Tuna pond culture strategy optimization method and system based on meteorological influence analysis

Abstract

The invention discloses a tuna pond culture strategy optimization method and system based on meteorological influence analysis, and the method comprises the steps: collecting regional meteorological data and pond culture monitoring data, and constructing a multi-modal pond ecological feature vector; establishing a weather-water body dynamic coupling model by using a space-time diagram neural network, and generating water body situation response prediction information; a breeding regulation and control strategy is made based on the prediction information, and strategy self-adaptive optimization is achieved through real-time monitoring data; evaluating the anti-disaster capability of the tuna after the meteorological influence is finished, screening high-quality germplasm resources by combining genetic analysis, and generating an anti-disaster breeding scheme. According to the method, full-chain optimization from meteorological early warning to culture regulation and control to genetic breeding is achieved, and the climate toughness and economic benefits of tuna pond culture are comprehensively improved.

Description

Technical Field

[0001] This invention relates to the field of tuna pond aquaculture technology, and in particular to a method and system for optimizing tuna pond aquaculture strategies based on meteorological influence analysis. Background Technology

[0002] As a high-value marine fish, tuna pond culture technology has received widespread attention and development in recent years. However, in actual aquaculture, dynamic changes in meteorological conditions have become a key factor affecting the success of the aquaculture. Extreme weather events such as typhoons, rainstorms, sustained high temperatures, and rapid temperature drops, as well as daily meteorological elements such as diurnal temperature variations and changes in light intensity, can directly or indirectly alter the physicochemical environment of pond water, thereby affecting the physiological state and behavioral patterns of tuna. Existing aquaculture management methods largely rely on the experience and judgment of aquaculture personnel, lacking a deep understanding of the complex correlation mechanisms between meteorological conditions and aquatic environmental parameters, and even more so, lacking a quantitative assessment system for fish stress responses.

[0003] Currently, the aquaculture monitoring systems commonly used in the industry mainly focus on real-time monitoring of basic water parameters (such as temperature, dissolved oxygen, and pH), triggering alarms or performing simple control operations when parameters exceed preset thresholds. These systems have significant limitations: First, their response mechanisms are lagging, often only taking action after environmental anomalies occur, failing to provide proactive early warning and intervention; second, each control subsystem (such as aeration, feeding, and water circulation) typically operates independently, lacking a synergistic optimization strategy; third, existing technologies fail to effectively integrate meteorological forecast information, making it impossible to predict the potential impact of future weather conditions on the pond ecosystem; and most importantly, traditional methods ignore the differences in how individual fish respond to environmental stress, failing to establish a complete technological chain from environmental management to genetic improvement.

[0004] Therefore, there is an urgent need in this field for a comprehensive solution that can deeply integrate meteorological data analysis, aquatic environment prediction, aquaculture strategy optimization, and germplasm resource screening. This would enable a shift from passive response to proactive adaptation, thereby comprehensively improving the climate resilience and economic benefits of tuna pond aquaculture. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. Its main purpose is to improve the climate resilience and economic benefits of tuna pond aquaculture.

[0006] To achieve the above objectives, the first aspect of this invention provides a method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, comprising: Regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond are acquired and preprocessed. After preprocessing, a multimodal pond ecological feature vector of the target aquaculture pond is constructed. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. The preset aquaculture management strategy for the target aquaculture pond is obtained, and the aquaculture control strategy under the current meteorological influence is formulated in combination with the water body situation response prediction information. The control strategy is adaptively optimized based on the water body monitoring data during the real-time control period. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis, and finally a disaster-resistant breeding recommendation program is obtained.

