure and training parameters of the neural network, such as the number of layers, the number of nodes, the activation function, etc., make the model achieve the optimal performance on the validation set.

[0035] Step 403: The trained neural network model is the water body change model, which can be used to predict the sampling analysis results under given environmental parameter conditions.

[0036] Among them, the specific implementation of step S50 is as follows:

[0037] Step 501: Input the environmental parameters and sampling analysis parameters of each group of project data obtained by expanding step S30 into the water body change model trained in step S40.

[0038] Step 502: Calculate the corresponding water body distribution data according to the model, which reflects the spatial distribution of each index of the water body under given environmental parameter conditions.

[0039] Among them, the specific implementation of step S60 is as follows:

[0040] Step 601: Based on the water body distribution data obtained in step S50, use the genetic algorithm to optimize the sampling parameters of the project data expanded in step S30.

[0041] Step 602: The objective function is to maximize the sampling coverage rate and minimize the sampling error. Through the iterative optimization of the genetic algorithm, the optimal sampling parameter scheme corresponding to each group of project data is obtained.

[0042] Among them, the specific implementation of step S70 is as follows:

[0043] Step 701: Use the optimal sampling parameters optimized in step S60 as the output and the environmental parameters obtained by expanding step S30 as the input to train a regression neural network model.

[0044] Step 702: Through training, learn the implicit law between the environmental parameters and the sampling parameters, and obtain the sampling parameter determination model.

[0045] Among them, the specific implementation of step S80 is as follows:

[0046] Step 801: Obtain the environmental parameters of the water area to be measured, including water depth, water quality, water temperature, flow velocity, etc.

[0047] Step 802: Input the environmental parameters obtained in step 801 into the sampling parameter determination model trained in step S70.

[0048] Step 803: The model outputs the best sampling parameters of the water area, including sampling time interval, sampling depth, sampling volume, etc., as the depth-stratified parameters of the multi-channel water sampler.

[0049] Optionally, the best sampling parameters output in step 803 meet the requirements of a sampling coverage rate of not less than 90% and a sampling error of not more than 5%.

[0050] Optionally, for the specific structure of the water body change model, a hybrid model form can be adopted, combining the advantages of statistical models and deep learning models. Specifically, the model can be divided into the following parts:

[0051] 1. Environmental parameter preprocessing module

[0052] Standardize or normalize the original environmental parameter data to eliminate the influence of dimensions.

[0053] Differencing or sliding window processing of time series data can be considered to extract time features.

[0054] 2. Statistical model module

[0055] Adopt classical statistical models, such as linear regression, polynomial regression, time series model (ARIMA), etc., to fit the basic relationship between environmental parameters and sampling analysis data.

[0056] This module can capture the linear and non-linear trends in the data and serve as the basis for the deep learning model.

[0057] 3. Deep learning module

[0058] Introduce a deep neural network to capture the complex non-linear relationship between environmental parameters and sampling analysis data.

[0059] The network structure can adopt a multi-layer perceptron (MLP) or a long short-term memory network (LSTM), etc.

[0060] The input is the preprocessed environmental parameter data, and the output is the corresponding sampling analysis index.

[0061] 4. Residual connection

[0062] Take the output of the statistical model as the residual term of the deep learning model to fuse the linear and non-linear parts.

[0063] This design can explicitly retain the basic trends captured by the statistical model, while using the deep model to learn the residuals to improve the overall fitting ability.

[0064] 5. Loss function

[0065] Adopt the mean square error (MSE) or the mean absolute error (MAE), etc. as the loss function to measure the deviation between the model prediction and the true value.

[0066] Introducing a regularization term can also be considered to avoid overfitting.

[0067] 6. Model training and optimization

[0068] Use optimization algorithms such as stochastic gradient descent to train the model parameters on the training set.

[0069] A validation set can be used for model selection to prevent overfitting.

[0070] For time series data, a rolling prediction method can be adopted for training and testing.

[0071] Through the design of the above hybrid model structure, the advantages of statistical models and deep learning models can be effectively integrated to capture the linear, non-linear, and time series features in water body change data, thereby improving the overall prediction accuracy. In practical applications, the model structure can be further adjusted and optimized according to the characteristics of the data and the requirements of the task.

[0072] Among them, the project data includes environmental parameters, sampling parameters, and sampling analysis data.

[0073] Furthermore, the environmental parameters include water depth, water quality, water temperature, and flow velocity.

[0074] Furthermore, the sampling parameters are provided by experts according to the environmental parameters, including sampling time interval, sampling depth, and sampling volume.

[0075] Furthermore, the sampling analysis data are the physical, biological, and chemical indicators of the sampled water.

[0076] Among them, the water body change model includes the following modules: environmental parameter preprocessing module, statistical model module, deep learning module, and residual connection module.

[0077] The environmental parameter preprocessing module is used to standardize or normalize the original environmental parameter data to eliminate the influence of dimensions.

[0078] Furthermore, the statistical model module uses a mathematical statistical model to fit the basic relationship between environmental parameters and sampling analysis data.

[0079] Furthermore, the deep learning module uses a multi-layer perceptron or a long short-term memory network.

[0080] The second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored, and when the program instructions run, they are used to execute the above method for determining the depth stratification parameters of a multi-channel water sampler.

[0081] The third aspect of the present invention provides a system for determining the depth stratification parameters of a multi-channel water sampler, which includes the above computer-readable storage medium.

