Multi-objective optimization decision control method for shrimp and silkworm polyculture environment parameters

Abstract

The invention discloses a multi-objective optimization decision control method for shrimp and silkworm polyculture environment parameters, and relates to the technical field of polyculture control. The method comprises the following steps: deploying a water quality sensor network, and combining prawn and nereis growth monitoring equipment; extracting criticality of environmental parameters and biological growth indexes by using historical breeding data and experimental data; constructing a dynamic coupling model, and establishing a dynamic model of the polyculture system based on a differential equation; developing a hybrid optimization algorithm to balance the competition and symbiotic relationship between the prawns and the nereis; and generating a control instruction set based on an optimization result, and adjusting the control equipment through a fuzzy PID controller. According to the invention, intelligent regulation and control of shrimp and silkworm polyculture environment parameters are realized, and economic benefits of shrimp and silkworm polyculture are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of polyculture control, and in particular relates to a multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm polyculture, which can not only purify the breeding environment and improve the survival rate of shrimp, but also harvest finished sandworms and improve the comprehensive economic benefits of breeding. Background Art

[0002] Prawns are popular with consumers for their tender texture and nutritious nutrition. However, due to their low feed utilization capacity and the gradual aging of pond substrates, high-density shrimp farming in ponds is prone to large-scale shrimp disease outbreaks, often causing huge losses to shrimp farmers.

[0003] When shrimp feed, they grasp food particles with their front paws and chew them, producing a lot of food debris that is lost in the water. Furthermore, their feces contain a high amount of organic matter. Studies have shown that only 19% of the nitrogen introduced into shrimp ponds through artificial feeding is converted into nitrogen in the shrimp's bodies. The remainder accumulates at the bottom of the pond in the form of leftover bait and feces, gradually transforming into toxic and harmful substances that breed pathogenic microorganisms and harm the shrimp's health. To improve the aquaculture environment, farmers often resort to large-scale water changes to regulate water quality. This can easily deteriorate the aquatic environment in the aquaculture ponds and surrounding waters, exacerbating the spread of infectious shrimp diseases and forcing farmers to use large quantities of medication repeatedly, which in turn increases the drug resistance of pathogens, creating a vicious cycle and leading to serious drug residue problems in shrimp.

[0004] Nereids, commonly known as sea earthworms, are known as "pond bottom cleaners" and "universal bait." Student team member Lin Lu explained that the environmentally friendly and efficient shrimp-silkworm polyculture model involves introducing nereids into the shrimp aquaculture environment. Because nereids inhabit burrows at the pond bottom and feed on mud containing shrimp bait and feces, they effectively remove these debris, maintaining a stable and healthy aquaculture environment and ensuring clean, safe, high-yield, and healthy shrimp farming. However, the high price and low yield of nereids have limited the expansion of this polyculture model. For a long time, nereids were primarily harvested from the seashore. However, excessive harvesting by fishermen and environmental changes have led to a significant decline in wild nereid production. This harvesting method also disrupts the ecological balance of the seashore. At the same time, with the development of various aquaculture industries both domestically and internationally and the rise of the fishing industry, market demand for nereids is increasing. Currently, one kilogram of nereids can fetch over $20. However, the success rate of raising large-sized nereid seedlings is less than 5%, far from meeting market demand. In addition, sandworms caught in the wild or raised in ponds outside the city also have a higher risk of carrying pathogens.

[0005] The present invention has developed a shrimp-silkworm mixed breeding model, which has a significant effect through decision-making and control of breeding environment parameters, and can increase income by 1,500 to 2,000 yuan per mu. Summary of the Invention

[0006] The purpose of the present invention is to provide a multi-objective optimization decision-making control method for the environmental parameters of shrimp and silkworm polyculture. By conducting in-depth research on the shrimp and silkworm polyculture system with circulating water and applying advanced sensor technology, machine learning methods and intelligent control systems, the problems of the existing shrimp and silkworm polyculture environmental parameter control being unintelligent and having low economic benefits are solved.

[0007] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0008] The present invention provides a multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture, comprising the following steps:

[0009] Step S1: deploying a water quality sensor network and combining it with shrimp and lugworm growth monitoring equipment;

[0010] Step S2: Using historical breeding data and experimental data, extract the key of environmental parameters and biological growth indicators;

[0011] Step S3: constructing a dynamic coupling model and establishing a polyculture system dynamics model based on differential equations;

[0012] Step S4: developing a hybrid optimization algorithm to balance the competition and symbiotic relationship between shrimp and nereids;

[0013] Step S5: Generate a control instruction set based on the optimization result, and adjust the control device through the fuzzy PID controller;

[0014] Step S6: Deploy the digital twin system to simulate the long-term impact of different control strategies and use Bayesian optimization to update the model parameters;

[0015] Step S7: Combined with sensor feedback on the biological disturbance behavior of the lugworm, the feeding amount and water change cycle are dynamically modified.

[0016] As a preferred technical solution, in step S1, the water quality sensor network is used to collect data on the dissolved oxygen, temperature, salinity, pH, ammonia nitrogen and nitrite content of the water body in the breeding environment; the growth monitoring equipment is an underwater high-definition camera, which is used to capture the morphological characteristics of shrimp and lugworms in real time, and automatically calculate the body length changes through an edge detection algorithm; the growth monitoring equipment obtains the body length, feeding rate, survival rate and metabolic rate of shrimp and lugworms.

[0017] The specific monitoring process of the growth monitoring equipment is as follows:

[0018] (1) Body length monitoring: The body length change is automatically calculated through edge detection algorithms (such as the YOLO model); the body length of shrimp is estimated based on the proportional relationship between the carapace length and the distance between the tail fans, with an accuracy error of ≤0.5mm; the growth curve model of lugworms is established using the linear relationship between the number of body segments and the distance between the segments, and the body cavity volume is measured using 3D scanning technology.

[0019] (2) Feeding rate monitoring: A weight sensor and RFID chip are integrated into the bait delivery device to record the type, weight, and location of bait delivered each time. The feeding amount per unit time is calculated by the difference between the amount of bait consumed and the weight of the remaining bait (the specific formula is: feeding rate = (amount fed - amount of remaining bait) / total weight of the organism × 100%). Underwater cameras are used to capture the behavioral characteristics of shrimp feeding on bait (such as the frequency of chelicerae grasping), and a LSTM network is used to establish a correlation model between feeding behavior and feeding amount.

[0020] (3) Survival rate monitoring: Use fluorescent labeling technology to inject harmless fluorescent dyes (such as Calcein-AM) into shrimp seedlings and sandworm larvae, and use a handheld fluorescence detector to achieve non-contact individual identification and survival statistics; or deploy a hydrophone array at the bottom of the culture pond to determine the survival density by analyzing the sound wave frequency (200-800Hz) generated by the sandworm's drilling activity.

[0021] (4) Metabolic rate monitoring: A movable isolation cabin is set up in the culture pond, and the oxygen consumption rate per unit time is continuously recorded by a dissolved oxygen sensor (accuracy ±0.1 mg / L) to calculate the standard metabolic rate (SMR) and activity metabolic rate (AMR); or organic matter labeled with the stable isotope 13C is regularly injected into the water body, and the enrichment degree of 13C in the body of the sandworm is detected by a mass spectrometer to infer the decomposition metabolic rate of the organic matter.

