A deep-sea farming yield prediction management method based on virtual simulation

Abstract

The application discloses a kind of based on virtual simulation deep-sea aquaculture output value prediction management method, belongs to deep-sea aquaculture management and output value prediction technical field.This method constructs virtual simulation environment by collecting multi-source data, extracts fish feeding activity and individual body length distribution, establishes the dynamic response function of feeding intensity to environmental factors, injects extreme event simulation to obtain survival rate attenuation coefficient, deduces biomass accumulation curve and generates output value prediction by combining market price, and outputs optimal management scheme by multi-scenario comparison.This application realizes the whole chain closed loop from environment perception to output optimization, significantly improves the accuracy of deep-sea aquaculture output value prediction and risk management capability.

Description

Technical Field

[0001] This invention relates to the field of deep-sea aquaculture management and output value prediction technology, and in particular to a deep-sea aquaculture output value prediction and management method based on virtual simulation. Background Technology

[0002] Deep-sea aquaculture is a crucial direction for the development of marine fisheries. With near-shore aquaculture space nearing saturation and increasing pressure on nearshore ecological environments, global marine aquaculture is rapidly expanding into deep-sea areas. Compared to near-shore aquaculture, deep-sea aquaculture offers significant advantages such as better water exchange conditions, larger farming capacity, and superior fish quality. However, deep-sea aquaculture also faces significant challenges, including complex environmental conditions, large investment scale, and high operational risks. Deep-sea aquaculture involves substantial investment, high costs, and high risks; the farmed species must be adaptable to the complex deep-sea environment and possess high economic value. Statistics show that approximately 60% of global typhoon activity is concentrated in the southeastern coastal region of my country. Deep-sea aquaculture equipment is constantly exposed to the complex marine environment, facing threats from various extreme marine dynamic events such as typhoons, strong currents, and storm surges. Some aquaculture farmers, lacking risk mitigation technologies, may suffer economic losses of millions of yuan from a single severe convective weather event.

[0003] In terms of deep-sea aquaculture output management, existing technologies mainly face the following prominent problems.

[0004] First, there is a lack of effective correlation between environmental monitoring data and biological growth data. Currently, deep-sea aquaculture platforms are generally equipped with multi-parameter water quality sensors that can collect core water quality parameters such as water temperature, dissolved oxygen, pH, ammonia nitrogen, and nitrite in real time. However, in actual aquaculture management, these environmental data are often only used for water quality early warning and simple adjustments to feeding amounts, without establishing a quantitative and dynamic response relationship with key biological indicators such as fish growth rate, survival rate, and biomass accumulation.

[0005] Second, existing methods for predicting aquaculture output value lack accuracy and are insufficiently supported by biological mechanisms. Traditional output value predictions are mostly based on empirical models or simple time-series statistical methods, which are ill-suited to the highly dynamic and complex deep-sea aquaculture environment. In recent years, although machine learning techniques have been introduced into the field of aquaculture output prediction, such as intelligent aquaculture output prediction models based on backpropagation neural networks, these methods typically rely solely on historical output data and do not fully integrate the causal relationships between environmental factors and biological growth processes, resulting in poor biological interpretability of the prediction results. Furthermore, research indicates that while growth prediction models based on physical information neural networks have achieved high prediction accuracy in small-scale recirculating aquaculture systems in the laboratory, their applicability in the complex environmental conditions of open seas has not yet been effectively verified.

[0006] Third, there is a lack of quantitative assessment methods for the impact of extreme marine environmental events on aquaculture output. Deep-sea aquaculture equipment is exposed to harsh sea conditions for extended periods. Extreme events such as typhoons and strong currents can not only damage net cage structures and cause fish escapes, but also lead to multiple negative effects in farmed fish, including stress responses, decreased feed intake, slowed growth rates, and increased susceptibility to disease. Netting is a critical but vulnerable component of the aquaculture system, subjected to significant hydrodynamic loads in the harsh and dynamic marine environment, potentially causing structural damage, leading to fish escapes, environmental impacts, and economic losses. However, current technologies lack effective methods to establish a quantitative mapping relationship between the risk probability and intensity level of extreme environmental events and aquaculture output losses.

[0007] Fourth, the integration of virtual simulation technology with deep-sea aquaculture output management is insufficient. While digital twin technology has made some progress in real-time monitoring of the dynamics of aquaculture cage structures, combining low-precision simulation data with limited high-precision on-site sensor measurements to achieve real-time prediction of cage displacement and mooring cable loads, its application is mainly concentrated in the field of structural safety monitoring and has not yet extended to the prediction and optimization management of aquaculture output. Specifically, existing technologies have failed to systematically integrate data such as fish growth processes, survival rate changes, and biomass accumulation obtained from virtual simulation environments with market price information and harvest timing decisions, thus failing to achieve dynamic prediction and proactive management of deep-sea aquaculture output throughout its entire lifecycle.

[0008] Fifth, deep-sea aquaculture management decisions lack a holistic optimization perspective focused on output value. Existing aquaculture management solutions often focus on optimizing individual aspects, such as water quality control, feeding strategy adjustments, and disease early warning, with different management aspects operating in isolation. While the industry recognizes the difficulty in coordinating feeding decisions with water quality risk control, a further bottleneck lies in the fact that even with coordinated feeding and environmental control, managers still cannot quantitatively assess the contribution of a particular management decision to the final aquaculture output value, nor can they conduct scientific comparisons and optimizations among multiple alternative management solutions based on output value. In other words, current technology lacks a complete technological chain from environmental perception to output value prediction and then to management decision optimization.

[0009] In conclusion, the deep-sea aquaculture industry urgently needs a method that can effectively integrate environmental monitoring data, biological growth processes, extreme event risk assessments, and market price information, and achieve dynamic prediction and proactive management of the entire aquaculture cycle's output value through virtual simulation. Summary of the Invention

[0010] To achieve the above objectives, this invention provides a method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation, comprising the following steps:

[0011] Step 1: Collect multi-source data of the target deep-sea aquaculture cages. The multi-source data includes water quality parameter data, meteorological parameter data, acoustic scanning data of the farmed fish, underwater image data of the farmed fish, anchor tension data, and cage motion response data. The multi-source data is transmitted to the land-based data processing center and processed for timestamp alignment and outlier removal to form a structured data base layer.

[0012] Step 2: Process the underwater image data to obtain the individual body length distribution data and feeding behavior activity data of the farmed fish population;

[0013] Step 3: Process the acoustic scanning data to obtain biomass estimation data of the fish population in the net cage;

[0014] Step 4: Perform time-series correlation analysis on the fish feeding behavior activity data, the water quality parameter data, and the meteorological parameter data to construct a dynamic response function of fish feeding intensity to environmental factors;

[0015] Step 5: Construct a virtual simulation environment for the target deep-sea aquaculture cage based on the multi-source data;

[0016] Step 6: Inject virtual environmental disturbance events into the virtual simulation environment to simulate the impact of extreme marine dynamic events on the biomass and survival rate of farmed fish, and obtain the survival rate attenuation coefficient of farmed fish.

[0017] Step 7: In the virtual simulation environment, the growth process of the farmed fish population is simulated and deduced using the dynamic response function and the survival rate decay coefficient as constraints, and the expected biomass accumulation curves at each time point in the farming cycle are obtained.

[0018] Step 8: Combine the expected biomass accumulation curve with market price data to generate the output value prediction result of the target deep-sea aquaculture cage;

[0019] Step 9: Based on the output value prediction results, construct a multi-scenario simulation scheme and conduct an output value comparison analysis to output the management scheme with the optimal output value;

[0020] Step 10: Apply the optimal management plan to actual aquaculture management and continuously update the output value prediction results based on real-time monitoring data.

[0021] Preferably, the water quality parameters in step 1 include dissolved oxygen concentration, water temperature, salinity, pH value, ammonia nitrogen concentration, and water flow velocity; the meteorological parameters include sea surface wind speed, wind direction, air temperature, atmospheric pressure, and rainfall. The multi-source data is collected using a multi-source data acquisition device deployed in the sea area of ​​the target deep-sea aquaculture cages. This device includes an underwater multi-parameter water quality sensor, a meteorological monitoring station, a panoramic multibeam sonar device, an underwater binocular stereo vision camera, a tension sensor, and an acceleration sensor. The underwater multi-parameter water quality sensor is installed at the center of the cage and at four directional points along its edge. Each sampling point is arranged in three layers vertically, with the three layers of sampling points located at a depth of 2 meters in the surface water, half the depth of the middle layer aquaculture water, and 2 meters above the bottom of the bottom layer aquaculture cage. The underwater multi-parameter water quality sensor collects data once per hour; the meteorological monitoring station is installed on the buoy platform outside the net cage, and its data collection frequency is once per hour; the panoramic multibeam scanning sonar device is installed on a fixed bracket on the inner wall of the net cage, and its data collection frequency is once per day; the underwater binocular stereo vision camera device is installed at the center of the top of the net cage, and its data collection frequency is once per day, with each collection lasting 10 minutes; the tension sensor is installed at key nodes of the net cage mooring system, and its data collection frequency is once per minute; the acceleration sensor is installed at key nodes of the net cage floating frame structure, and its data collection frequency is once per minute.

[0022] Preferably, the specific process of obtaining the individual body length distribution data and feeding behavior activity data of the farmed fish population in step 2 is as follows: For the underwater image video sequence acquired by the underwater binocular stereo vision camera device, extract single-frame images at a frequency of 1 frame per second to form an image sequence sample set; for each frame image in the image sequence sample set, use a fish target detection model based on a deep convolutional neural network to detect and identify fish targets, outputting the bounding box coordinates of each identified fish in the image; for each fish target detected in each frame image, calculate the three-dimensional Euclidean distance between the two endpoints of the fish target bounding box using the parallax method based on the calibration parameters of the underwater binocular stereo vision camera device as the estimated body length value; statistically analyze the estimated body length values ​​of all fish targets in the same frame image and record the body length distribution histogram; and finally, analyze the body length distribution of all frames within a day. Histograms are aggregated and averaged to obtain the individual body length distribution data of the farmed fish population on that day. The individual body length distribution data of the farmed fish population on that day includes the mean body length, median body length, standard deviation of body length, and distribution skewness of body length. The image sequence sample set is subjected to a fish feeding behavior recognition model based on a three-dimensional convolutional neural network for behavioral index quantification analysis. The behavioral index quantification analysis includes extracting the movement speed of the fish centroid in each frame of the image, calculating the average curvature of the movement trajectory of the fish centroid between 5 consecutive frames, and calculating the reciprocal of the average Euclidean distance between any two individuals in the fish population as the fish population aggregation index. After normalizing the movement speed of the fish centroid, the average curvature of the movement trajectory of the fish centroid, and the fish population aggregation index, a weighted summation method is used to calculate the fish feeding behavior activity index. The value range of the fish feeding behavior activity index is [0, 1].

[0023] Preferably, the specific process for obtaining the biomass estimation data of the fish population in the net cage in step 3 is as follows: Beamforming processing and backscattering intensity calculation are performed on the acoustic scanning data acquired by the panoramic multibeam scanning sonar device. The detection range of the panoramic multibeam scanning sonar device is no less than 30 meters, the horizontal scanning angle is 360°, the vertical scanning angle is 65°, the horizontal resolution is 0.5°, the vertical resolution is 2°, and the single complete scan cycle is 8 to 16 seconds. Three-dimensional point cloud data is generated, and the attributes of each point in the three-dimensional point cloud data include spatial three-dimensional coordinates and acoustic backscattering intensity values. The three-dimensional point cloud data is divided into multiple water layers according to the depth of the aquaculture water body, with each water layer being 1 meter thick. Density-based spatial clustering algorithms are used to cluster the three-dimensional point cloud data within each water layer to identify clusters belonging to the fish population. For each identified fish population cluster, a cluster package is calculated. The method involves: 1) determining the number of point clouds and extracting the average acoustic backscatter intensity value of all points within each cluster; 2) establishing a mapping model between the acoustic backscatter intensity value and the biomass of individual fish; 3) sampling and weighing the initial fish fry at the beginning of the rearing cycle to obtain the average weight of individual fish; 4) performing a standard scan of the initial fish population using a panoramic multibeam sonar device and recording the average acoustic backscatter intensity value obtained from the scan; 5) using a fish pump to extract sample fish from the net cage for actual weighing every 30 days during the rearing cycle; 6) performing regression fitting between the actual weighing results and the acoustic scan results during the same period to correct the parameters in the mapping model; 7) calculating the estimated biomass of the fish population in each water layer based on the mapping model and the number of point clouds and the average acoustic backscatter intensity value of each fish population cluster in each water layer; and 8) summing the estimated biomass of the fish population in all water layers to obtain the estimated biomass data of the fish population in the net cage.

[0024] Preferably, the specific process of constructing the dynamic response function of fish feeding intensity to environmental factors in step 4 is as follows: Dissolved oxygen concentration, water temperature, water flow velocity, and sea surface wind speed are used as four environmental factors affecting fish feeding intensity, and the fish feeding behavior activity index is used as the dependent variable. A nonlinear regression model between the dependent variable and the four environmental factors is established using hourly data over 30 consecutive days as the sample time window. The nonlinear regression model adopts the form...