[0082] Compared with the prior art, the beneficial effects of a method, medium, and system for determining depth-stratified parameters of a multi-channel water sampler provided by the present invention are as follows:

[0083] 1. Improve the representativeness and reliability of sampling data: This method makes full use of historical water quality monitoring data to establish a mathematical relationship model between water environment parameters and sampling analysis indicators. This model can accurately describe the variation law of the water body, providing a basis for the optimization of subsequent sampling parameters. Compared with the selection of sampling parameters based on manual experience, the automatic optimization based on the model can better cover the overall distribution characteristics of the water body and improve the representativeness of the sampling results. At the same time, the optimized sampling parameters can also reduce errors in the sampling process, thereby improving data reliability.

[0084] 2. Enhance the immediacy and response ability of water quality monitoring: This method uses technologies such as machine learning to quickly determine the optimal sampling parameter scheme. Once the real-time environmental parameters of the water area to be measured are obtained, the key parameters such as the sampling depth, volume, and time interval required for this water area can be quickly output through the pre-trained parameter determination model. Compared with the traditional method of manual experience selection, this method significantly shortens the time from obtaining environmental information to completing the sampling plan, providing strong support for the rapid response to water quality abnormal accidents.

[0085] 3. Improve the intelligent level of water quality monitoring: This method integrates advanced technical means such as mathematical modeling and machine learning to achieve automatic optimization of sampling parameters. Compared with the traditional method relying on manual experience, this method has stronger adaptability and automation capabilities. On the one hand, by establishing a water body variation law model, it can adapt to the characteristics of different water areas and dynamically adjust sampling parameters; on the other hand, based on the optimization process of genetic algorithms, it can independently explore the best sampling plan and reduce manual intervention. This not only improves the monitoring efficiency but also lays a technical foundation for realizing the intelligence of water quality monitoring.

[0086] Therefore, the solution of the present invention solves the technical problem that the prior art often relies on the experience of operators to determine the depth-stratified parameters of the water sampler, and the representativeness and reliability of sampling data are insufficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0088] Figure 1 It is a flowchart of the method provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0089] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0090] As Figure 1 shown, it is a flowchart of a method for determining depth-stratified parameters of a multi-channel water sampler provided by the present invention. This method includes the following steps:

[0091] S10. Collect multiple sets of historical water sampling project data, where each set of water sampling project data is specifically the project data of historical water sampling projects of a successful multi-channel water sampler in different regions of the same water body;

[0092] S20. For the environmental parameters and sampling analysis data in each set of project data, establish and fit a basic change model of the water body;

[0093] S30. Based on the basic change model, expand each set of project data to obtain corresponding multiple sets of expanded project data;

[0094] S40. Use each set of water sampling project data and its corresponding expanded project data as the training set, use multiple sets of water sampling project data as the validation set, and train a multi-layer perceptron neural network model to obtain a water body change model;

[0095] S50. For each set of water sampling project data in the expanded project data, input the environmental parameters and sampling analysis parameters therein into the water body change model to obtain water body distribution data;

[0096] S60. For each set of water sampling project data in the expanded project data, use the sampling parameters as the initial population, use the maximization of sampling coverage rate and the minimization of sampling error as the objective function, and use the obtained water body distribution to calculate the optimal sampling parameters by using a genetic algorithm;

[0097] S70. For all the augmented project data, establish a training data set, where the input for training is the environmental parameters and the output for training is the sampling parameters, and train a regression neural network to obtain a sampling parameter determination model;

[0098] S80. Obtain the environmental parameters of the currently measured water area, input them into the sampling parameter determination model, and obtain the sampling parameters as the depth-stratified parameters of the multi-channel water sampler for the currently measured water area.

[0099] The following will describe in detail the specific implementation manners of the above steps:

[0100] Step S10: Collect multiple sets of historical water sampling project data

[0101] In this step, it is first necessary to collect historical project data of different regions of the same water body that have been successfully sampled using a multi-channel water sampler. These data include three categories: environmental parameters, sampling parameters, and sampling analysis data.

[0102] Environmental parameters include water depth, water quality, water temperature, and flow velocity, etc. Sampling parameters are provided by experts based on environmental parameters, mainly including sampling time interval, sampling depth, and sampling volume. Sampling analysis data is the test results of physical, biological, and chemical indicators of the collected water samples.

[0103] The purpose of collecting these historical project data is to establish a basic change model for this water body and provide a basis for subsequent parameter optimization. By analyzing the historical data of different regions of the same water body, the internal relationship between the environmental parameters and sampling analysis indicators of this water body can be obtained.

[0104] Step S20: Establish and fit the basic change model of the water body

[0105] Based on the environmental parameters and sampling analysis data collected in step S10, use mathematical modeling methods to establish and fit the basic change model of this water body. Here, methods such as regression analysis and time series analysis can be used to establish a mathematical relationship model between environmental parameters and sampling analysis indicators.

[0106] By fitting the model, the change law of the environmental parameters of this water body over time and space can be described, providing a basis for subsequent parameter optimization. The accuracy of the model directly affects the accuracy of the final optimized parameters. Therefore, it is necessary to select appropriate modeling methods and fully verify the model.

[0107] Step S30: Augment the project data

[0108] Based on the basic change model established in step S20, augment each group of project data collected in step S10. Through augmentation, more sample data can be obtained, providing a sufficient training set for subsequent neural network training.

[0109] The augmentation methods can be statistical methods such as Monte Carlo random simulation and Bootstrap sampling. Through these methods, more diverse simulated data can be generated while retaining the characteristics of the original data. The augmented data set will become the input for the subsequent steps.

[0110] Step S40: Train the multi-layer perceptron neural network model

[0111] Next, use the original project data in step S10 and the data augmented in step S30 to train a multi-layer perceptron (MLP) neural network model. The purpose of this model is to learn the complex non-linear mapping relationship between environmental parameters and sampling analysis data.

