A three-dimensional ecological cultivation optimization method for seaweed Gracilaria

Abstract

The present invention discloses a method for optimizing the three-dimensional ecological cultivation of seaweed Gracilaria, the method comprising: obtaining growth data of Gracilaria and target marine organisms under different cultivation conditions; establishing constraints and objective functions based on the growth data; wherein the objective function is established by weighted concentration of resource utilization and ecological health index; based on the constraints and objective functions, generating the optimal cultivation density of Gracilaria and target marine organisms according to an ecological cultivation model, and generating a three-dimensional ecological cultivation plan of Gracilaria according to the optimal cultivation density. The present invention proposes a method for optimizing the three-dimensional ecological cultivation of seaweed Gracilaria, the generated three-dimensional ecological cultivation plan can reduce the negative impact on the environment and promote ecological balance; and, by reasonably controlling the cultivation density and species matching, competition and conflict between organisms can be avoided, the occurrence and spread of diseases can be reduced, and the cultivation efficiency of Gracilaria can be improved, thereby solving the problem of difficulty in improving the cultivation efficiency of Gracilaria.

Description

Technical Field

[0001] The invention relates to the technical field of seaweed cultivation, and in particular to a three-dimensional ecological cultivation optimization method of the seaweed Gracilaria. Background Art

[0002] Existing methods for cultivating Gracilaria focus on selecting cultivation areas with sufficient light, clean water, appropriate salinity, suitable temperature and good bottom quality to provide favorable conditions for the growth and reproduction of Gracilaria; or by optimizing the selection, cultivation and management of seedlings to improve the health and stress resistance of Gracilaria seedlings, thereby improving their growth and reproduction efficiency.

[0003] However, the breeding environment is a complex and changeable system, involving multiple factors such as light and water quality. Although existing methods focus on selecting conditions such as sufficient light and clean water, it is often difficult to fully and accurately grasp the subtle changes in all environmental factors; and, on the one hand, seedling cultivation requires a lot of manpower, material and financial resources, and the cost is relatively high; on the other hand, the effect of seedling cultivation is affected by many factors, such as genetic characteristics, cultivation technology, etc., and it is difficult to ensure that high-quality seedlings can be obtained each time they are cultivated. Therefore, these factors lead to low breeding efficiency of Gracilaria. Summary of the invention

[0004] The invention provides a three-dimensional ecological cultivation optimization method of seaweed Gracilaria, so as to solve the problem that it is difficult to improve the cultivation efficiency of Gracilaria.

[0005] To achieve the above objectives, the present application provides a three-dimensional ecological cultivation optimization method of seaweed Gracilaria, comprising:

[0006] Obtain growth data of Gracilaria and target marine organisms under different culture conditions;

[0007] Establishing constraint conditions and an objective function based on the growth data; wherein the objective function is established by weighted centralization of resource utilization and ecological health index;

[0008] Based on the constraints and the objective function, the optimal breeding density of the Gracilaria and the target marine organisms is generated according to the ecological breeding model, and the three-dimensional ecological breeding plan of the Gracilaria is generated according to the optimal breeding density; wherein the ecological breeding model is based on an adversarial learning mechanism, and is obtained by iteratively training a generator to simulate and approximate the density distribution of a teacher model, and training a student model according to the generated density samples; the student model is obtained by training several basis learners according to a comprehensive loss function, and the comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space.

[0009] The present invention can consider both economic and ecological benefits by incorporating resource utilization and ecological health index into the objective function and weighting and concentrating them; this comprehensive consideration helps to reduce resource waste and environmental pollution while pursuing high yields, thereby improving the sustainability of aquaculture. A three-dimensional ecological aquaculture scheme for Gracilaria is generated according to the optimal aquaculture density. This scheme can make full use of aquaculture space and water resources, increase aquaculture density and yield, and reduce negative impacts on the environment to achieve sustainable development. For the ecological aquaculture model, an ecological aquaculture model based on an adversarial learning mechanism is adopted. This model simulates and approximates the density distribution of the teacher model through iterative training of the generator, so as to capture the complex growth laws of Gracilaria and target marine organisms under different aquaculture conditions; at the same time, the student model is trained according to the generated density samples, which further improves the prediction ability and generalization performance of the model. The student model is obtained by training several basis learners based on a comprehensive loss function. The comprehensive loss function considers the sample error distribution in the reproducing kernel Hilbert space, which enables the model to better adapt to the complex and changeable aquaculture environment and improve the accuracy and robustness of the prediction.

[0010] Compared with the prior art, the three-dimensional ecological breeding scheme of the present invention can reduce the negative impact on the environment and promote ecological balance; and, by reasonably controlling the breeding density and species combination, it can avoid competition and conflict between organisms, reduce the occurrence and spread of diseases, and improve the breeding efficiency of Gracilaria, thereby solving the problem of difficulty in improving the breeding efficiency of Gracilaria.

[0011] As a preferred solution, the objective function is established by weighted centralization of resource utilization and ecological health index, specifically:

[0012] Based on the growth data, an attenuation coefficient is introduced to simulate the phenomenon that the light intensity gradually decreases, and a light intensity constraint is established in combination with a preset minimum value and a maximum value of the appropriate light intensity;

[0013] Water temperature constraints are established based on the sensitivity of growth rate to temperature changes and a preset growth rate threshold;

[0014] Establish spatial distribution constraints based on light intensity and water temperature at different geographical locations;

[0015] The light intensity constraint, the water temperature constraint and the spatial distribution constraint are weighted and concentrated according to a weight set to obtain the constraint condition; wherein the weight set is established by using fuzzy logic to quantify the satisfaction degree of different constraints.

[0016] This preferred solution can more realistically reflect the changing pattern of light intensity in nature by introducing an attenuation coefficient to simulate the gradual decrease in light intensity, thereby improving the accuracy of simulating the plant growth environment. Using a preset growth rate threshold to establish a water temperature constraint allows the model to be adjusted according to the needs of specific biological processes or species, enhancing the versatility and adaptability of the model. Considering the differences in light intensity and water temperature in different geographical locations to establish spatial distribution constraints, the model can be applied to a wider geographical area, improving the practicality of the model.

[0017] As a preferred solution, the weight set is established by using fuzzy logic to quantify the satisfaction degree of different constraints, specifically:

[0018] Based on the constraint set, fuzzy sets are defined respectively to represent the satisfaction degree of each constraint, and a plurality of fuzzy sets are obtained; wherein the constraint set includes the light intensity constraint, the water temperature constraint and the spatial distribution constraint;

[0019] Establishing a number of membership functions based on the fuzzy sets;

[0020] The weight set is calculated based on the initial weight value set and the weight values ​​corresponding to the membership functions; wherein the initial weight value set is calculated based on the information entropy corresponding to the constraint set according to the entropy weight method.