[0025]

[0026] in The activity index of fish feeding behavior at hour (t) is given. The maximum feeding activity of farmed fish under optimal environmental conditions is set to 1. The dissolved oxygen concentration at hour (t) is... Let (t) be the water temperature at hour (t). Let (t) be the water flow velocity at hour (t). Let (t) be the sea surface wind speed at hour (t). This is the dissolved oxygen response function. Let the water temperature response function be... Let be the flow velocity response function. The wind speed response function; the dissolved oxygen response function The function is piecewise; when the dissolved oxygen concentration is below a set critical threshold, feeding activity decreases exponentially with increasing dissolved oxygen concentration; when the dissolved oxygen concentration is within a set suitable range, feeding activity maintains its maximum value; and when the dissolved oxygen concentration exceeds a set upper limit, feeding activity decreases slowly. The water temperature response function... The function is Gaussian, with its peak value located at the optimal growth temperature for farmed fish fry; the flow velocity response function The wind speed response function is an inverse proportional function; the higher the flow velocity, the lower the feeding activity. The function is an exponential decay function; the response function parameter values ​​obtained by calibration through the nonlinear regression model are stored in the model parameter library of the land-based data processing center as the dynamic response function of fish feeding intensity to environmental factors.

[0027] Preferably, the specific process of constructing the virtual simulation environment for the deep-sea aquaculture target cage in step 5 is as follows: A virtual simulation platform is built in a land-based data processing center. The core computing engine of the virtual simulation platform is constructed based on computational fluid dynamics solvers and multibody dynamics solvers. Water depth and topographic data, seabed sediment type data, cage structure geometric parameters, mooring system configuration parameters, and netting material physical parameters of the sea area where the deep-sea aquaculture target cage is located are input into the virtual simulation platform as basic simulation parameters. The cage structure geometric parameters include cage perimeter, cage depth, cage volume, floating frame structure dimensions, and weight distribution. The mooring system configuration parameters include anchor chain segment length, anchor chain diameter, anchor chain material grade, and anchor point spatial coordinates. The netting material physical parameters include netting mesh size, netting wire diameter, netting elastic modulus, and netting damping coefficient. The real-time collected sea surface wind speed is used as the input. Data, wind direction data, and water flow velocity data are used as external environmental excitation boundary conditions to drive the computational fluid dynamics solver to calculate the flow field distribution and hydrodynamic load distribution in the sea area surrounding the cage. The calculated hydrodynamic load distribution is used as input to drive the multibody dynamics solver to calculate the motion response of the cage's floating structure and the tension response of the mooring system. The motion response of the cage's floating structure is used as input to calculate the deformation state and stress distribution of the netting under the action of water flow. The motion response data of the cage and the mooring tension data output by the virtual simulation platform are compared and verified hourly with the motion response data of the cage collected by the accelerometer and the mooring tension data collected by the tension sensor. When the root mean square error between the simulation results and the actual monitoring data exceeds a preset threshold, the parameter correction program of the simulation model is automatically triggered to adjust the damping coefficient of the netting and the stiffness coefficient of the mooring system until the simulation accuracy meets the preset requirements.

[0028] Preferably, the specific process for obtaining the survival rate decay coefficient of the farmed fish population in step 6 is as follows: when the virtual simulation platform detects that the sea surface wind speed exceeds the first preset threshold for 3 consecutive hours, a typhoon precursor warning state is triggered; when the virtual simulation platform detects that the sea surface wind speed exceeds the second preset threshold for 1 consecutive hour, a typhoon passage simulation event is triggered. The first preset threshold is 15 meters per second, and the second preset threshold is 24.5 meters per second. After the typhoon passage simulation event is triggered, the biomass estimation data and individual body length distribution data of the farmed fish population in the cage at the current moment are obtained as the initial state of the simulation. Based on the real-time sea surface wind speed data collected by the meteorological monitoring station, the maximum expected wind speed and the corresponding maximum expected wind speed within the next 12-hour time window, the next 24-hour time window, the next 48-hour time window, and the next 72-hour time window are calculated using an extreme value distribution model. The probability of exceeding the limit is calculated; the maximum expected wind speed is input into the verified virtual simulation environment to drive the virtual simulation environment to simulate the cage motion response, net deformation state, and internal flow field changes of the aquaculture water under the corresponding wind speed conditions; based on the cage motion response amplitude and internal flow velocity distribution data output by the virtual simulation environment, combined with the stress response experimental data of farmed fish under severe water disturbance conditions, the stress intensity index is calculated. The stress intensity index is calculated by normalizing the cage motion response amplitude and the internal flow velocity and taking the weighted average; based on the stress intensity index, the survival rate decay coefficient of farmed fish during the typhoon event is calculated. The mapping relationship between the survival rate decay coefficient and the stress intensity index is calibrated through post-disaster survey data of historical typhoon events; the survival rate decay coefficient is applied to the biomass estimation of the initial simulation state to obtain the expected remaining biomass after the typhoon event ends.

[0029] Preferably, the specific process of obtaining the expected biomass accumulation curve for each time node within the aquaculture cycle in step 7 is as follows: the aquaculture cycle is divided into discrete time steps with 1 day as the unit; within each time step, the dissolved oxygen concentration, water temperature, water flow velocity, sea surface wind speed, and individual body length distribution data of the farmed fish are read; the dynamic response function of the fish feeding intensity to environmental factors is called to calculate the fish feeding behavior activity index corresponding to the current time step; the actual feeding amount is calculated using a bioenergetics feeding model based on the fish feeding behavior activity index and the total biomass of the fish at the current time step. The bioenergetics feeding model uses the standard metabolizable energy requirement of the farmed fish, the feeding activity correction coefficient, the current biomass, and the current water temperature to jointly determine the daily feeding rate. The basic daily feeding rate is determined by looking up the average fish weight and water temperature in a table and then multiplied by the fish feeding rate. The actual daily feeding rate is obtained by correcting the feeding behavior activity index. The fish biomass increment at the current time step is calculated based on the actual daily feeding rate, current water temperature conditions, and feed conversion ratio parameters of the cultured fish species. The feed conversion ratio parameters vary with fish size and water temperature and are expressed in piecewise function form. The system checks whether the current time step triggers a typhoon passage simulation event. If not, the total biomass of the fish population is updated with the fish biomass increment. If it has been triggered, the current biomass is reduced based on the survival rate decay coefficient before calculating the growth increment for subsequent time steps. The updated total biomass of the fish population and the individual body length distribution data of the cultured fish population obtained by back-calculating from the total biomass of the fish population through the body length-weight allometric growth relationship are used as the initial state for the next time step. This process is repeated until the daily simulation of the entire culture cycle is completed, generating the expected biomass accumulation curve for each time node within the culture cycle.

[0030] Preferably, the specific process for generating the output value prediction result of the target deep-sea aquaculture cage in step 8 is as follows: Historical market price sequence data of farmed fish species is obtained from the market price database connected to the land-based data processing center. The time span of the historical market price sequence data is the past 3 to 5 years, and the time granularity is daily. A seasonal autoregressive integral moving average model is used to model and analyze the historical market price sequence data, identifying the trend component, seasonal component, and random fluctuation component in the historical market price sequence data, and generating the expected market unit price for each future date within the aquaculture cycle. The expected output value for each time node on the expected biomass accumulation curve is multiplied by the expected market unit price corresponding to that time node to obtain the expected output value for that time node. The calculation of the expected output value uses the formula...

[0031]

[0032] in The expected output value on day (t) The expected biomass on day (t) The expected market unit price on day (t) is used; the expected output value of each day during the aquaculture cycle is connected to form a curve to generate the output value prediction result of the target deep-sea aquaculture cage, and the output value prediction result of the target deep-sea aquaculture cage is visualized in the form of a chart on the monitoring terminal of the land-based data processing center.

[0033] Preferably, the specific process of constructing a multi-scenario simulation scheme and performing output value comparison analysis in step 9 to output the management scheme with the optimal output value is as follows: At least three alternative management scenarios are set in the virtual simulation environment. Each alternative management scenario consists of a combination of adjustable management parameters, including seedling density, initial seedling size, harvest time, feed type and grade, and a batch thinning strategy. The batch thinning strategy includes thinning time and thinning ratio. Steps 6 to 8 are repeated for each alternative management scenario to generate the expected biomass accumulation curve and output value prediction results for each alternative management scenario. The output value prediction results for all alternative management scenarios are compared and displayed in the same coordinate system. The comparative analysis includes the peak output value of each alternative management scenario, the time node when the output value reaches its peak, the shape of the output value accumulation curve during the breeding cycle, and the fluctuation characteristics of the output value over time. Among all alternative management scenarios, the alternative management scenario with the largest area under the output value accumulation curve is selected as the management scheme with the optimal output value.

[0034] The specific process of applying the optimal production value management plan to actual aquaculture management and continuously updating the production value prediction results based on real-time monitoring data in step 10 is as follows: The optimal production value management plan, including seedling density, initial seedling specifications, harvest time nodes, feed type and grade, and batch thinning strategies, is distributed to the on-site management terminal of the target deep-sea aquaculture cage; during the implementation of the optimal production value management plan, the data collection, processing, and prediction process of steps 1 to 8 is continuously executed; every 30 days, the biomass estimation data of the cultured fish population in the cage obtained by the panoramic multibeam scanning sonar device is compared with the expected biomass data of the same period obtained by the simulation in step 7. When the relative deviation between the biomass estimation data of the cultured fish population in the cage and the expected biomass data of the same period exceeds ±15%, the recalibration process of the simulation model is triggered. The current biomass estimation data of the cultured fish population in the cage is used as the new initial state of the simulation, and the model parameters of the virtual simulation environment are adjusted in step 5 and steps 6 to 9 are re-executed to generate the corrected production value prediction results and update the management plan recommendations.

[0035] The beneficial effects of this invention are:

[0036] 1. This invention deeply integrates real-time environmental monitoring data, fish acoustic biomass data, and underwater visual behavior data in a virtual simulation environment, constructing a closed-loop chain from environmental perception to output prediction. The virtual simulation environment drives hydrodynamic calculations and fish growth projections based on real-time sea conditions, achieving for the first time high-precision dynamic prediction of deep-sea aquaculture output, thus solving the core problem of the disconnect between environmental factors and biological growth processes, and insufficient prediction accuracy.

[0037] 2. This invention injects extreme marine dynamic events into a virtual simulation environment, calculates the stress intensity index through the simulated cage motion response and water disturbance, and then quantifies the survival rate attenuation coefficient. For the first time, it achieves a quantitative assessment of the impact of extreme events such as typhoons on aquaculture output, filling the technical gap in the scientific quantification of extreme losses in deep-sea aquaculture risk management.

[0038] 3. This invention embeds the dynamic response function of fish feeding intensity to environmental factors into a bioenergetics feeding model and links it with market price forecasting. Through multi-scenario simulation comparison, it outputs the optimal management plan for output value, realizing full-cycle proactive management from data-driven, simulation pre-run to decision optimization, which significantly improves aquaculture efficiency and has outstanding practical value and industrial application prospects. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0041] Figure 2 This is a flowchart illustrating the specific steps involved in obtaining the survival rate attenuation coefficient of farmed fish in step 6 of the method of the present invention.

[0042] Figure 3 This is a flowchart illustrating the specific steps involved in generating the predicted output value of the target deep-sea aquaculture cage in step 8 of the method of the present invention. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0044] Please see Figures 1-3 This invention provides a method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation. The method uses a multi-source data acquisition device deployed in deep-sea aquaculture cages as the sensing front end and a land-based data processing center as the computing back end. Through the integration of the entire chain of environmental perception, biological behavior recognition, virtual simulation, market price prediction and management scheme optimization, it realizes dynamic prediction and proactive management of the output value of deep-sea aquaculture throughout the entire cycle.

[0045] Step 1: Collect multi-source data of target deep-sea aquaculture cages to build a data foundation layer.

[0046] Multi-source data acquisition devices are deployed in the sea area where the target deep-sea aquaculture cages are located. These devices include underwater multi-parameter water quality sensors, a meteorological monitoring station, a panoramic multibeam sonar device, an underwater binocular stereo vision camera, a tension sensor, and an acceleration sensor.

[0047] The underwater multi-parameter water quality sensor is installed at the center of the net cage and at four directional points along the edge of the cage. Each sampling point is arranged in three layers vertically, with the three layers of sampling points located at a depth of 2 meters in the surface layer, halfway down the middle layer of the net cage, and 2 meters above the bottom of the net cage. Each underwater multi-parameter water quality sensor simultaneously measures six water quality parameters: dissolved oxygen concentration, water temperature, salinity, pH value, ammonia nitrogen concentration, and water flow velocity. The underwater multi-parameter water quality sensor collects data once per hour.

[0048] The meteorological monitoring station is installed on a buoy platform outside the net cage. It is equipped with wind speed and direction sensors, temperature sensors, atmospheric pressure sensors, and rainfall sensors to collect five meteorological parameters: sea surface wind speed, wind direction, temperature, atmospheric pressure, and rainfall. The station collects data once per hour.