[0112] Specifically, the original project data is used as the training set to input into the neural network model for learning, and the augmented data is used as the validation set for evaluating the model performance. By repeatedly adjusting the structure and training parameters of the neural network, such as the number of layers, the number of nodes, the activation function, etc., until the model achieves the optimal performance on the validation set.

[0113] The trained neural network model is the water body change model, which can be used to predict the sampling analysis results under given environmental parameter conditions. This provides an important basis for subsequent parameter optimization.

[0114] Step S50: Predict the water body distribution based on the water body change model

[0115] After obtaining the water body change model trained in step S40, for each set of augmented project data obtained in step S30, input its environmental parameters and sampling analysis parameters into this model, and the corresponding water body distribution data can be obtained.

[0116] The water body distribution data reflects the spatial distribution of each index of the water body under given environmental parameter conditions. This provides an important decision-making basis for subsequent parameter optimization and can guide how to arrange sampling points to cover the water body spatial distribution to the greatest extent.

[0117] Step S60: Optimize the sampling parameters using the genetic algorithm

[0118] Based on the water body distribution data obtained in step S50, the genetic algorithm is used to optimize the sampling parameters of each set of augmented project data. The objective function is to maximize the sampling coverage rate and minimize the sampling error.

[0119] The genetic algorithm is a heuristic optimization algorithm that simulates the biological evolution process. Through operations such as selection, crossover, and mutation, the optimal solution of the objective function can be found. Here, the decision variables of the genetic algorithm are the sampling parameters, including the sampling time interval, sampling depth, and sampling volume.

[0120] Through the iterative optimization of the genetic algorithm, the optimal sampling parameter scheme corresponding to each set of augmented project data can be obtained, providing a basis for the training of the subsequent parameter determination model.

[0121] Step S70: Train the sampling parameter determination model

[0122] With the optimal sampling parameters optimized in step S60, a regression neural network model can be established next to fit the mapping relationship between the environmental parameters and the sampling parameters.

[0123] Specifically, all the project data obtained by expanding step S30 are used as the training set, where the input is the environmental parameters and the output is the optimized sampling parameters. By training a regression neural network, the implicit law between the environmental parameters and the sampling parameters can be learned.

[0124] The trained regression model is the sampling parameter determination model, which can be used to predict the optimal sampling parameters under given environmental parameter conditions. This provides a basis for parameter determination in practical applications.

[0125] Step S80: Determine the sampling parameters of the current water area

[0126] In practical applications, when it is necessary to determine the depth stratification parameters of a multi-channel water sampler for a certain water area to be measured, first obtain the environmental parameters of the water area, including water depth, water quality, water temperature, flow velocity, etc.

[0127] Then, input these environmental parameters into the sampling parameter determination model trained in step S70, and the model can output the optimal sampling parameters of the water area, including sampling time interval, sampling depth, sampling volume, etc.

[0128] These optimized sampling parameters are the depth stratification parameters of the multi-channel water sampler for the water area, which can guide the actual sampling operation and ensure the accuracy and representativeness of the sampling results.

[0129] Optionally, for the specific structure of the water body change model, its model structure:

[0130] 1. Input layer:

[0131] The input layer receives environmental parameter data, including water depth, water quality, water temperature, flow velocity, etc.

[0132] For time series data, the sliding window method can be used to take historical data as part of the input.

[0133] For categorical data (such as water quality level), One-Hot encoding is required.

[0134] 2. Embedding Layer:

[0135] For continuous numerical data, a fully connected layer is used for embedding to map the original features to a low-dimensional embedding space.

[0136] For categorical data, an embedding layer is used to map the One-Hot encoding to a low-dimensional embedding vector.

[0137] 3. Statistical model part:

[0138] A multi-layer perceptron (MLP) network is adopted to perform a non-linear transformation on the embedded input data.

[0139] The output layer uses linear regression or other statistical models to capture linear and non-linear trends in the data.

[0140] The output of this part serves as the residual term of the deep learning model.

[0141] 4. Deep learning module:

[0142] Recurrent neural network structures such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) are used.

[0143] Multiple layers of LSTM or GRU units can be stacked to improve the expressive power of the model.

[0144] The input is the embedded data, and the output is the corresponding sampling analysis metrics.

[0145] 5. Residual connection:

[0146] The output of the statistical model part is added to the output of the deep learning module as the final model output.

[0147] This residual connection can explicitly retain the basic trends captured by the statistical model while leveraging the deep model to learn the residuals.

[0148] 6. Output layer:

[0149] For regression tasks, the output layer is a fully connected layer that outputs continuous sampling analysis metric values.

[0150] For classification tasks, the output layer is a Softmax layer that outputs the probability distribution for each class.

[0151] Model training:

[0152] 1. Data preprocessing:

[0153] The original data is standardized or normalized to eliminate the influence of measurement units.

[0154] The training set, validation set, and test set are divided according to a fixed ratio.

[0155] For time series data, samples can be generated using a sliding window approach.

[0156] 2. Loss function:

[0157] For regression tasks, the mean squared error (MSE) or mean absolute error (MAE) is used as the loss function.

[0158] For classification tasks, the cross-entropy loss function is used.

[0159] Regularization terms (such as L1 or L2 regularization) can be considered to prevent overfitting.

[0160] 3. Optimization Algorithm:

[0161] Use Stochastic Gradient Descent (SGD), Adam, or other optimization algorithms to calculate the gradients of the parameters according to the loss function.

[0162] Set an appropriate learning rate scheduling strategy, such as exponential decay or cosine annealing.

[0163] 4. Model Training:

[0164] Perform iterative training for multiple epochs on the training set and update the parameters using Mini-Batch GD.

