Fusion method of water quality regulation and control and disease control in seedling cultivation

Abstract

The invention relates to the field of seedling cultivation, and discloses a method for fusing water quality regulation and disease control in seedling cultivation, which comprises the following steps of: acquiring multi-field coupling space-time continuous data of a water quality field, a biological and physiological field, a medicament diffusion field and an environment auxiliary field in a seedling cultivation pond; performing fractional order multi-field data preprocessing on the multi-field coupling space-time continuous data to obtain a feature vector containing a long-time memory feature and a coupling feature; and constructing a fractional order LSTM-SPDE coupling prediction model formed by coupling the fractional order LSTM and a random partial differential equation based on the feature vector, and predicting water quality evolution data in a future preset duration by taking the real-time feature vector as input. Multi-field coupling space-time continuous data is acquired, information containing long-term memory features and multi-field coupling features is accurately extracted through fractional order multi-field data preprocessing, and the future water quality evolution trend and disease risk level are pre-judged in advance, so that dynamic cooperation, pre-judgment and long-term adaptation of water quality regulation and control and disease prevention and control are realized.

Description

Technical Field

[0001] This invention relates to the field of seedling cultivation technology, specifically to a method for integrating water quality control and disease prevention in seedling cultivation. Background Technology

[0002] Seedling cultivation is a fundamental step in aquaculture, and its quality directly determines the economic and ecological benefits of subsequent aquaculture production, occupying a crucial position in the aquaculture industry chain. As aquaculture transforms towards large-scale, refined, and green practices, the requirements for water quality stability and timely disease control during seedling cultivation are becoming increasingly stringent. Water quality, as a core environmental factor for seedling growth, directly affects the physiological metabolism and immune function of seedlings through the dynamic balance of parameters such as dissolved oxygen concentration, pH value, and ammonia nitrogen concentration. Disease control is a key means to avoid large-scale seedling mortality and ensure survival rates; together, these two aspects constitute the core system of risk control in the seedling cultivation process.

[0003] Currently, in the existing seedling cultivation process, water quality control mainly involves maintaining water quality parameters within a preset range through regular water changes, mechanical aeration, and the addition of chemical regulators. In terms of disease prevention and control, the probability of disease occurrence is reduced by regularly feeding antibiotics and spraying disinfectants throughout the pond. However, the timing of these operations is disconnected from the control objectives, failing to effectively capture the dynamic transmission relationship between changes in water quality parameters and the physiological state of seedlings and the risk of disease occurrence. Water quality control measures do not consider their auxiliary role in disease prevention and control, and disease prevention and control methods do not take into account water quality stability. This easily leads to the contradiction of neglecting disease prevention while adjusting water quality, and damaging water quality while controlling disease, resulting in low control effectiveness. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for integrating water quality control and disease prevention in seedling cultivation. This method resolves the problem that existing seedling cultivation processes often result in a contradiction between neglecting disease prevention during water management and the damage to water quality caused by disease prevention, leading to low control effectiveness.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for integrating water quality control and disease prevention in seedling cultivation, comprising the following steps: Acquire continuous spatiotemporal data of multiple fields coupled together, including water quality field, biological physiological field, reagent diffusion field, and environmental auxiliary field, within the seedling pond; Fractional-order multi-field data preprocessing is performed on multi-field coupled spatiotemporal continuous data to obtain feature vectors containing long-term memory features and coupling features; A fractional LSTM-SPDE coupled prediction model is constructed based on feature vectors and coupled with stochastic partial differential equations. The real-time feature vectors are used as input to predict water quality evolution data, seedling physiological status data and disease risk level within a preset time period, and the prediction data is obtained. Based on fractional-order control theory, a multi-objective optimization function is constructed using predicted data, and a fractional-order fusion control strategy is formulated in combination with preset constraints. A fractional-order fusion control strategy is implemented in the seedling pond, and the measured data in the seedling pond after implementation are collected. The deviation between the measured data and the predicted data is calculated. The parameters of the fractional-order LSTM-SPDE coupled prediction model and the fractional-order fusion control strategy are optimized through closed-loop feedback iteration.

[0006] By adopting the above technical solution, multi-field coupled spatiotemporal continuous data is acquired, and information containing long-term memory features and multi-field coupling features is accurately extracted through fractional-order multi-field data preprocessing. Then, the dynamic transmission relationship between water quality, physiology, and disease is quantified through a fractional-order LSTM-SPDE coupled prediction model, which predicts the future water quality evolution trend and disease risk level in advance. Based on fractional-order control theory, a multi-objective optimization function is constructed, and a synergistic regulation strategy that takes into account water quality stability, disease prevention and control, cost optimization, and seedling safety is formulated in combination with constraints. The model parameters and regulation scheme are continuously corrected through closed-loop feedback iteration, thereby realizing the dynamic synergy, prediction, and long-term adaptation of water quality regulation and disease prevention and control. This solves the problem that in the seedling cultivation process, water regulation may neglect disease prevention, and disease prevention may damage water quality, resulting in low prevention and control effects.