[0049] The panoramic multibeam scanning sonar device is mounted on a fixed bracket on the inner wall of the net cage. It is used to acquire acoustic scanning data of the fish population within the net cage, with a data acquisition frequency of once daily.

[0050] The underwater binocular stereo vision camera is installed at the center of the top of the fish cage. It is used to acquire underwater image data of the farmed fish, with a daily acquisition frequency of 10 minutes per video sequence.

[0051] Tension sensors are installed at key nodes of the gabion mooring system to collect mooring tension data. The tension sensors collect data once per minute.

[0052] Accelerometers are installed at key nodes of the cage structure to collect motion response data. The accelerometers collect data once per minute.

[0053] The acquired multi-source data is transmitted in real time to the land-based data processing center via a 5G mobile communication network. The land-based data processing center performs timestamp alignment on all multi-source data, using the hourly acquisition time of the underwater multi-parameter water quality sensor as the baseline timestamp, and aligning other data to the nearest hourly time. Simultaneously, outlier removal is performed; values ​​deviating from the average of the three adjacent data points by more than three standard deviations are identified as outliers and removed. Linear interpolation is used to fill the data gaps after removal. The processed multi-source data forms a structured data foundation layer, providing a unified and reliable data input for subsequent biological behavior analysis, environmental correlation modeling, and virtual simulation.

[0054] Step 2: Process the underwater image data collected in Step 1 to obtain the individual body length distribution data and feeding behavior activity data of the farmed fish population.

[0055] For the underwater image and video sequence acquired by the underwater binocular stereo vision camera in step 1, extract single-frame images at a frequency of 1 frame per second to form an image sequence sample set.

[0056] For each frame of the image sequence sample set, a fish target detection model based on a deep convolutional neural network is used to detect and identify fish targets. This deep convolutional neural network outputs the bounding box coordinates of each identified fish in the image.

[0057] For each fish target detected in each frame of an image, based on the calibration parameters of the underwater binocular stereo vision camera, the three-dimensional Euclidean distance between the two endpoints of the fish target's bounding box is calculated using the parallax method. This three-dimensional Euclidean distance is used as the estimated body length of the fish. The estimated body lengths of all fish targets in the same frame are statistically analyzed, and a histogram of the body length distribution of individual fish in that frame is recorded. The histograms of the body length distribution of all frames within a day are aggregated and averaged to obtain the individual body length distribution data of the farmed fish population for that day. The individual body length distribution data of the farmed fish population for that day includes the mean, median, standard deviation, and skewness of the body length distribution.

[0058] A fish feeding behavior recognition model based on a 3D convolutional neural network was used to quantify and analyze behavioral indicators on the image sequence sample set. The quantitative analysis included: extracting the velocity of the fish centroid in each frame; calculating the average curvature of the fish centroid trajectory over five consecutive frames; and calculating the reciprocal of the average Euclidean distance between any two individuals in the fish group as a fish aggregation index. After normalizing the three indicators—velocity of the fish centroid, average curvature of the fish centroid trajectory, and fish aggregation index—a weighted summation method was used to calculate the fish feeding behavior activity index. The fish feeding behavior activity index ranged from 0 to 1, where 0 indicates the fish group is completely inactive or extremely dispersed, and 1 indicates the fish group is highly aggregated and swimming rapidly, indicating a strong feeding state. This step achieved non-contact quantification of individual fish size and feeding intensity using underwater vision technology, providing key behavioral parameters for subsequent growth projections and feeding decisions.

[0059] Step 3: Process the acoustic scanning data collected in Step 1 to obtain biomass estimation data of the fish population in the net cage.

[0060] The acoustic scanning data acquired in step 1 using the panoramic multibeam scanning sonar device undergoes beamforming processing and backscatter intensity calculation. The detection range of the panoramic multibeam scanning sonar device is no less than 30 meters, with a horizontal scanning angle of 360° and a vertical scanning angle of 65°. The horizontal resolution is 0.5°, the vertical resolution is 2°, and the cycle time for a single complete scan is 8 to 16 seconds. The scan generates three-dimensional point cloud data, where each point's attributes include its spatial three-dimensional coordinates and acoustic backscatter intensity value.

[0061] The 3D point cloud data was divided into multiple water layers along the depth of the aquaculture water, with each layer being 1 meter thick. For the 3D point cloud data within each water layer, a density-based spatial clustering algorithm was used to cluster the data, identifying clusters belonging to the fish population. For each identified fish population cluster, the number of points contained within the cluster was calculated, and the average acoustic backscattering intensity value of all points within the cluster was extracted.

[0062] A mapping model was established between acoustic backscattering intensity and individual fish biomass. The model was established as follows: at the beginning of the rearing cycle, initial fish fry were sampled and weighed to obtain the average weight of individual fish; a standard scan of the initial fish population was performed using a panoramic multibeam sonar device, and the average acoustic backscattering intensity was recorded; during the rearing cycle, a calibration period of 30 days was used, and sample fish were extracted from the net cages using a fish pump for actual weighing. The actual weighing results were then regressed and fitted with the acoustic scan results from the same period to correct the parameters in the mapping model.

[0063] Based on the mapping relationship model and the point cloud count and average acoustic backscattering intensity of each fish cluster within each water layer, the estimated fish biomass for each water layer is calculated. The estimated biomass of all water layers is then summed to obtain the estimated biomass data for the cultured fish in the net cage. This step utilizes acoustic scanning technology to achieve continuous and non-destructive monitoring of the total biomass of the fish in the net cage, overcoming the shortcomings of traditional manual sampling and weighing methods, such as high stress on the fish and insufficient sample representativeness.

[0064] Step 4: Perform time-series correlation analysis on the fish feeding behavior activity data obtained in Step 2 and the water quality and meteorological parameter data collected in Step 1 to construct a dynamic response function of fish feeding intensity to environmental factors.

[0065] Four environmental factors—dissolved oxygen concentration, water temperature, water flow velocity, and sea surface wind speed—were used as independent variables influencing fish feeding intensity, while the fish feeding behavior activity index was used as the dependent variable. A nonlinear regression model was established between the dependent variable and the four environmental factors, using hourly data from 30 consecutive days as the sample time window.

[0066] The nonlinear regression model adopts the form

[0067]

[0068] in The activity index of fish feeding behavior at hour (t) is given. The maximum feeding activity of farmed fish under optimal environmental conditions is set to 1. The dissolved oxygen concentration at hour (t) is... Let (t) be the water temperature at hour (t). Let (t) be the water flow velocity at hour (t). Let (t) be the sea surface wind speed at hour (t). This is the dissolved oxygen response function. Let the water temperature response function be... Let be the flow velocity response function. This is the wind speed response function.

[0069] Dissolved oxygen response function The function is piecewise: when the dissolved oxygen concentration is below the set critical threshold of 3.5 mg / L, the feeding activity decreases exponentially with the dissolved oxygen concentration; when the dissolved oxygen concentration is within the set suitable range of 3.5 mg / L to 8.0 mg / L, the feeding activity maintains its maximum value; when the dissolved oxygen concentration exceeds the set upper limit of 8.0 mg / L, the feeding activity decreases slowly.

[0070] Water temperature response function It is a Gaussian function, and its peak value is located at the optimal growth temperature of 24 degrees Celsius for farmed fish.

[0071] Flow response function It is an inverse proportional function; the higher the flow rate, the lower the feeding activity.

[0072] Wind speed response function It is an exponentially decaying function.

[0073] The response function parameter values ​​obtained through the above nonlinear regression model calibration are used as the dynamic response function of fish feeding intensity to environmental factors and stored in the model parameter library of the land-based data processing center. This step establishes a quantitative mathematical relationship between environmental factors and fish feeding behavior, enabling subsequent virtual simulations to realistically reflect the dynamic impact of changes in the deep-sea aquaculture environment on fish growth.

[0074] Step 5: Based on the multi-source data collected in Step 1, construct a virtual simulation environment for the target deep-sea aquaculture cage.

[0075] A virtual simulation platform is built in the land-based data processing center. The core computing engine of the virtual simulation platform is built based on computational fluid dynamics solvers and multibody dynamics solvers.

[0076] The water depth and topography data, seabed sediment type data, cage structure geometric parameters, mooring system configuration parameters, and netting material physical parameters of the sea area where the target deep-sea aquaculture cages are located are input into the virtual simulation platform as the basic simulation parameters. The cage structure geometric parameters include cage perimeter, cage depth, cage volume, floating frame structure dimensions, and weight distribution. The mooring system configuration parameters include anchor chain segment length, anchor chain diameter, anchor chain material grade, and anchor point spatial coordinates. The netting material physical parameters include netting mesh size, netting wire diameter, netting elastic modulus, and netting damping coefficient.

[0077] The virtual simulation platform uses real-time collected sea surface wind speed, wind direction, and water flow velocity data as external environmental excitation boundary conditions to drive a computational fluid dynamics solver to calculate the flow field distribution and hydrodynamic load distribution in the sea area surrounding the gabion. The calculated hydrodynamic load distribution is then used as input to drive a multibody dynamics solver to calculate the motion response of the gabion's floating structure and the tension response of the mooring system. Finally, the motion response of the gabion's floating structure is used as input to further calculate the deformation state and stress distribution of the gabion under the action of water flow.

[0078] The motion response data and mooring tension data of the net cages output by the virtual simulation platform are compared and verified hourly with the motion response data of the net cages collected by the accelerometer and the mooring tension data collected by the tension sensor. When the root mean square error between the simulation results and the actual monitoring data exceeds a preset threshold, the parameter correction program of the simulation model is automatically triggered to adjust the net damping coefficient and the mooring system stiffness coefficient until the simulation accuracy meets the preset requirements. The virtual simulation environment constructed in this step can reproduce the structural response and hydrodynamic behavior of deep-sea aquaculture net cages under actual sea conditions with high fidelity, providing an accurate digital twin foundation for subsequent extreme event simulation and growth process extrapolation.

[0079] Step 6: In the virtual simulation environment constructed in Step 5, inject virtual environmental disturbance events to simulate the impact of extreme marine dynamic events on the biomass and survival rate of farmed fish, and obtain the survival rate decay coefficient of farmed fish.

[0080] When the virtual simulation platform detects that the sea surface wind speed exceeds the first preset threshold of 15 meters per second for three consecutive hours, it triggers a typhoon precursor warning. When the virtual simulation platform detects that the sea surface wind speed exceeds the second preset threshold of 24.5 meters per second for one consecutive hour, it triggers a typhoon passage simulation event.

[0081] After the typhoon passage simulation event is triggered, the biomass estimation data and individual body length distribution data of the farmed fish in the net cage at the current moment are obtained as the initial state of the simulation.

[0082] Based on real-time sea surface wind speed data collected by meteorological monitoring stations, the maximum expected wind speed and the corresponding exceedance probability of the current typhoon event within the next 12-hour, 24-hour, 48-hour, and 72-hour time windows are calculated using an extreme value distribution model.

[0083] The maximum expected wind speed is input into the verified virtual simulation environment, which then drives the virtual simulation environment to simulate the cage's motion response, net deformation state, and internal flow field changes in the aquaculture water under the corresponding wind speed conditions.

[0084] Based on the data on the cage motion response amplitude and water velocity distribution output from the virtual simulation environment, combined with experimental data on the stress response of farmed fish under conditions of severe water disturbance, a stress intensity index is calculated. The stress intensity index is calculated by normalizing the cage motion response amplitude and the water velocity and then taking a weighted average.

[0085] The survival rate attenuation coefficient of farmed fish during typhoon events was calculated based on the stress intensity index. The mapping relationship between the survival rate attenuation coefficient and the stress intensity index was calibrated using post-disaster survey data from historical typhoon events.

[0086] By applying the survival rate attenuation coefficient to the biomass estimate at the initial state of the simulation, the expected remaining biomass after the typhoon event is obtained. This step is the first to achieve a quantitative assessment of the survival rate attenuation of farmed fish populations under the impact of extreme marine dynamic events, filling the gap in the ability of existing aquaculture management methods to quantitatively assess extreme risks.

[0087] Step 7: In the virtual simulation environment constructed in Step 5, the growth process of the farmed fish population is simulated and deduced using the dynamic response function of the fish feeding intensity calibrated in Step 4 and the survival rate decay coefficient obtained in Step 6 as constraints, and the expected biomass accumulation curves at each time point in the farming cycle are obtained.

[0088] The aquaculture cycle was divided into discrete time steps of one day. Within each time step, data on dissolved oxygen concentration, water temperature, water flow velocity, sea surface wind speed, and individual body length distribution of the farmed fish were collected.

[0089] The dynamic response function of fish feeding intensity to environmental factors is invoked to calculate the fish feeding behavior activity index corresponding to the current time step.

[0090] The actual feeding amount is calculated using a bioenergetics feeding model based on the fish feeding activity index and the total biomass of the fish at the current time step. This bioenergetics feeding model determines the daily feeding rate using the standard metabolizable energy requirement of the cultured fish, a feeding activity correction coefficient, the current biomass, and the current water temperature. The baseline daily feeding rate is determined by referring to a table based on the average fish weight and water temperature, and then multiplied by the fish feeding activity index for correction to obtain the actual daily feeding rate.