[0007] In this scheme, the acquisition and preprocessing of regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond, followed by the construction of a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing, specifically includes: Regional meteorological data, including real-time observations and short-term forecasts, is obtained through meteorological stations deployed in aquaculture areas and external forecast interfaces. Pond aquaculture monitoring data, including water environment parameters and fish monitoring images, is collected through heterogeneous sensor arrays deployed in ponds, forming an asynchronous multi-source data stream and undergoing preprocessing. For regional meteorological data and water environment parameters, the local outlier detection algorithm is used to identify outliers. The relative density of each data point in its local neighborhood is calculated according to a preset time window, and the density is compared with that of neighboring data points to obtain the local outlier score. When the score exceeds the preset score threshold, it is marked as an outlier data point. The time series parameter sequence after marking outlier data points is obtained and the time series parameter sequence is decomposed into trend term, seasonal term and residual term. The trend term captures the long-term change direction of the data, the seasonal term extracts the periodic change pattern, and the residual term contains random fluctuations and outlier information, generating sequence decomposition information. After completing the sequence decomposition, the normal time period data segment is extracted based on the sequence decomposition information. The blank long short-term memory network model is called and trained through the residual sequence of the normal time period data segment. After training is completed, the residual terms corresponding to the normal data before and after the abnormal time period are input to predict the residual value of the abnormal time period. Finally, the predicted residual value is recombined with the corresponding trend term and seasonal term. The reconstructed regional meteorological data sequence and aquatic environmental parameter sequence are time-aligned using a dynamic time warping algorithm. Simultaneously, the fish monitoring images are enhanced and filtered, and finally, a preprocessed asynchronous multi-source dataset is output.

[0008] In this scheme, the step of acquiring regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond and preprocessing them, and then constructing a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing, further includes: Asynchronous multi-source dataset is obtained. Preprocessed fish monitoring images are extracted from the asynchronous multi-source dataset and imported into Gaussian mixture model for background modeling and moving target segmentation. Based on the color distribution features of each pixel in consecutive frames, pixels that conform to the background distribution are classified as background, while pixels that do not conform to the background distribution are judged as foreground moving targets. After completing the segmentation of moving targets, the YOLOv5 deep neural network architecture is used to identify and locate fish bodies in the segmented foreground targets. The target detection is performed on multi-scale feature maps through its anchor box mechanism. Multi-level features are extracted by the backbone network and combined with the feature pyramid network to enhance the detection capability, and fish body target detection information is output. Based on the fish target detection information, the DeepSORT multi-target tracking algorithm is used to construct the fish school movement trajectory. Kalman filtering is used to predict the target position in the current frame based on the trajectory state of the previous frame. By fusing the correlation metric of appearance features and motion information, the detection box and trajectory are matched to obtain the fish movement trajectory information. Based on the fish movement trajectory information, fish school behavior characteristics including average swimming speed, movement path tortuosity, spatial distribution density and group aggregation are calculated. At the same time, the principal component analysis method is used to reduce the dimensionality of multidimensional water quality parameters and extract the principal component features whose cumulative contribution rate reaches the predetermined contribution threshold as key water body environmental factors characterizing the water body state, and finally the water body environmental feature vector is obtained. The asynchronous multi-source dataset is used to extract the temporal features of meteorological elements, calculate the mutual information value between fish behavior features and key environmental factors of the water body to construct environmental behavior interaction features, and combine the water body environmental feature vector and the temporal features of meteorological elements to perform multi-dimensional vector splicing to generate a multimodal pond ecological feature vector.