[0021] This preferred solution uses fuzzy sets to represent the degree of satisfaction of each constraint, which can handle the uncertainty of the constraint conditions more flexibly. Establishing a membership function can quantify the degree of satisfaction of each constraint condition, thereby providing more accurate information for subsequent processing and decision-making. Since information entropy reflects the amount of information contained in the constraint condition, and the entropy weight method can assign weights according to the amount of information, this makes the weight assignment of the initial weight value set more reasonable and objective. Combined with the weight value corresponding to the membership function, the weight set can be further adjusted and optimized so that the final weight assignment is more in line with the actual situation and needs.

[0022] As a preferred solution, the comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space, specifically:

[0023] According to the difference between the actual stocking density and the stocking density predicted by the model, the first loss function is established by penalizing the square or absolute value of the parameter error;

[0024] Penalizing the square of the parameter error according to a preset method to establish a second loss function; wherein the preset method is determined based on the first loss function, and the preset method is used to maintain the consistency of the loss function;

[0025] Mapping the growth data samples of Gracilaria from the original space to the reproducing kernel Hilbert space, and establishing a third loss function based on the growth data sample error distribution in the reproducing kernel Hilbert space;

[0026] The intensity of the gradient penalty term is controlled by introducing the gradient penalty coefficient, and the fourth loss function is established in combination with the gradient norm;

[0027] The comprehensive loss function is established according to the first loss function, the second loss function, the third loss function and the fourth loss function.

[0028] In this preferred embodiment, the first loss function, which is an error-based penalty mechanism, can ensure that the model continuously reduces the prediction error during the training process, thereby improving the reliability of the prediction results. The second loss function determines the preset method based on the first loss function, which is used to penalize the square of the parameter error. This approach maintains the consistency between the loss functions, so that the model can balance the influence of different loss terms during the optimization process and avoid overfitting or underfitting problems. The third loss function is established by mapping the growth data samples of Gracilaria to the reproducing kernel Hilbert space, which can capture the nonlinear relationship between data samples and improve the model's ability to handle complex data, thereby enhancing the generalization ability of the model. The fourth loss function is established by introducing a gradient penalty coefficient to control the intensity of the gradient penalty term. This mechanism can prevent the model from generating excessive gradients during training and maintain the stability of training.

[0029] As a preferred solution, the ecological farming model is based on an adversarial learning mechanism, which is obtained by iteratively training the generator to simulate and approximate the density distribution of the teacher model, and training the student model according to the generated density samples, specifically:

[0030] Design a generator and several discriminators; wherein the generator is used to generate distribution characteristics of the aquaculture density of Gracilaria and target marine organisms that are close to those presented by the teacher model, and the several discriminators are used to identify and distinguish the difference between the density distribution generated by the generator and the actual density distribution of the teacher model;

[0031] Under the condition that the parameters of the generator are fixed, the discriminators are trained according to the real density distribution data of the teacher model and the fake data generated by the generator; under the condition that the parameters of the discriminators are fixed, the generator is trained with the goal of making the density distribution generated by the generator closer to the real density distribution of the teacher model; the density samples generated by the generator are used as the training data of the student model, and the student model is trained to approximate the output of the teacher model;

[0032] The alternating training process of the generator and the plurality of discriminators is iterated to obtain the ecological farming model.

[0033] Through continuous training, the generator of this preferred solution can generate fake data that is closer and closer to the real distribution, which is of great significance for simulating and understanding the dynamic changes of biological density in aquaculture ecosystems. In addition, through training, the discriminator can continuously improve its recognition ability, thereby more effectively guiding the training process of the generator and ensuring that the generated density distribution gradually approaches the real distribution.

[0034] As a preferred solution, the combined loss function of the plurality of discriminators is established based on the loss functions of different discriminators;

[0035] Among them, the combined loss function is ：

[0036] ；

[0037]

[0038] in, represents the combined loss function, Indicates The loss function of the discriminator is Represents samples sampled from the real data distribution After Discriminator The expected output after represents the noise sampled from the noise distribution Through the generator Generated samples After the Discriminator The expected output after is the gradient penalty term, Indicates Discriminator In the interpolation sample The gradient at is the gradient penalty coefficient, is the interpolation between the real sample and the generated sample, It is The weight of the discriminator, is the total number of discriminators.

[0039] As a preferred solution, based on the constraints and the objective function, the optimal culture density of the Gracilaria and the target marine organisms is generated according to the ecological culture model, specifically:

[0040] Randomly generate a breeding density solution that satisfies the constraint condition to obtain a first breeding density;

[0041] Based on the first breeding density, generating a new solution by adding random perturbations to the linear programming problem regarding the constraints and the objective function, thereby obtaining a second breeding density;

[0042] According to the first breeding density and the second breeding density, respectively calculating a first objective function value and a second objective function value;

[0043] If the first objective function value is less than the second objective function value, the second breeding density is used as the current solution, and the current solution is updated through data iteration, and the current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density;

[0044] Based on the constraint conditions, the objective function and the auxiliary optimal breeding density, the optimal breeding density of the Gracilaria and the target marine organisms is generated according to the ecological breeding model.

[0045] This preferred solution starts iterating by randomly generating a solution for the stocking density that satisfies the constraints, which increases the diversity of the search space and helps to explore more possible solutions. By adding random perturbations to the linear programming problem about the constraints and the objective function to generate new solutions, this method can introduce changes based on the current solution, thereby improving the quality of the solution.

[0046] As a preferred solution, the teacher model is established based on the breeding density, biological species matching and environmental control parameters in shelf-type multi-layer breeding and floor-type three-dimensional breeding.

[0047] The teacher model of this preferred solution is designed for specific breeding modes and can more accurately reflect the actual conditions under these breeding modes. This targeting makes the model more accurate and reliable in predicting and optimizing breeding density, biological species matching and environmental control parameters.

[0048] As a preferred solution, the objective function is established based on the ecological health index by quantifying the degree to which the actual breeding efficiency is close to the optimal state and the degree to which the actual resource utilization is close to the optimal state.

[0049] The objective function of this optimization scheme is centered on the ecological health index, which reflects the concern and protection of the ecological environment; by optimizing aquaculture efficiency and resource utilization, it aims to achieve a win-win situation of economic and ecological benefits and promote the sustainable development of the aquaculture industry.

[0050] As a preferred solution, the objective function is specifically:

[0051]

[0052] in, , , and is the preset weight coefficient, Indicates that at a given stocking density , ecological factors and time The breeding efficiency under represents the upper limit of farming efficiency under optimal conditions, Indicates that at a given stocking density , Resource Factor and time The resource utilization rate under represents the upper limit of resource utilization under optimal conditions, Indicates that at a given stocking density , ecological factors , Resource Factor , spatial distribution parameters and time Ecological health index under Indicates that at a given stocking density , cost factor and time The breeding costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of a flow chart of a three-dimensional ecological cultivation optimization method of seaweed Gracilaria provided in an embodiment of the present application;

[0054] Figure 2 It is a structural schematic diagram of a three-dimensional ecological cultivation optimization system of seaweed Gracilaria provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] In the description of the present application, it should be understood that the terms "first", "second", "third" and "fourth" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first", "second", "third" and "fourth" may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, "several" means two or more.