[0091] The increase in fish biomass at the current time step is calculated based on the actual daily feeding rate, current water temperature, and feed conversion ratio (FCR) parameters of the cultured fish fry. The FCR parameters vary with fish size and water temperature and are expressed as a piecewise function.

[0092] Check if the current time step has triggered a typhoon passage simulation event. If not, update the total biomass of the fish population with the increase in fish biomass; if it has been triggered, first reduce the current biomass based on the survival rate decay coefficient, and then calculate the growth increment for subsequent time steps based on this.

[0093] The updated total biomass of the fish population and the individual body length distribution data of the cultured fish population obtained by back-calculating from the total biomass through the body length-weight allometric growth relationship are used as the initial state for the next time step.

[0094] Repeat the above calculations until the daily simulation of the entire breeding cycle is completed, generating the expected biomass accumulation curves for each time point within the breeding cycle. This step dynamically couples the environmental response function, the extreme event attenuation coefficient, and the bioenergetics feeding model in a virtual simulation environment, ensuring that the expected biomass accumulation curves not only reflect normal growth patterns but also accurately include the impact of environmental fluctuations and extreme events on the growth process.

[0095] Step 8: Combine the expected biomass accumulation curve obtained in Step 7 with market price data to generate the output value prediction results of the target deep-sea aquaculture cages.

[0096] Historical market price sequence data for farmed fish species are obtained from a market price database connected to a land-based data processing center. The historical market price sequence data spans 3 to 5 years, with a daily time granularity.

[0097] A seasonal autoregressive integral moving average model is used to model and analyze historical market price series data, identify trend components, seasonal components and random fluctuation components in the historical market price series data, and generate the expected market unit price for each future date within the breeding cycle.

[0098] The expected biomass value at each time point on the expected biomass accumulation curve is multiplied by the expected market price at that time point to obtain the expected output value at that time point. The expected output value is calculated using the following formula:

[0099]

[0100] in The expected output value on day (t) The expected biomass on day (t) Let be the expected market unit price on day (t).

[0101] By connecting the expected output values ​​for each day during the aquaculture cycle into a curve, a production value prediction result for the target deep-sea aquaculture cages is generated. This prediction result is then visualized in chart form on the monitoring terminal of the land-based data processing center. This step organically integrates biomass prediction results with market price information, enabling aquaculture managers to intuitively grasp the expected returns at various future time points before stocking or during the aquaculture process.

[0102] Step 9: Based on the output value prediction results generated in Step 8, construct a multi-scenario simulation scheme and conduct output value comparison analysis to output the management scheme with the optimal output value.

[0103] At least three alternative management scenarios are set up in the virtual simulation environment. Each alternative management scenario consists of a combination of adjustable management parameters, including seedling density, initial seedling size, harvest time, feed type and grade, and a batch thinning strategy. The batch thinning strategy includes the thinning time point and the thinning ratio.

[0104] For each alternative management scenario, steps 6 to 8 are repeated to generate the expected biomass accumulation curve and output prediction results for each alternative management scenario.

[0105] The output value forecasts for all alternative management scenarios are compared and displayed on the same coordinate system. The comparative analysis includes the peak output value of each alternative management scenario, the time point when the output value reaches the peak, the cumulative output value curve shape during the breeding cycle, and the fluctuation characteristics of output value over time.

[0106] Among all alternative management scenarios, the one with the largest area under the cumulative output curve is selected as the optimal management plan. This step, through multi-scenario simulation comparison, transforms aquaculture management decision-making from experience-driven to data- and model-driven scientific comparison, ensuring the optimal output of the selected management plan under given constraints.

[0107] Step 10: Apply the optimal management plan for output value output from Step 9 to actual aquaculture management, and continuously update the output value prediction results based on real-time monitoring data.

[0108] The optimal management plan, which includes seedling density, initial seedling specifications, harvest time, feed type and grade, and batch thinning strategy, will be distributed to the on-site management terminal of the target deep-sea aquaculture cages.

[0109] During the implementation of the management plan that optimizes output, the data collection, processing, and forecasting process from step 1 to step 8 is continuously executed.

[0110] Every 30 days, the estimated biomass data of the fish population in the net cages, acquired using a panoramic multibeam scanning sonar device, is compared with the expected biomass data obtained from the simulation in step 7. When the relative deviation between the estimated biomass data and the expected biomass data exceeds ±15%, the simulation model recalibration process is triggered. Using the current estimated biomass data of the fish population in the net cages as the new initial state for the simulation, the process returns to step 5 to adjust the model parameters of the virtual simulation environment, and steps 6 to 9 are re-executed to generate the corrected output value prediction results and update the management plan recommendations. This process constitutes a closed-loop iterative mechanism from real-time monitoring to model correction, from prediction updates to plan adjustments, ensuring that the prediction results and management recommendations remain highly consistent with the actual situation at the aquaculture site throughout the entire aquaculture cycle.

[0111] This invention, through the organic combination of the aforementioned 10 steps, constructs a complete technical chain from environmental perception, biological behavior recognition, structural dynamics simulation, growth process deduction, market information integration to management scheme optimization. Each step forms a closed loop with sequential coupling, linked by data flow and model parameters. The output of one technical step serves as the input of the next, resulting in a significant synergistic effect. This effectively solves core technical problems in existing technologies, such as insufficient accuracy in predicting deep-sea aquaculture output, weak correlation between environment and organisms, difficulty in quantifying extreme risks, and a lack of a global optimization perspective in management decisions.

[0112] Example

[0113] This specific embodiment uses a 120m circumference, 15m depth HDPE gravity cage deployed in a deep-sea aquaculture demonstration area in the South my country Sea as the target cage for deep-sea aquaculture, with large yellow croaker as the cultured fish species, to illustrate the method of the present invention in detail. The average annual water depth of this aquaculture area is 28m, the seabed is sandy and muddy, and it is significantly affected by the East Asian monsoon, with typhoon activity occurring from June to October each year.

[0114] Step 1: Collect multi-source data of target deep-sea aquaculture cages to build a data foundation layer.

[0115] Multi-source data acquisition devices are deployed in the sea area where the target deep-sea aquaculture cages are located. The multi-source data acquisition devices specifically include:

[0116] The underwater multi-parameter water quality sensor has five sampling points: one at the center of the net cage and four points along the edges of the net cage in the east, south, west, and north directions. Each sampling point is arranged in three layers vertically: at a surface water depth of 2m, at half the depth of the middle net cage (7.5m), and at a bottom water depth of 13m, 2m above the bottom of the net cage. Each underwater multi-parameter water quality sensor can simultaneously measure six water quality parameters: dissolved oxygen concentration, water temperature, salinity, pH value, ammonia nitrogen concentration, and water flow velocity. The measurement range for dissolved oxygen concentration is 0 mg / L to 20 mg / L, with an accuracy of ±0.1 mg / L; the measurement range for water temperature is -5℃ to 45℃, with an accuracy of ±0.05℃; the measurement range for salinity is 0 PSU to 50 PSU, with an accuracy of ±0.1 PSU; the measurement range for pH value is 0 to 14, with an accuracy of ±0.01; the measurement range for ammonia nitrogen concentration is 0 mg / L to 10 mg / L, with an accuracy of ±0.01 mg / L; and the measurement range for water flow velocity is 0 m / s to 3 m / s, with an accuracy of ±0.01 m / s. All underwater multi-parameter water quality sensors are uniformly set to a sampling frequency of once per hour.

[0117] The meteorological monitoring station is installed on a buoy platform outside the net cage, which is anchored 50m from the edge of the net cage on the northwest side. The station is equipped with wind speed and direction sensors, air temperature sensors, atmospheric pressure sensors, and rainfall sensors. The measurement range for sea surface wind speed is 0m / s to 60m / s, with an accuracy of ±0.2m / s; the measurement range for wind direction is 0° to 360°, with an accuracy of ±2°; the measurement range for air temperature is -20℃ to 50℃, with an accuracy of ±0.1℃; the measurement range for atmospheric pressure is 800hPa to 1100hPa, with an accuracy of ±0.5hPa; and the measurement resolution for rainfall is 0.2mm. The meteorological monitoring station is set to collect data once per hour.

[0118] The panoramic multibeam scanning sonar device is mounted on a fixed bracket on the inner wall of the net cage, at a water depth of 10m. The device has a detection range of 35m, a horizontal scanning angle of 360°, a vertical scanning angle of 65°, a horizontal resolution of 0.5°, a vertical resolution of 2°, and a single complete scan cycle of 12s. The scanning frequency is 500kHz. Data is acquired once daily, with each acquisition lasting three complete scan cycles.

[0119] The underwater binocular stereo vision camera system is installed at the center of the top of the net cage, fixed 2 meters below the cage's floating frame by a stainless steel bracket. The system consists of two industrial-grade underwater cameras with a baseline distance of 25cm. Each camera has an image resolution of 1920×1080 pixels and a frame rate of 30 frames per second, and is equipped with an underwater LED supplemental lighting source. The system captures video once daily, with each capture session consisting of a 10-minute video sequence, from 8:00 AM to 8:10 AM daily.

[0120] Tension sensors are installed at key nodes in the gabion mooring system, specifically at the shackles connecting the mooring cables and anchor chains. A total of eight tension sensors are installed, each corresponding to one of the eight mooring cables of the gabion. The tension sensors have a range of 0 kN to 200 kN, a measurement accuracy of ±0.5 kN, and a sampling frequency of once per minute.

[0121] Accelerometers are installed at key nodes of the gabion floating structure, specifically at eight nodes evenly distributed circumferentially along the gabion. Each node is equipped with a triaxial accelerometer, capable of simultaneously measuring acceleration in the X, Y, and Z axes. The accelerometers have a range of ±4g, a measurement accuracy of ±0.005g, and a sampling frequency of once per minute.

[0122] Multi-source data acquired by the multi-source data acquisition device is transmitted in real time to the land-based data processing center via a 5G mobile communication network. The land-based data processing center deploys data receiving servers and data processing servers. It performs timestamp alignment on all received multi-source data, using the hourly acquisition time of the underwater multi-parameter water quality sensor as the baseline timestamp and aligning the timestamps of the remaining data to the nearest hourly time. Simultaneously, outlier removal is performed. The outlier removal rules are as follows: for continuously acquired data, values ​​deviating from the average of the three adjacent data points by more than three times the standard deviation are identified as outliers and removed. Linear interpolation is used to fill the data gaps after removal. The processed data is then stored hierarchically in the time-series database of the land-based data processing center according to date and time, forming a structured data foundation layer.

[0123] Step 2: Process the underwater image data collected in Step 1 to obtain the individual body length distribution data and feeding behavior activity data of the farmed fish population.

[0124] The underwater image and video sequences acquired in step 1 using the underwater binocular stereo vision camera are processed. Using a 10-minute video sequence acquired daily as the processing unit, single-frame images are extracted at a rate of one frame per second, resulting in 600 frames extracted from the 10-minute video to form the image sequence sample set for that day.

[0125] For each frame of the image sequence sample set, a fish target detection model based on a deep convolutional neural network (DCNN) is used to detect and identify fish targets. In this embodiment, the DCNN uses the YOLOv5 architecture, which consists of three parts: a backbone feature extraction network, a neck feature fusion network, and a detection head. The backbone feature extraction network uses the CSPDarknet53 structure, and the neck feature fusion network uses the PANet structure. The training dataset for the DCNN is a pre-labeled underwater image sample set of large yellow croaker, containing 5000 images with labeled large yellow croaker bounding boxes. The DCNN outputs the bounding box coordinates of each identified fish in the image, with the bounding box coordinates formatted as the top-left and bottom-right pixel coordinates.

[0126] For each fish target detected in each frame of the image, based on the calibration parameters of the underwater binocular stereo vision camera, the spatial coordinates and estimated body length of the fish target in three-dimensional space are calculated using the principle of binocular stereo vision. The underwater binocular stereo vision camera was calibrated in a land-based pool before use. The calibration parameters include the intrinsic parameter matrix of the left camera, the intrinsic parameter matrix of the right camera, the rotation matrix, and the translation vector. The method for calculating the estimated body length is as follows: based on the calibration parameters, the spatial three-dimensional coordinates of the two endpoints of the fish target's bounding box, namely the snout and the end of the caudal fin, are calculated using the parallax method. The three-dimensional Euclidean distance between the two endpoints is then calculated as the estimated body length of the fish, in cm. The estimated body length values ​​of all fish targets in the same frame are statistically analyzed, and body length intervals are divided according to a group interval of 2 cm. The body length distribution histogram of the fish in that frame is recorded. The arithmetic mean of the body length distribution histograms of all 600 frames of images within a day is taken for each body length interval to obtain the individual body length distribution data of the farmed fish for that day. The specific data on the individual body length distribution of the farmed fish on that day include: the mean body length, the median body length, the standard deviation of body length, and the skewness of the body length distribution. The skewness of the body length distribution is calculated using the formula...