[0009] In this scheme, a dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. This model is then combined with the analysis of the multimodal pond ecological feature vectors to determine the response of the pond water body to meteorological changes, thereby obtaining water body situation response prediction information. Specifically, this includes: A multimodal pond ecological feature vector is obtained. Based on the geospatial layout of the pond, the aquaculture water area is discretized into a regular grid structure. Each grid cell is defined as a node of the graph. Based on the multimodal pond ecological feature vector, node features corresponding to the spatial location are assigned. The node features consist of environmental parameters, meteorological elements and fish behavior indicators of the corresponding location. The K-nearest neighbor algorithm based on spatial distance is used to establish physical proximity edges. Historical water flow velocity and direction data are used to calculate the hydrodynamic correlation between nodes. Functional connection edges are established for nodes whose correlation exceeds the threshold, and a dynamic graph structure containing geometric spatial proximity and water movement correlation is constructed. A dynamic coupling model of meteorology and water bodies is constructed using a spatiotemporal graph neural network. The dynamic graph structure is input and stacked spatiotemporal convolutional blocks are used to extract hierarchical spatiotemporal features. Based on Chebyshev multinomial approximation spectral graph convolution, the spatial dependencies between nodes are captured by multi-order neighborhood feature aggregation, and high-order spatial feature representations are obtained by stacking multi-layer graph convolutions. The high-order spatial feature representation is input into the gated recurrent unit network for time dynamic modeling. The degree of retention of historical state information is controlled by updating the gate, and the combination of the current input and historical state is determined by the reset gate. The importance weight of each time step in the hidden state sequence is calculated, and adaptive focusing is performed on the impact period of key meteorological events. Finally, the spatiotemporal feature representation is output. The temporal features of meteorological elements are extracted from the multimodal pond ecological feature vector and converted into meteorological context vectors by an encoder. The cross-attention weights between the meteorological context and the state of all nodes are calculated using a multi-head attention mechanism and fused with the spatiotemporal hidden state of each node to obtain weighted node features. The weighted node features are input into the decoder, which consists of a temporal deconvolution layer and a fully connected layer, to map the high-dimensional features into the predicted environmental parameters of each grid node in the future time step. The predicted environmental parameters of all nodes in each time step are reconstructed into two-dimensional field data according to spatial coordinates. The Kriging spatial interpolation algorithm is used to generate a spatial distribution map of the water body situation. Based on the spatial distribution map of the water body situation, the parameter change rate, extreme values ​​and duration at each spatial location are extracted to generate water body situation response prediction information, which characterizes the spatiotemporal evolution of the pond water environment under future meteorological scenarios.

[0010] In this solution, the acquisition of the preset aquaculture management strategy for the target aquaculture pond, combined with the water body status response prediction information, to formulate an aquaculture control strategy under the current meteorological influence, specifically includes: Obtain the preset aquaculture management and control strategy for the target aquaculture pond, extract the preset water state benchmark features of the target aquaculture pond as the initial control sequence through the preset aquaculture management and control strategy, initialize the model prediction control framework, and construct a dual-objective optimization function and set constraints with the objectives of maximizing economic benefits and minimizing aquaculture risks. The water body situation response prediction information is obtained, and the water body situation response prediction information is initialized in the model prediction control framework and a multi-dimensional search space is generated. A multi-objective reptile algorithm is introduced to perform population initialization in the multi-dimensional search space to generate an initial crocodile population P0 for optimization of aquaculture regulation strategy. The fitness value of each crocodile individual in the population is calculated based on the initial crocodile population P0 to obtain an initial target matrix F0 containing the fitness values ​​of all individuals; the position is updated during the encirclement phase based on the initial crocodile population P0 and the initial target matrix F0. During the encirclement process, a high-speed movement strategy simulating the rapid movement of crocodiles and a local search strategy simulating the slow crawling of crocodiles close to the ground are used. The choice of movement strategy is determined by the ratio of the current iteration number to the total number of iterations for the corresponding crocodile individual. After all crocodile individuals have completed the position update, a new crocodile population P1 is output. Based on the new crocodile population P1, the location update during the hunting phase is carried out using coordinated predation and cooperative hunting strategies. When the coordinated predation strategy is executed, individuals move towards the center point determined by the positions of multiple elite individuals in the current population. When the cooperative hunting strategy is executed, individuals move towards the position of another randomly selected elite individual. In the hunting phase, dynamic sensitivity parameters and random perturbation factors are introduced to avoid getting trapped in local optima and to adaptively adjust the search focus range. The dynamic sensitivity parameters adaptively decay as the number of iterations increases, and the hunting phase is completed to finally output the crocodile population P2. Based on the crocodile population P2, non-dominated sorting and unset screening are performed. Pareto dominance relationships between individuals are established by traversing and comparing the fitness values ​​of all individuals in population P2. All individuals are divided into non-dominated fronts of different levels. For individuals located in the same non-dominated front, the crowding distance on the corresponding front is calculated and sorted in combination with the Pareto level. Based on the ranking results, individuals with the highest rank and crowding greater than a preset threshold are stored as high-quality solutions in the external archive. An elite retention strategy is adopted to maintain the external archive. Non-dominated optimal solutions generated in each iteration are stored in the external archive. When the new generation of population is generated, elite individuals in the archive are selected first. Through repeated iterations, the Pareto optimal solution set is output from the external archive, and finally the optimal aquaculture control strategy under the current weather influence is generated.