[0057] Embodiment 1:

[0058] See also Figure 1 The embodiment of the present application provides a three-dimensional ecological cultivation optimization method of seaweed Gracilaria, including S1~S3, and the specific implementation steps are as follows:

[0059] S1. Obtain growth data of Gracilaria and target marine organisms under different culture conditions;

[0060] Step S1 of the embodiment of the present application is specifically as follows:

[0061] The growth data of Gracilaria and target marine organisms under different culture conditions were collected, and the growth data were preprocessed by cleaning, removing outliers and standardizing. The growth data included light intensity, water temperature, nutrient concentration and corresponding culture density. The target marine organisms could be fish, shellfish and other algae. Gracilaria is a seaweed belonging to Rhodophyta, Eurychophyceae, Taxodium and Gracilariaceae.

[0062] S2. Establishing constraint conditions and objective functions based on growth data; wherein the objective function is established by weighted centralization of resource utilization and ecological health index;

[0063] Step S2 of the embodiment of the present application includes S2.1 to S2.3, specifically:

[0064] S2.1. Based on the growth data, the attenuation coefficient is introduced to simulate the phenomenon of gradual decrease in light intensity, and the light intensity constraint is established by combining the preset minimum and maximum values ​​of the appropriate light intensity;

[0065] Water temperature constraints are established based on the sensitivity of growth rate to temperature changes and a preset growth rate threshold;

[0066] Establish spatial distribution constraints based on light intensity and water temperature at different geographical locations.

[0067] The light intensity constraint is:

[0068]

[0069] The water temperature constraint is:

[0070]

[0071] The spatial distribution constraints are:

[0072]

[0073] Among them, in the light intensity constraint, and These are the minimum and maximum values ​​of suitable light intensity. For a breeding environment, the light intensity needs to be kept within this range to ensure the normal growth and development of the breeding organisms; is an attenuation coefficient, which is used to describe the rate at which light intensity decays with increasing stocking density; is the preset stocking density;

[0074] Under the water temperature constraint, It is a preset growth rate threshold to ensure that organisms can grow at suitable water temperatures; is the maximum possible value of the growth rate (i.e., the maximum value that the growth rate can reach when the water temperature is at the optimal condition); is the steepness of the curve, which determines the sensitivity of the growth rate to temperature changes; is the optimal growth temperature, i.e. the water temperature at which the growth rate reaches its maximum value; is the current water temperature;

[0075] In the spatial distribution constraint, and They are ecological factors (light intensity and water temperature ), Indicates geographic location, and Geographical location The light intensity and water temperature, is a constant term, It is the lowest breeding efficiency.

[0076] In this embodiment S2.1, by introducing an attenuation coefficient to simulate the gradual decrease in light intensity, the changing pattern of light intensity in nature can be more realistically reflected, thereby improving the accuracy of the simulation of the plant growth environment. Using a preset growth rate threshold to establish a water temperature constraint allows the model to be adjusted according to the needs of a specific biological process or species, thereby enhancing the versatility and adaptability of the model. Considering the differences in light intensity and water temperature in different geographical locations to establish spatial distribution constraints, the model can be applied to a wider geographical area, thereby improving the practicality of the model.

[0077] S2.2. Based on the constraint set, fuzzy sets are defined to represent the satisfaction degree of each constraint, and several fuzzy sets are obtained; among them, the constraint set includes light intensity constraint, water temperature constraint and spatial distribution constraint; for example, light intensity constraint can be defined as three fuzzy sets of "low", "medium" and "high"; water temperature constraint can be defined as fuzzy sets such as "too low", "suitable" and "too high"; spatial distribution constraint can be defined as fuzzy sets such as "uniform", "relatively uniform" and "uneven".

[0078] According to several fuzzy sets, several membership functions are established. Based on the functions, specific constraint values ​​in the constraint set can be mapped to memberships in the interval [0,1]. A membership value approaching 1 indicates a high degree of constraint satisfaction, while a membership value approaching 0 indicates a low degree of constraint satisfaction.

[0079] Standardize the light intensity constraints, water temperature constraints, and spatial distribution constraints and convert them into dimensionless values;

[0080] The information entropy of each constraint indicator is calculated based on the transformed data; the information entropy is used to reflect the discrete degree of the indicator value. The greater the discrete degree, the greater the role of the indicator in the evaluation, that is, the greater the weight should be;

[0081] According to the size of information entropy, the entropy weight method formula is used to calculate the weight of each constraint indicator in the constraint set to obtain the initial weight value set.

[0082] A weight set is calculated based on an initial weight value set and a number of weight values ​​corresponding to a number of membership functions.

[0083] The light intensity constraint, water temperature constraint and spatial distribution constraint are weighted and concentrated according to the weight set to obtain the constraint conditions.

[0084] The present embodiment S2.2 uses fuzzy sets to represent the degree of satisfaction of each constraint, which can handle the uncertainty of the constraint conditions more flexibly. Establishing a membership function can quantify the degree of satisfaction of each constraint condition, thereby providing more accurate information for subsequent processing and decision-making. Since the information entropy reflects the amount of information contained in the constraint condition, and the entropy weight method can allocate weights according to the amount of information, this makes the weight allocation of the initial weight value set more reasonable and objective. Combined with the weight value corresponding to the membership function, the weight set can be further adjusted and optimized so that the final weight allocation is more in line with the actual situation and needs.

[0085] S2.3. Based on the ecological health index, the objective function is established by quantifying the degree to which the actual breeding efficiency is close to the optimal state and the degree to which the actual resource utilization is close to the optimal state.

[0086] Among them, the objective function is:

[0087]

[0088] in, , , and It is a preset weight coefficient used to balance the relative importance of different objectives; Indicates that at a given stocking density , ecological factors and time The breeding efficiency under represents the upper limit of aquaculture efficiency under optimal conditions (i.e. without considering the limitation of aquaculture density), Indicates that at a given stocking density , Resource Factor and time The resource utilization rate under represents the upper limit of resource utilization under optimal conditions, Indicates that at a given stocking density , ecological factors , Resource Factor , spatial distribution parameters and time Ecological health index under Indicates that at a given stocking density , cost factor and time The breeding costs.

[0089] The objective function of S2.3 of this embodiment is centered on the ecological health index, which reflects the concern and protection of the ecological environment; by optimizing aquaculture efficiency and resource utilization, it aims to achieve a win-win situation of economic and ecological benefits and promote the sustainable development of the aquaculture industry.

[0090] S3. Based on the constraints and the objective function, the optimal breeding density of Gracilaria and the target marine organisms is generated according to the ecological breeding model, and the three-dimensional ecological breeding plan of Gracilaria is generated according to the optimal breeding density; wherein, the ecological breeding model is based on the adversarial learning mechanism, and is obtained by iteratively training the generator to simulate and approximate the density distribution of the teacher model, and training the student model according to the generated density samples; the student model is obtained by training several basis learners according to the comprehensive loss function, and the comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space.