[0165] After each epoch, evaluate the model performance on the validation set and select the model with the best validation metrics as the final model.

[0166] For time series data, rolling prediction can be used for training, that is, using historical data to predict the value of the next time step.

[0167] 5. Early Stopping:

[0168] Monitor the loss value or other metrics of the validation set during training.

[0169] When the validation set metrics do not improve for several consecutive epochs, terminate the training early to avoid overfitting.

[0170] 6. Model Ensemble:

[0171] Multiple models can be trained and their outputs can be ensembled to improve the generalization ability.

[0172] Common ensemble methods include average ensemble, weighted ensemble, and stacking ensemble, etc.

[0173] 7. Hyperparameter Optimization:

[0174] For the hyperparameters of the model, such as the number of layers, the number of neurons, the dropout rate, etc., methods such as grid search or Bayesian optimization can be used for automatic tuning.

[0175] Evaluate the model performance on the validation set under different hyperparameter combinations and select the optimal configuration.

[0176] Through the above model structure and training strategy, the advantages of statistical models and deep learning models can be effectively integrated to capture the linear, non-linear, and time-series features in water body change data, thereby improving the overall prediction accuracy. In practical applications, the model structure and training process can be further adjusted and optimized according to the characteristics of the data and task requirements to obtain better performance.

[0177] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run, they are used to execute the above method for determining the depth stratification parameters of a multi-channel water sampler.

[0178] The third aspect of the present invention provides a system for determining the depth stratification parameters of a multi-channel water sampler, which includes the above computer-readable storage medium.

[0179] To better understand and implement the present invention, the following provides a specific Example 1 of the present invention in a computer-readable storage medium or computer scenario for a computer program: For this Example 1, in step S10, it is necessary to collect historical project data of successful sampling using a multi-channel water sampler in different regions of the same water body. These data include three categories: environmental parameters, sampling parameters, and sampling analysis data.

[0180] The environmental parameters include water depth D, water quality Q, water temperature T, flow velocity V, etc. These environmental parameters can be represented by a vector E:

[0181] E = [D, Q, T, V]

[0182] The sampling parameters are provided by experts according to the environmental parameters, and mainly include sampling time interval Δt, sampling depth Δz, and sampling volume V s etc. These sampling parameters are represented by a vector S:

[0183] S = [Δt, Δz, V s

[0184] The sampling analysis data is the detection results of physical index P, biological index B, and chemical index C for the collected water samples. These sampling analysis data can be represented by a vector A:

[0185] A = [P, B, C]

[0186] In summary, each set of historical water sampling project data can be represented as:

[0187] X = [E, S, A]

[0188] In step S20, it is necessary to establish and fit the basic change model of the water body. According to the historical project data X, the mathematical relationship model between the environmental parameter E and the sampling analysis data A can be established by using the method of regression analysis:

[0189] A = f(E) + ∈

[0190] where f(·) represents an unknown functional relationship, and ∈ is a random error term. By fitting the training sample data, the specific form of the function f(·) can be obtained, thus establishing the basic change model of the water body. Commonly used fitting methods include linear regression, polynomial regression, neural networks, etc.

[0191] In step S30, it is necessary to expand each group of historical project data collected in step S10 to obtain more sample data. Here, the method of Monte Carlo random simulation can be adopted. Based on the water body change model f(·) established in step S20, more diverse environmental parameters E and sampling analysis data A can be generated. Specifically, the probability distribution p(E) of the environmental parameter E can be set, and then a random number generator is used to sample according to this distribution to obtain multiple realization values of E, and then substitute them into the model f(·) to calculate the corresponding A values. Repeating this iteration can obtain a large amount of expanded project data.

[0192] In step S40, it is necessary to train a multi-layer perceptron (MLP) neural network model. The input of this model is the environmental parameter E, and the output is the sampling analysis data A. Its structure can be expressed as:

[0193]

[0194] where, represents the non-linear mapping function of the neural network, and Θ is the parameter of the network. The backpropagation algorithm can be used to optimize the network parameter Θ with the goal of minimizing the mean square error between the training samples and the network output, so as to train and obtain the water body change model.

[0195] In step S50, it is necessary to predict the water body distribution corresponding to each group of data in the expanded project data based on the water body change model trained in step S40 Specifically, for the i-th expanded project data X i = [E i , S i , A i , the environmental parameter E i can be input into the model to obtain the predicted sampling analysis data Then, can be compared with the actual A i to analyze the distribution characteristics of the water body under this environmental parameter condition.

[0196] In step S60, it is necessary to optimize the sampling parameter S of each group of augmented project data by using the genetic algorithm. The genetic algorithm is a heuristic optimization algorithm that simulates the biological evolution process. Its basic idea is to iteratively optimize and solve the optimal solution of the objective function through operations such as selection, crossover, and mutation. Here, the objective function includes two indicators:

[0197] 1. Sampling coverage rate R c , which represents the proportion of sampling points within the water body distribution area, and the larger the value, the better.

[0198] 2. Sampling error E s , which represents the root mean square error between the actual sampling data and the predicted water body distribution, and the smaller the value, the better.

[0199] These two indicators can be combined to form the objective function J, and the genetic algorithm is used for optimization:

[0200] minJ = -R c + λE s

[0201] where λ is the weight coefficient used to balance the importance of the two indicators. Through the iteration of the genetic algorithm, the optimal sampling parameter S* corresponding to each group of augmented project data can be obtained.

[0202] In step S70, it is necessary to train a regression neural network model to learn the mapping relationship between the environmental parameter E and the optimized sampling parameter S*. The input of this model is E, and the output is S*, which can be expressed as:

[0203]

[0204] where represents the non-linear mapping function of the regression neural network, and Φ is the network parameter. Similarly, the backpropagation algorithm can be used to optimize the network parameter Φ with the goal of minimizing the mean square error between the training samples and the network output, so as to train and obtain the sampling parameter determination model.

