Soil dispersion coefficient identification method based on longhorn beetle algorithm improved neural network

Abstract

A soil dispersion coefficient identification method based on a longhorn beetle algorithm improved neural network aims at the problem of low accuracy of an existing soil dispersion coefficient identification method, and identifies a soil dispersion coefficient by combining a longhorn beetle search algorithm with a deep neural network. The problem of low accuracy caused by the fact that the soil dispersion coefficient is determined through empirical values in an existing identification method is solved. And the global optimization capability of the longhorn beetle search algorithm is utilized, so that the defect that a single deep neural network is easy to fall into local optimum is effectively avoided, and high-precision fitting prediction is remarkably improved and realized.

Description

Technical Field

[0001] This application relates to the field of soil pollutant detection technology, specifically a method for identifying soil dispersion coefficients based on an improved neural network using the longhorn beetle whisker algorithm. Background Technology

[0002] The dispersion coefficient is a key parameter used to study the diffusion behavior of pollutants in environmental media such as soil and water, and to assess the migration range and fate of pollutants. It quantifies the diffusion capacity of pollutants in environmental media. A higher dispersion coefficient indicates a stronger diffusion capacity of the pollutant, meaning the pollutant can migrate a greater distance within the same time frame. In soil-polluted areas, the possible diffusion direction and extent of pollutants are calculated based on the soil dispersion coefficient. Monitoring points are then placed in key locations to monitor the dynamic changes of pollutants in a timely and accurate manner, providing data support for adjusting pollution control measures and a theoretical basis for soil environmental pollution control and remediation. While empirical values ​​for soil dispersion coefficients for different media types have been analyzed and summarized based on previous experiments and practical cases, demonstrating a certain degree of universality, actual soil environmental media and pollutant migration and diffusion conditions vary greatly. Therefore, existing methods for identifying soil dispersion coefficients suffer from low accuracy. Summary of the Invention

[0003] The purpose of this invention is to address the problem of low accuracy in existing soil dispersion coefficient identification methods by providing a soil dispersion coefficient identification method based on an improved neural network using the longhorn beetle whisker algorithm.

[0004] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0005] A method for identifying soil dispersion coefficients based on an improved neural network using the longhorn beetle whisker algorithm includes the following steps:

[0006] For the soil to be identified, a tracer is injected to obtain... Time point The tracer concentration at the location The longitudinal dispersion coefficient was obtained using the beetle whisker algorithm. Optimal range and lateral dispersion coefficient The optimal range for the tracer concentration was determined, and then the tracer concentration was adjusted. , Time point, Longitudinal dispersion coefficient Optimal range and lateral dispersion coefficient The optimal range is input into the neural network to obtain the optimal longitudinal dispersion coefficient of the neural network output. and optimal transverse dispersion coefficient .

[0007] Furthermore, the obtained Time point The tracer concentration at the location The specific steps are as follows:

[0008] Step 1: Select the application point for the soil to be identified;

[0009] Step 2: Inject tracer into the release point at a rate of 0.9-1.0 cm / day;

[0010] Step 3: Obtain conductivity according to the set time interval;

[0011] Step 4: Based on the change in conductivity, obtain the change in tracer concentration, and then obtain the tracer concentration.

[0012] Furthermore, the tracer concentration Represented as:

[0013] ,

[0014] in, The mass of tracer injected per unit time. The thickness of the confined aquifer. Effective porosity, dimensionless. The longitudinal dispersion coefficient is... The lateral dispersion coefficient is... For water flow velocity, For the zeroth-order modified Bessel function of the second kind, For a first-order overflow system well function, The diffusion parameter is defined with the location where the tracer was placed as the origin. To monitor the x-coordinate of the location, The vertical coordinate of the monitored location, For monitoring time.

[0015] Furthermore, the diffusion parameters Represented as:

[0016] ,

[0017] With the location where the tracer was placed as the origin, To monitor the x-coordinate of the location, The vertical coordinate of the monitoring location.

[0018] Furthermore, the well function of the first type of overflow system is expressed as:

[0019] .