[0022] As a preferred technical solution, in step S2, the collected data needs to be cleaned and outlier processed, specifically as follows: deploy water quality sensors such as dissolved oxygen, pH, salinity, and ammonia nitrogen to record environmental parameters in real time, with a sampling frequency of ≥15 minutes / time; synchronously collect biological indicator data such as shrimp body length, survival rate, and sandworm density, and achieve individual tracking through RFID tags or image recognition technology; and integrate laboratory control test data (such as the impact of different salinity gradients on shrimp survival rate) and historical breeding logs (feeding amount, water change cycle, etc.);

[0023] When filling missing values, linear interpolation or LSTM-based time series prediction is used to fill in short-term sensor interruption data; when detecting outliers, the 3σ principle or isolation forest algorithm is applied to identify outliers (such as a sudden drop in salinity caused by heavy rain), combined with manual review of the aquaculture log.

[0024] The key steps for extracting environmental parameters and biological growth indicators are as follows:

[0025] Step S21: Perform sliding window statistics on water quality parameters and align them with the biological growth stage. First, perform device time calibration, aligning the trigger signals of the camera and water quality sensor to the millisecond level through a hardware synchronization interface (such as GPIO). The device clock needs to be automatically calibrated daily to avoid timestamp drift due to battery loss.

[0026] Aggregate hourly biological activity data (e.g., nereid food intake) and water quality parameters (ammonia nitrogen concentration) by time window (e.g., 00:00-01:00) and calculate the mean or peak value. Interpolation is used to fill in the gaps in asynchronous data (e.g., manual sampling), and the LSTM model is used to predict parameters for missing time periods.

[0027] Constructing time series features: The hourly change rate of ammonia nitrogen concentration (ΔNH3 / Δt) was extracted and associated with the frequency of shrimp molting, and a matching data set was generated through sliding window statistics (window size = 6 hours, step size = 1 hour).

[0028] The peak feeding period of lugworms (e.g., 18:00-22:00) was aligned with the ammonia nitrogen degradation rate by time stamp, and the lag effect of feeding intensity on water quality parameters was calculated (e.g., 1-3 h delay);

[0029] Step S22: Pearson correlation coefficient is used to analyze the linear association between environmental parameters and biological indicators, and the generalized additive model is used to capture nonlinear relationships;

[0030] Step S23: using the LSTM network to capture the temporal characteristics of environmental parameters and predict the changing trend of the specific growth rate of shrimp;

[0031] Step S24: combining the random forest algorithm to screen key parameters and inferring the causal chain between water quality parameters and biological indicators through the Leaf-Bassian network;

[0032] Step S25: K-fold cross validation is used to evaluate the generalization ability of the model and calculate the parameter sensitivity index.

[0033] As a preferred technical solution, in step S22, the Pearson correlation coefficient formula is:

[0034]

[0035] Where x i ,y i represent the observed values of water quality parameters and biological indicators, represent the mean values of water quality parameters and biological indicators, respectively;

[0036] When |r|>0.8, it indicates that the correlation between water quality parameters and biological indicators is strong; when 0.5≤|r|≤0.8, it indicates that the correlation between water quality parameters and biological indicators is moderate; when |r|<0.5, it indicates that the correlation between water quality parameters and biological indicators is weak;

[0037] The significance of the correlation results was verified by testing whether the correlation coefficient t was significant. The calculation formula for the correlation coefficient was as follows:

[0038]

[0039] Where n represents the sample size. When p < 0.05, it is considered that there is a correlation between water quality parameters and biological indicators. p is obtained by looking up the table based on the verified correlation coefficient t, which is used to determine whether the linear correlation between variables is statistically significant.

[0040] By the formula It can be seen that the larger the sample size, the weaker the amplification effect of the denominator on the result, but the easier it is to achieve statistical significance; the larger the absolute value of r, the stronger the amplification effect of the numerator, the larger the t value, and the higher the significance;

[0041] When the generalized additive model captures nonlinear relationships, it fits the nonlinear relationship between variables through a nonparametric smoothing function; the model form is:

[0042] g(E(Y))=β0+f1(x1)+f2(x2)+......+f p (x p );

[0043] Where Y represents the biological indicator, x p represents the pth water quality parameter, f p (·) represents a nonlinear smooth function that captures the independent effects of each variable;

[0044] Thin plate splines or cubic splines are used for continuous variables, the degrees of freedom of the smoothing term are set to prevent overfitting, and generalized cross-validation is used for optimization. To facilitate visual observation, an effect diagram can be drawn with the horizontal axis representing the environmental parameter value and the vertical axis representing the contribution value to the biological indicator. The stability of the effect can be judged by the confidence interval. The critical point of the change in the effect direction can then be determined by derivative calculation or visual interpretation.

[0045] As a preferred technical solution, in step S23, the LSTM network architecture is designed as follows:

[0046] Input layer: Contains time series features of environmental parameters (6 dimensions) and historical growth rates (1 dimension), with a time step size of 72 (i.e., 3 days of data, sampled at 1-hour intervals).

[0047] Hidden layer: uses a bidirectional LSTM structure (64 neurons) to capture the forward / reverse dependencies of environmental parameters; superimposes an attention mechanism layer to dynamically weight key periods.

[0048] Output layer: Connect to the fully connected layer to output the growth rate forecast for the next 5 days and calculate the confidence interval.

[0049] As a preferred technical solution, in step S24, a random forest regression is constructed with the biological indicator as the target variable and the water quality parameter as the feature, the Gini importance or permutation importance is calculated, parameters whose importance exceeds a threshold are screened, and parameters whose importance is below the threshold are iteratively eliminated; based on the conditional independence test, the V-structure rule is used to determine the causal direction; then, the do-operator is used to simulate the intervention experiment, and the change in the posterior distribution of the biological indicator is calculated to verify the causal strength;

[0050] When calculating Gini importance, the random forest algorithm is used to train the model. When each decision tree splits a node, the feature with the largest decrease in Gini impurity is selected for splitting. For each feature, the sum of the reduction in Gini impurity when all nodes in the tree are split is calculated. The Gini impurity reduction of the feature in all decision trees is averaged and normalized (the sum is 1) to obtain the Gini importance value. The specific formula is as follows:

[0051]

[0052] Where S f is the set of nodes that feature f participates in the split, ΔGini(s) is the decrease in Gini impurity after node S is split;

[0053] When calculating permutation importance, a random forest model is used to calculate benchmark performance indicators (such as accuracy, RMSE, etc.) on a validation set (such as out-of-bag data OOB). For each feature f, the values of all its samples are randomly shuffled, while other features remain unchanged. The perturbed data set is used to re-predict and calculate the change in performance indicators. The perturbation is repeated N times for each feature (usually N ≥ 30), and the average of the performance changes is taken as the permutation importance value.

[0054] The V-structure is defined as follows: In a causal directed acyclic graph (DAG), a V-structure is represented by two variables (A and B) that jointly point to a third variable (C), with no direct connection between A and B (i.e., no edge or reverse edge). Mathematically, this is expressed as A→C←B, forming a "common effect" structure.