[0205] In step S80, when it is necessary to determine the depth stratification parameter of the multi-channel water sampler for a certain water area to be measured, first obtain the environmental parameter E 0 , including water depth D 0 , water quality Q 0 , water temperature T 0 and flow velocity V 0 etc. Then, input E 0 into the sampling parameter determination model trained in step S70 to obtain the best sampling parameter for this water area

[0206]

[0207] These optimized sampling parameters are the depth-stratified parameters of the multi-channel water sampler for this water area, which can guide actual sampling operations and ensure the accuracy and representativeness of sampling results.

[0208] Specifically, the principle of the present invention is as follows:

[0209] 1. Establish a water body change rule model based on historical sampling data: In actual water quality monitoring, historical sampling data from different regions of the same water body contain the internal rules of the environmental changes of this water body. The present invention first collects this historical data, including environmental parameters (water depth, water quality, water temperature, flow velocity, etc.), sampling parameters (sampling time interval, sampling depth, sampling volume, etc.), and sampling analysis results (physical and chemical indicators, biological indicators, etc.).

[0210] By analyzing and modeling these historical data, a mathematical relationship model between water body environmental parameters and sampling analysis indicators can be obtained. Common modeling methods include regression analysis, time series analysis, etc. Establishing such a water body change rule model can more accurately describe the overall situation of this water body under different environmental conditions and provide a basis for subsequent parameter optimization.

[0211] 2. Optimize sampling parameters based on machine learning: After having the water body change rule model, the present invention further uses machine learning technology to optimize sampling parameters. First, the historical data is extended and augmented to generate more diverse simulated data. Then, a multi-layer perceptron neural network model is trained to learn the complex mapping relationship between environmental parameters and sampling analysis data. Next, based on the trained neural network model, combined with a genetic algorithm, the sampling parameters (sampling depth, sampling volume, sampling time interval, etc.) are optimized. The genetic algorithm is a heuristic optimization algorithm that searches for the optimal solution of the objective function by simulating the process of biological evolution. In the present invention, the objective function includes two indicators: maximizing sampling coverage and minimizing sampling error. Through the iterative optimization of the genetic algorithm, the optimal sampling parameter scheme under various environmental conditions can be obtained, meeting the maximum coverage of the overall water body distribution and the minimum measurement error. These optimized parameters are the depth-stratified parameters of the multi-channel water sampler, which can guide actual water quality sampling work and improve the representativeness of sampling results.

[0212] 3. Realize adaptive determination of sampling parameters: Finally, the present invention trains a regression neural network model to learn the mapping relationship between environmental parameters and optimized sampling parameters. In this way, as long as the real-time environmental parameters of the water area to be measured are obtained, the optimal sampling parameter scheme can be quickly predicted through this regression model.

[0213] This adaptive parameter determination method can automatically adjust the sampling plan according to the actual water conditions without manual experience intervention. Compared with the traditional method relying on manual judgment, this method has stronger adaptability and automation capabilities, which not only improves the monitoring efficiency but also ensures the scientificity and representativeness of the sampling results.

[0214] To better understand the present invention, the following provides Embodiment 2 of the present invention applied in a specific scenario: A certain reservoir in a certain city has long been responsible for the drinking water supply task of the city, so its water quality has always been highly valued. To better grasp the dynamic changes in the water quality of this reservoir, the environmental protection department decides to deploy a multi-channel water sampler in the reservoir to monitor water quality indicators in real time.

[0215] Based on the analysis of past water quality monitoring data, this reservoir has some typical water quality change characteristics:

[0216] 1) The water temperature changes significantly with seasons. The water temperature is higher in summer and lower in winter, and the average annual water temperature is about 15°C;

[0217] 2) The reservoir has obvious stratification, and there are significant differences in water quality between the surface layer and the bottom layer, especially the dissolved oxygen content;

[0218] 3) Affected by upstream agricultural non-point source pollution, there are differences in the water quality of the reservoir during the wet season and the dry season, and the nitrogen and phosphorus indicators fluctuate greatly;

[0219] 4) Affected by weather factors, there are certain intra-day and intra-month changes in the water level and flow rate of the reservoir.

[0220] To effectively monitor these water quality dynamic characteristics, the environmental protection department requires that the multi-channel water sampler needs to meet the following technical indicators:

[0221] 1) The sampling depth covers the entire water layer of the reservoir and can reflect the water quality conditions of the surface layer, middle layer, and bottom layer;

[0222] 2) The sampling time interval does not exceed 4 hours to capture short-term changes in water quality;

[0223] 3) The single sampling volume does not exceed 2 liters to minimize the disturbance to the water body;

[0224] 4) The entire monitoring system can achieve remote monitoring and automatic data upload.

[0225] According to these technical requirements, the environmental protection department decides to adopt the multi-channel water sampler depth and layer parameter determination method proposed by the present invention to design the monitoring plan for this reservoir. The specific implementation steps are as follows:

[0226] Step S10: Collect multiple groups of historical water sampling project data

[0227] First, the environmental protection department collected the historical water quality monitoring data of the reservoir for the past 5 years. These data include:

[0228] 1) Environmental parameters: water depth (3 - 25 m), water temperature (10 - 25 °C), dissolved oxygen (4 - 12 mg / L), pH value (6.5 - 8.5), turbidity (5 - 50 NTU), nutrient salts (nitrogen and phosphorus indicators, 0.1 - 2.0 mg / L), flow velocity (0.1 - 0.6 m / s), etc.;

[0229] 2) Sampling parameters: sampling depth (surface layer, middle layer, bottom layer), sampling time interval (1 h, 2 h, 4 h, 8 h), sampling volume (1 L, 1.5 L, 2 L), etc.;

[0230] 3) Sampling analysis data: physical and chemical indicators (pH value, dissolved oxygen, turbidity, conductivity, etc.), biological indicators (chlorophyll a, plankton quantity, etc.), nutrient salt indicators (ammonia nitrogen, nitrate, phosphate, etc.).