[0091] Step S3 of the embodiment of the present application includes S3.1 to S3.4, specifically:

[0092] S3.1. Use sensors or monitoring equipment to regularly collect data on ecological factors such as light intensity, water temperature, nutrient concentration, and growth data samples of Gracilaria in the breeding area;

[0093] The collected raw data were cleaned to remove outliers, missing values ​​and other data that did not meet the requirements, and the data were normalized so that the numerical ranges of different ecological factors were on the same scale to obtain the data set.

[0094] Based on the data set, according to the difference between the actual stocking density and the stocking density predicted by the model, the first loss function is established by penalizing the square or absolute value of the parameter error;

[0095] Penalizing the square of the parameter error according to a preset method to establish a second loss function; wherein the preset method is determined based on the first loss function, and the preset method is used to maintain the consistency of the loss function;

[0096] Mapping the growth data samples of Gracilaria from the original space to the reproducing kernel Hilbert space, and establishing a third loss function based on the error distribution of the growth data samples in the reproducing kernel Hilbert space;

[0097] The intensity of the gradient penalty term is controlled by introducing the gradient penalty coefficient, and the fourth loss function is established in combination with the gradient norm;

[0098] A comprehensive loss function is established according to the first loss function, the second loss function, the third loss function and the fourth loss function.

[0099] According to the data set and the comprehensive loss function, several preset base learners (such as decision trees, random forests, gradient boosting trees, etc.) are trained to minimize the comprehensive loss function, thereby improving the prediction performance and obtaining a base learner set;

[0100] Bagging, Boosting or other ensemble learning methods are used to integrate the trained base learner set to construct an integrated version of the student model to obtain the final student model.

[0101] Among them, the first loss function is:

[0102]

[0103] The second loss function is:

[0104]

[0105] The third loss function is:

[0106]

[0107] The fourth loss function is:

[0108]

[0109] in, is the actual stocking density, is the stocking density predicted by the model, is a threshold parameter.

[0110] is the sample size, is the optimal stocking density obtained through linear programming, Represents an index;

[0111] is the mapping function that maps data to the Reproducing Kernel Hilbert Space (RKHS), is the norm in the reproducing kernel Hilbert space, represents samples drawn from the true data distribution, Represented by the generator Generated samples;

[0112] is the gradient penalty coefficient, Indicates the sample The expected value of Denotes the discriminator D for the input sample The L2 norm of the gradient of .

[0113] In this embodiment S3.1, the first loss function, which is an error-based penalty mechanism, can ensure that the model continuously reduces the prediction error during the training process, thereby improving the reliability of the prediction results. The second loss function determines a preset method based on the first loss function, which is used to penalize the square of the parameter error. This approach maintains the consistency between the loss functions, so that the model can balance the influence of different loss terms during the optimization process and avoid overfitting or underfitting problems. The third loss function is established by mapping the growth data samples of Gracilaria to the reproducing kernel Hilbert space, which can capture the nonlinear relationship between data samples and improve the model's ability to handle complex data, thereby enhancing the generalization ability of the model. The fourth loss function is established by introducing a gradient penalty coefficient to control the intensity of the gradient penalty term. This mechanism can prevent the model from generating excessive gradients during training and maintain the stability of training.

[0114] S3.2. Establish a teacher model based on the stocking density, biological species matching and environmental control parameters in shelf-type multi-layer stocking and floor-type three-dimensional stocking;

[0115] Among them, "shelf-type multi-layer farming" is an innovative farming model, the core design concept of which is to construct the farming facilities into a three-dimensional structure similar to a shelf. In this model, multiple farming units (such as farming ponds or farming cages) are cleverly stacked vertically to form a multi-layer farming system that efficiently utilizes space;

[0116] "Floor-type three-dimensional farming" is a highly intensive farming model that integrates the design principles of building floors and divides the farming space vertically into multiple independent floors or functional areas. Each floor or area has the conditions and capabilities for independent farming operations.

[0117] The teacher model S3.2 of this embodiment is designed for specific breeding modes and can more accurately reflect the actual conditions under these breeding modes. This targeting makes the model more accurate and reliable in predicting and optimizing breeding density, biological species matching and environmental control parameters.

[0118] S3.3, design a generator and several discriminators; wherein the generator is used to generate distribution characteristics of the aquaculture density of Gracilaria and target marine organisms that are close to those presented by the teacher model; several discriminators are used to identify and distinguish the difference between the density distribution generated by the generator and the actual density distribution of the teacher model, and each discriminator adopts a convolutional neural network architecture, but the parameters are independent;

[0119] Based on the generative adversarial network, with the parameters of the generator fixed, several discriminators are trained according to the real density distribution data of the teacher model and the fake data generated by the generator; with the parameters of several discriminators fixed, the generator is trained with the goal of making the density distribution generated by the generator closer to the real density distribution of the teacher model; the density samples generated by the generator are used as the training data of the student model, and the student model is trained to approximate the output of the teacher model; and a consistency regularization term is introduced to encourage the judgment of consistency of the discriminator output at different scales;

[0120] Iterate the alternating training process of the generator and several discriminators until the predetermined number of training times is reached or the preset stop condition is met, and obtain the trained generator and student model, and form an ecological farming model based on the trained generator and student model; among them, for the ecological farming model, the generator is responsible for generating density distribution samples, and the student model is responsible for making predictions or decisions based on these samples;

[0121] Among them, the combined loss function of several discriminators is established based on the loss functions of different discriminators;

[0122] Among them, the combined loss function of several discriminators is:

[0123] ;

[0124]

[0125] For a pair of discriminators among several discriminators , and its consistency regularization term is:

[0126]

[0127] For all discriminators of several discriminators, the average consistency regularization term is:

[0128]

[0129] The loss function of the generator is:

[0130]

[0131] in, represents the combined loss function, Indicates The loss function of the discriminator is Represents samples sampled from the real data distribution After Discriminator The expected output after represents the noise sampled from the noise distribution Through the generator Generated samples After the Discriminator The expected output after is the gradient penalty term, Indicates Discriminator In the interpolation sample The gradient at is the gradient penalty coefficient, is the interpolation between the real sample and the generated sample, It is The weight of the discriminator, is the total number of discriminators;

[0132] and Respectively represent the outputs of the two discriminators for the same input sample, Represents a sample randomly drawn from the joint distribution of real samples and generated samples;

[0133] is from The number of combinations of two pairs selected from the discriminators is is the weight;

[0134] Representation Generator Input noise The output, Indicates The discriminator is used to analyze the samples generated by the generator. Rating, Represents the noise vector From the prior distribution The expected value of the sample.

[0135] In this embodiment S3.3, through continuous training, the generator can generate fake data that is closer and closer to the real distribution, which is of great significance for simulating and understanding the dynamic changes of biological density in aquaculture ecosystems. In addition, through training, the discriminator can continuously improve its recognition ability, thereby more effectively guiding the training process of the generator and ensuring that the generated density distribution gradually approaches the real distribution.

[0136] S3.4, randomly generate a stocking density solution that meets the constraint conditions, and obtain the first stocking density as the starting point of optimization;

[0137] Starting from the first stocking density, a new solution is generated by adding random perturbations to the linear programming problem about constraints and objective functions, where the "random perturbations" may be small adjustments to the decision variables (i.e., stocking density) or slight changes in the coefficients of the objective function.