[0055] In structure learning algorithms (such as the PC algorithm), edges are forced to be oriented as A→C←B if the following conditions are met:

[0056] There is no direct edge connection between A and B;

[0057] There exists a set of variables S (excluding C) such that A and B are independent given S;

[0058] But A and B are no longer independent given S∪{C}.

[0059] By constructing a causal diagram of biological systems, marking direct / indirect causal paths, using the do-operator to force variable values or modify variable distributions, and integrating experimental design with causal reasoning, the intensity of causal effects in cross-species aquaculture can be effectively quantified, providing a theoretical basis for optimizing biological polyculture systems.

[0060] As a preferred technical solution, in step S25, the specific process of using K-fold cross validation to evaluate the generalization ability of the model and calculate the parameter sensitivity index is as follows:

[0061] Step S251, data preprocessing: clean the data and standardize the features to ensure that the data distribution of each fold is consistent;

[0062] Step S252, K-fold random partitioning: randomly divide the data set into K mutually exclusive subsets (K=5 or 10), with the sample size of each subset being approximately equal;

[0063] Step S253, cyclic training: each time K-1 subsets are taken and merged as the training set, and the remaining subset is used as the validation set, and the process is repeated K times;

[0064] Step S254, performance index recording: record the evaluation indicators of each iteration (such as accuracy, F1 value, mean square error), and calculate the average and standard deviation of K results;

[0065] Step S255: Hierarchical optimization: Hierarchical K-folding is used for the classification task to ensure that the category distribution of each subset is consistent with the original data and to avoid data variance;

[0066] K-fold cross-validation is used to quantify model stability (standard deviation) and parameter sensitivity (volatility), providing dual guidance for model optimization. Under the same hardware conditions, 5-fold cross-validation saves computing time compared to the leave-one-out method (LOOCV).

[0067] As a preferred technical solution, in step S3, the specific process of establishing a polyculture system dynamics model based on differential equations is as follows:

[0068] Step S31: Determine the core variables and interaction mechanisms of shrimp and silkworm polyculture, including biological variables, environmental variables and control variables; wherein the biological variables include: shrimp biomass B s , Nereid biomass B w , shrimp feeding rate F s , ammonia nitrogen degradation rate of nereid w The environmental variables include dissolved oxygen DO, ammonia nitrogen concentration NH3, temperature T, pH value, sediment organic carbon content C org ; The control variable: aerator power Water exchange rate Q water , feeding amount F free ;

[0069] The specific interaction relationships are as follows:

[0070] Shrimp excretion → ammonia nitrogen concentration (increases) → nereid degradation activity (increases) → dissolved oxygen (decreases) → shrimp respiration is obstructed.

[0071] Nereids disturb the bottom mud → organic carbon oxidation → dissolved oxygen fluctuations → impact on microbial communities → changes in ammonia nitrogen conversion efficiency.

[0072] Step S32: establishing a kinetic model of the differential equations; the kinetic model includes biological metabolism equations and environmental parameter dynamic equations;

[0073] Biological metabolic equation: such as shrimp growth and metabolic equation:

[0074]

[0075] Where a s is the feed conversion efficiency, f(T) is the temperature effect function, g(NH3) is the ammonia nitrogen toxicity inhibition function, and h(DO) is the hypoxia stress function;

[0076] For example, the ammonia nitrogen degradation equation of nereids:

[0077]

[0078] Where η S is the ammonia nitrogen excretion coefficient of shrimp, η S is the maximum degradation rate of nereid, is the half-saturation constant, K(T) is the temperature dependence coefficient [0, 1];

[0079] Dynamic equations of environmental parameters, such as the dissolved oxygen balance equation:

[0080]

[0081] Where, is the oxygenation efficiency, φ air (T) is the air dissolution rate, μ S 、μ W are the oxygen consumption coefficients of shrimp and nereid, δ org is the sediment oxygen consumption rate;

[0082] Sediment organic carbon change equation:

[0083]

[0084] Where, is the residual bait sinking rate, ζW is the feeding efficiency of sandworms, is the natural oxidation rate;

[0085] Step S33: Calibrate the parameters, perform sensitivity analysis, and validate the model; obtain the metabolic parameters of the above-mentioned shrimp and nereid in indoor simulation experiments, use the Morris method or Sobol index to identify the dominant variables, use ODE45 (MATLAB) or Scipy.integrate (Python) to solve the differential equations, and compare the actual breeding data to verify the accuracy of the model;

[0086] Step S34: Integrate and optimize the model; embed the differential equation model into the model predictive control (MPC) framework and generate Q water The optimal control sequence is calculated and fuzzy logic is introduced to dynamically adjust the weights. The model status is updated based on real-time sensor data. The long-term trends of shrimp survival rate and ammonia nitrogen concentration under different control schemes are simulated to support decision optimization.

[0087] As a preferred technical solution, in step S4, the specific process of the hybrid optimization algorithm is as follows:

[0088] Step S41: setting the prawn yield, nereid biomass, and energy efficiency as maximization targets, the ammonia nitrogen peak value and water change frequency as minimization targets, and introducing water quality parameters (dissolved oxygen, pH) as constraints;

[0089] Step S42: Initialize the population and randomly generate an initial population containing a combination of polyculture parameters (such as dissolved oxygen set value, nereid density, feeding amount, etc.);

[0090] Step S43: performing non-dominated sorting (Pareto ranking) on the population, and associating individuals to preset reference points (evenly distributed in the target space) by normalizing the target space to ensure population diversity;

[0091] Step S44: Dynamically adjust the target weight according to the real-time environmental status; input variables: dissolved oxygen deviation (low / medium / high), ammonia nitrogen concentration (low / medium / high); output variable: objective function weight (e.g., shrimp production weight increased by 20%);

[0092] Step S45: In the cross-mutation phase, the weight coefficient of the objective function is adjusted according to the fuzzy rules to guide the algorithm to prioritize satisfying the current key constraints;

[0093] Step S46: abstracting the ecological interaction between shrimp and nereids into a non-cooperative game and defining a payoff function;

[0094] The strategy spaces of both parties are:

[0095] Shrimp strategy: Increase feeding rate to accelerate growth, but increase ammonia nitrogen excretion.

[0096] Nereid strategy: Increase density to degrade more ammonia nitrogen, but may compete for dissolved oxygen resources.

[0097] The payment function is defined as:

[0098] Shrimp benefits: growth rate - ammonia nitrogen toxicity penalty.

[0099] Benefits of sandworms: ammonia nitrogen degradation - loss due to competition for dissolved oxygen.

[0100] In each generation of the population, solutions that meet the Nash equilibrium conditions are screened (i.e., there is no unilateral motivation for deviation in the strategies of both parties).

[0101] Step S47: Generate an initial population and associate it with reference points for iterative optimization, merge the parent and child populations, and select the next generation of elite individuals through the reference point association mechanism.