[0231] These historical data cover the sampling situations of different time periods and different water layers of the reservoir, laying a foundation for the subsequent establishment of a water body change rule model.

[0232] Step S20: Establish and fit the basic change model of the water body

[0233] Based on the historical data collected in step S10, the environmental protection department used the method of multiple linear regression to establish a relationship model between the water quality indicators of the reservoir and the environmental parameters. Taking dissolved oxygen (DO) as an example, the regression model can be expressed as:

[0234] DO = 9.82 - 0.15 × water temperature + 0.24 × flow velocity - 0.36 × turbidity + 0.41 × pH value

[0235] Through model fitting, R^2 reaches 0.78, indicating that the model can better describe the variation law of DO in the reservoir with environmental parameters. Similarly, regression models between other water quality indicators (such as ammonia nitrogen, total phosphorus, etc.) and environmental parameters are also established.

[0236] In addition, the environmental protection department also tried to establish a time series model of water quality indicators to describe the dynamic changes of water quality over time. For example, the ARIMA model was used to fit the time series of dissolved oxygen, and the model parameters obtained were:

[0237]

[0238] Through this model, the change trend of dissolved oxygen in a certain future period can be predicted, providing a basis for the optimization of sampling parameters.

[0239] Generally speaking, these water body change models established based on historical data provide important reference information for subsequent parameter determination and can accurately describe the law of water quality change in this reservoir with environmental changes.

[0240] Step S30: Augment the project data

[0241] Based on the water body change model established in step S20, the environmental protection department augmented the historical project data by using the Monte Carlo simulation method.

[0242] Specifically, they first performed distribution fitting on the historical observation data of environmental parameters (water temperature, flow velocity, turbidity, etc.) to obtain the probability density function of each parameter. Then, using a random number generator, a large number of simulations of environmental parameters were carried out according to these probability distributions, and more diverse combinations of environmental parameters were obtained.

[0243] Next, these simulated environmental parameters were substituted into the water quality index regression model and time series model established in step S20 to calculate the corresponding water quality analysis data. By repeating such simulations many times, a large number of augmented project data were finally generated, providing rich samples for subsequent neural network training.

[0244] Through this augmentation method, the environmental protection department obtained a total of 3,000 groups of simulated project data, which is 6 times the original historical data. These data not only retain the statistical characteristics of the real data but also cover a wider range of environmental conditions, providing a good foundation for subsequent optimization.

[0245] Step S40: Train the multi-layer perceptron neural network model

[0246] With rich training samples, the environmental protection department started to train a multi-layer perceptron (MLP) neural network model to learn the complex non-linear mapping relationship between environmental parameters and water quality indicators.

[0247] The specific network structure settings are as follows:

[0248] - Input layer: 7 nodes, corresponding to environmental parameters such as water temperature, flow velocity, turbidity, pH value, ammonia nitrogen, nitrate, and phosphate respectively

[0249] - Hidden layer 1: 64 nodes, using the ReLU activation function

[0250] - Hidden layer 2: 32 nodes, using the ReLU activation function

[0251] - Output layer: 9 nodes, corresponding to water quality indicators such as dissolved oxygen, chemical oxygen demand, total nitrogen, total phosphorus, chlorophyll a, and plankton quantity respectively

[0252] The standard backpropagation algorithm is used for network training. The mean absolute error (MAE) is used as the loss function to optimize the network parameters. During the training process, the 2000 groups of data generated in step S30 are used as the training set, and the remaining 1000 groups of data are used as the validation set. By adjusting the network structure and hyperparameters, the performance index of the mean absolute error less than 5% is finally achieved on the validation set.

[0253] The trained MLP model is the water body change model, which can accurately predict the distribution of various water quality indicators of the reservoir under given environmental conditions. This model provides an important basis for subsequent parameter optimization.

[0254] Step S50: Predict the water body distribution based on the water body change model

[0255] With the water body change model trained in step S40, the environmental protection department inputs the environmental parameters of the 3000 groups of expansion project data in step S30 into this model to calculate the corresponding water quality index distribution.

[0256] Taking dissolved oxygen as an example, the prediction results of the model show that: under the environmental conditions of water temperature 15°C, flow velocity 0.3 m / s, and turbidity 20 NTU, the dissolved oxygen in the surface layer of the reservoir is about 9.5 mg / L, about 7.2 mg / L in the middle layer, and about 5.8 mg / L in the bottom layer. Similarly, the distribution of other water quality indicators in different water layers is also obtained.

[0257] By analyzing these water body distribution data, it can be found that:

[0258] 1) Factors such as water temperature and flow velocity have a significant impact on the spatial distribution of dissolved oxygen in the entire water body, and the dissolved oxygen in the bottom layer is generally low;

[0259] 2) There are significant differences in nutrient salt indicators (ammonia nitrogen, phosphate) between the surface layer and the bottom layer, reflecting the stratification characteristics of the reservoir;

[0260] 3) Phytoplankton indicators (chlorophyll a) are relatively abundant in the middle and upper water layers, and less in the bottom layer.

[0261] These water quality distribution characteristics provide a basis for the optimization of subsequent sampling parameters and can guide how to reasonably arrange sampling points to cover the overall water body conditions.