[0007] Preferably, the water quality field data includes dissolved oxygen concentration, pH value, ammonia nitrogen concentration, nitrite concentration and total alkalinity; the biological physiological field data includes seedling swimming speed, feeding frequency, superoxide dismutase activity and stress hormone concentration; the drug diffusion field data includes probiotic concentration and traditional Chinese medicine extract concentration; and the environmental auxiliary field data includes water temperature, light intensity, water flow velocity and oxygenation intensity.

[0008] Preferably, the fractional-order multi-field data preprocessing specifically includes the following steps: The Caputo fractional difference method is used to denoise time series data in multi-field coupled spatiotemporal continuous data, while preserving the long-term memory features of the data to obtain feature data. Based on the fractional covariance function, the Kriging interpolation method is used to complete the spatial gap data in the feature data and generate a three-dimensional global data distribution map. Fractional cross-correlation analysis was performed on the three-dimensional global data distribution map to screen out coupling features, and feature vectors were constructed based on the coupling features. The coupling features include water quality-physiology coupling features, water quality-pharmaceutical coupling features, and physiology-pharmaceutical coupling features.

[0009] Preferably, obtaining the predicted data specifically includes the following steps: Based on the long-term memory characteristics and multi-field coupling correlation characteristics of eigenvectors, and by incorporating the Caputo fractional derivative and the spatial Laplace operator into the gating mechanism, a fractional LSTM unit is constructed. Based on the multi-field coupled state vector containing the water quality field, biological physiological field and the drug diffusion field in the seedling pond, a fractional-order stochastic partial differential equation is constructed. The fractional-order stochastic partial differential equations are spatially discretized using the finite element method and time-discretized using the Milstein scheme, resulting in discretized fractional-order stochastic partial differential equations. Fractional LSTM units are coupled with discretized fractional stochastic partial differential equations to form a fractional LSTM-SPDE coupled prediction model. The feature vector is input into the fractional LSTM unit in the fractional LSTM-SPDE coupled prediction model to extract the spatial-temporal fusion features. The spatial-temporal fusion features are input into the discretized fractional-order stochastic partial differential equations in the fractional-order LSTM-SPDE coupled prediction model. The prediction data, which includes water quality evolution data, seedling physiological status data, and disease risk level, are obtained through numerical solution.

[0010] Preferably, the disease risk level is determined by using the Sigmoid function to convert the seedling physiological state data into disease risk values ​​within a preset range, and then dividing the disease risk values ​​into several disease risk levels according to a preset threshold.

[0011] Preferably, the multi-objective optimization function includes a disease risk control objective, a water quality stability objective, a regulation cost control objective, and a seedling stress control objective. The disease risk control objective corresponds to the disease risk level in the predicted data, the water quality stability objective corresponds to the deviation between the water quality evolution data in the predicted data and the preset optimal water quality range, the regulation cost control objective corresponds to the regulation input parameters, and the seedling stress control objective corresponds to the seedling stress index in the predicted data.

[0012] Preferably, the preset constraints include hard constraints and soft constraints. The hard constraints include upper limits for drug dosage, upper limits for equipment operating parameters, and water quality safety thresholds. The soft constraints include upper limits for seedling stress index and upper limits for single regulation cost.

[0013] Preferably, the formulation of the fractional-order fusion control strategy specifically includes the following steps: Based on the Pontryagin minimum principle in fractional control theory, a costate variable is introduced, and a multi-objective optimization function and preset constraints are incorporated to construct a fractional Hamiltonian function. Construct a fractional-order inverse costate equation based on the fractional-order Hamiltonian function, and obtain the control law by solving the fractional-order inverse costate equation; The fractional-order fusion control strategy is generated based on the control law. The fractional-order fusion control strategy includes the dosage of probiotics, the dosage of traditional Chinese medicine extracts, the oxygenation intensity, and the water exchange flow rate.

[0014] Preferably, the execution of the fractional-order fusion control strategy specifically includes the following steps: The fractional-order fusion control strategy is converted into a control signal that can be recognized by IoT devices, including a drug delivery device, a variable frequency oxygenator, and a smart water exchange valve. The disease risk level is used to implement a graded control mechanism, including a combination of automatic control and manual confirmation, a combination of automatic execution and real-time monitoring, and a combination of emergency execution and manual intervention.