[0102] Iterative optimization involves crossover and mutation, fuzzy weight updating, and game equilibrium screening. Crossover and mutation utilize simulated binary crossover (SBX) and multi-locus mutation, incorporating an adaptive mutation rate (dynamically adjusted based on population diversity). Fuzzy weight updating adjusts the objective function weights based on current water quality parameters. During game equilibrium screening, individuals that deviate from the Nash equilibrium are removed from the non-dominated solution set.

[0103] As a preferred technical solution, in step S5, water quality parameters (dissolved oxygen, pH, salinity, etc.) collected in real time by the sensor network are integrated with equipment status data (aerator power, water pump flow rate) to construct a dynamic data set; a control instruction set (such as dissolved oxygen target value ±0.2 mg / L, water change cycle adjustment range) is output through a multi-objective optimization model and mapped as input parameters of the PID controller; fuzzy rules are defined based on expert experience (such as "when dissolved oxygen is less than 4 mg / L and ammonia nitrogen is greater than 0.3 mg / L", the aerator power is increased to 90%), the membership function is trained in combination with historical data, and the PID parameters are updated using an online learning algorithm to achieve precise tracking of nonlinear processes (such as dissolved oxygen fluctuation suppression error <5%).

[0104] The present invention has the following beneficial effects:

[0105] (1) The present invention deploys a water quality sensor network, constructs a dynamic coupling model, and establishes a polyculture system dynamics model based on differential equations; develops a hybrid optimization algorithm to balance the competition and symbiotic relationship between shrimp and sandworms, generates a control instruction set based on the optimization results, and adjusts the control equipment through a fuzzy PID controller to achieve intelligent regulation of environmental parameters of shrimp and silkworm polyculture, thereby improving the economic benefits of shrimp and silkworm polyculture.

[0106] The present invention automatically adjusts the priority of environmental parameters according to the growth stage of shrimp (such as larval stage and adult shrimp stage), introduces the intensity of biological disturbance of sandworms as an intermediary variable, analyzes its regulatory effect on the water quality-growth correlation, and realizes the precise management of environmental-biological coordinated regulation.

[0107] The present invention uses an LSTM network to capture the temporal characteristics of environmental parameters and predict the changing trend of the specific growth rate of shrimp. At different stages of the breeding cycle (seedling stage / growing stage), the gated recurrent unit (GRU) automatically switches the model parameter weights to adapt to the stage-by-stage changes in the interaction between the environment and organisms, extracting the spatial correlation of environmental parameters of adjacent breeding ponds. This overcomes the limitation of traditional methods that only rely on the time series data of a single pond and improves the accuracy of the prediction of the changing trend of the specific growth rate of shrimp.

[0108] The present invention reduces computational complexity through random forest screening, extracts causal chains through Bayesian networks, performs dynamic causal recommendations, supports real-time data updating of network parameters, adapts to environmental mutations, and uses DAG to intuitively display the conduction path of "water quality parameters → biological response" to assist manual decision-making.

[0109] The present invention uses biological indicators as target variables and water quality parameters as features, constructs a random forest regression, calculates Gini importance or permutation importance, and uses the V-structure rule based on conditional independence test to determine the causal direction. Then, through the do-operator simulation intervention experiment, the posterior distribution change of the biological indicators is calculated to verify the causal strength. This method can effectively quantify the causal effect strength in cross-species aquaculture and provide a theoretical basis for optimizing biological polyculture systems.

[0110] The present invention develops a hybrid optimization algorithm that combines a multi-objective genetic algorithm with fuzzy logic decision-making, dynamically adjusts the algorithm weights, and introduces game theory ideas to balance the "competition-symbiosis" relationship between shrimp and sandworms. It can effectively balance the ecological needs of shrimp and sandworms, and dynamically respond to environmental changes, providing a theoretical basis for the precise regulation of polyculture systems.

[0111] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0113] Figure 1 This is a flow chart of a multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to the present invention. DETAILED DESCRIPTION

[0114] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0115] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0116] To make the purpose, technical solutions and advantages of this application clearer, Figure 1 The implementation methods of this application are described in further detail.

[0117] Before introducing the embodiments of the present application, the co-culture technology of whiteleg shrimp and dicynoglossum is first described.

[0118] The environment and facilities for the polyculture of whiteleg shrimp and Perineuridae are as follows:

[0119] Pond conditions: convenient water inlet and outlet, muddy and sandy bottom, water storage depth can reach 2.0m, area should be 3 to 15 mu, length-to-width ratio should be 2:1 to 3:1, and the long axis should be parallel to the main wind direction during the production season.

[0120] Inlet and drainage facilities: A 40-mesh screen is installed on the outside of the inlet gate, a gate is placed in the middle, and an 80-mesh sleeve filter is installed on the inside: a gate is placed on the outside of the drainage gate, and a 20-30-mesh arc filter is installed on the inside.

[0121] Aeration facilities: One aerator with a power of 0.75KW to 1.0KW is provided for every 2 to 3 mu of aquaculture water surface.

[0122] (1) Preparation stage before stocking:

[0123] Pond maintenance: remove silt from the pond, expose the pond bottom to the sun, repair the embankment, plow (10cm to 20cm deep) and harrow.

[0124] Pond disinfection: Dry the pond or add water (fill the pond with 10cm to 20cm of water) and spray the entire pond with drugs for disinfection. The use of drugs should comply with the provisions of NY5071-2002.

[0125] Water injection and disinfection: 10 to 15 days before stocking the seedlings, inject seawater with a depth of 60 cm to 80 cm and a salinity of 18 to 25 into the pond, and use 15 mg / L to 20 mg / L of effective chlorine for disinfection. The inlet water quality should comply with the requirements of GB11607-1989.

[0126] Fertilizer and water: Apply special compound fertilizer for cultivating single-cell algae. The usage and dosage are operated according to the instructions, and the aerator is turned on for 2h~3h at noon every day to promote the reproduction of single-cell algae.

[0127] (2) Stocking stage

[0128] Stocking density: 3-4 segmented planktonic larvae stocking density is 1000 / m 2 ~2000 / m 2 Or large-sized seedlings with about 40 joints and a body length of about 1 cm, with a stocking density of 50 fish / m 2 ~100 tails / m 2

[0129] Stocking time: Stock the fry when the water temperature is stable above 18℃.

[0130] Stocking method: Soak the plastic bag containing sandworm seedlings in the breeding pond water for about 30 minutes, and then slowly add pond water into the bag to allow the sandworm seedlings to gradually adapt to the temperature and salinity of the breeding pond water.

[0131] Only whiteleg shrimp compound feed was fed during the entire breeding process.

[0132] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0133] See also Figure 1 As shown, the present invention is a multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture, comprising the following steps:

[0134] Step S1: deploying a water quality sensor network and combining it with shrimp and lugworm growth monitoring equipment;

[0135] Step S2: Using historical breeding data and experimental data, extract the key of environmental parameters and biological growth indicators;

[0136] Step S3: constructing a dynamic coupling model and establishing a polyculture system dynamics model based on differential equations;

[0137] Step S4: developing a hybrid optimization algorithm to balance the competition and symbiotic relationship between shrimp and nereids;

[0138] Step S5: Generate a control instruction set based on the optimization result, and adjust the control device through the fuzzy PID controller;

[0139] Step S6: Deploy the digital twin system to simulate the long-term impact of different control strategies and use Bayesian optimization to update the model parameters;

[0140] Step S7: Combined with sensor feedback on the biological disturbance behavior of the lugworm, the feeding amount and water change cycle are dynamically modified.