[0127]

[0128] The calculation is performed, where (S) is the distribution skewness and (n) is the total number of fish targets detected that day. Let (i) be the estimated body length of the (i)th fish. The average body length This represents the standard deviation of body length.

[0129] For an image sequence sample set, a fish feeding behavior recognition model based on a 3D convolutional neural network was used to quantitatively analyze the feeding behavior of fish schools. The 3D convolutional neural network adopts the I3D architecture, which expands the 2D convolutional kernel into a 3D convolutional kernel, and can simultaneously extract the spatial and temporal features of the image sequence. The training dataset for the fish feeding behavior recognition model consists of manually labeled video samples of fish feeding behavior, with labels including strong feeding, weak feeding, and no feeding. The fish feeding behavior recognition model performs quantitative analysis on the fish behavior in the image sequence sample set using three indicators:

[0130] Extract the velocity of the fish school's centroid in each frame. The spatial coordinates of the centroid are obtained by calculating the arithmetic mean of the spatial coordinates of all identified fish targets in that frame. The velocity of the centroid is calculated by dividing the Euclidean distance between the centroids of adjacent frames by the inter-frame time interval, in cm / s.

[0131] Calculate the average curvature of the fish school's centroid trajectory across 5 consecutive frames. For the spatial coordinate sequence of the fish school's centroid across 5 consecutive frames, use cubic spline interpolation to fit the spatial trajectory curve, and calculate the curvature at each point on the curve. Curvature is defined as the derivative of the tangent angle with respect to the arc length, expressed in rad / cm. The arithmetic mean of the absolute values ​​of the curvature at each point across the 5 frames is taken as the average curvature for that 5-frame interval.

[0132] Calculate the fish school aggregation index. For the spatial coordinates of all identified fish targets in each frame of the image, calculate the Euclidean distance between any two individuals, take the arithmetic mean of the Euclidean distances of all individual pairs to obtain the average individual spacing, and take the reciprocal of the average individual spacing as the fish school aggregation index, with the unit being m⁻¹.

[0133] Three indicators—the movement speed of the fish school's centroid, the average curvature of the fish school's centroid trajectory, and the fish school aggregation index—were normalized. The normalization method was as follows: for movement speed, the maximum movement speed observed and recorded for this farmed fish species under feeding conditions was used as the normalization baseline value; in this example, the maximum movement speed of the large yellow croaker was observed to be 45 cm / s. For average curvature, 0.2 rad / cm was used as the normalization baseline value; and for the fish school aggregation index, 5 m⁻¹ was used as the normalization baseline value. The normalized values ​​of the three indicators were then weighted and summed to calculate the fish school feeding behavior activity index. The weighting coefficients for the weighted summation were extracted from the training dataset using principal component analysis; in this example, the weight for movement speed was 0.4, the weight for average curvature was 0.3, and the weight for the fish school aggregation index was 0.3. The value range of the fish feeding behavior activity index is from 0 to 1. 0 indicates that the fish are in a static or extremely dispersed state where they are not feeding at all, while 1 indicates that the fish are in a strong feeding state where they are highly gathered and swimming rapidly.

[0134] Step 3: Process the acoustic scanning data collected in Step 1 to obtain biomass estimation data of the fish population in the net cage.

[0135] The acoustic scanning data acquired in step 1 using the panoramic multibeam scanning sonar device is processed. The panoramic multibeam scanning sonar device emits acoustic pulses through a transducer array and receives echo signals. Beamforming processing is performed on the received echo signals using a delay-summing algorithm to generate three-dimensional point cloud data within the acoustic scanning area. Each point in the three-dimensional point cloud data contains five attributes: X-axis spatial coordinates, Y-axis spatial coordinates, Z-axis spatial coordinates, acoustic backscattering intensity value, and target intensity value. The X-axis and Y-axis spatial coordinates are horizontal plane coordinates with the center of the net cage as the origin, and the Z-axis spatial coordinates are depth coordinates with sea level as the reference, both in meters (m). The acoustic backscattering intensity value is in dB. The target intensity value is calculated based on the echo characteristics of a single target and is used to distinguish fish from non-biological targets.

[0136] The 3D point cloud data was divided into multiple water layers according to the depth of the aquaculture water. The water depth of the net cage aquaculture was 15m, and from the water depth of 0m to 15m, it was divided into 15 water layers with a thickness of 1m each: 0m-1m layer, 1m-2m layer, ..., 14m-15m layer. For the 3D point cloud data within each water layer, the density-based spatial clustering algorithm, DBSCAN algorithm, was used for clustering. The neighborhood radius parameter of the DBSCAN algorithm was set to 1.5m, and the minimum number of neighboring points parameter was set to 8. Clustering was performed on the point cloud data within each water layer to identify clusters belonging to fish groups. Discrete points that did not belong to any cluster were considered noise points and filtered out. Noise points may originate from the acoustic reflection of netting, floating structures, or suspended particles in the water. For each identified fish group cluster, the number of point clouds contained in the cluster was counted, and the arithmetic mean of the acoustic backscattering intensity values ​​of all points within the cluster was extracted to obtain the average acoustic backscattering intensity value of the cluster.

[0137] A mapping model was established between acoustic backscattering intensity and individual fish biomass. At the start of the rearing cycle, initial fish fry were sampled and weighed. In this example, the initial stocking date was April 10, 2025, with a total of 25,000 large yellow croaker fry stocked. 200 sample fish were randomly selected from the fry and weighed individually using an electronic balance with an accuracy of 0.1g, calculating the average weight of each individual fish to be 125.6g. The day after stocking, on April 11, 2025, a standard scan of the initial fish population was performed using a panoramic multibeam sonar device, recording the average acoustic backscattering intensity as -42.3dB. A baseline correspondence between initial biomass and acoustic intensity was established: an average individual fish weight of 125.6g corresponds to an average acoustic backscattering intensity of -42.3dB. During the rearing cycle, a calibration period of 30 days was used, during which sample fish were extracted from the net cages using a fish suction pump for actual weighing. The fish suction pump is a vacuum-type live fish transfer device, extracting 80 to 120 sample fish at a time. The average weight of the sample fish, obtained from actual weighing, is regressed against the average acoustic backscattering intensity value acquired by the panoramic multibeam sonar device on the same day, using a power function model.

[0138]

[0139] The fitting is performed, where (W) is the weight of a single fish in g; The average acoustic backscattering intensity is expressed in dB; (a) and (b) are the parameters to be fitted. Parameters (a) and (b) are continuously corrected through regression fitting in each calibration period to ensure the mapping model remains accurate throughout the entire breeding cycle. In this embodiment, the parameter values ​​obtained after fitting in the first calibration period, i.e., day 30 of breeding, are (a=0.023) and (b=0.78).

[0140] Based on the mapping relationship model and the point cloud count and average acoustic backscattering intensity of each fish population cluster within each water layer, the estimated fish biomass for each water layer is calculated. For the (j)th water layer, the estimated fish biomass for that water layer is... According to the formula

[0141]

[0142] Calculation, where Let (j) be the number of fish clusters identified in the (j)th water layer. Let be the number of point clouds representing the (k)th fish cluster in the (j)th water layer. Let be the average acoustic backscattering intensity value of the (k)th fish cluster in the (j)th water layer. The mapping relationship model provides a transformation function for the weight of individual fish, in grams. The total biomass estimate of the fish population in the net cage is obtained by summing the biomass estimates of all water layers. The unit is kg, that is

[0143]

[0144] Step 4: Perform time-series correlation analysis on the fish feeding behavior activity data obtained in Step 2 and the water quality and meteorological parameter data collected in Step 1 to construct a dynamic response function of fish feeding intensity to environmental factors.

[0145] Four environmental factors—dissolved oxygen concentration, water temperature, water flow velocity, and sea surface wind speed—collected in step 1, were used as independent variables influencing the feeding intensity of the fish. Dissolved oxygen concentration and water temperature data were obtained from underwater multi-parameter water quality sensors at a depth of 7.5m in the middle layer of the net cage, as this location best represents the water quality within the net cage. Water flow velocity data were obtained as the arithmetic mean of flow velocity measurements at 7.5m in the middle layer at four locations along the edge of the net cage. Sea surface wind speed data were obtained from wind speed sensors at a meteorological monitoring station. The fish feeding behavior activity index obtained in step 2 was used as the dependent variable. A nonlinear regression model was established between the dependent variable and the four environmental factors, using hourly data over 30 consecutive days as the sample time window.

[0146] The nonlinear regression model uses a product form:

[0147]

[0148] in The activity index of fish feeding behavior at hour (t) is given. The maximum feeding activity of farmed fish under optimal environmental conditions is set to 1. The dissolved oxygen concentration at hour (t) is expressed in mg / L. The water temperature at hour (t) is expressed in °C. The water flow velocity at hour (t) is expressed in m / s. The wind speed at the sea surface in hour (t) is expressed in m / s.

[0149] All four response functions were fitted and calibrated using measured data.

[0150] Dissolved oxygen response function This is a piecewise function. For large yellow croaker, the set critical threshold for dissolved oxygen concentration is 3.5 mg / L, the lower limit of the suitable range is 5.0 mg / L, and the upper limit of the suitable range is 8.0 mg / L. The specific form of the piecewise function is:

[0151]

[0152] in It has a low oxygen attenuation coefficient. The high oxygen decay coefficient was obtained by fitting 30 days of sample data. , When the dissolved oxygen concentration is below 3.5 mg / L, feeding activity decreases exponentially with increasing dissolved oxygen concentration; when the dissolved oxygen concentration is in the range of 3.5 mg / L to 8.0 mg / L, feeding activity remains at its maximum value of 1; when the dissolved oxygen concentration exceeds 8.0 mg / L, feeding activity decreases slowly, mainly due to the feeding inhibition effect caused by the risk of gas bubble disease in fish caused by supersaturated dissolved oxygen.

[0153] Water temperature response function This is a Gaussian function. The optimal growth temperature range for large yellow croaker is 22℃ to 26℃; in this embodiment, the optimal growth temperature is taken as 24℃.

[0154]

[0155] in The optimal growth temperature is set at 24℃. The temperature tolerance width parameter characterizes the sensitivity of farmed fish fry to water temperature deviations. It was obtained by fitting data from 30 days of samples. The function reaches its maximum value of 1 when the water temperature is 24℃; when the water temperature deviates from 24℃, the feeding activity decreases according to the Gaussian curve.

[0156] Flow response function This is an inverse proportional function. Increased water flow velocity increases the swimming energy consumption of fish and disrupts their schooling behavior, leading to a decrease in feeding activity. The inverse proportional function takes the form of:

[0157]

[0158] in The strength coefficient is affected by the flow velocity. The nonlinear exponent is influenced by flow velocity. It was obtained by fitting data from 30 days of sample data. , The function value is 1 when the water flow velocity is 0 m / s; as the flow velocity increases, the feeding activity gradually decreases.

[0159] Wind speed response function This is an exponentially decaying function. Sea surface wind speed indirectly affects the disturbance of the water within the net cage by generating waves and wind-driven currents, thus interfering with the feeding behavior of fish. The form of the exponentially decaying function is:

[0160]

[0161] in This is the wind speed attenuation coefficient. It was obtained by fitting 30 days of sample data. The function takes a value of 1 when the wind speed at the sea surface is 0 m / s; as the wind speed increases, the feeding activity index decreases.

[0162] All response function parameter values ​​obtained through the above nonlinear regression model calibration are used as the dynamic response function of the fish feeding intensity of the aquaculture cage to environmental factors, and stored in the model parameter library of the land-based data processing center.

[0163] Step 5: Based on the multi-source data collected in Step 1, construct a virtual simulation environment for the target deep-sea aquaculture cage.

[0164] A virtual simulation platform was built at the land-based data processing center. The core computing engine of the virtual simulation platform is based on a computational fluid dynamics (CFD) solver and a multibody dynamics solver. The CFD solver uses the OpenFOAM open-source framework and solves the Reynolds-averaged Navier-Stokes equations using the finite volume method. The turbulence model uses the k-ωSST model. The multibody dynamics solver uses the OrcaFlex software solver and models the anchor chains and mooring cables using the lumped mass method, and models the gabion floating structure using the rigid body motion equations.

[0165] The water depth and topography data, seabed sediment type data, cage structure geometric parameters, anchoring system configuration parameters, and netting material physical parameters of the sea area where the deep-sea aquaculture target cages are located are input into the virtual simulation platform as the basic simulation parameters.

[0166] The water depth and topography data are obtained from nautical chart water depth measurement data. In this embodiment, the water depth at the location of the net cage is 28m. The water depth variation within a radius of 200m centered on the net cage does not exceed ±1.5m, and the seabed topography is relatively flat.

[0167] The seabed sediment type data comes from seabed geological sampling and analysis. In this embodiment, the sediment is sandy and muddy, with a median particle size of 0.15 mm and a sediment friction coefficient of 0.35.