[0015] Preferably, the closed-loop feedback iterative optimization specifically includes the following steps: Collect measured data in the seedling pond after execution; the measured data is of the same type as multi-field coupled spatiotemporal continuous data. Calculate the deviation vector between the measured data and the predicted data, where the deviation vector is the difference between the measured data and the predicted data; Using the bias vector as input, the parameters of the fractional LSTM-SPDE coupled prediction model are solved by the projection shrinkage algorithm based on the stochastic variational inequality. By using the online Bayesian update method, the solved parameters are substituted into the fractional LSTM-SPDE coupled prediction model to correct the coupling coefficient matrix and the random fluctuation coefficient. Predictive data is regenerated based on the modified fractional-order LSTM-SPDE coupled prediction model, and the fractional-order fusion control strategy is updated iteratively by combining the deviation vector.

[0016] This invention provides a method for integrating water quality control and disease prevention in seedling cultivation. It has the following beneficial effects: 1. This invention acquires multi-field coupled spatiotemporal continuous data, accurately extracts information containing long-term memory features and multi-field coupling features through fractional-order multi-field data preprocessing, and then quantifies the dynamic transmission relationship between water quality, physiology, and disease through a fractional-order LSTM-SPDE coupled prediction model. This allows for the prediction of future water quality evolution trends and disease risk levels. Based on fractional-order control theory, a multi-objective optimization function is constructed, and a synergistic control strategy that balances water quality stability, disease prevention and control, cost optimization, and seedling safety is formulated in combination with constraints. Through closed-loop feedback iteration, the model parameters and control scheme are continuously corrected, thereby achieving dynamic synergy, prediction, and long-term adaptation between water quality control and disease prevention and control, and improving the overall prevention and control effect.

[0017] 2. This invention incorporates fractional calculus theory into the prediction model to capture the long-term memory effect of water quality parameters and the fractional response characteristics of seedling physiological state. It combines stochastic partial differential equations to characterize the stochastic fluctuation characteristics of multi-field coupling, thereby reducing prediction errors and effectively extending the early warning time window. This allows for the early prediction of potential disease risks, giving operators sufficient time to take mild and efficient prevention and control measures, avoiding passive responses after disease outbreaks, and reducing seedling loss rates.

[0018] 3. This invention comprehensively covers disease risk control, water quality stability, regulation costs, and seedling stress through a multi-objective optimization function. It ensures the feasibility and safety of the regulation strategy by pre-setting hard and soft constraints, and solves the optimal control law by using the Pontryagin minimum principle. The generated regulation strategy can effectively reduce disease risk and maintain water quality stability while reducing ineffective consumption of chemicals and seedling stress, avoiding cost waste and seedling damage caused by over-regulation, and improving the economic efficiency and quality of seedling cultivation. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method for integrating water quality control and disease prevention in seedling cultivation proposed in this invention; Figure 2 This is a diagram of the integrated system for water quality control and disease prevention in seedling cultivation proposed in Embodiment 2 of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: In the first embodiment of the present invention, the present invention provides a method for integrating water quality control and disease prevention in seedling cultivation, such as... Figure 1 As shown, it includes the following steps: Acquire continuous spatiotemporal data of multiple fields coupled together, including water quality field, biological physiological field, reagent diffusion field, and environmental auxiliary field, within the seedling pond; Furthermore, the water quality field data includes dissolved oxygen concentration, pH value, ammonia nitrogen concentration, nitrite concentration and total alkalinity; the biological and physiological field data includes seedling swimming speed, feeding frequency, superoxide dismutase activity and stress hormone concentration; the drug diffusion field data includes probiotic concentration and traditional Chinese medicine extract concentration; and the environmental auxiliary field data includes water temperature, light intensity, water flow velocity and oxygenation intensity.

[0022] Specifically, it is necessary to acquire continuous spatiotemporal data on multiple coupled fields within the seedling rearing pond, including water quality, biophysiological fields, reagent diffusion fields, and environmental auxiliary fields, to provide foundational data support for constructing multi-field coupled analysis. Water quality data includes dissolved oxygen concentration, pH value, ammonia nitrogen concentration, nitrite concentration, and total alkalinity. Biophysiological field data includes seedling swimming speed, feeding frequency, superoxide dismutase activity, and stress hormone concentration. Reagent diffusion field data can include probiotic concentration and traditional Chinese medicine extract concentration. Environmental auxiliary field data includes water temperature, light intensity, water flow velocity, and oxygenation intensity. Data is collected at a fixed time frequency by deploying corresponding sensors at a preset density in the bottom, middle, and surface layers of the seedling rearing pond, ensuring comprehensive and continuous data coverage.