[0141] In step S1, the water quality sensor network is used to collect data on the dissolved oxygen, temperature, salinity, pH, ammonia nitrogen and nitrite content of the water body in the breeding environment; the growth monitoring equipment is an underwater high-definition camera, which is used to capture the morphological characteristics of shrimp and lugworms in real time, and automatically calculate the body length changes through the edge detection algorithm; the growth monitoring equipment obtains the body length, feeding rate, survival rate and metabolic rate of shrimp and lugworms.

[0142] The specific monitoring process of the growth monitoring equipment is as follows:

[0143] (1) Body length monitoring: The body length change is automatically calculated through edge detection algorithms (such as the YOLO model); the body length of shrimp is estimated based on the proportional relationship between the carapace length and the distance between the tail fans, with an accuracy error of ≤0.5mm; the growth curve model of lugworms is established using the linear relationship between the number of body segments and the distance between the segments, and the body cavity volume is measured using 3D scanning technology.

[0144] (2) Feeding rate monitoring: A weight sensor and RFID chip are integrated into the bait delivery device to record the type, weight, and location of bait delivered each time. The feeding amount per unit time is calculated by the difference between the amount of bait consumed and the weight of the remaining bait (the specific formula is: feeding rate = (amount fed - amount of remaining bait) / total weight of the organism × 100%). Underwater cameras are used to capture the behavioral characteristics of shrimp feeding on bait (such as the frequency of chelicerae grasping), and a LSTM network is used to establish a correlation model between feeding behavior and feeding amount.

[0145] (3) Survival rate monitoring: Use fluorescent labeling technology to inject harmless fluorescent dyes (such as Calcein-AM) into shrimp seedlings and sandworm larvae, and use a handheld fluorescence detector to achieve non-contact individual identification and survival statistics; or deploy a hydrophone array at the bottom of the culture pond to determine the survival density by analyzing the sound wave frequency (200-800Hz) generated by the sandworm's drilling activity.

[0146] Metabolic rate monitoring: A movable isolation cabin is set up in the breeding pond, and the oxygen consumption rate per unit time is continuously recorded through a dissolved oxygen sensor (accuracy ±0.1 mg / L) to calculate the standard metabolic rate (SMR) and activity metabolic rate (AMR); or organic matter labeled with the stable isotope 13C is regularly injected into the water body, and the enrichment level of 13C in the sandworm body is detected by a mass spectrometer to infer the decomposition metabolic rate of the organic matter.

[0147] The above monitoring data is transmitted to the cloud via LoRa wireless, and a decision support system is built with the following modules:

[0148] Growth prediction model: Based on the positive correlation between the body length of sandworms and the ammonia nitrogen degradation efficiency (r=0.82), the optimal harvest period is predicted.

[0149] Abnormal warning: When the shrimp survival rate drops by more than 5% for three consecutive days, the water quality parameter adjustment mechanism is automatically triggered.

[0150] In step S2, the collected data needs to be cleaned and outliers processed, specifically as follows: deploy water quality sensors such as dissolved oxygen, pH, salinity, and ammonia nitrogen to record environmental parameters in real time, with a sampling frequency of ≥15 minutes / time; synchronously collect biological indicator data such as shrimp body length, survival rate, and nereid density, and achieve individual tracking through RFID tags or image recognition technology; and integrate laboratory control test data (such as the impact of different salinity gradients on shrimp survival rate) and historical breeding logs (feeding amount, water change cycle, etc.);

[0151] When filling missing values, linear interpolation or LSTM-based time series prediction is used to fill in short-term sensor interruption data; when detecting outliers, the 3σ principle or isolation forest algorithm is applied to identify outliers (such as a sudden drop in salinity caused by heavy rain), combined with manual review of the aquaculture log.

[0152] The key steps for extracting environmental parameters and biological growth indicators are as follows:

[0153] Step S21: Perform sliding window statistics on water quality parameters and align them with the biological growth stage. First, perform device time calibration, aligning the trigger signals of the camera and water quality sensor to the millisecond level through a hardware synchronization interface (such as GPIO). The device clock needs to be automatically calibrated daily to avoid timestamp drift due to battery loss.

[0154] Aggregate hourly biological activity data (e.g., nereid food intake) and water quality parameters (ammonia nitrogen concentration) by time window (e.g., 00:00-01:00) and calculate the mean or peak value. Interpolation is used to fill in the gaps in asynchronous data (e.g., manual sampling), and the LSTM model is used to predict parameters for missing time periods.

[0155] Constructing time series features: The hourly change rate of ammonia nitrogen concentration (ΔNH3 / Δt) was extracted and associated with the frequency of shrimp molting, and a matching data set was generated through sliding window statistics (window size = 6 hours, step size = 1 hour).

[0156] The peak feeding period of lugworms (e.g., 18:00-22:00) was aligned with the ammonia nitrogen degradation rate by time stamp, and the lag effect of feeding intensity on water quality parameters was calculated (e.g., 1-3 h delay);

[0157] Step S22: Pearson correlation coefficient is used to analyze the linear association between environmental parameters and biological indicators, and the generalized additive model is used to capture nonlinear relationships;

[0158] Step S23: using the LSTM network to capture the temporal characteristics of environmental parameters and predict the changing trend of the specific growth rate of shrimp;

[0159] Step S24: combining the random forest algorithm to screen key parameters and inferring the causal chain between water quality parameters and biological indicators through the Leaf-Bassian network;

[0160] Step S25: K-fold cross validation is used to evaluate the generalization ability of the model and calculate the parameter sensitivity index.

[0161] As a preferred technical solution, in step S22, the Pearson correlation coefficient formula is:

[0162]

[0163] Where x i ,y i represent the observed values of water quality parameters and biological indicators, represent the mean values of water quality parameters and biological indicators, respectively;

[0164] When |r|>0.8, it indicates that the correlation between water quality parameters and biological indicators is strong; when 0.5≤|r|≤0.8, it indicates that the correlation between water quality parameters and biological indicators is moderate; when |r|<0.5, it indicates that the correlation between water quality parameters and biological indicators is weak;

[0165] The significance of the correlation results was verified by testing whether the correlation coefficient t was significant. The calculation formula for the correlation coefficient was as follows:

[0166]

[0167] Where n represents the sample size. When p < 0.05, it is considered that there is a correlation between water quality parameters and biological indicators. p is obtained by looking up the table based on the verified correlation coefficient t, which is used to determine whether the linear correlation between variables is statistically significant.