[0020] Furthermore, the longhorn beetle whisker algorithm is specifically as follows:

[0021] Step 1: Randomly generate the initial position vector of the longhorn beetle, and set the algorithm step size and number of iterations;

[0022] Step 2: The initial position of the longhorn beetle's center of mass is , For spatial dimensions, For the left beard, For the right whisker, The distance between the two whiskers. To define the spatial search range, the initial direction vector for the longhorn beetle's movement is represented as:

[0023] ,

[0024] In the formula, rand() represents a random function;

[0025] The coordinates of the longhorn beetle's left and right antennae are expressed in the following way:

[0026] Left coordinates: ,

[0027] Right coordinates: ,

[0028] in, For iteration The position of the center of mass of the second longhorn beetle. For iteration The position of the left whiskers of the second longhorn beetle. For iteration The position of the right whiskers of the second longhorn beetle;

[0029] The update of the centroid coordinates of the longhorn beetle is represented as follows:

[0030] ,

[0031] in, Indicates the first Step size in the next iteration , and For the first Secondary centroid and left and right coordinates For the first Secondary centroid coordinates For the fitness function,

[0032] The root mean square error (MSE) is used as the fitness evaluation function.

[0033] Step 3: Repeat Step 2 until the gradient is sufficiently small, then stop iterating to complete the global optimal weight and threshold search.

[0034] Step 4: Obtain the optimal and The range of intervals is used as the output of the initial calculation stage.

[0035] Furthermore, the fitness evaluation function is expressed as:

[0036] ,

[0037] in, The number of training samples. For the first Output values ​​of each training sample. For the first The actual values ​​of each training sample.

[0038] Furthermore, the algorithm step size is an adaptive step size, which is expressed as:

[0039] ,

[0040] in, For the first Step size in the next iteration This represents the current iteration number. This represents the maximum number of iterations.

[0041] Furthermore, the neural network is trained with a topology structure using the inverse rank-2 quasi-Newton method.

[0042] The beneficial effects of this invention are:

[0043] This application combines the longhorn beetle whisker search algorithm with a deep neural network to identify soil dispersion coefficients, avoiding the low accuracy problem caused by existing methods that rely on empirical values ​​to determine soil dispersion coefficients. Furthermore, this application utilizes the global optimization capability of the longhorn beetle whisker search algorithm to effectively avoid the tendency of a single deep neural network to get trapped in local optima, significantly improving the accuracy of high-precision fitting and prediction.

[0044] In addition, this application enables rapid and accurate calculation of soil dispersion coefficient, which can support soil-groundwater pollution prediction and remediation scheme optimization, provide key parameters for simulating pollution hydrogeological processes, and help prevent and control environmental risks in soil and groundwater. Attached Figure Description

[0045] Figure 1 This is the overall flowchart of this application;

[0046] Figure 2 A comparison chart of RMSE (Recovery Rate of Received Messages) between a single deep neural network and the iterative process of this application;

[0047] Figure 3 This is a comparison diagram of the MAE iterative process of a single deep neural network and that of this application. Detailed Implementation

[0048] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.

[0049] Specific Implementation Method 1: The soil dispersion coefficient identification method based on the longhorn beetle whisker algorithm-improved neural network described in this implementation method includes:

[0050] 1. Data preparation stage

[0051] A sand box experimental setup with dimensions of 45cm×30cm×35cm (length×width×height) was constructed. Tracers were injected into the injection point (point A) at a rate of 0.9-1.0cm / day. Points were taken at certain time intervals to monitor the conductivity of each observation well. The change in conductivity reflects the change in tracer concentration. Concentration data were obtained, and the soil dispersion coefficient in the solute transport model was solved according to the following equation.

[0052] ,

[0053] ,

[0054] ,

[0055] In the formula, for Time point The tracer concentration at the location, in g / L; The mass of tracer injected per unit time, in kg / d; The thickness of the confined aquifer is in meters (m). Effective porosity, dimensionless; The longitudinal dispersion coefficient is m² / d. The lateral dispersion coefficient is m² / d. The velocity of the water flow is expressed in m / d. This is a zero-order trimmed Bessel function of the second kind; For a first-order overflow system well function, its essence is a generalized integral.

[0056] 2. Initial Calculation Stage of Dispersion Coefficient Based on the Longhorn Beetle Algorithm

[0057] (1) Initialize the individual beetle positions and algorithm parameters: Randomly generate the initial position vector of the beetles and set key parameters such as the algorithm step size and number of iterations. The formula for calculating the adaptive step size is as follows:

[0058] ,

[0059] in, For the first Step size in the next iteration This represents the current iteration number. This is the maximum number of iterations, which can be preset.