[0023] Fractional-order multi-field data preprocessing is performed on multi-field coupled spatiotemporal continuous data to obtain feature vectors containing long-term memory features and coupling features; Further, fractional-order multi-field data preprocessing is performed, specifically including the following steps: The Caputo fractional difference method is used to denoise time series data in multi-field coupled spatiotemporal continuous data, while preserving the long-term memory features of the data to obtain feature data. Based on the fractional covariance function, the Kriging interpolation method is used to complete the spatial gap data in the feature data and generate a three-dimensional global data distribution map. Fractional cross-correlation analysis was performed on the three-dimensional global data distribution map to screen out coupling features, and feature vectors were constructed based on the coupling features. The coupling features include water quality-physiology coupling features, water quality-pharmaceutical coupling features, and physiology-pharmaceutical coupling features.

[0024] Specifically, in order to fully preserve the long-term memory features in multi-field coupled spatiotemporal continuous data and remove noise interference, while mining the coupling correlation between multiple fields, fractional-order multi-field data preprocessing is required to obtain feature vectors from the multi-field coupled spatiotemporal continuous data.

[0025] First, the Caputo fractional difference method is used to process the time series data. Its expression is: ,in, For the denoised first Time-series feature data, The coefficients are fractional binomial coefficients. For the first stage before preprocessing Multi-field time series data, This is the fractional-order memory order. Using ammonia nitrogen time-series data from the water quality field as input, the denoised ammonia nitrogen feature data is calculated using this formula, achieving noise removal and long-term memory feature retention.

[0026] Based on fractional covariance function The spatial gap data is completed using Kriging interpolation. For space monitoring points and Fractional covariance between For data variance, Generate a three-dimensional global data distribution map for the variable range parameters.

[0027] Through fractional cross-correlation coefficients Filter coupling features, among which, for and The fractional cross-correlation coefficient, For water quality field data, For biological physiological field data, For the time lag, and Fractional covariance and variance were used to select three types of coupling features: water quality-physiology, water quality-pharmaceutical, and physiology-pharmaceutical. Based on these features, feature vectors containing long-term memory and coupling features were constructed to provide input for subsequent coupling prediction models.

[0028] A fractional LSTM-SPDE coupled prediction model is constructed based on feature vectors and coupled with stochastic partial differential equations. The real-time feature vectors are used as input to predict water quality evolution data, seedling physiological status data and disease risk level within a preset time period, and the prediction data is obtained. Furthermore, obtaining the predicted data specifically includes the following steps: Based on the long-term memory characteristics and multi-field coupling correlation characteristics of eigenvectors, and by incorporating the Caputo fractional derivative and the spatial Laplace operator into the gating mechanism, a fractional LSTM unit is constructed. Based on the multi-field coupled state vector containing the water quality field, biological physiological field and the drug diffusion field in the seedling pond, a fractional-order stochastic partial differential equation is constructed. The fractional-order stochastic partial differential equations are spatially discretized using the finite element method and time-discretized using the Milstein scheme, resulting in discretized fractional-order stochastic partial differential equations. Fractional LSTM units are coupled with discretized fractional stochastic partial differential equations to form a fractional LSTM-SPDE coupled prediction model. The feature vector is input into the fractional LSTM unit in the fractional LSTM-SPDE coupled prediction model to extract the spatial-temporal fusion features. The spatial-temporal fusion features are input into the discretized fractional-order stochastic partial differential equations in the fractional-order LSTM-SPDE coupled prediction model. The prediction data, which includes water quality evolution data, seedling physiological status data, and disease risk level, are obtained through numerical solution.

[0029] Furthermore, the disease risk level is determined by using the Sigmoid function to convert the seedling physiological state data into disease risk values ​​within a preset range, and then dividing the disease risk values ​​into several disease risk levels according to a preset threshold.

[0030] Specifically, in order to predict the evolution of water quality, seedling physiological status, and disease risk within a preset time period, a fractional-order LSTM-SPDE coupled prediction model is constructed based on the feature vector containing long-term memory features and coupling features obtained from the aforementioned preprocessing. The prediction data is obtained by using the real-time feature vector as input.

[0031] Generally, when constructing fractional LSTM units, it is necessary to combine the long-term memory characteristics and multi-field coupling correlation characteristics of feature vectors, and incorporate the Caputo fractional derivative and the spatial Laplacian operator into the gating mechanism. The forgetting gate expression is as follows: ,in, Output for the forget gate. It is the Sigmoid activation function. This is the weight matrix. The state was hidden in the previous moment. For feature vectors, For bias vectors, for Caputo fractional derivative, For spatial feature weighting coefficients, This is a spatial Laplacian operator for multi-field coupled state vectors. Using a feature vector composed of water quality-physiological coupling features and water quality-pharmaceutical coupling features as input, this forgetting gate is used to calculate and retain key historical features, achieving collaborative extraction of long-term memory and spatially coupled features.