[0168] The specific geometric parameters of the cage structure include: a cage perimeter of 120m, a cage depth of 15m, a cage volume of approximately 16200m³, a floating frame structure composed of double rows of HDPE pipes, with a single pipe outer diameter of 315mm, a wall thickness of 23mm, a center distance of 1.8m between the double rows of pipes, and a total weight of 18600kg, with the weight evenly distributed along the circumference of the floating frame.

[0169] The specific configuration parameters of the mooring system include: the net cage adopts an 8-point mooring system, with a total of 8 anchor chains. Each anchor chain consists of two parts: a surface mooring cable and a submerged anchor chain. The surface mooring cable is a 48mm diameter polyester fiber cable with a length of 35m, and the submerged anchor chain is a 38mm diameter sluged anchor chain with a length of 120m. The anchor chain material grade is R3, and the breaking load is 980kN. The spatial coordinates of the 8 anchor points, with the center of the net cage as the origin, are as follows: Anchor point 1 is located at 0° east of north. Anchor point 2 is located 150m away in the direction of 45° east of north, anchor point 3 is located 150m away in the direction of 90° east of north, anchor point 4 is located 150m away in the direction of 135° east of north, anchor point 5 is located 150m away in the direction of 180° east of north, anchor point 6 is located 150m away in the direction of 225° east of north, anchor point 7 is located 150m away in the direction of 270° east of north, and anchor point 8 is located 150m away in the direction of 315° east of north.

[0170] The specific physical parameters of the mesh material include: the mesh is made of ultra-high molecular weight polyethylene fiber woven mesh, the mesh size is 25mm, the mesh wire diameter is 2.8mm, and the elastic modulus of the mesh is 1.2×10⁻⁶. 8 Pa, the mesh damping coefficient is 0.18.

[0171] The virtual simulation platform operates as follows: The sea surface wind speed, wind direction, and water flow velocity data collected in real-time during step 1 are used as external environmental excitation boundary conditions. In this embodiment, wind direction and sea surface wind speed data are used together to determine the magnitude and direction of the wind load. The wind load is calculated according to the wind load calculation formula in the CCS "Marine Mobile Platform Classification Specification" and acts on the part of the gabion floating structure above the water surface. The water flow velocity data is used to drive the computational fluid dynamics solver to calculate the flow field distribution and hydrodynamic load distribution in the sea area surrounding the gabion. The flow field calculation domain is a 200m × 200m × 30m cuboid region centered on the gabion, discretized using a structured hexahedral mesh with approximately 3.8 million grids. The computational fluid dynamics solver outputs the velocity distribution, pressure distribution, and hydrodynamic load distribution acting on the gabion surrounding the gabion. The calculated hydrodynamic load distribution is used as input to drive the multibody dynamics solver to calculate the motion response of the gabion floating structure and the tension response of the mooring system. The multibody dynamics solver uses the time-domain integration method with a time step of 0.02 s and a simulation duration of 600 s for each simulation condition. The motion response data of the gabion float structure is used as input to further calculate the deformation state and stress distribution of the gabion under water flow. The gabion deformation calculation employs a gabion structure model based on the finite element method.

[0172] The motion response data and mooring tension data of the cage output by the virtual simulation platform are compared and verified hourly with the actual motion response data and mooring tension data of the cage collected by the accelerometer and tension sensor in step 1. The indicators for comparing the motion response data of the cage include three degrees of freedom of the floating frame: roll angle, pitch angle, and heave displacement. When the root mean square error between the simulation result and the actual monitoring data exceeds a preset threshold, the parameter correction program of the simulation model is automatically triggered. In this embodiment, the root mean square error thresholds for roll angle, pitch angle, and heave displacement are 0.8°, 0.12m, and 8kN, respectively. The parameter correction program adopts the particle swarm optimization algorithm, with the mesh damping coefficient and the mooring system stiffness coefficient as the variables to be optimized, and minimizing the root mean square error between the simulation result and the actual monitoring data as the optimization objective. The mesh damping coefficient and the mooring system stiffness coefficient are adjusted iteratively until the simulation accuracy meets the preset requirements.

[0173] Step 6: In the virtual simulation environment constructed in Step 5, inject virtual environmental disturbance events to simulate the impact of extreme marine dynamic events on the biomass and survival rate of farmed fish, and obtain the survival rate decay coefficient of farmed fish.

[0174] Based on the meteorological parameter data collected in real time in step 1, the virtual simulation platform continuously monitors changes in sea surface wind speed. When the virtual simulation platform detects that the sea surface wind speed continuously exceeds the first preset threshold of 15 m / s for 3 consecutive hours, it triggers a typhoon precursor warning state. At this time, the typhoon emergency response process is initiated, including increasing the data collection frequency to once every 30 minutes and notifying aquaculture management personnel to prepare for typhoon resistance. When the virtual simulation platform detects that the sea surface wind speed exceeds the second preset threshold of 24.5 m / s for 1 consecutive hour, it triggers a typhoon passage simulation event.

[0175] After the typhoon simulation event is triggered, the virtual simulation platform executes the following simulation process:

[0176] First, obtain the estimated total biomass of the fish population in the net cage calculated in step 3 at the current moment. The individual body length distribution data of the farmed fish population obtained in step 2 are used as the initial state for simulation.

[0177] Secondly, based on real-time sea surface wind speed data collected by meteorological monitoring stations, the maximum expected wind speed and its corresponding exceedance probability for the current typhoon event within the next 12, 24, 48, and 72-hour time windows are calculated using an extreme value distribution model. The extreme value distribution model used in this embodiment is the generalized Pareto distribution, and the cumulative distribution function of the generalized Pareto distribution is...

[0178]

[0179] in For position parameters, For scale parameters, These are shape parameters. The three parameters were obtained through maximum likelihood estimation of historical typhoon wind speed data for this sea area, as described in this embodiment. , , Based on real-time wind speed data and generalized Pareto distribution parameters, the maximum expected wind speed for the next 12 hours is calculated to be 32.5 m / s with a 15% probability of exceeding the speed limit; the maximum expected wind speed for the next 24 hours is 38.2 m / s with an 8% probability of exceeding the speed limit; the maximum expected wind speed for the next 48 hours is 41.6 m / s with a 4% probability of exceeding the speed limit; and the maximum expected wind speed for the next 72 hours is 36.4 m / s with a 10% probability of exceeding the speed limit.

[0180] Then, the maximum expected wind speeds within the next 12, 24, 48, and 72-hour time windows are sequentially input into the virtual simulation environment verified in step 5. This drives the virtual simulation environment to simulate the cage's motion response, net deformation state, and internal flow field changes in the aquaculture water under the corresponding wind speed conditions. The simulation conditions for each time window use the maximum expected wind speed within that time window as a constant wind speed input, while also considering the combined effects of wind and waves. The wave spectrum used is the JONSWAP spectrum.

[0181] Based on the data of net cage motion response amplitude and internal water velocity distribution output from the virtual simulation environment, combined with the stress response experimental data of farmed fish under severe water disturbance conditions, a stress intensity index was calculated. The net cage motion response amplitude was represented by the root mean square values ​​of the roll angle and heave displacement. The stress intensity index was calculated as follows: the root mean square value of the net cage motion response amplitude was compared and normalized with the roll angle threshold of 12° and the heave displacement threshold of 0.5m determined in the large yellow croaker stress response experiment. The normalized baseline value for the roll angle was 12°, and the normalized baseline value for the heave displacement was 0.5m. The simulated output value of the internal water velocity was compared and normalized with the velocity threshold of 0.8m / s determined in the large yellow croaker stress response experiment. The weighted average of the above three normalized values ​​was taken as the stress intensity index, with weighting coefficients of 0.4 for roll angle, 0.3 for heave displacement, and 0.3 for internal water velocity.

[0182] The survival rate decline coefficient of farmed fish during typhoon events was calculated based on the stress intensity index. The mapping relationship between the survival rate decline coefficient and the stress intensity index was calibrated using post-typhoon survey data from historical typhoon events. This embodiment collected statistical data on fish mortality after seven typhoon events in the aquaculture area from 2020 to 2024, and established an empirical fitting curve between the stress intensity index and the survival rate decline. The fitting adopted the form of a Logistic function:

[0183]

[0184] in This is the survival rate attenuation coefficient, with a value ranging from 0 to 1, where 0 indicates no mortality and 1 indicates all deaths. (r) is the stress intensity index, ranging from 0 to 1; (r) is the attenuation sensitivity parameter. The threshold for half-lethal stress intensity was determined through fitting (r=8.5). That is, when the stress intensity index reaches 0.62, the survival rate decay coefficient is 0.5, indicating that the mortality rate of the fish population is 50%.

[0185] The calculated survival rate attenuation coefficient Biomass estimation applied to the initial state of the simulation The expected remaining biomass after the typhoon event ends is obtained as follows:

[0186]

[0187] The survival rate decay coefficient and expected remaining biomass data are stored in the land-based data processing center.

[0188] Step 7: In the virtual simulation environment constructed in Step 5, the growth process of the farmed fish population is simulated and deduced using the dynamic response function of the fish feeding intensity calibrated in Step 4 and the survival rate decay coefficient obtained in Step 6 as constraints, and the expected biomass accumulation curves at each time point in the farming cycle are obtained.

[0189] The farming cycle is divided into discrete time steps of one day. In this embodiment, the farming cycle for large yellow croaker is set to 240 days, from stocking on April 10, 2025 to harvesting on December 6, 2025. Within each time step, the virtual simulation platform performs the following calculations:

[0190] Read the dissolved oxygen concentration, water temperature, water flow velocity, and sea surface wind speed data collected in step 1 within this time step. In this embodiment, the arithmetic mean of the data from 0:00 to 23:00 on the same day is used as the representative value of the environmental factors for that day. At the same time, read the individual body length distribution data of the farmed fish obtained in step 2, and take the mean body length for subsequent calculations.

[0191] Using the dynamic response function of the fish feeding intensity to environmental factors, which was calibrated in step 4, calculate the fish feeding behavior activity index corresponding to the current time step. The dissolved oxygen concentration for the day Water temperature Water flow velocity and sea surface wind speed Substituting the daily average into the dynamic response function

[0192]

[0193] The activity index of the fish feeding behavior on that day was calculated.

[0194] Based on the activity index of fish feeding behavior The actual feeding amount is calculated using a bioenergetics feeding model based on the total biomass of the fish population at the current time step. The bioenergetics feeding model determines the daily feeding rate based on the standard metabolizable energy requirement (MAU) of the cultured fish species, a feeding activity correction factor, the current biomass, and the current water temperature. The relationship between the MAU and body weight of large yellow croaker is as follows: Where (SMR) is standard metabolizable energy in kJ / d and (W) is the weight of an individual fish in g. The basal daily feeding rate was determined by referring to a table based on the average weight of the fish and the water temperature. The table data was obtained from the Large Yellow Croaker Farming Technical Specifications. In this example, when the average weight of an individual fish was 200g and the water temperature was 22℃, the basal daily feeding rate was 2.8%. The basal daily feeding rate was multiplied by the fish feeding behavior activity index. After adjustments, the actual daily feeding rate is obtained. The formula for calculating the actual feeding amount is:

[0195]

[0196] in This represents the actual feeding amount on day (t), in kg. The total biomass of the fish population on day (t-1) is expressed in kg. The base daily feeding rate is expressed as a decimal. This represents the activity level of a fish's feeding behavior.

[0197] Based on the actual daily feeding rate, current water temperature, and feed conversion ratio (FCR) parameters of the cultured fish fry, the increase in fish biomass at the current time step is calculated. The FCR of large yellow croaker varies with fish size and water temperature, and is expressed as a piecewise function. When the weight of an individual fish is between 100g and 300g and the water temperature is between 20℃ and 26℃, the FCR is linearly interpolated between 1.45 and 1.65; when the weight of an individual fish is greater than 300g, the FCR increases linearly between 1.65 and 1.85. The formula for calculating the increase in fish biomass is:

[0198]

[0199] in This represents the increase in fish biomass on day (t), in kg. The feed conversion ratio is determined based on the current average weight of individual fish and water temperature.

[0200] Check if the typhoon passage simulation event defined in step 6 is triggered at the current time step. If not, update the total biomass of the fish population by the increase in fish biomass:

[0201]

[0202] If it has been triggered, first calculate the survival rate attenuation coefficient obtained in step 6. Reduce the current biomass, and then calculate the growth increment for subsequent time steps. The handling procedure for the day a typhoon event is triggered is as follows:

[0203]

[0204] The calculation of normal growth increments resumed the day after the typhoon ended, and the calculation was based on... As new initial biomass.

[0205] Updated total biomass data of the fish population This serves as the initial state for the next time step. Simultaneously, the individual body length distribution data of the cultured fish population is calculated from the total biomass using the body length-weight allometric growth relationship. The body length-weight allometric growth relationship for large yellow croaker is as follows: Where (W) is the weight of an individual fish in g; and (L) is the body length in cm. The updated average weight of an individual fish is calculated by dividing the updated total biomass of the fish population by the total number of fish in the population. Then, the updated mean body length is calculated by substituting the mean body length into the allometric growth formula. Assuming that the ratio of the standard deviation of the body length distribution to the mean remains constant, the individual body length distribution data of the cultured fish population is reconstructed.