[0060] (2) Optimization direction and step size update: Adjust the movement direction according to the difference in root mean square error values ​​that the left and right sides need to perceive, and optimize towards the side with smaller error;

[0061] The initial position of the centroid of the longhorn beetle is defined as follows: , For spatial dimensions, For the left beard, For the right whisker, The distance between the two whiskers. The spatial search range is defined. The formula for calculating the initial direction vector of the longhorn beetle's movement is as follows:

[0062] ,

[0063] In the formula, rand() represents a random function.

[0064] The coordinates of the longhorn beetle's left and right antennae are expressed in the following way:

[0065] Left coordinates: ,

[0066] Right coordinates: .

[0067] in, For iteration The position of the center of mass of the second longhorn beetle. For iteration The position of the left whiskers of the second longhorn beetle. For iteration The position of the right whiskers of the second longhorn beetle.

[0068] The update of the centroid coordinates of the longhorn beetle is calculated using the following formula.

[0069] ,

[0070] In the formula, Indicates the first Step size in the next iteration , and For the first Secondary centroid and left and right coordinates For the first Secondary centroid coordinates This is the fitness function.

[0071] Using the root mean square error (RMSE) as the fitness evaluation function, the following formula is used for calculation.

[0072] ,

[0073] in, The number of training samples. For the first Output values ​​of each training sample. For the first The actual values ​​of each training sample;

[0074] (3) Iterative optimization: Repeat step (2) The iteration stops when the gradient is sufficiently small, completing the global optimal weight and threshold search.

[0075] (4) Obtain the optimal D L and D T The range of intervals is used as the output of the initial calculation stage.

[0076] 3. High-order calculation stage of diffusion coefficient based on deep neural network

[0077] (1) Data processing stage: Obtain relevant data in the diffusion equation and divide the data into training set and test set in a ratio of 4:1;

[0078] (2) Construct a deep neural network model: Determine the number of neurons in the input layer, hidden layer, and output layer. The input of this model is the optimal value obtained in step 2. and Interval range, tracer location and time ,concentration Output , The optimal determined value.

[0079] (3) Determine the initial neural network weights and threshold combinations for the model, and train the neural network topology using the inverse rank 2 quasi-Newton method (Broyden-Fletcher-Goldfarb-Shanno method, abbreviated as BFGS method). At this time, the input... and In fact, it is the optimal interval range obtained in the initial calculation stage;

[0080] (4) Repeat step (2) The iteration stops when the objective function changes steadily.

[0081] (5) Obtain the optimal and .

[0082] The parameter optimization results are highly accurate, with small errors between predicted and actual values. The core advantage of deep neural networks lies in their nonlinear fitting and adaptive learning capabilities, enabling them to accurately capture the complex nonlinear mapping relationship between soil property parameters and dispersion coefficients. They can autonomously optimize prediction accuracy without requiring a pre-set mathematical model. This invention proposes a soil dispersion coefficient optimization method based on an adaptive beetle whisker algorithm-improved deep neural network. Utilizing the global optimization capability of the adaptive beetle whisker search algorithm, it effectively avoids the tendency of a single deep neural network to get trapped in local optima, significantly improving the root mean square error (RMSE) and mean absolute error (MAE) of the coupled model, achieving high-precision fitting and prediction.

[0083] The simulation runs quickly. The beetle whisker search algorithm is a highly efficient algorithm that determines the optimization direction solely by the fitness difference perceived by the left and right whiskers. It achieves global optimization by simply setting a few parameters such as adaptive step size and number of iterations. The improved deep neural network based on the beetle whisker algorithm proposed in this invention significantly improves the running speed while ensuring the accuracy of the prediction results.

[0084] The dispersion coefficient value was optimized using the above steps, and the specific conclusions are as follows:

[0085] Figure 2 , 3 This paper presents the trends of RMSE and MAE with the number of iterations in solving the diffusion coefficient using both a standalone deep neural network and an adaptive beetle whisker algorithm-improved deep neural network optimization method. The results show that the adaptive beetle whisker algorithm-improved deep neural network optimization method significantly outperforms the standalone deep neural network model in both convergence efficiency and accuracy. The adaptive beetle whisker algorithm-improved deep neural network optimization method requires only about 10 iterations for the RMSE to rapidly decrease and stabilize at a low level of around 0.12. In contrast, the standalone deep neural network model not only converges more slowly, but also has a higher final stable RMSE (around 0.13) than the beetle whisker algorithm-improved deep neural network, and exhibits significant error fluctuations during iteration. The core reason for this difference lies in the global optimization effect of the adaptive beetle whisker algorithm. Standalone deep neural networks are prone to getting trapped in local optima due to randomized initial parameters, resulting in low convergence efficiency. The beetle whisker algorithm, through its search capability, optimizes the initial weights and thresholds of the deep neural network, helping the model approach a globally optimal solution more quickly. Therefore, in the diffusion coefficient solving task, the adaptive beetle whisker algorithm-improved deep neural network model demonstrates superior accuracy and stability.