[0032] In some embodiments, a multi-field coupled state vector containing water quality field data, biological physiological field data, and reagent diffusion field data is used. Construct fractional-order stochastic partial differential equations ,in, for Caputo fractional derivative, For fractional-order elliptic operators, For random fluctuation coefficients, This is the increment for the Wiener process. The fractional-order stochastic partial differential equations are spatially discretized using the finite element method, and time-discretized using the Milstein scheme to obtain the discretized equations. The output of the fractional-order LSTM unit is connected to the input of the discretized equations to form a complete coupled prediction model.

[0033] The feature vector is input into a fractional-order LSTM unit to extract spatial-temporal fusion features. These features are then fed into a discretized fractional-order stochastic partial differential equation, and water quality evolution data and seedling physiological state data are obtained through numerical solution. The sigmoid function is used to transform the seedling physiological state data into disease risk values ​​within a preset interval; its expression is as follows: ,in, This represents the disease risk value. For the mapping weight matrix, For seedling physiological status data, As a bias term, the disease risk value is divided into several disease risk levels according to a preset threshold. Combined with water quality evolution data and seedling physiological status data, predictive data is formed to provide a basis for subsequent formulation of fractional-order optimal fusion control strategies.

[0034] Based on fractional-order control theory, a multi-objective optimization function is constructed using predicted data, and a fractional-order fusion control strategy is formulated in combination with preset constraints. Furthermore, the multi-objective optimization function includes disease risk control objectives, water quality stability objectives, regulation cost control objectives, and seedling stress control objectives. The disease risk control objective corresponds to the disease risk level in the predicted data, the water quality stability objective corresponds to the deviation between the water quality evolution data in the predicted data and the preset optimal water quality range, the regulation cost control objective corresponds to the regulation input parameters, and the seedling stress control objective corresponds to the seedling stress index in the predicted data.

[0035] Furthermore, the preset constraints include hard constraints and soft constraints. Hard constraints include upper limits for reagent dosage, upper limits for equipment operating parameters, and water quality safety thresholds. Soft constraints include upper limits for seedling stress index and upper limits for single regulation cost.

[0036] Furthermore, a fractional-level fusion control strategy is formulated, which specifically includes the following steps: Based on the Pontryagin minimum principle in fractional control theory, a costate variable is introduced, and a multi-objective optimization function and preset constraints are incorporated to construct a fractional Hamiltonian function. Construct a fractional-order inverse costate equation based on the fractional-order Hamiltonian function, and obtain the control law by solving the fractional-order inverse costate equation; The fractional-order fusion control strategy is generated based on the control law. The fractional-order fusion control strategy includes the dosage of probiotics, the dosage of traditional Chinese medicine extracts, the oxygenation intensity, and the water exchange flow rate.

[0037] Specifically, in order to achieve the synergistic goals of disease risk control, water quality stability regulation cost optimization, and seedling stress minimization, a multi-objective optimization function is constructed based on the predicted data obtained in the aforementioned steps, and a fractional-order optimal fusion regulation strategy is formulated in combination with preset constraints.

[0038] The expression for the multi-objective optimization function is: ,in, To optimize the objective function value, These are respectively: disease risk weight, water quality stability weight, regulation cost weight, and seedling stress weight. To predict the disease risk level in the data, For water quality evolution data in the prediction data, To preset the optimal water quality range, To adjust the input parameters, This is used to predict the seedling stress index in the data. Using predicted ammonia nitrogen evolution data, dissolved oxygen evolution data, and disease risk levels as inputs, this optimization function balances various objectives to achieve a global equilibrium across multiple goals.

[0039] The preset constraints include hard constraints and soft constraints. Hard constraints include the upper limit of the dosage of the agent, the upper limit of the operating parameters of the Internet of Things device, and the water quality safety threshold. Soft constraints include the upper limit of the seedling stress index and the upper limit of the cost of a single regulation, so as to ensure the feasibility and safety of the regulation strategy.