[0168] By the formula It can be seen that the larger the sample size, the weaker the amplification effect of the denominator on the result, but the easier it is to achieve statistical significance; the larger the absolute value of r, the stronger the amplification effect of the numerator, the larger the t value, and the higher the significance;

[0169] When the generalized additive model captures nonlinear relationships, it fits the nonlinear relationship between variables through a nonparametric smoothing function; the model form is:

[0170] g(E(Y))=β0+f1(x1)+f2(x2)+......+f p(x p );

[0171] Where Y represents the biological indicator, x p represents the pth water quality parameter, f p (·) represents a nonlinear smooth function that captures the independent effects of each variable;

[0172] Thin plate splines or cubic splines are used for continuous variables, the degrees of freedom of the smoothing term are set to prevent overfitting, and generalized cross-validation is used for optimization. To facilitate visual observation, an effect diagram can be drawn with the horizontal axis representing the environmental parameter value and the vertical axis representing the contribution value to the biological indicator. The stability of the effect can be judged by the confidence interval. The critical point of the change in the effect direction can then be determined by derivative calculation or visual interpretation.

[0173] In step S23, the LSTM network architecture is designed as follows:

[0174] Input layer: Contains time series features of environmental parameters (6 dimensions) and historical growth rates (1 dimension), with a time step size of 72 (i.e., 3 days of data, sampled at 1-hour intervals).

[0175] Hidden layer: uses a bidirectional LSTM structure (64 neurons) to capture the forward / reverse dependencies of environmental parameters; superimposes an attention mechanism layer to dynamically weight key periods.

[0176] Output layer: Connect to the fully connected layer to output the growth rate forecast for the next 5 days and calculate the confidence interval.

[0177] In step S24, a random forest regression is constructed using the biological indicator as the target variable and the water quality parameter as the feature. The Gini importance or permutation importance is calculated, and parameters with importance exceeding the threshold are screened. Parameters with importance below the threshold are iteratively eliminated. Based on the conditional independence test, the V-structure rule is used to determine the causal direction. Then, the do-operator is used to simulate the intervention experiment, calculate the change in the posterior distribution of the biological indicator, and verify the causal strength.

[0178] When calculating Gini importance, the random forest algorithm is used to train the model. When each decision tree splits a node, the feature with the largest decrease in Gini impurity is selected for splitting. For each feature, the sum of the reduction in Gini impurity when all nodes in the tree are split is calculated. The Gini impurity reduction of the feature in all decision trees is averaged and normalized (the sum is 1) to obtain the Gini importance value. The specific formula is as follows:

[0179]

[0180] Where S f is the set of nodes that feature f participates in the split, ΔGini(s) is the decrease in Gini impurity after node S is split;

[0181] When calculating permutation importance, a random forest model is used to calculate benchmark performance indicators (such as accuracy, RMSE, etc.) on a validation set (such as out-of-bag data OOB). For each feature f, the values of all its samples are randomly shuffled, while other features remain unchanged. The perturbed data set is used to re-predict and calculate the change in performance indicators. The perturbation is repeated N times for each feature (usually N ≥ 30), and the average of the performance changes is taken as the permutation importance value.

[0182] The V-structure is defined as follows: In a causal directed acyclic graph (DAG), a V-structure is represented by two variables (A and B) that jointly point to a third variable (C), with no direct connection between A and B (i.e., no edge or reverse edge). Mathematically, this is expressed as A→C←B, forming a "common effect" structure.

[0183] In structure learning algorithms (such as the PC algorithm), edges are forced to be oriented as A→C←B if the following conditions are met:

[0184] There is no direct edge connection between A and B;

[0185] There exists a set of variables S (excluding C) such that A and B are independent given S;

[0186] But A and B are no longer independent given S∪{C}.

[0187] By constructing a causal diagram of biological systems, marking direct / indirect causal paths, using the do-operator to force variable values or modify variable distributions, and integrating experimental design with causal reasoning, the intensity of causal effects in cross-species aquaculture can be effectively quantified, providing a theoretical basis for optimizing biological polyculture systems.

[0188] In step S25, the K-fold cross validation is used to evaluate the generalization ability of the model and calculate the parameter sensitivity index. The specific process is as follows:

[0189] Step S251, data preprocessing: clean the data and standardize the features to ensure that the data distribution of each fold is consistent;

[0190] Step S252, K-fold random partitioning: randomly divide the data set into K mutually exclusive subsets (K=5 or 10), with the sample size of each subset being approximately equal;

[0191] Step S253, cyclic training: each time K-1 subsets are taken and merged as the training set, and the remaining subset is used as the validation set, and the process is repeated K times;

[0192] Step S254, performance index recording: record the evaluation indicators of each iteration (such as accuracy, F1 value, mean square error), and calculate the average and standard deviation of K results;

[0193] Step S255: Hierarchical optimization: Hierarchical K-folding is used for the classification task to ensure that the category distribution of each subset is consistent with the original data and to avoid data variance;

[0194] K-fold cross-validation is used to quantify model stability (standard deviation) and parameter sensitivity (volatility), providing dual guidance for model optimization. Under the same hardware conditions, 5-fold cross-validation saves computing time compared to the leave-one-out method (LOOCV).

[0195] In step S3, the specific process of establishing the polyculture system dynamics model based on differential equations is as follows:

[0196] Step S31: Determine the core variables and interaction mechanisms of shrimp-silkworm polyculture, including biological variables, environmental variables and control variables; biological variables include: shrimp biomass B s , Nereid biomass B w , shrimp feeding rate F s , ammonia nitrogen degradation rate of nereid w Environmental variables include dissolved oxygen (DO), ammonia nitrogen concentration (NH3), temperature (T), pH value, and sediment organic carbon content (C). org ; Control variable: Aerator power Water exchange rate Q water , feeding amount F free ;

[0197] The specific interaction relationships are as follows:

[0198] Shrimp excretion → ammonia nitrogen concentration (increases) → nereid degradation activity (increases) → dissolved oxygen (decreases) → shrimp respiration is obstructed.

[0199] Nereids disturb the bottom mud → organic carbon oxidation → dissolved oxygen fluctuations → impact on microbial communities → changes in ammonia nitrogen conversion efficiency.

[0200] Step S32: establishing a kinetic model of the differential equations; the kinetic model includes biological metabolism equations and environmental parameter dynamic equations;

[0201] Biological metabolic equation: such as shrimp growth and metabolic equation:

[0202]

[0203] Where a s is the feed conversion efficiency, f(T) is the temperature effect function, g(NH3) is the ammonia nitrogen toxicity inhibition function, and h(DO) is the hypoxia stress function;

[0204] For example, the ammonia nitrogen degradation equation of nereids:

[0205]

[0206] Where η S is the ammonia nitrogen excretion coefficient of shrimp, η S is the maximum degradation rate of nereid, is the half-saturation constant, K(T) is the temperature dependence coefficient [0, 1];

[0207] Dynamic equations of environmental parameters, such as the dissolved oxygen balance equation:

[0208]

[0209] Where, is the oxygenation efficiency, φ air (T) is the air dissolution rate, μ S 、μ W are the oxygen consumption coefficients of shrimp and nereid, δ org is the sediment oxygen consumption rate;

[0210] Sediment organic carbon change equation:

[0211]

[0212] Where, is the residual bait sinking rate, ζ W is the feeding efficiency of sandworms, is the natural oxidation rate;

[0213] Step S33: Calibrate the parameters, perform sensitivity analysis, and validate the model; obtain the metabolic parameters of the above-mentioned shrimp and nereid in indoor simulation experiments, use the Morris method or Sobol index to identify the dominant variables, use ODE45 (MATLAB) or Scipy.integrate (Python) to solve the differential equations, and compare the actual breeding data to verify the accuracy of the model;

[0214] Step S34: Integrate and optimize the model; embed the differential equation model into the model predictive control (MPC) framework and generate Q water The optimal control sequence is calculated and fuzzy logic is introduced to dynamically adjust the weights. The model status is updated based on real-time sensor data. The long-term trends of shrimp survival rate and ammonia nitrogen concentration under different control schemes are simulated to support decision optimization.