[0011] In this scheme, the adaptive optimization of the control strategy based on water monitoring data during real-time control specifically includes: The optimal aquaculture regulation strategy under the current meteorological influence is obtained. The optimal regulation command sequence for aquaculture ponds in the current time domain is generated through the model predictive control framework and sent to various execution agencies to regulate the water body of aquaculture ponds. During the regulation process, the first control command of the optimal regulation command sequence for the aquaculture pond is executed first, and the actual water body regulation response data is obtained by real-time water environment monitoring through the deployed heterogeneous sensor network. The data is then compared with the preset water body state benchmark characteristics of the target aquaculture pond to obtain the deviation sequence. The deviation sequence is compared with the preset tolerance range. If it is not within the preset tolerance range, the water body response situation prediction information at the current moment is obtained based on the actual water body regulation response data. This information is then input into the model prediction control framework for sample update and re-optimization of the aquaculture regulation strategy. The updated optimal regulation command sequence for the aquaculture pond is then obtained for the next moment's aquaculture pond regulation.

[0012] In this plan, after the current weather impact ends, the disaster resistance capacity of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened, and a parent selection matrix is ​​constructed by combining genetic pedigree analysis. Finally, a recommended disaster-resistant breeding program is obtained, which specifically includes: After the target meteorological impact event ends, environmental monitoring data and fish behavior response data throughout the entire meteorological event are collected, and after standardization and preprocessing, a meteorological response dataset is generated. Based on the meteorological response dataset, the principal component analysis method is used to reduce the dimensionality of all environmental feature parameters to obtain the principal component scores of each environmental feature parameter. The main environmental feature parameters are then selected as environmental stress feature vectors by using the principal component scores. Subsequently, fish behavior response data were used to evaluate the environmental adaptability, stress recovery speed and growth stability of each individual during the current weather impact. Combined with the environmental stress feature vector, the TOPSIS comprehensive evaluation method was used to calculate the comprehensive evaluation value of the disaster resistance of each tuna individual. Based on the comprehensive evaluation value of each tuna individual's disaster resistance ability, the fish population is divided into three levels: strong disaster resistance, medium disaster resistance, and weak disaster resistance using a set grading threshold. Strong disaster resistance individuals are selected as candidate superior disaster resistance germplasm resources. Genome analysis was performed on the selected candidate individuals with superior disaster resistance. Biological samples were collected from the candidates and their SNP genotype data were obtained through whole-genome sequencing. Genome-wide association analysis was used to correlate disaster resistance phenotype data with genotype data. A mixed linear model was used to control for the influence of population structure and to identify SNP sites that were significantly associated with disaster resistance traits. Genotype feature vectors related to disaster resistance are constructed for each individual using the identified significant SNP loci to generate genotype feature spectra. Parental selection and optimization are carried out by combining the comprehensive evaluation value of disaster resistance ability of each tuna individual. The best linear unbiased prediction method is used to estimate the breeding value of each individual. By calculating the kinship coefficient and genetic distance between individuals, a parental selection affinity matrix is ​​established. At the same time, considering the heritability and selection intensity of disaster resistance traits, the gene flow algorithm is used to optimize the parental combination scheme and generate a parental selection matrix that maximizes the genetic gain of offspring. Based on the parental mating matrix, the candidate superior disaster-resistant germplasm resources are screened again to obtain the optimal disaster-resistant individuals and mark them, generating a recommended individual table for tuna breeding at the current time.

[0013] A second aspect of this invention provides a system for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. The system includes a memory, a processor, and a communication interface. The memory contains a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When executed by the processor, the program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis performs the following steps: Regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond are acquired and preprocessed. After preprocessing, a multimodal pond ecological feature vector of the target aquaculture pond is constructed. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. The preset aquaculture management strategy for the target aquaculture pond is obtained, and the aquaculture control strategy under the current meteorological influence is formulated in combination with the water body situation response prediction information. The control strategy is adaptively optimized based on the water body monitoring data during the real-time control period. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis, and finally a disaster-resistant breeding recommendation program is obtained.