[0040] The control strategy is formulated based on the Pontryagin minimum principle in fractional-order control theory. First, costate variables are introduced. Constructing fractional Hamiltonian functions ,in, It is a fractional Hamiltonian function. As a core term of the multi-objective optimization function, For multi-field coupled state vectors, For costate variables, for Caputo fractional derivative, A fractional-order elliptic operator. A fractional-order inverse costate equation is constructed based on the fractional-order Hamiltonian function. ,in, Costate variables The Caputo fractional right derivative is used to solve the equation, yielding the optimal control law. By integrating the weights and constraints of the multi-objective optimization function into the equation-solving process, the resulting optimal control law is directly mapped to the specific values ​​of the control input parameters. Based on this optimal control law, a fractional optimal fusion control strategy is generated. This strategy includes the dosage of probiotics, the dosage of traditional Chinese medicine extracts, the oxygenation intensity, and the water exchange flow rate, providing a clear basis for the precise control of the subsequent seedling pond.

[0041] A fractional-order fusion control strategy was implemented in the seedling pond, and measured data in the seedling pond after implementation were collected. The deviation between the measured data and the predicted data was calculated. The parameters of the fractional-order LSTM-SPDE coupled prediction model and the fractional-order fusion control strategy were optimized through closed-loop feedback iteration.

[0042] Furthermore, the fractional-order fusion control strategy is implemented, specifically including the following steps: The fractional-order fusion control strategy is transformed into control signals that can be recognized by IoT devices, including drug delivery devices, variable frequency oxygenators, and smart water exchange valves. The disease risk level is used to implement a graded control mechanism, including a combination of automatic control and manual confirmation, a combination of automatic execution and real-time monitoring, and a combination of emergency execution and manual intervention.

[0043] Furthermore, the closed-loop feedback iterative optimization specifically includes the following steps: The measured data collected in the seedling pond after the execution were consistent with the type of multi-field coupled spatiotemporal continuous data. Calculate the deviation vector between the measured data and the predicted data. The deviation vector is the difference between the measured data and the predicted data. Using the bias vector as input, the parameters of the fractional LSTM-SPDE coupled prediction model are solved by the projection shrinkage algorithm based on the stochastic variational inequality. By using the online Bayesian update method, the solved parameters are substituted into the fractional LSTM-SPDE coupled prediction model to correct the coupling coefficient matrix and the random fluctuation coefficient. Predictive data is regenerated based on the modified fractional-order LSTM-SPDE coupled prediction model, and the fractional-order fusion control strategy is updated iteratively by combining the deviation vector.

[0044] Specifically, in order to implement the aforementioned fractional-order optimal fusion control strategy and continuously optimize the adaptability of the model and strategy through actual test feedback, the control strategy needs to be implemented in the seedling pond, while collecting actual test data for closed-loop feedback iterative optimization.

[0045] First, the fractional-order optimal fusion control strategy is transformed into control signals recognizable by IoT devices, including a precision drug delivery device, a variable-frequency aerator, and an intelligent water exchange valve, ensuring accurate transmission and execution of control commands. A tiered execution mechanism is adopted based on the disease risk level in the predicted data. For low-risk levels, a combination of automatic control and manual confirmation is used; for medium- and high-risk levels, a combination of automatic execution and real-time monitoring, and emergency execution and manual intervention are used respectively, balancing control efficiency and safety. During and after the control process, measured data from the seedling pond are continuously collected. The measured data is consistent with the aforementioned multi-field coupled spatiotemporal continuous data, covering water quality, biological and physiological fields, drug diffusion fields, and environmental auxiliary field data, ensuring the completeness and comparability of the feedback data.

[0046] Generally, the expression for calculating the deviation vector between measured data and predicted data is: ,in, This is the multi-field coupled state deviation vector. for Multi-field measured data vectors at time points, for The multi-field prediction data vector at time 1 is used to quantify the difference between the predicted and actual values ​​using this formula. Taking the bias vector as input, an optimization objective is constructed based on the stochastic variational inequality, expressed as follows: ,in, This represents the optimal parameter vector for the corrected model. This is the current model parameter vector. For inner product operations, the direction of parameter correction is constrained by this inequality to ensure that the deviation is minimized.

[0047] The parameters of the fractional-order LSTM-SPDE coupled prediction model are solved using the projection shrinkage algorithm, and its iterative formula is as follows: ,in, These are the model parameters after iteration. For the parameter feasible region projection operator, The parameters for the current iteration round, The step size factor is used to progressively adjust the parameters, guided by the bias vector. The model parameters are further optimized using an online Bayesian update method, the expression of which is: ,in, Let be the posterior probability distribution of the parameter. The likelihood probability of the bias vector based on the current parameters. Given the prior probability distribution of the parameters, the optimal parameters obtained by solving are substituted into the fractional LSTM-SPDE coupled prediction model to correct the coupling coefficient matrix and random fluctuation coefficient of the model.

[0048] Predictive data is regenerated based on the modified fractional LSTM-SPDE coupled prediction model. The fractional optimal fusion control strategy is then iteratively updated by combining the deviation vector, making the control strategy more consistent with the actual state of the seedling pond. This forms a closed loop of execution-feedback-optimization, improving the accuracy and dynamic adaptability of subsequent control.