[0215] In step S4, the specific process of the hybrid optimization algorithm is as follows:

[0216] Step S41: setting the prawn yield, nereid biomass, and energy efficiency as maximization targets, the ammonia nitrogen peak value and water change frequency as minimization targets, and introducing water quality parameters (dissolved oxygen, pH) as constraints;

[0217] Step S42: Initialize the population and randomly generate an initial population containing a combination of polyculture parameters (such as dissolved oxygen set value, nereid density, feeding amount, etc.);

[0218] Step S43: performing non-dominated sorting (Pareto ranking) on the population, and associating individuals to preset reference points (evenly distributed in the target space) by normalizing the target space to ensure population diversity;

[0219] Step S44: Dynamically adjust the target weight according to the real-time environmental status; input variables: dissolved oxygen deviation (low / medium / high), ammonia nitrogen concentration (low / medium / high); output variable: objective function weight (e.g., shrimp production weight increased by 20%);

[0220] Step S45: In the cross-mutation phase, the weight coefficient of the objective function is adjusted according to the fuzzy rules to guide the algorithm to prioritize satisfying the current key constraints;

[0221] Step S46: abstracting the ecological interaction between shrimp and nereids into a non-cooperative game and defining a payoff function;

[0222] The strategy spaces of both parties are:

[0223] Shrimp strategy: Increase feeding rate to accelerate growth, but increase ammonia nitrogen excretion.

[0224] Nereid strategy: Increase density to degrade more ammonia nitrogen, but may compete for dissolved oxygen resources.

[0225] The payment function is defined as:

[0226] Shrimp benefits: growth rate - ammonia nitrogen toxicity penalty.

[0227] Benefits of sandworms: ammonia nitrogen degradation - loss due to competition for dissolved oxygen.

[0228] In each generation of the population, solutions that meet the Nash equilibrium conditions are screened (i.e., there is no unilateral motivation for deviation in the strategies of both parties).

[0229] Step S47: Generate an initial population and associate it with reference points for iterative optimization, merge the parent and child populations, and select the next generation of elite individuals through the reference point association mechanism.

[0230] Iterative optimization involves crossover and mutation, fuzzy weight updating, and game equilibrium screening. Crossover and mutation utilize simulated binary crossover (SBX) and multi-locus mutation, incorporating an adaptive mutation rate (dynamically adjusted based on population diversity). Fuzzy weight updating adjusts the objective function weights based on current water quality parameters. During game equilibrium screening, individuals that deviate from the Nash equilibrium are removed from the non-dominated solution set.

[0231] In step S5, water quality parameters (dissolved oxygen, pH, salinity, etc.) collected in real time by the sensor network are integrated with equipment status data (aerator power, water pump flow rate) to construct a dynamic data set; a control instruction set (such as dissolved oxygen target value ±0.2 mg / L, water change cycle adjustment range) is output through a multi-objective optimization model and mapped as input parameters of the PID controller; fuzzy rules are defined based on expert experience (such as "when dissolved oxygen is less than 4 mg / L and ammonia nitrogen is greater than 0.3 mg / L", the aerator power is increased to 90%), and the membership function is trained with historical data. The PID parameters are updated using an online learning algorithm to achieve accurate tracking of nonlinear processes (such as dissolved oxygen fluctuation suppression error <5%).

[0232] For example, emergency decision-making for sudden pollution incidents (taking a sudden drop in salinity as an example):

[0233] For example, when identifying an emergency state, the salinity change rate is monitored in real time through the edge computing node (such as a drop of >10% within 5 minutes), triggering a threshold alarm and activating the emergency module. Combined with the Bayesian network to infer the source of pollution (such as the probability of rainstorm runoff pollution >80%), a risk level assessment report is generated. When performing reinforcement learning (DQN) dynamic strategy optimization, the state space is defined: including the current value of salinity, the rate of decline, and the shrimp stress behavior index (such as the degree of decline in feeding rate). Multi-dimensional reward weights are set based on comprehensive ecological restoration costs (such as water change volume), equipment energy consumption, and biological survival rate improvement goals. Build a historical emergency case library and select the optimal action through the Q network (such as starting the seawater replenishment system + reducing the feeding amount by 50%); update the strategy every 5 minutes until the salinity returns to a safe threshold (25-30‰).

[0234] In step S6, a digital twin system is deployed to simulate the long-term impact of different control strategies, and the model parameters are updated using Bayesian optimization. A Gaussian process (GP) is used to establish a proxy model of the objective function (such as the comprehensive benefit index) and the control parameters, and a kernel function is defined to capture nonlinear relationships. The optimal parameter combination is iteratively selected through the expected improvement (EI) acquisition function, and the model parameters are updated every 24 hours. The convergence threshold of the optimization process is set to RMSE < 0.05. The optimized parameters are deployed to the physical breeding system, and the actual operation data is recorded through the blockchain notarization module of the digital twin platform to reversely correct the model deviation. The conditional probability table (CPT) is updated using Markov chain Monte Carlo (MCMC) sampling to adjust the parameter weights in real time (such as the dissolved oxygen control priority changes dynamically with the growth stage of the shrimp).

[0235] In step S7, the feeding amount and water change cycle are dynamically modified in combination with the sensor feedback of the nereid bioturbation behavior; specifically as follows:

[0236] A disturbance behavior-water quality feedback mechanism was established. When the ORP value increased to above +50 mV and the organic particle concentration dropped to below 400 ppm, the nereid disturbance was judged to be effective (the threshold value was calibrated based on experimental data).

[0237] Constructing the Disturbance Intensity Index (DBI):

[0238]

[0239] The lugworm disturbance intensity index reflects the efficiency of lugworms in improving the bottom soil per unit time.

[0240] The LSTM neural network is used to learn the correlation between DBI and water quality parameters (such as ammonia nitrogen degradation rate and dissolved oxygen fluctuation) in historical data to predict the water quality change trend in the next 24 hours.

[0241] Dynamic correction of feeding amount:

[0242]

[0243] Where, F t is the feeding amount at time t, α is the adjustment coefficient, the default value is 0.3; DBI 阈值 represents the optimal perturbation critical value determined experimentally.

[0244] When DBI is >0.8 for 2 hours continuously, the automatic feeding reduction mechanism is triggered, reducing the feeding amount by 10%-20% (because the sandworms have effectively decomposed the leftover bait); if DBI is <0.5 and the ammonia nitrogen concentration is >0.2mg / L, the frequency of feeding suspension is increased to avoid the accumulation of leftover bait.