[0138] According to the first breeding density and the second breeding density, a first objective function value and a second objective function value are calculated respectively;

[0139] Compare the first objective function value and the second objective function value: ① If the first objective function value is less than the second objective function value, it means that the second breeding density performs better on the objective function. Then the second breeding density is used as the current solution, and the current solution is updated through data iteration. The current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density; ② If the first objective function value is greater than or equal to the second objective function value, decide whether to accept the second objective function value as the new solution based on the probability function related to the difference between the current temperature and the objective function value, so as to jump out of the local optimal solution and continue to explore the global optimal solution. The current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density;

[0140] Based on the constraints, objective function and auxiliary optimal breeding density, the optimal breeding density of Gracilaria and target marine organisms is generated according to the ecological breeding model, and the three-dimensional ecological breeding plan of Gracilaria is generated according to the optimal breeding density; among them, the auxiliary optimal breeding density is used as a starting point to provide a basis and direction for subsequent optimization, specifically: the ecological breeding model starts searching from the auxiliary optimal breeding density to find the optimal breeding density that meets the constraints and objective function.

[0141] In this embodiment S3.4, the iteration starts from randomly generating a farming density solution that satisfies the constraints, which increases the diversity of the search space and helps to explore more possible solutions. By adding random perturbations to the linear programming problem about the constraints and the objective function to generate new solutions, this method can introduce changes based on the current solution, thereby improving the quality of the solution.

[0142] Overall, this embodiment has the following beneficial effects:

[0143] This application can consider both economic and ecological benefits by incorporating resource utilization and ecological health index into the objective function and weighting and concentrating them; this comprehensive consideration helps to reduce resource waste and environmental pollution while pursuing high yields, thereby improving the sustainability of aquaculture. A three-dimensional ecological aquaculture plan for Gracilaria is generated based on the optimal aquaculture density. This plan can make full use of aquaculture space and water resources, increase aquaculture density and yield, while reducing negative impacts on the environment and achieving sustainable development. For the ecological aquaculture model, an ecological aquaculture model based on an adversarial learning mechanism is adopted. This model simulates and approximates the density distribution of the teacher model through iterative training of the generator, so as to capture the complex growth laws of Gracilaria and target marine organisms under different aquaculture conditions; at the same time, the student model is trained according to the generated density samples, which further improves the model's prediction ability and generalization performance. The student model is obtained by training several basis learners based on a comprehensive loss function. The comprehensive loss function takes into account the sample error distribution in the reproducing kernel Hilbert space, which enables the model to better adapt to the complex and changeable aquaculture environment and improve the accuracy and robustness of the prediction;

[0144] In summary, the three-dimensional ecological breeding scheme of the present application can reduce the negative impact on the environment and promote ecological balance; and, by reasonably controlling the breeding density and species combination, it can avoid competition and conflict between organisms, reduce the occurrence and spread of diseases, and improve the breeding efficiency of Gracilaria, thereby solving the problem of difficulty in improving the breeding efficiency of Gracilaria.

[0145] Embodiment 2:

[0146] See also Figure 2, the embodiment of the present application provides a three-dimensional ecological cultivation optimization system for seaweed Gracilaria, including a data module 10, a calculation module 20 and a solution module 30;

[0147] Wherein, the data module 10 is used to obtain the growth data of Gracilaria and target marine organisms under different breeding conditions;

[0148] A calculation module 20, for establishing constraint conditions and an objective function based on the growth data; wherein the objective function is established by weighted centralization of resource utilization and ecological health index;

[0149] The solution module 30 is used to generate the optimal breeding density of Gracilaria and target marine organisms according to the ecological breeding model based on the constraints and the objective function, and generate a three-dimensional ecological breeding plan of Gracilaria according to the optimal breeding density; wherein the ecological breeding model is based on an adversarial learning mechanism, and is obtained by iteratively training a generator to simulate and approximate the density distribution of a teacher model, and training a student model according to the generated density samples; the student model is obtained by training several basis learners according to a comprehensive loss function, and the comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space.

[0150] In one embodiment, the data module 10 is specifically:

[0151] The growth data of Gracilaria and target marine organisms under different culture conditions were collected, and the growth data were preprocessed by cleaning, removing outliers and standardizing. The growth data included light intensity, water temperature, nutrient concentration and corresponding culture density. The target marine organisms could be fish, shellfish and other algae. Gracilaria is a seaweed belonging to Rhodophyta, Eurychophyceae, Taxodium and Gracilariaceae.

[0152] In one embodiment, the calculation module 20 includes a constraint unit, a condition unit and a function unit;

[0153] The constraint unit is used to simulate the phenomenon of gradually decreasing light intensity by introducing an attenuation coefficient based on growth data, and to establish a light intensity constraint in combination with a preset minimum value and a maximum value of the appropriate light intensity;

[0154] The constraint unit is further used to establish a water temperature constraint according to the sensitivity of the growth rate to temperature change and a preset growth rate threshold;

[0155] The constraint unit is also used to establish spatial distribution constraints based on light intensity and water temperature at different geographical locations.

[0156] The light intensity constraint is:

[0157]

[0158] The water temperature constraint is:

[0159]

[0160] The spatial distribution constraints are:

[0161]

[0162] Among them, in the light intensity constraint, and These are the minimum and maximum values ​​of suitable light intensity. For a breeding environment, the light intensity needs to be kept within this range to ensure the normal growth and development of the breeding organisms; is an attenuation coefficient, which is used to describe the rate at which light intensity decays with increasing stocking density; is the preset stocking density;

[0163] Under the water temperature constraint, It is a preset growth rate threshold to ensure that organisms can grow at suitable water temperatures; is the maximum possible value of the growth rate (i.e., the maximum value that the growth rate can reach when the water temperature is at the optimal condition); is the steepness of the curve, which determines the sensitivity of the growth rate to temperature changes; is the optimal growth temperature, i.e. the water temperature at which the growth rate reaches its maximum value; is the current water temperature;

[0164] In the spatial distribution constraint, and They are ecological factors (light intensity and water temperature ), Indicates geographic location, and Geographical location The light intensity and water temperature, is a constant term, It is the lowest breeding efficiency.

[0165] The constraint unit of this embodiment can more realistically reflect the changing law of light intensity in nature by introducing an attenuation coefficient to simulate the gradual decrease in light intensity, thereby improving the accuracy of simulating the plant growth environment. Using a preset growth rate threshold to establish a water temperature constraint allows the model to be adjusted according to the needs of a specific biological process or species, thereby enhancing the versatility and adaptability of the model. Considering the differences in light intensity and water temperature in different geographical locations to establish spatial distribution constraints, the model can be applied to a wider geographical area, thereby improving the practicality of the model.