[0049] Example 2: In a second embodiment of the present invention, the present invention provides an integrated system for water quality control and disease prevention in seedling cultivation, such as... Figure 2 As shown, it includes the following modules: The integrated system for water quality control and disease prevention in seedling cultivation includes: Data acquisition module: used to acquire multi-field coupled spatiotemporal continuous data of water quality field, biological physiological field, chemical diffusion field and environmental auxiliary field in the seedling pond; Preprocessing module: used to perform fractional-order multi-field data preprocessing on multi-field coupled spatiotemporal continuous data to obtain feature vectors containing long-term memory features and coupling features; Coupled prediction module: It is used to construct a fractional LSTM-SPDE coupled prediction model based on feature vectors and the coupling of fractional LSTM and stochastic partial differential equations. It uses real-time feature vectors as input to predict water quality evolution data, seedling physiological status data and disease risk level within a preset time period to obtain prediction data. The module for constructing and formulating is used to construct a multi-objective optimization function based on fractional-order control theory and predictive data, and to formulate a fractional-order fusion control strategy in combination with preset constraints. The execution optimization module is used to implement a fractional-order fusion control strategy on the seedling pond, collect the measured data in the seedling pond after execution, calculate the deviation with the predicted data, and optimize the parameters of the fractional-order LSTM-SPDE coupled prediction model and the fractional-order fusion control strategy through closed-loop feedback iteration.

[0050] A large-scale shrimp breeding base faced a disconnect between water quality control and disease prevention during the breeding process. It was unable to accurately capture the correlation between water quality changes and shrimp physiological state and disease risk, nor could it predict disease occurrence in advance. Control decisions relied on manual experience, often resulting in excessive water changes causing stress to the seedlings and improper drug administration damaging water quality, leading to low seedling survival rates and unstable quality. To solve these problems, the integrated water quality control and disease prevention system for seedling breeding, as described in this embodiment, was adopted. Its architecture is as follows: Figure 2 As shown. The specific implementation process of this system is as follows: The data acquisition module deploys multiple types of sensors throughout the entire nursery pond to simultaneously collect water quality parameters, shrimp physiological signals, drug concentrations, and environmental data, enabling comprehensive acquisition of multi-field coupled spatiotemporal continuous data. The preprocessing module performs fractional-order denoising, interpolation completion, and coupled feature extraction on the collected data to generate feature vectors containing long-term memory features; The coupled prediction module constructs a coupled prediction model based on feature vectors. By inputting real-time feature vectors, it obtains future water quality evolution, shrimp physiological status, and disease risk levels. The formulation module is based on the predicted data and combined with the constraints to construct a multi-objective optimization function, and formulate an integrated control strategy that takes into account both water quality stability and disease prevention and control. The execution optimization module executes the control strategy through IoT devices, collects measured data to calculate deviations, and iteratively optimizes model parameters and control strategies in a closed loop to achieve precise and coordinated control.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for integrating water quality control and disease prevention in seedling cultivation, characterized in that, Includes the following steps: Acquire continuous spatiotemporal data of multiple fields coupled together, including water quality field, biological physiological field, reagent diffusion field, and environmental auxiliary field, within the seedling pond; Fractional-order multi-field data preprocessing is performed on multi-field coupled spatiotemporal continuous data to obtain feature vectors containing long-term memory features and coupling features; A fractional LSTM-SPDE coupled prediction model is constructed based on feature vectors and coupled with stochastic partial differential equations. The real-time feature vectors are used as input to predict water quality evolution data, seedling physiological status data and disease risk level within a preset time period, and the prediction data is obtained. Based on fractional-order control theory, a multi-objective optimization function is constructed using predicted data, and a fractional-order fusion control strategy is formulated in combination with preset constraints. A fractional-order fusion control strategy is implemented in the seedling pond, and the measured data in the seedling pond after implementation are collected. The deviation between the measured data and the predicted data is calculated. The parameters of the fractional-order LSTM-SPDE coupled prediction model and the fractional-order fusion control strategy are optimized through closed-loop feedback iteration.

2. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The water quality field data includes dissolved oxygen concentration, pH value, ammonia nitrogen concentration, nitrite concentration, and total alkalinity; the biological physiological field data includes seedling swimming speed, feeding frequency, superoxide dismutase activity, and stress hormone concentration; the drug diffusion field data includes probiotic concentration and traditional Chinese medicine extract concentration; and the environmental auxiliary field data includes water temperature, light intensity, water flow velocity, and oxygenation intensity.

3. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The fractional-order multi-field data preprocessing specifically includes the following steps: The Caputo fractional difference method is used to denoise time series data in multi-field coupled spatiotemporal continuous data, while preserving the long-term memory features of the data to obtain feature data. Based on the fractional covariance function, the Kriging interpolation method is used to complete the spatial gap data in the feature data and generate a three-dimensional global data distribution map. Fractional cross-correlation analysis was performed on the three-dimensional global data distribution map to screen out coupling features, and feature vectors were constructed based on the coupling features. The coupling features include water quality-physiology coupling features, water quality-pharmaceutical coupling features, and physiology-pharmaceutical coupling features.

4. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The process of obtaining the predicted data specifically includes the following steps: Based on the long-term memory characteristics and multi-field coupling correlation characteristics of eigenvectors, and by incorporating the Caputo fractional derivative and the spatial Laplace operator into the gating mechanism, a fractional LSTM unit is constructed. Based on the multi-field coupled state vector containing the water quality field, biological physiological field and the drug diffusion field in the seedling pond, a fractional-order stochastic partial differential equation is constructed. The fractional-order stochastic partial differential equations are spatially discretized using the finite element method and time-discretized using the Milstein scheme, resulting in discretized fractional-order stochastic partial differential equations. Fractional LSTM units are coupled with discretized fractional stochastic partial differential equations to form a fractional LSTM-SPDE coupled prediction model. The feature vector is input into the fractional LSTM unit in the fractional LSTM-SPDE coupled prediction model to extract the spatial-temporal fusion features. The spatial-temporal fusion features are input into the discretized fractional-order stochastic partial differential equations in the fractional-order LSTM-SPDE coupled prediction model. The prediction data, which includes water quality evolution data, seedling physiological status data, and disease risk level, are obtained through numerical solution.

5. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 4, characterized in that: The disease risk level is determined by using the Sigmoid function to convert the seedling physiological state data into disease risk values ​​within a preset range, and then dividing the disease risk values ​​into several disease risk levels according to a preset threshold.

6. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The multi-objective optimization function includes disease risk control objective, water quality stability objective, regulation cost control objective, and seedling stress control objective. The disease risk control objective corresponds to the disease risk level in the predicted data, the water quality stability objective corresponds to the deviation between the water quality evolution data in the predicted data and the preset optimal water quality range, the regulation cost control objective corresponds to the regulation input parameters, and the seedling stress control objective corresponds to the seedling stress index in the predicted data.

7. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The preset constraints include hard constraints and soft constraints. The hard constraints include the upper limit of the drug dosage, the upper limit of the equipment operating parameters, and the water quality safety threshold. The soft constraints include the upper limit of the seedling stress index and the upper limit of the cost of a single regulation.

8. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The formulation of the fractional-order fusion control strategy specifically includes the following steps: Based on the Pontryagin minimum principle in fractional control theory, a costate variable is introduced, and a multi-objective optimization function and preset constraints are incorporated to construct a fractional Hamiltonian function. Construct a fractional-order inverse costate equation based on the fractional-order Hamiltonian function, and obtain the control law by solving the fractional-order inverse costate equation; The fractional-order fusion control strategy is generated based on the control law. The fractional-order fusion control strategy includes the dosage of probiotics, the dosage of traditional Chinese medicine extracts, the oxygenation intensity, and the water exchange flow rate.

9. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The execution of the fractional-order fusion control strategy specifically includes the following steps: The fractional-order fusion control strategy is converted into a control signal that can be recognized by IoT devices, including a drug delivery device, a variable frequency oxygenator, and a smart water exchange valve. The disease risk level is used to implement a graded control mechanism, including a combination of automatic control and manual confirmation, a combination of automatic execution and real-time monitoring, and a combination of emergency execution and manual intervention.

10. The method for integrating water quality control and disease prevention in seedling cultivation according to claim 1, characterized in that: The closed-loop feedback iterative optimization specifically includes the following steps: Collect measured data in the seedling pond after execution; the measured data is of the same type as multi-field coupled spatiotemporal continuous data. Calculate the deviation vector between the measured data and the predicted data, where the deviation vector is the difference between the measured data and the predicted data; Using the bias vector as input, the parameters of the fractional LSTM-SPDE coupled prediction model are solved by the projection shrinkage algorithm based on the stochastic variational inequality. By using the online Bayesian update method, the solved parameters are substituted into the fractional LSTM-SPDE coupled prediction model to correct the coupling coefficient matrix and the random fluctuation coefficient. Predictive data is regenerated based on the modified fractional-order LSTM-SPDE coupled prediction model, and the fractional-order fusion control strategy is updated iteratively by combining the deviation vector.

Images