[0262] Step S60: Optimize sampling parameters using genetic algorithm

[0263] Based on the water body distribution data predicted in step S50, the environmental protection department uses the genetic algorithm to optimize the sampling parameters. The objective function includes two indicators:

[0264] 1) Sampling coverage rate R c: Reflects the proportion of the water body distribution area covered by the sampling points, with a value range of [0, 1], and the larger the better.

[0265] 2) Sampling error E s : Reflects the error between the actual sampling data and the prediction of the water body distribution, and the smaller the value, the better.

[0266] These two indicators are combined to form the following objective function:

[0267] J = -R c + 0.2E s

[0268] where R c The calculation formula is:

[0269]

[0270] where A i is the area of the water body distribution area covered by the i-th sampling point, and V j is the total volume of the water body. The sampling error E s Then the root mean square error between the predicted value and the actual observed value is used.

[0271] The coding method of the genetic algorithm is:

[0272] Decision variables: Sampling depth (0 - 25m), sampling time interval (1h, 2h, 4h), sampling volume (1L, 1.5L, 2L)

[0273] Population size: 100

[0274] Number of evolutionary generations: 200

[0275] Through the iterative optimization of the genetic algorithm, the final optimal sampling parameter scheme obtained is:

[0276] Sampling depth: Surface layer 2m, middle layer 10m, bottom layer 20m

[0277] Sampling time interval: 4 hours

[0278] Sampling volume: 1.5L

[0279] Under this scheme, the sampling coverage rate R c reaches 92%, and the sampling error E s is controlled within 3%, meeting the technical requirements of the environmental protection department.

[0280] Step S70: Train the sampling parameter determination model

[0281] With the optimal sampling parameter scheme optimized in step S60, the environmental protection department then trained a regression neural network model to learn the mapping relationship between environmental parameters and sampling parameters.

[0282] The input of the regression model is: 7 environmental parameters such as water temperature, flow rate, turbidity, pH value, and nutrients; the output is: 3 sampling parameters: sampling depth, sampling time interval, and sampling volume. The network structure is set as follows:

[0283] - Input layer: 7 nodes

[0284] -Hidden layer 1: 32 nodes, using ReLU activation function

[0285] -Hidden layer 2: 16 nodes, using ReLU activation function

[0286] -Output layer: 3 nodes

[0287] The back propagation algorithm is also used for training, with mean absolute error (MAE) as the loss function. After 200 epochs of training, the model achieved a performance indicator of less than 5% mean absolute error on the validation set.

[0288] The trained regression model is the sampling parameter determination model, which can quickly predict the optimal sampling parameter plan required for the water area based on the environmental parameters obtained in real time.

[0289] Step S80: Determine the sampling parameters of the current water area

[0290] When actual sampling of the reservoir is required, the multi-channel water sampler equipment first obtains the current environmental parameters, including water temperature (15°C), flow rate (0.4m / s), turbidity (25NTU), pH value (7.2), nutrients (ammonia nitrogen 0.5mg / L, nitrate 1.2mg / L, phosphate 0.3mg / L), etc.

[0291] Then, these environmental parameters are input into the sampling parameter determination model trained in step S70, and the model can output the optimal sampling scheme:

[0292] - Sampling depth: 2m for surface layer, 10m for middle layer, 20m for bottom layer

[0293] - Sampling time interval: 4 hours

[0294] -Sampling volume: 1.5L

[0295] These optimized sampling parameters can not only cover the overall distribution of water bodies (coverage rate of 92%), but also control the measurement error at a low level (within 3%), meeting the technical requirements of environmental protection departments.

[0296] The multi-channel water sampler automatically completed the regular sampling of the reservoir based on these parameter settings. After the collected water samples were transported back to the laboratory, the test results showed:

[0297] - The dissolved oxygen in the surface layer is 9.3 mg / L, 7.5 mg / L in the middle layer, and 6.1 mg / L in the bottom layer.

[0298] - The ammonia nitrogen is 0.48 mg / L, nitrate is 1.1 mg / L, and phosphate is 0.28 mg / L.

[0299] - The chlorophyll a concentration is 12 μg / L, and the number of plankton is 3,500 per liter.

[0300] These monitoring results are basically consistent with the water quality distribution predicted in step S50, indicating that the sampling parameters determined by this method are reasonable and feasible, and can effectively reflect the actual water quality status of the reservoir.

[0301] To further verify the effectiveness of the method of the present invention, the environmental protection department has operated this multi-channel water sampler in the reservoir for a long time and collected a large amount of water quality monitoring data. Through the analysis of these data, it is found that this method performs excellently in practical applications, specifically reflected in the following aspects:

[0302] 1. The representativeness and reliability of the sampling results are greatly improved: The sampling parameters determined by the method of the present invention can better cover the overall water quality distribution of the reservoir. Taking dissolved oxygen as an example, the monitoring results of the surface layer, middle layer and bottom layer reflect obvious stratification characteristics, which are consistent with the predicted distribution in step S50. At the same time, the dispersion degree of the monitoring data of different water layers is also small, indicating that the error in the sampling process is effectively controlled.

[0303] Compared with the previous situation of selecting sampling parameters relying on manual experience, the sampling scheme determined by this method can better represent the overall water quality of the reservoir and provide reliable basic data for subsequent water quality evaluation.