[0166] The conditional unit is used to define fuzzy sets based on the constraint set to represent the satisfaction degree of each constraint, thereby obtaining several fuzzy sets; wherein the constraint set includes light intensity constraint, water temperature constraint and spatial distribution constraint; for example, the light intensity constraint can be defined as three fuzzy sets of "low", "medium" and "high"; the water temperature constraint can be defined as fuzzy sets such as "too low", "suitable" and "too high"; the spatial distribution constraint can be defined as fuzzy sets such as "uniform", "relatively uniform" and "uneven".

[0167] The conditional unit is also used to establish several membership functions based on several fuzzy sets. Based on the function, specific constraint values ​​in the constraint set can be mapped to memberships in the interval [0,1]. Among them, a membership value approaching 1 indicates a high degree of constraint satisfaction, and a membership value approaching 0 indicates a low degree of constraint satisfaction.

[0168] The conditional unit is also used to standardize light intensity constraints, water temperature constraints, and spatial distribution constraints, converting them into dimensionless values;

[0169] The conditional unit is also used to calculate the information entropy of each constraint indicator based on the transformed data; the information entropy is used to reflect the discrete degree of the indicator value. The greater the discrete degree, the greater the role of the indicator in the evaluation, that is, the greater the weight should be;

[0170] The conditional unit is also used to calculate the weight of each constraint indicator in the constraint set according to the size of the information entropy using the entropy weight method formula to obtain an initial weight value set.

[0171] The conditional unit is also used to calculate a weight set according to an initial weight value set and a number of weight values ​​corresponding to a number of membership functions.

[0172] The condition unit is also used to perform weighted centralized processing on the light intensity constraint, water temperature constraint and spatial distribution constraint according to the weight set to obtain the constraint condition.

[0173] The condition unit of this embodiment uses fuzzy sets to represent the satisfaction degree of each constraint, which can handle the uncertainty of the constraint conditions more flexibly. Establishing a membership function can quantify the satisfaction degree of each constraint condition, thereby providing more accurate information for subsequent processing and decision-making. Since the information entropy reflects the amount of information contained in the constraint condition, and the entropy weight method can allocate weights according to the amount of information, this makes the weight allocation of the initial weight value set more reasonable and objective. Combined with the weight value corresponding to the membership function, the weight set can be further adjusted and optimized so that the final weight allocation is more in line with the actual situation and needs.

[0174] The function unit is used to establish the objective function based on the ecological health index by quantifying the degree to which the actual breeding efficiency is close to the optimal state and the degree to which the actual resource utilization is close to the optimal state.

[0175] Among them, the objective function is:

[0176]

[0177] in, , , and It is a preset weight coefficient used to balance the relative importance of different objectives; Indicates that at a given stocking density , ecological factors and time The breeding efficiency under represents the upper limit of aquaculture efficiency under optimal conditions (i.e. without considering the limitation of aquaculture density), Indicates that at a given stocking density , Resource Factor and time The resource utilization rate under represents the upper limit of resource utilization under optimal conditions, Indicates that at a given stocking density , ecological factors , Resource Factor , spatial distribution parameters and time Ecological health index under Indicates that at a given stocking density , cost factor and time The breeding costs.

[0178] The objective function of the function unit in this embodiment is centered on the ecological health index, which reflects the concern and protection of the ecological environment; by optimizing aquaculture efficiency and resource utilization, it aims to achieve a win-win situation of economic and ecological benefits and promote the sustainable development of the aquaculture industry.

[0179] In one embodiment, the solution module 30 includes a student unit, a teacher unit, a training unit, and a solution unit;

[0180] The student unit is used to use sensors or monitoring equipment to regularly collect data on ecological factors such as light intensity, water temperature, and nutrient concentration in the breeding area, as well as growth data samples of Gracilaria;

[0181] The student unit is also used to clean the collected raw data, remove outliers, missing values ​​and other data that do not meet the requirements, normalize the data so that the numerical ranges of different ecological factors are on the same scale, and obtain a data set.

[0182] The student unit is also used to establish a first loss function based on the data set, according to the difference between the actual breeding density and the breeding density predicted by the model, by penalizing the square or absolute value of the parameter error;

[0183] The student unit is further used to penalize the square of the parameter error according to a preset method to establish a second loss function; wherein the preset method is determined according to the first loss function, and the preset method is used to maintain the consistency of the loss function;

[0184] The student unit is also used to map the growth data samples of Gracilaria from the original space to the reproducing kernel Hilbert space, and establish a third loss function based on the error distribution of the growth data samples in the reproducing kernel Hilbert space;

[0185] The student unit is also used to control the intensity of the gradient penalty term by introducing a gradient penalty coefficient and to establish the fourth loss function in combination with the gradient norm;

[0186] The student unit is also used to establish a comprehensive loss function based on the first loss function, the second loss function, the third loss function and the fourth loss function.

[0187] The student unit is also used to train several preset base learners (such as decision trees, random forests, gradient boosting trees, etc.) according to the data set and the comprehensive loss function, so that they can minimize the comprehensive loss function, thereby improving the prediction performance and obtaining the base learner set;

[0188] The student unit is also used to integrate the trained base learner set using bagging, boosting or other integrated learning methods to construct an integrated version of the student model to obtain the final student model.

[0189] Among them, the first loss function is:

[0190]

[0191] The second loss function is:

[0192]

[0193] The third loss function is:

[0194]

[0195] The fourth loss function is:

[0196]

[0197] in, is the actual stocking density, is the stocking density predicted by the model, is a threshold parameter.

[0198] is the sample size, is the optimal stocking density obtained through linear programming, Represents an index;

[0199] is the mapping function that maps data to the Reproducing Kernel Hilbert Space (RKHS), is the norm in the reproducing kernel Hilbert space, represents samples drawn from the true data distribution, Represented by the generator Generated samples;

[0200] is the gradient penalty coefficient, Indicates the sample The expected value of Denotes the discriminator D for the input sample The L2 norm of the gradient of .

[0201] In the student unit of this embodiment, the first loss function, which is an error-based penalty mechanism, can ensure that the model continuously reduces the prediction error during the training process, thereby improving the reliability of the prediction results. The second loss function determines the preset method based on the first loss function, which is used to penalize the square of the parameter error. This approach maintains the consistency between the loss functions, so that the model can balance the influence of different loss terms during the optimization process and avoid overfitting or underfitting problems. The third loss function is established by mapping the growth data samples of Gracilaria to the reproducing kernel Hilbert space, which can capture the nonlinear relationship between data samples and improve the model's ability to handle complex data, thereby enhancing the generalization ability of the model. The fourth loss function is established by introducing a gradient penalty coefficient to control the intensity of the gradient penalty term. This mechanism can prevent the model from generating excessive gradients during training and maintain the stability of training.

[0202] The teacher unit is used to establish a teacher model based on the stocking density, biological species matching and environmental control parameters in shelf-type multi-layer stocking and floor-type three-dimensional stocking;

[0203] Among them, "shelf-type multi-layer farming" is an innovative farming model, the core design concept of which is to construct the farming facilities into a three-dimensional structure similar to a shelf. In this model, multiple farming units (such as farming ponds or farming cages) are cleverly stacked vertically to form a multi-layer farming system that efficiently utilizes space;

[0204] "Floor-type three-dimensional farming" is a highly intensive farming model that integrates the design principles of building floors and divides the farming space vertically into multiple independent floors or functional areas. Each floor or area has the conditions and capabilities for independent farming operations.