[0206] Repeat the above calculation steps until the daily extrapolation of the entire 240-day breeding cycle is completed, ultimately generating the expected biomass accumulation curve for each time point within the breeding cycle. The expected biomass accumulation curve, with the number of breeding days as the horizontal axis and the expected biomass as the vertical axis, records the expected biomass value for each day within the breeding cycle.

[0207] Step 8: Combine the expected biomass accumulation curve obtained in Step 7 with market price data to generate the output value prediction results of the target deep-sea aquaculture cages.

[0208] Historical market price series data for large yellow croaker is obtained from the market price database connected to the land-based data processing center. The market price database is linked to the transaction data systems of major aquatic product wholesale markets nationwide, updating the wholesale market price of large yellow croaker daily. In this embodiment, the historical market price series data spans from January 1, 2022 to April 9, 2025, totaling 1195 daily data points, with a daily time granularity.

[0209] A seasonal autoregressive integral moving average (SARIMA) model is used to model and analyze historical market price series data. The parameters of the SARIMA model are denoted as SARIMA(p, d, q)(P, D, Q)_s, where p is the non-seasonal autoregressive order, d is the non-seasonal differencing order, q is the non-seasonal moving average order, P is the seasonal autoregressive order, D is the seasonal differencing order, Q is the seasonal moving average order, and s is the seasonal cycle length. In this embodiment, the model order is determined using the AIC information criterion, and the optimal model is identified as SARIMA(2, 1, 1)(1, 1, 1)_365. This means that the non-seasonal part includes a 2nd-order autoregressive, a 1st-order differencing, and a 1st-order moving average, while the seasonal part has a 365-day cycle and includes a 1st-order seasonal autoregressive, a 1st-order seasonal differencing, and a 1st-order seasonal moving average. The determined optimal SARIMA model is then used to fit the historical market price series data to identify the trend component, seasonal component, and stochastic fluctuation component in the historical market price series data. The trend component reflects the long-term direction of large yellow croaker market prices, the seasonal component reflects the regular price fluctuations during specific periods each year, such as around the Spring Festival, and the stochastic fluctuation component reflects price disturbances caused by temporary changes in market supply and demand. Based on the fitted SARIMA model, the expected market unit price for each future date within the breeding cycle is generated, namely, the expected market unit price sequence for 240 days from April 10, 2025 to December 6, 2025, with the unit of expected market unit price being yuan / kg.

[0210] Multiply the expected biomass value at each time point on the expected biomass accumulation curve generated in step 7 by the expected market price at that time point to obtain the expected output value at that time point. The formula for calculating the expected output value is as follows:

[0211]

[0212] in The expected output value on day (t) is expressed in yuan. The expected biomass on day (t) is expressed in kg. The expected market price on day (t) is expressed in yuan / kg.

[0213] The expected output value for each day during the aquaculture cycle is plotted as a curve to generate a forecast of the output value of the target deep-sea aquaculture cages. This forecast is then visualized in chart form on the monitoring terminal of the land-based data processing center. The visualization includes: a forecast output value curve, a dual-axis graph of the expected biomass accumulation curve and the expected market price curve, and a pie chart showing the monthly output value contribution percentage.

[0214] Step 9: Based on the output value prediction results generated in Step 8, construct a multi-scenario simulation scheme and conduct output value comparison analysis to output the management scheme with the optimal output value.

[0215] In the virtual simulation environment constructed in step 5, three alternative management scenarios are set up. The three alternative management scenarios consist of a combination of adjustable management parameters, including seedling density, initial seedling size, harvest time, feed type and grade, and batch thinning strategy.

[0216] The parameters for Alternative Management Scenario A, i.e., the baseline scenario, are set as follows: stocking density of 1.54 fish / m³, corresponding to a total stocking quantity of 25,000 fish; initial stocking size of 125.6g / fish; harvest time of the 240th day of culture, i.e., December 6, 2025; feed type and grade of ordinary compound feed, with a basic feed conversion ratio of 1.55 corresponding to a daily feeding rate of 2.8% under the conditions of individual fish weight of 200g and water temperature of 22℃; no batch thinning strategy is implemented, i.e., harvesting is carried out in one go throughout the entire process.

[0217] The parameters for Alternative Management Scenario B, namely the high-density batch harvesting scenario, are set as follows: stocking density of 2.16 fish / m³, corresponding to a total stocking quantity of 35,000 fish; initial stocking size of 125.6g / fish; two harvesting time nodes are set, the first harvesting time node is the 180th day of cultivation, i.e., October 7, 2025, and the second harvesting time node is the 240th day of cultivation, i.e., December 6, 2025; the feed type and grade is premium grade compound feed, and the corresponding basic feed conversion ratio parameter is 1.48 feed conversion ratio for a daily feeding rate of 2.6% under the same conditions; the batch thinning strategy is to thin 40% of the fish at the first harvesting time node, that is, to harvest and market 40% of the fish biomass in advance, and continue to cultivate the remaining 60% of the fish biomass until the second harvesting time node.

[0218] The parameters for alternative management scenario C, namely the low-density, large-size scenario, are set as follows: stocking density is 1.23 fish / m³, corresponding to a total stocking quantity of 20,000 fish; initial stocking size is 200.0g / fish, i.e., larger-sized fish are stocked; the harvest time is the 210th day of culture, i.e., November 6, 2025; the feed type and grade is premium-grade compound feed, and the corresponding basic feed conversion ratio parameter is 1.48 feed conversion ratio for a daily feeding rate of 2.6% under the same conditions; no batch thinning strategy is implemented.

[0219] For alternative management scenarios A, B, and C, steps 6 through 8 are repeated. During the repeated execution, the typhoon passage simulation event in step 6 is driven by the same historical meteorological data to ensure the comparability of the impact of extreme ocean dynamic events on different management scenarios; the growth process simulation in step 7 incorporates the seedling density, initial seedling size, feed type and grade, and batch thinning strategy parameters corresponding to each management scenario; the output value prediction in step 8 uses the same expected market unit price sequence.

[0220] Generate the expected biomass accumulation curve and output value prediction results for alternative management scenario A, alternative management scenario B, and alternative management scenario C.

[0221] The output value prediction results corresponding to alternative management scenarios A, B, and C are compared and displayed on the same coordinate system. The comparative analysis includes: the peak output value of each alternative management scenario, the time point when the output value reaches the peak, the cumulative output value curve shape during the breeding cycle, and the fluctuation characteristics of output value over time.

[0222] Table 1 shows a comparison of the cumulative output value of each alternative management scenario over the breeding cycle.

[0223] Table 1 Comparison of Cumulative Output Value During Different Alternative Management Scenarios and Breeding Cycles

[0224] Management Scenario Total number of seedlings released (tails) Seedling size (g / tail) Harvest time (number of days of raising) Total biomass at the end of the period (kg) Cumulative output value (ten thousand yuan) Scenario A 25000 125.6 240 19850 176.8 Scenario B 35000 125.6 180+240 in batches 24320 212.5 Scenario C 20000 200.0 210 17560 168.3

[0225] Among alternative management scenarios A, B, and C, the scenario with the largest area under the cumulative output value curve is selected as the management plan with the optimal output value. As shown in Table 1, the cumulative output value of alternative management scenario B is 2.125 million yuan, which is higher than that of alternative management scenario A (1.768 million yuan) and alternative management scenario C (1.683 million yuan). Therefore, alternative management scenario B is selected as the management plan with the optimal output value.

[0226] Step 10: Apply the optimal management plan for output value output from Step 9 to actual aquaculture management, and continuously update the output value prediction results based on real-time monitoring data.

[0227] The adjustable management parameters included in the optimal management plan selected in step 9, i.e., alternative management scenario B, are sent to the on-site management terminal of the target deep-sea aquaculture cage. Specific parameters sent include: stocking density of 2.16 fish / m³ (total stocking quantity 35,000 fish); initial stocking size of 125.6g / fish; first harvest date of October 7, 2025, with a thinning ratio of 40%; second harvest date of December 6, 2025, with a thinning ratio of 60%; feed type and grade of premium compound feed; and a batch thinning strategy implemented according to the set time points and thinning ratios.

[0228] During the implementation of the management plan to optimize output, the data acquisition, processing, and forecasting process from steps 1 to 8 was continuously executed. Daily water quality parameter data, meteorological parameter data, acoustic scanning data, underwater image data, mooring tension data, and cage motion response data were all collected, transmitted, and processed normally.

[0229] Every 30 days, the estimated biomass data of the fish population in the net cages obtained in step 3 using the panoramic multibeam scanning sonar device is compared with the expected biomass data for the same period obtained from the simulation in step 7. The comparison uses a relative deviation index, calculated using the following formula:

[0230]

[0231] in This is a relative deviation. These are estimated data for measured biomass, in kg. The data represents the expected biomass for the same period, in kg.

[0232] When the absolute value of the relative deviation exceeds a preset threshold of ±15%, the recalibration process of the simulation model is triggered. After triggering the recalibration process, the estimated biomass data of the fish population in the current net cage is used as the new initial state for the simulation. The process returns to step 5 to adjust the model parameters of the virtual simulation environment. The adjustment method is to use the latest 30-day measured net cage motion response data from the accelerometer and the measured anchor tension data from the tension sensor, and re-execute the parameter correction procedure in step 5 to update the net damping coefficient and the anchor system stiffness coefficient. After completing the parameter correction in step 5, steps 6 to 9 are re-executed to simulate and extrapolate the growth process of the remaining aquaculture cycle with the new model parameters and the new initial state, generating a revised output value prediction result and updating the management plan recommendations. The updated management plan recommendations include whether to adjust the original batch thinning time nodes and thinning ratios, and whether to adjust the harvest time nodes, etc.

[0233] In this embodiment, during the periodic comparison on the 120th day of breeding, i.e. August 8, 2025, the actual measured biomass estimate was 9850 kg, while the simulated expected biomass was 10240 kg, with a relative deviation of -3.8%, which did not exceed the ±15% threshold. Therefore, the recalibration process was not triggered, and the original optimal production value management plan continued to be implemented.

[0234] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0235] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation, characterized in that, Includes the following steps: Step 1: Collect multi-source data of the target deep-sea aquaculture cages. The multi-source data includes water quality parameter data, meteorological parameter data, acoustic scanning data of the farmed fish, underwater image data of the farmed fish, anchor tension data, and cage motion response data. The multi-source data is transmitted to the land-based data processing center and processed for timestamp alignment and outlier removal to form a structured data base layer. Step 2: Process the underwater image data to obtain the individual body length distribution data and feeding behavior activity data of the farmed fish population; Step 3: Process the acoustic scanning data to obtain biomass estimation data of the fish population in the net cage; Step 4: Perform time-series correlation analysis on the fish feeding behavior activity data, water quality parameter data, and meteorological parameter data to construct a dynamic response function of fish feeding intensity to environmental factors; Step 5: Construct a virtual simulation environment for the target deep-sea aquaculture cage based on the multi-source data; Step 6: Inject virtual environmental disturbance events into the virtual simulation environment to simulate the impact of extreme marine dynamic events on the biomass and survival rate of farmed fish, and obtain the survival rate attenuation coefficient of farmed fish. Step 7: In the virtual simulation environment, the growth process of the farmed fish population is simulated and deduced using the dynamic response function and the survival rate decay coefficient as constraints, and the expected biomass accumulation curves at each time point in the farming cycle are obtained. Step 8: Combine the expected biomass accumulation curve with market price data to generate the output value prediction result of the target deep-sea aquaculture cage; Step 9: Based on the output value prediction results, construct a multi-scenario simulation scheme and conduct an output value comparison analysis to output the management scheme with the optimal output value; Step 10: Apply the optimal management plan to actual aquaculture management and continuously update the output value prediction results based on real-time monitoring data.

2. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 1, characterized in that, The water quality parameters mentioned in step 1 include dissolved oxygen concentration, water temperature, salinity, pH value, ammonia nitrogen concentration, and water flow velocity. The meteorological parameters include sea surface wind speed, wind direction, air temperature, atmospheric pressure, and rainfall. The multi-source data is collected using a multi-source data acquisition device deployed in the deep-sea aquaculture target cage deployment area. This device includes an underwater multi-parameter water quality sensor, a meteorological monitoring station, a panoramic multibeam sonar device, an underwater binocular stereo vision camera, a tension sensor, and an acceleration sensor. The underwater multi-parameter water quality sensor is installed at the center of the cage and at four directional points along the edge of the cage. Each sampling point is arranged in three layers vertically. The three layers of sampling points are located at a surface water depth of 2 meters, halfway through the water depth in the middle layer of the cage, and 2 meters above the bottom of the bottom layer of the cage. The underwater multi-parameter water quality sensor collects data once per hour; the meteorological monitoring station is installed on the buoy platform outside the net cage, and its data collection frequency is once per hour; the panoramic multibeam scanning sonar device is installed on a fixed bracket on the inner wall of the net cage, and its data collection frequency is once per day; the underwater binocular stereo vision camera device is installed at the center of the top of the net cage, and its data collection frequency is once per day, with each collection consisting of a 10-minute video sequence; the tension sensor is installed at a key node of the net cage mooring system, and its data collection frequency is once per minute; the acceleration sensor is installed at a key node of the net cage floating structure, and its data collection frequency is once per minute.

3. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 2, characterized in that, The specific process for obtaining individual body length distribution data and feeding behavior activity data of farmed fish in step 2 is as follows: For the underwater image video sequence acquired by the underwater binocular stereo vision camera, extract single-frame images at a frequency of 1 frame per second to form an image sequence sample set; for each frame in the image sequence sample set, use a fish target detection model based on a deep convolutional neural network to detect and identify fish targets, outputting the bounding box coordinates of each identified fish in the image; for each fish target detected in each frame, calculate the three-dimensional Euclidean distance between the two endpoints of the fish target bounding box using the parallax method based on the calibration parameters of the underwater binocular stereo vision camera as the estimated body length value; statistically analyze the estimated body length values ​​of all fish targets in the same frame and record the body length distribution histogram; and then analyze the body length distribution histogram of all frames within a day. The image sequence sample set is aggregated and averaged to obtain the individual body length distribution data of the farmed fish group on that day. The individual body length distribution data of the farmed fish group on that day includes the mean body length, median body length, standard deviation of body length, and distribution skewness of body length. The image sequence sample set is subjected to fish feeding behavior recognition model based on three-dimensional convolutional neural network for behavioral index quantification analysis. The behavioral index quantification analysis includes extracting the movement speed of the fish centroid in each frame of the image, calculating the average curvature of the movement trajectory of the fish centroid between 5 consecutive frames, and calculating the reciprocal of the average Euclidean distance between any two individuals in the fish group as the fish group aggregation index. After normalizing the movement speed of the fish centroid, the average curvature of the movement trajectory of the fish centroid, and the fish group aggregation index, the fish feeding behavior activity index is calculated by weighted summation. The value range of the fish feeding behavior activity index is [0, 1].

4. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 3, characterized in that, The specific process for obtaining biomass estimation data of the fish population in the net cage in step 3 is as follows: Beamforming processing and backscattering intensity calculation are performed on the acoustic scanning data acquired by the panoramic multibeam scanning sonar device. The detection range of the panoramic multibeam scanning sonar device is no less than 30 meters, the horizontal scanning angle is 360°, the vertical scanning angle is 65°, the horizontal resolution is 0.5°, the vertical resolution is 2°, and the single complete scan cycle is 8 to 16 seconds. The scan generates three-dimensional point cloud data. The attributes of each point in the three-dimensional point cloud data include spatial three-dimensional coordinates and acoustic backscattering intensity values. The three-dimensional point cloud data is divided into multiple water layers according to the depth direction of the aquaculture water body, with each water layer having a thickness of 1 meter. For the three-dimensional point cloud data within each water layer, density-based spatial... Clustering algorithms are used to perform clustering processing, identifying clusters belonging to the fish population. For each identified fish population cluster, the number of point clouds contained in the cluster is calculated, and the average acoustic backscatter intensity value of all points within the cluster is extracted. A mapping relationship model between the acoustic backscatter intensity value and the biomass of individual fish is established. The method for establishing the mapping relationship model is as follows: at the beginning of the breeding cycle, the initial fish fry to be released are sampled and weighed to obtain the average weight of individual fish. A panoramic multibeam scanning sonar device is used to perform a standard scan of the initial fish population and record the average acoustic backscatter intensity value obtained from the scan. During the breeding cycle, a calibration cycle is performed every 30 days. Sample fish are extracted from the net cage using a fish suction pump for actual weighing. The actual weighing results are regressed and fitted with the acoustic scanning results of the same period to correct the parameters in the mapping relationship model. Based on the mapping relationship model and the point cloud number and average acoustic backscattering intensity of each fish population cluster in each water layer, the estimated biomass of the fish population in each water layer is calculated. The estimated biomass of the fish population in all water layers is then summed to obtain the estimated biomass data of the fish population in the net cage.

5. The deep-sea aquaculture output prediction and management method based on virtual simulation according to claim 4, characterized in that, The specific process for constructing the dynamic response function of fish feeding intensity to environmental factors in step 4 is as follows: Dissolved oxygen concentration, water temperature, water flow velocity, and sea surface wind speed are used as four environmental factors affecting fish feeding intensity, and the fish feeding behavior activity index is used as the dependent variable. A nonlinear regression model between the dependent variable and the four environmental factors is established using hourly data over 30 consecutive days as the sample time window. The nonlinear regression model adopts the form... in The activity index of fish feeding behavior at hour (t) is given. The maximum feeding activity of farmed fish under optimal environmental conditions is set to 1. The dissolved oxygen concentration at hour (t) is... Let (t) be the water temperature at hour (t). Let (t) be the water flow velocity at hour (t). Let (t) be the sea surface wind speed at hour (t). This is the dissolved oxygen response function. Let the water temperature response function be... Let be the flow velocity response function. The wind speed response function; the dissolved oxygen response function The function is piecewise; when the dissolved oxygen concentration is below a set critical threshold, feeding activity decreases exponentially with increasing dissolved oxygen concentration; when the dissolved oxygen concentration is within a set suitable range, feeding activity maintains its maximum value; and when the dissolved oxygen concentration exceeds a set upper limit, feeding activity decreases slowly. The water temperature response function... The function is Gaussian, with its peak value located at the optimal growth temperature for farmed fish fry; the flow velocity response function The wind speed response function is an inverse proportional function; the higher the flow velocity, the lower the feeding activity. The function is an exponential decay function; the response function parameter values ​​obtained by calibration through the nonlinear regression model are stored in the model parameter library of the land-based data processing center as the dynamic response function of fish feeding intensity to environmental factors.

6. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 5, characterized in that, The specific process of constructing the virtual simulation environment for the deep-sea aquaculture target cage in step 5 is as follows: A virtual simulation platform is built in a land-based data processing center. The core computing engine of the virtual simulation platform is built based on computational fluid dynamics solvers and multibody dynamics solvers. The water depth and topography data, seabed sediment type data, cage structure geometric parameters, mooring system configuration parameters, and netting material physical parameters of the sea area where the deep-sea aquaculture target cage is located are input into the virtual simulation platform as the basic simulation parameters. The cage structure geometric parameters include cage perimeter, cage depth, cage volume, floating frame structure dimensions, and weight distribution. The mooring system configuration parameters include anchor chain segment lengths. The anchor chain diameter, anchor chain material grade, and anchor point spatial coordinates are specified. The physical parameters of the netting material include netting mesh size, netting wire diameter, netting elastic modulus, and netting damping coefficient. Real-time collected sea surface wind speed data, wind direction data, and water flow velocity data are used as external environmental excitation boundary conditions to drive the computational fluid dynamics solver to calculate the flow field distribution and hydrodynamic load distribution in the sea area surrounding the net cage. The calculated hydrodynamic load distribution is used as input to drive the multibody dynamics solver to calculate the motion response of the net cage floating structure and the tension response of the mooring system. The motion response of the net cage floating structure is used as input to calculate the deformation state and stress distribution of the netting under the action of water flow. The motion response data of the cage and the anchor tension data output by the virtual simulation platform are compared and verified hourly with the motion response data of the cage collected by the accelerometer and the anchor tension data collected by the tension sensor. When the root mean square error between the simulation results and the actual monitoring data exceeds the preset threshold, the parameter correction program of the simulation model is automatically triggered to adjust the mesh damping coefficient and the anchor system stiffness coefficient until the simulation accuracy meets the preset requirements.

7. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 6, characterized in that, The specific process for obtaining the survival rate attenuation coefficient of farmed fish in step 6 is as follows: When the virtual simulation platform detects that the sea surface wind speed exceeds the first preset threshold for 3 consecutive hours, it triggers the typhoon precursor warning state. When the virtual simulation platform detects that the sea surface wind speed exceeds the second preset threshold for 1 consecutive hour, it triggers the typhoon passage simulation event. The first preset threshold is 15 meters per second, and the second preset threshold is 24.5 meters per second. After the typhoon simulation event is triggered, the biomass estimation data and individual body length distribution data of the farmed fish in the cage at the current moment are obtained as the initial state of the simulation. Based on real-time sea surface wind speed data collected by meteorological monitoring stations, the maximum expected wind speed and the corresponding exceedance probability of the current typhoon event within the next 12-hour, 24-hour, 48-hour, and 72-hour time windows are calculated using an extreme value distribution model. The maximum expected wind speed is input into the verified virtual simulation environment, which drives the virtual simulation environment to simulate the cage motion response, net deformation state, and internal flow field changes of the aquaculture water under the corresponding wind speed conditions. Based on the data of cage motion response amplitude and water velocity distribution output by the virtual simulation environment, and combined with the stress response experimental data of farmed fish under severe water disturbance conditions, the stress intensity index is calculated. The stress intensity index is calculated by normalizing the cage motion response amplitude and water velocity and then taking the weighted average. The survival rate attenuation coefficient of farmed fish during typhoon events is calculated based on the stress intensity index. The mapping relationship between the survival rate attenuation coefficient and the stress intensity index is calibrated using post-disaster survey data of historical typhoon events. Applying the survival rate attenuation coefficient to the biomass estimate in the initial state of the simulation yields the expected remaining biomass after the typhoon event ends.

8. The method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation according to claim 7, characterized in that, The specific process for obtaining the expected biomass accumulation curve at each time point within the aquaculture cycle in step 7 is as follows: the aquaculture cycle is divided into discrete time steps with 1 day as the unit. Within each time step, the dissolved oxygen concentration, water temperature, water flow velocity, sea surface wind speed, and individual body length distribution data of the farmed fish are read. The dynamic response function of the fish feeding intensity to environmental factors is called to calculate the fish feeding behavior activity index corresponding to the current time step. The actual feeding amount is calculated using a bioenergetics feeding model based on the fish feeding activity index and the total biomass of the fish at the current time step. The bioenergetics feeding model determines the daily feeding rate by considering the standard metabolizable energy requirement of the cultured fish, the feeding activity correction coefficient, the current biomass, and the current water temperature. The baseline daily feeding rate is determined by referring to a table based on the average fish weight and water temperature, and then multiplied by the fish feeding activity index to obtain the actual daily feeding rate. The increase in fish biomass at the current time step is calculated based on the actual daily feeding rate, the current water temperature, and the feed conversion ratio parameter of the cultured fish. The feed conversion ratio parameter varies with fish size and water temperature and is expressed as a piecewise function. The system checks whether a typhoon passage simulation event has been triggered at the current time step. If not, the total biomass of the fish population is updated with the increase in fish biomass. If it has been triggered, the current biomass is reduced based on the survival rate decay coefficient before calculating the growth increment for subsequent time steps. The updated total biomass of the fish population and the individual body length distribution data of the cultured fish population, which are calculated from the total biomass of the fish population through the body length-weight allometric growth relationship, are used as the initial state for the next time step. This process is repeated until the daily simulation of the entire culture cycle is completed, generating the expected biomass accumulation curve for each time node within the culture cycle.

9. A method for predicting and managing the output value of deep-sea aquaculture based on virtual simulation as described in claim 8, characterized in that, The specific process of constructing a multi-scenario simulation scheme and performing output value comparison analysis to output the management scheme with the optimal output value in step 9 is as follows: At least three alternative management scenarios are set in the virtual simulation environment. Each alternative management scenario consists of a combination of adjustable management parameters, including seedling density, initial seedling size, harvest time, feed type and grade, and a batch thinning strategy. The batch thinning strategy includes thinning time and thinning ratio. Steps 6 to 8 are repeated for each alternative management scenario to generate the expected biomass accumulation curve and output value prediction results corresponding to each alternative management scenario. The output value prediction results corresponding to all alternative management scenarios are compared and displayed in the same coordinate system. The comparative analysis includes the peak output value of each alternative management scenario, the time node when the output value reaches the peak, the cumulative curve shape of the output value during the breeding cycle, and the fluctuation characteristics of the output value over time. Among all alternative management scenarios, the management scenario with the largest area under the cumulative output curve is selected as the management plan with the optimal output. The specific process of applying the optimal production value management plan to actual aquaculture management and continuously updating the production value prediction results based on real-time monitoring data in step 10 is as follows: The optimal production value management plan, including seedling density, initial seedling specifications, harvest time nodes, feed type and grade, and batch thinning strategies, is distributed to the on-site management terminal of the target deep-sea aquaculture cage; during the execution of the optimal production value management plan, the data collection, processing, and prediction process of steps 1 to 8 is continuously executed; every 30 days, the biomass estimation data of the cultured fish population in the cage obtained by the panoramic multibeam scanning sonar device is compared with the expected biomass data of the same period obtained by the simulation in step 7. When the relative deviation between the biomass estimation data of the cultured fish population in the cage and the expected biomass data of the same period exceeds ±15%, the recalibration process of the simulation model is triggered. The current biomass estimation data of the cultured fish population in the cage is used as the new initial state of the simulation, and the model parameters of the virtual simulation environment are adjusted in step 5 and steps 6 to 9 are re-executed to generate the corrected production value prediction results and update the management plan recommendations.

Images