[0304] 2. The response efficiency to water quality abnormal events is greatly improved: During the operation of the reservoir, an accident of chemical industrial wastewater flowing into the lake occurred, resulting in a drastic change in water quality in a short time. Since the sampling time interval determined by the method of the present invention is short (4 hours), the multi-channel water sampler can quickly capture this water quality anomaly and upload the abnormal alarm information within 2 hours after the accident. The environmental protection department quickly organized emergency disposal, and by retrieving the recent water quality monitoring data, quickly identified the pollution source and took effective measures, minimizing the environmental loss to the greatest extent. If the traditional sampling scheme of once a day is adopted, it is very likely that the problem will not be discovered until 1-2 days later, missing the best disposal opportunity.

[0305] Therefore, the method of the present invention greatly improves the timeliness and response ability of the water quality monitoring system and plays an important role in dealing with water environment emergencies.

[0306] 3. Significantly reduced operating costs and more intelligent and efficient operation and maintenance: First, the method of the present invention makes full use of historical monitoring data and does not require additional manpower for on-site sampling. At the same time, the optimized sampling parameters can cover the water body conditions to the greatest extent, reduce unnecessary repeated sampling, and greatly reduce the consumable and labor costs. Second, the adaptive adjustment function of the sampling parameters enables the multi-channel water sampler to autonomously optimize the sampling scheme according to the water body changes without manual intervention. This not only improves the automation level of monitoring but also reduces the dependence on professional technical personnel, greatly reducing the operation and maintenance pressure. In addition, the system realizes the remote automatic upload of water quality data, which greatly facilitates the information management and early warning analysis work of environmental protection departments. Compared with the traditional manual inspection and on-site recording methods, this intelligent water quality monitoring mode greatly improves the work efficiency.

[0307] In summary, through long-term practice, the environmental protection department has fully verified the excellent performance of the method of the present invention in the application of multi-channel water samplers. This method not only improves the scientific nature and response ability of water quality monitoring but also significantly reduces the operating costs, providing strong support for the realization of intelligent water environment monitoring. In the future, the environmental protection department plans to promote the application of this technology in more water bodies to better protect the regional water environment quality.

[0308] In order to further demonstrate the performance of the method of the present invention, the long-term monitoring data of the reservoir are now sorted into a table and Table 1 as follows:

[0309] Table 1 Statistical results of dissolved oxygen monitoring in the reservoir

[0310] Water layer Average value (mg / L) Standard deviation Surface layer 9.1 0.4 Middle layer 7.4 0.6 Bottom layer 6.0 0.5

[0311] As can be seen from Table 1, the sampling parameters determined by the method of the present invention can better reflect the differences in dissolved oxygen in the vertical stratification of the reservoir. The dissolved oxygen in the surface layer is relatively high, while that in the bottom layer is relatively low, which conforms to the stratification characteristics of the reservoir. The standard deviations of the monitoring results of each water layer are also small, indicating that the stability and reliability of the sampling data are relatively high.

[0312] The 4-hour sampling interval determined by the method of the present invention can better capture this short-term dynamic change and provide valuable data support for water quality early warning.

[0313] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A method for determining depth-determining stratification parameters of a multi-channel water sampler, characterized in that: The following steps are involved: S10, collecting multiple groups of historical water sampling project data, wherein each group of water sampling project data is specifically project data of historical water sampling projects of successful multi-channel water samplers in different areas of the same water body; the project data includes environmental parameters, sampling parameters and sampling analysis data; S20, for each set of environmental parameters and sampling analysis data in the project data, establishing and fitting the basic change model of the water body; S30, based on the basic change model, each set of project data is augmented and expanded to obtain corresponding multiple sets of augmented and expanded project data; S40, using each set of water collection project data and its corresponding augmented project data as a training set, using the multiple sets of water collection project data as a validation set, and training a multi-layer perceptron neural network model to obtain a water body change model; the water body change model includes the following modules: an environmental parameter preprocessing module, a statistical model module, a deep learning module, and a residual connection module; S50, for each set of water sampling project data in the augmented project data, inputting the environmental parameters and sampling analysis parameters therein into the water body change model to obtain water body distribution data; S60, for each group of water sampling project data in the augmented project data, taking the sampling parameters as the initial population, taking maximizing the sampling coverage and minimizing the sampling error as the objective function, using the obtained water body distribution, and using a genetic algorithm to calculate the optimal sampling parameters; S70, for all augmented project data, establish a training data set, wherein the input of the training is the environmental parameter, the output of the training is the sampling parameter, and train a regression neural network to obtain a sampling parameter determination model; S80, obtaining environmental parameters of the current water area to be tested, and inputting them into the sampling parameter determination model to obtain sampling parameters as depth-determining and stratification parameters of the multi-channel water sampler of the current water area to be tested.

2. The method for determining the depth-setting stratification parameters of a multi-channel water sampler according to claim 1, characterized in that: The environmental parameters include water depth, water quality, water temperature and flow rate.

3. The method for determining the depth-setting stratification parameters of a multi-channel water sampler according to claim 2, characterized in that: The sampling parameters are provided by experts according to the environmental parameters, including sampling time interval, sampling depth and sampling volume.

4. The method for determining the depth-setting stratification parameters of a multi-channel water sampler according to claim 3, characterized in that: The sampling and analysis data include physical indicators, biological indicators and chemical indicators of the sampled water.

5. A method for determining depth-determining stratification parameters of a multi-channel water sampler according to claim 4, characterized in that: The statistical model module adopts a mathematical statistical model to fit the basic relationship between environmental parameters and sampling analysis data.

6. A method for determining depth-determining stratification parameters of a multi-channel water sampler according to claim 5, characterized in that: The deep learning module uses a multi-layer perceptron or a long short-term memory network.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for determining the fixed-depth stratification parameters of a multi-channel water sampler according to any one of claims 1 to 6.

8. A system for determining depth and stratification parameters of a multi-channel water sampler, characterized in that: A computer-readable storage medium comprising the method of claim 7.

Images