[0086] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.

Claims

1. A method for identifying soil diffusion coefficients based on an improved neural network using the longhorn beetle whisker algorithm, characterized in that... Includes the following steps: For the soil to be identified, a tracer is injected to obtain... Time point The tracer concentration at the location The longitudinal dispersion coefficient was obtained using the beetle whisker algorithm. Optimal range and lateral dispersion coefficient The optimal range for the tracer concentration was determined, and then the tracer concentration was adjusted. , Time point, Longitudinal dispersion coefficient Optimal range and lateral dispersion coefficient The optimal range is input into the neural network to obtain the optimal longitudinal dispersion coefficient of the neural network output. and optimal transverse dispersion coefficient .

2. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 1, characterized in that... The obtained Time point The tracer concentration at the location The specific steps are as follows: Step 1: Select the application point for the soil to be identified; Step 2: Inject tracer into the release point at a rate of 0.9-1.0 cm / day; Step 3: Obtain conductivity according to the set time interval; Step 4: Based on the change in conductivity, obtain the change in tracer concentration, and then obtain the tracer concentration.

3. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 1, characterized in that... The tracer concentration Represented as: ， in, The mass of tracer injected per unit time. The thickness of the confined aquifer. Effective porosity, dimensionless. The longitudinal dispersion coefficient is... The lateral dispersion coefficient is... For water flow velocity, For the zeroth-order trimmed Bessel function of the second kind, For a first-order overflow system well function, The diffusion parameter is defined with the location where the tracer was placed as the origin. To monitor the x-coordinate of the location, The vertical coordinate of the monitored location, For monitoring time.

4. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 3, characterized in that... The diffusion parameters Represented as: ， With the location where the tracer was placed as the origin, To monitor the x-coordinate of the location, The vertical coordinate of the monitoring location.

5. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 4, characterized in that... The well function of the first type of overflow system is expressed as: 。 6. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 1, characterized in that... The specific algorithm for the longhorn beetle whiskers is as follows: Step 1: Randomly generate the initial position vector of the longhorn beetle, and set the algorithm step size and number of iterations; Step 2: The initial position of the longhorn beetle's center of mass is , For spatial dimensions, For the left beard, For the right whisker, The distance between the two whiskers. To define the spatial search range, the initial direction vector for the longhorn beetle's movement is represented as: ， In the formula, rand() represents a random function; The coordinates of the longhorn beetle's left and right antennae are expressed in the following way: Left coordinates: , Right coordinates: , in, For iteration The position of the center of mass of the second longhorn beetle. For iteration The position of the left whiskers of the second longhorn beetle. For iteration The position of the right whiskers of the second longhorn beetle; The update of the centroid coordinates of the longhorn beetle is represented as follows: ， in, Indicates the first Step size in the next iteration , and For the first Secondary centroid and left and right coordinates For the first Secondary centroid coordinates For the fitness function, The root mean square error (MSE) is used as the fitness evaluation function. Step 3: Repeat Step 2 until the gradient is sufficiently small, then stop iterating to complete the global optimal weight and threshold search. Step 4: Obtain the optimal and The range of intervals is used as the output of the initial calculation stage.

7. The method for identifying soil diffusion coefficient based on the improved neural network using the longhorn beetle whisker algorithm according to claim 6, characterized in that... The fitness evaluation function is expressed as: ， in, The number of training samples. For the first Output values ​​of each training sample. For the first The actual values ​​of each training sample.

8. The method for identifying soil diffusion coefficients based on the improved neural network using the longhorn beetle whisker algorithm according to claim 7, characterized in that... The algorithm step size is an adaptive step size, which is expressed as follows: ， in, For the first Step size in the next iteration This represents the current iteration number. This represents the maximum number of iterations.

9. The method for identifying soil diffusion coefficient based on the improved neural network using the longhorn beetle whisker algorithm according to claim 1, characterized in that... The neural network is trained using the inverse rank-2 quasi-Newton method.

Images