[0205] The teacher model of the teacher unit in this embodiment is designed for specific breeding modes and can more accurately reflect the actual conditions under these breeding modes. This targeting makes the model more accurate and reliable in predicting and optimizing breeding density, biological species matching and environmental control parameters.

[0206] A training unit is used to design a generator and a plurality of discriminators; wherein the generator is used to generate distribution characteristics of the aquaculture density of Gracilaria and target marine organisms close to that presented by the teacher model; the plurality of discriminators are used to identify and distinguish the difference between the density distribution generated by the generator and the actual density distribution of the teacher model, and each discriminator adopts a convolutional neural network architecture, but has independent parameters;

[0207] The training unit is also used to train several discriminators based on the generative adversarial network according to the real density distribution data of the teacher model and the fake data generated by the generator when the parameters of the generator are fixed; the generator is trained with the goal of making the density distribution generated by the generator closer to the real density distribution of the teacher model when the parameters of several discriminators are fixed; the density samples generated by the generator are used as training data of the student model, and the student model is trained to approximate the output of the teacher model; and a consistency regularization term is introduced to encourage the judgment of consistency of the discriminator output at different scales;

[0208] The training unit is also used to iterate the alternating training process of the generator and several discriminators until a predetermined number of training times is reached or a preset stop condition is met, so as to obtain the trained generator and student model, and form an ecological farming model based on the trained generator and student model; wherein, for the ecological farming model, the generator is responsible for generating density distribution samples, and the student model is responsible for making predictions or decisions based on these samples;

[0209] Among them, the combined loss function of several discriminators is established based on the loss functions of different discriminators;

[0210] Among them, the combined loss function of several discriminators is:

[0211] ;

[0212]

[0213] For a pair of discriminators among several discriminators , and its consistency regularization term is:

[0214]

[0215] For all discriminators of several discriminators, the average consistency regularization term is:

[0216]

[0217] The loss function of the generator is:

[0218]

[0219] in, represents the combined loss function, Indicates The loss function of the discriminator is Represents samples sampled from the real data distribution After Discriminator The expected output after represents the noise sampled from the noise distribution Through the generator Generated samples After the Discriminator The expected output after is the gradient penalty term, Indicates Discriminator In the interpolation sample The gradient at is the gradient penalty coefficient, is the interpolation between the real sample and the generated sample, It is The weight of the discriminator, is the total number of discriminators;

[0220] and Respectively represent the outputs of the two discriminators for the same input sample, Represents a sample randomly drawn from the joint distribution of real samples and generated samples;

[0221] is from The number of combinations of two pairs selected from the discriminators is is the weight;

[0222] Representation Generator Input noise The output, Indicates The discriminator is used to analyze the samples generated by the generator. Rating, Represents the noise vector From the prior distribution The expected value of the sample.

[0223] Through continuous training of the training unit in this embodiment, the generator can generate fake data that is closer and closer to the real distribution, which is of great significance for simulating and understanding the dynamic changes of biological density in aquaculture ecosystems. In addition, through training, the discriminator can continuously improve its recognition ability, thereby more effectively guiding the training process of the generator and ensuring that the generated density distribution gradually approaches the real distribution.

[0224] A solution unit is used to randomly generate a stocking density solution that meets the constraint conditions, and obtain a first stocking density as a starting point for optimization;

[0225] The solving unit is further used to generate a new solution starting from the first stocking density by adding random perturbations to the linear programming problem about the constraints and the objective function, so as to obtain the second stocking density; wherein the "random perturbations" may be a slight adjustment to the decision variable (i.e., the stocking density) or a slight change in the coefficient of the objective function;

[0226] The solving unit is further used to calculate the first objective function value and the second objective function value according to the first breeding density and the second breeding density respectively;

[0227] The solution unit is also used to compare the first objective function value and the second objective function value: ① If the first objective function value is less than the second objective function value, it means that the second breeding density performs better on the objective function, then the second breeding density is used as the current solution, and the current solution is updated through data iteration, and the current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density; ② If the first objective function value is greater than or equal to the second objective function value, then it is determined whether to accept the second objective function value as a new solution based on a probability function related to the difference between the current temperature and the objective function value, so as to jump out of the local optimal solution and continue to explore the global optimal solution, and the current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density;

[0228] The solving unit is also used to generate the optimal breeding density of Gracilaria and target marine organisms according to the ecological breeding model based on the constraints, objective function and auxiliary optimal breeding density, and generate a three-dimensional ecological breeding plan of Gracilaria according to the optimal breeding density; wherein the auxiliary optimal breeding density is used as a starting point to provide a basis and direction for subsequent optimization, specifically: the ecological breeding model searches from the auxiliary optimal breeding density to find the optimal breeding density that meets the constraints and objective function.

[0229] The solution unit of this embodiment starts iterating by randomly generating a farming density solution that satisfies the constraints, which increases the diversity of the search space and helps to explore more possible solutions. By adding random perturbations to the linear programming problem about the constraints and the objective function to generate new solutions, this method can introduce changes based on the current solution, thereby improving the quality of the solution.

[0230] Overall, this embodiment has the following beneficial effects:

[0231] This application can consider both economic and ecological benefits by incorporating resource utilization and ecological health index into the objective function and weighting and concentrating them; this comprehensive consideration helps to reduce resource waste and environmental pollution while pursuing high yields, thereby improving the sustainability of aquaculture. A three-dimensional ecological aquaculture plan for Gracilaria is generated based on the optimal aquaculture density. This plan can make full use of aquaculture space and water resources, increase aquaculture density and yield, while reducing negative impacts on the environment and achieving sustainable development. For the ecological aquaculture model, an ecological aquaculture model based on an adversarial learning mechanism is adopted. This model simulates and approximates the density distribution of the teacher model through iterative training of the generator, so as to capture the complex growth laws of Gracilaria and target marine organisms under different aquaculture conditions; at the same time, the student model is trained according to the generated density samples, which further improves the model's prediction ability and generalization performance. The student model is obtained by training several basis learners based on a comprehensive loss function. The comprehensive loss function takes into account the sample error distribution in the reproducing kernel Hilbert space, which enables the model to better adapt to the complex and changeable aquaculture environment and improve the accuracy and robustness of the prediction;

[0232] In summary, the three-dimensional ecological breeding scheme of the present application can reduce the negative impact on the environment and promote ecological balance; and, by reasonably controlling the breeding density and species combination, it can avoid competition and conflict between organisms, reduce the occurrence and spread of diseases, and improve the breeding efficiency of Gracilaria, thereby solving the problem of difficulty in improving the breeding efficiency of Gracilaria.