[0245] Based on the composite index of DBI and ammonia nitrogen concentration, the water exchange model is as follows:

[0246]

[0247] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0248] In addition, those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiments can be accomplished by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0249] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture, characterized in that: The steps include: Step S1: deploying a water quality sensor network and combining it with shrimp and lugworm growth monitoring equipment; Step S2: Using historical breeding data and experimental data, extract the key of environmental parameters and biological growth indicators; Step S3: constructing a dynamic coupling model and establishing a polyculture system dynamics model based on differential equations; Step S4: developing a hybrid optimization algorithm to balance the competition and symbiotic relationship between shrimp and nereids; Step S5: Generate a control instruction set based on the optimization result, and adjust the control device through the fuzzy PID controller; Step S6: Deploy the digital twin system to simulate the long-term impact of different control strategies and use Bayesian optimization to update the model parameters; Step S7: Combined with sensor feedback on the biological disturbance behavior of the lugworm, the feeding amount and water change cycle are dynamically modified.

2. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S1, the water quality sensor network is used to collect data on the dissolved oxygen, temperature, salinity, pH, ammonia nitrogen and nitrite content of the water body in the breeding environment; the growth monitoring equipment is an underwater high-definition camera, which is used to capture the morphological characteristics of shrimp and lugworms in real time and automatically calculate the body length changes through an edge detection algorithm; the growth monitoring equipment obtains the body length, feeding rate, survival rate and metabolic rate of shrimp and lugworms.

3. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S2, the key steps of extracting environmental parameters and biological growth indicators are as follows: Step S21: Perform sliding window statistics on water quality parameters and align them with biological growth stages; Step S22: Pearson correlation coefficient is used to analyze the linear association between environmental parameters and biological indicators, and the generalized additive model is used to capture nonlinear relationships; Step S23: using the LSTM network to capture the temporal characteristics of environmental parameters and predict the changing trend of the specific growth rate of shrimp; Step S24: combining the random forest algorithm to screen key parameters and inferring the causal chain between water quality parameters and biological indicators through the Leaf-Bassian network; Step S25: K-fold cross validation is used to evaluate the generalization ability of the model and calculate the parameter sensitivity index.

4. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 3 is characterized in that: In step S22, the correlation between the water quality parameter and the biological indicator is determined based on the Pearson correlation coefficient r; when |r|>0.8, it indicates that the water quality parameter and the biological indicator are strongly correlated; when 0.5≤|r|≤0.8, it indicates that the water quality parameter and the biological indicator are moderately correlated; when |r|<0.5, it indicates that the water quality parameter and the biological indicator are weakly correlated; The significance of the correlation results was verified by testing whether the correlation coefficient t was significant. The calculation formula for the correlation coefficient was as follows: Where n represents the sample size. When p < 0.05, it is considered that there is a correlation between water quality parameters and biological indicators. p is obtained by looking up the table based on the verified correlation coefficient t, which is used to determine whether the linear correlation between variables is statistically significant. When the generalized additive model captures nonlinear relationships, it fits the nonlinear relationships between variables through nonparametric smoothing functions; Thin plate splines or cubic splines were used for continuous variables, the degrees of freedom of the smoothing term were set to prevent overfitting, and generalized cross-validation was used for optimization.

5. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S24, a random forest regression is constructed with the biological indicator as the target variable and the water quality parameter as the feature, the Gini importance or permutation importance is calculated, parameters whose importance exceeds a threshold are screened, and parameters whose importance is below the threshold are iteratively eliminated; based on the conditional independence test, the V-structure rule is used to determine the causal direction; then, the do-operator is used to simulate the intervention experiment, and the change in the posterior distribution of the biological indicator is calculated to verify the causal strength.

6. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 5 is characterized in that: When calculating the Gini importance, a random forest algorithm is used to train the model. When each decision tree splits a node, the feature with the largest decrease in Gini impurity is selected for splitting. For each feature, the sum of the reductions in Gini impurity when all nodes in the tree are split is calculated. The reductions in Gini impurity of the feature in all decision trees are averaged and normalized to obtain the Gini importance value. When calculating the permutation importance, a random forest model is used to calculate the benchmark performance index on the validation set. For each feature, the values of all its samples are randomly shuffled, while other features are kept unchanged. The perturbed data set is used to re-predict and calculate the change in performance index. The perturbation is repeated N times for each feature, and the average of the performance changes is taken as the permutation importance value.

7. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 3 is characterized in that: In step S25, the K-fold cross validation is used to evaluate the generalization ability of the model and calculate the parameter sensitivity index. The specific process is as follows: Step S251, data preprocessing: clean the data and standardize the features to ensure that the data distribution of each fold is consistent; Step S252, K-fold random partitioning: randomly divide the data set into K mutually exclusive subsets, with the sample size of each subset being approximately equal; Step S253, cyclic training: each time K-1 subsets are taken and merged as the training set, and the remaining subset is used as the validation set, and the process is repeated K times; Step S254, performance index recording: record the evaluation index of each iteration, and calculate the average and standard deviation of K results; Step S255: Hierarchical optimization: Hierarchical K-fold is used for the classification task to ensure that the category distribution of each subset is consistent with the original data.

8. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S3, the specific process of establishing the polyculture system dynamics model based on differential equations is as follows: Step S31: Determine the core variables and interaction mechanisms of shrimp and silkworm polyculture, including biological variables, environmental variables and control variables; wherein the biological variables include: shrimp biomass B s , Nereid biomass B w , shrimp feeding rate F s , ammonia nitrogen degradation rate of nereid w The environmental variables include dissolved oxygen DO, ammonia nitrogen concentration NH3, temperature T, pH value, sediment organic carbon content C org ; The control variable: aerator power P O2 , water exchange rate Q water , feeding amount F free ; Step S32: establishing a dynamic model of the differential equations; Step S33: calibrate the parameters, perform sensitivity analysis, and verify the model; Step S34: Integrate and optimize the model.

9. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S4, the specific process of the hybrid optimization algorithm is as follows: Step S41: setting the shrimp production, nereid biomass, and energy efficiency as maximization targets, the ammonia nitrogen peak value and water change frequency as minimization targets, and introducing water quality parameters as constraints; Step S42: Initializing the population and randomly generating an initial population containing a combination of polyculture parameters; Step S43: performing non-dominated sorting on the population and associating individuals to preset reference points by normalizing the target space; Step S44: Dynamically adjust the target weight according to the real-time environmental status; Step S45: In the cross-mutation phase, the weight coefficient of the objective function is adjusted according to the fuzzy rules to guide the algorithm to prioritize satisfying the current key constraints; Step S46: abstracting the ecological interaction between shrimp and nereids into a non-cooperative game and defining a payoff function; Step S47: Generate an initial population and associate it with reference points for iterative optimization, merge the parent and child populations, and select the next generation of elite individuals through the reference point association mechanism.

10. The multi-objective optimization decision-making control method for environmental parameters of shrimp and silkworm co-culture according to claim 1 is characterized in that: In step S5, water quality parameters and equipment status data collected in real time by the sensor network are integrated to construct a dynamic data set; a control instruction set is output through a multi-objective optimization model and mapped as input parameters of a PID controller; fuzzy rules are defined based on expert experience, membership functions are trained in combination with historical data, and PID parameters are updated using an online learning algorithm to achieve accurate tracking of nonlinear processes.

Images