[0233] Embodiment three:

[0234] The embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the three-dimensional ecological cultivation optimization method of the seaweed Gracilaria;

[0235] Wherein, the three-dimensional ecological cultivation optimization method of the seaweed Gracilaria, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0236] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A three-dimensional ecological cultivation optimization method for seaweed Gracilaria, characterized in that: include: Obtain growth data of Gracilaria and target marine organisms under different culture conditions; Establishing constraint conditions and an objective function based on the growth data; wherein the objective function is established by weighted centralization of resource utilization and ecological health index; Based on the constraints and the objective function, the optimal breeding density of the Gracilaria and the target marine organisms is generated according to the ecological breeding model, and the three-dimensional ecological breeding plan of the Gracilaria is generated according to the optimal breeding density; wherein the ecological breeding model is based on an adversarial learning mechanism, and is obtained by iteratively training a generator to simulate and approximate the density distribution of a teacher model, and training a student model according to the generated density samples; the student model is obtained by training several basis learners according to a comprehensive loss function, and the comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space.

2. The three-dimensional ecological cultivation optimization method of a seaweed Gracilaria as claimed in claim 1, characterized in that: The objective function is established by weighting resource utilization and ecological health index, specifically: Based on the growth data, an attenuation coefficient is introduced to simulate the phenomenon that the light intensity gradually decreases, and a light intensity constraint is established in combination with a preset minimum value and a maximum value of the appropriate light intensity; Water temperature constraints are established based on the sensitivity of growth rate to temperature changes and a preset growth rate threshold; Establish spatial distribution constraints based on light intensity and water temperature at different geographical locations; The light intensity constraint, the water temperature constraint and the spatial distribution constraint are weighted and concentrated according to a weight set to obtain the constraint condition; wherein the weight set is established by using fuzzy logic to quantify the satisfaction degree of different constraints.

3. The three-dimensional ecological cultivation optimization method of a seaweed Gracilaria as claimed in claim 2, characterized in that: The weight set is established by using fuzzy logic to quantify the satisfaction of different constraints, specifically: Based on the constraint set, fuzzy sets are defined respectively to represent the satisfaction degree of each constraint, and a plurality of fuzzy sets are obtained; wherein the constraint set includes the light intensity constraint, the water temperature constraint and the spatial distribution constraint; Establishing a number of membership functions based on the fuzzy sets; The weight set is calculated based on the initial weight value set and the weight values ​​corresponding to the membership functions; wherein the initial weight value set is calculated based on the information entropy corresponding to the constraint set according to the entropy weight method.

4. The three-dimensional ecological cultivation optimization method of a seaweed Gracilaria according to claim 1, characterized in that: The comprehensive loss function is established based on the sample error distribution in the reproducing kernel Hilbert space, specifically: According to the difference between the actual stocking density and the stocking density predicted by the model, the first loss function is established by penalizing the square or absolute value of the parameter error; Penalizing the square of the parameter error according to a preset method to establish a second loss function; wherein the preset method is determined based on the first loss function, and the preset method is used to maintain the consistency of the loss function; Mapping the growth data samples of Gracilaria from the original space to the reproducing kernel Hilbert space, and establishing a third loss function based on the growth data sample error distribution in the reproducing kernel Hilbert space; The intensity of the gradient penalty term is controlled by introducing the gradient penalty coefficient, and the fourth loss function is established in combination with the gradient norm; The comprehensive loss function is established according to the first loss function, the second loss function, the third loss function and the fourth loss function.

5. The three-dimensional ecological cultivation optimization method of the seaweed Gracilaria as claimed in claim 1, characterized in that: The ecological farming model is based on an adversarial learning mechanism, which is obtained by iteratively training the generator to simulate and approximate the density distribution of the teacher model, and training the student model according to the generated density samples, specifically: Design a generator and several discriminators; wherein the generator is used to generate distribution characteristics of the aquaculture density of Gracilaria and target marine organisms that are close to those presented by the teacher model, and the several discriminators are used to identify and distinguish the difference between the density distribution generated by the generator and the actual density distribution of the teacher model; Under the condition that the parameters of the generator are fixed, the discriminators are trained according to the real density distribution data of the teacher model and the fake data generated by the generator; under the condition that the parameters of the discriminators are fixed, the generator is trained with the goal of making the density distribution generated by the generator closer to the real density distribution of the teacher model; the density samples generated by the generator are used as the training data of the student model, and the student model is trained to approximate the output of the teacher model; The alternating training process of the generator and the plurality of discriminators is iterated to obtain the ecological farming model.

6. The three-dimensional ecological cultivation optimization method of the seaweed Gracilaria as claimed in claim 5, characterized in that: The combined loss function of the plurality of discriminators is established based on the loss functions of different discriminators; Among them, the combined loss function is ： ； ； in, represents the combined loss function, Indicates The loss function of the discriminator is Represents samples sampled from the real data distribution After Discriminator The expected output after represents the noise sampled from the noise distribution Through the generator Generated samples After the Discriminator The expected output after is the gradient penalty term, Indicates Discriminator In the interpolation sample The gradient at is the gradient penalty coefficient, is the interpolation between the real sample and the generated sample, It is The weight of the discriminator, is the total number of discriminators.

7. The three-dimensional ecological cultivation optimization method of the seaweed Gracilaria as claimed in claim 1, characterized in that: Based on the constraints and the objective function, the optimal culture density of the Gracilaria and the target marine organisms is generated according to the ecological culture model, specifically: Randomly generate a breeding density solution that satisfies the constraint condition to obtain a first breeding density; Based on the first breeding density, generating a new solution by adding random perturbations to the linear programming problem regarding the constraints and the objective function, thereby obtaining a second breeding density; According to the first breeding density and the second breeding density, respectively calculating a first objective function value and a second objective function value; If the first objective function value is less than the second objective function value, the second breeding density is used as the current solution, and the current solution is updated through data iteration, and the current solution when the maximum number of iterations is reached is defined as the auxiliary optimal breeding density; Based on the constraint conditions, the objective function and the auxiliary optimal breeding density, the optimal breeding density of the Gracilaria and the target marine organisms is generated according to the ecological breeding model.

8. The three-dimensional ecological cultivation optimization method of the seaweed Gracilaria as claimed in claim 1, characterized in that: The teacher model is established based on the breeding density, biological species matching and environmental control parameters in shelf-type multi-layer breeding and floor-type three-dimensional breeding.

9. The three-dimensional ecological cultivation optimization method of the seaweed Gracilaria as claimed in claim 1, characterized in that: The objective function is established based on the ecological health index by quantifying the degree to which the actual breeding efficiency is close to the optimal state and the degree to which the actual resource utilization is close to the optimal state.

10. A three-dimensional ecological cultivation optimization method for Gracilaria seaweed according to any one of claims 1 to 9, characterized in that: The objective function is specifically: ； in, , , and is the preset weight coefficient, Indicates that at a given stocking density , ecological factors and time The breeding efficiency under represents the upper limit of farming efficiency under optimal conditions, Indicates that at a given stocking density , Resource Factor and time The resource utilization rate under represents the upper limit of resource utilization under optimal conditions, Indicates that at a given stocking density , ecological factors , Resource Factor , spatial distribution parameters and time Ecological health index under Indicates that at a given stocking density , cost factor and time The breeding costs.

Images