[0014] A third aspect of the present invention provides a computer-readable storage medium comprising a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When the program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis is executed by a processor, it implements the steps of the method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis as described in any of the preceding claims. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0016] Figure 1 A flowchart of the first method for optimizing tuna pond aquaculture strategy based on meteorological impact analysis, provided in an embodiment of the present invention; Figure 2 A flowchart of the second method for optimizing tuna pond aquaculture strategy based on meteorological impact analysis, provided in an embodiment of the present invention; Figure 3 This is a block diagram of a tuna pond aquaculture strategy optimization system based on meteorological impact analysis, provided in an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] Figure 1 A flowchart of the first method for optimizing tuna pond aquaculture strategy based on meteorological impact analysis, provided in an embodiment of the present invention; like Figure 1 As shown, the present invention provides a first method flowchart for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, comprising: S102, acquire regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond and preprocess them, and construct a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing; S104. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. S106, Obtain the preset aquaculture management strategy for the target aquaculture pond, formulate the aquaculture control strategy under the current meteorological influence state in combination with the water body situation response prediction information, and adaptively optimize the control strategy based on the water body monitoring data during the real-time control period. S108. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis. Finally, a disaster-resistant breeding recommendation scheme is obtained.

[0020] Furthermore, in a preferred embodiment of the present invention, the step of acquiring regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond and preprocessing them, and then constructing a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing, specifically includes: Regional meteorological data, including real-time observations and short-term forecasts, is obtained through meteorological stations deployed in aquaculture areas and external forecast interfaces. Pond aquaculture monitoring data, including water environment parameters and fish monitoring images, is collected through heterogeneous sensor arrays deployed in ponds, forming an asynchronous multi-source data stream and undergoing preprocessing. For regional meteorological data and water environment parameters, the local outlier detection algorithm is used to identify outliers. The relative density of each data point in its local neighborhood is calculated according to a preset time window, and the density is compared with that of neighboring data points to obtain the local outlier score. When the score exceeds the preset score threshold, it is marked as an outlier data point. The time series parameter sequence after marking outlier data points is obtained and the time series parameter sequence is decomposed into trend term, seasonal term and residual term. The trend term captures the long-term change direction of the data, the seasonal term extracts the periodic change pattern, and the residual term contains random fluctuations and outlier information, generating sequence decomposition information. After completing the sequence decomposition, the normal time period data segment is extracted based on the sequence decomposition information. The blank long short-term memory network model is called and trained through the residual sequence of the normal time period data segment. After training is completed, the residual terms corresponding to the normal data before and after the abnormal time period are input to predict the residual value of the abnormal time period. Finally, the predicted residual value is recombined with the corresponding trend term and seasonal term. The reconstructed regional meteorological data sequence and aquatic environmental parameter sequence are time-aligned using a dynamic time warping algorithm. Simultaneously, the fish monitoring images are enhanced and filtered, and finally, a preprocessed asynchronous multi-source dataset is output.

[0021] It should be noted that, firstly, regional meteorological data is acquired through weather stations deployed in the aquaculture area and external forecast interfaces. Simultaneously, a heterogeneous sensor array deployed in the ponds collects pond aquaculture data, including aquatic environmental parameters (such as water temperature, dissolved oxygen, and pH) and fish monitoring images, forming an initial asynchronous multi-source data stream. For the acquired regional meteorological data and aquatic environmental parameter data, a local outlier detection algorithm is used to identify outliers. This algorithm sets a local neighborhood for each data point and calculates an outlier score by determining the relative density relationship between that point and other points in its neighborhood. When the score of a data point exceeds a preset threshold, it indicates a significant difference in distribution pattern from surrounding data points, and is thus marked as an outlier. After outlier marking, the time-series parameter sequence is decomposed into three components: trend, seasonal, and residual. The trend reflects the long-term direction of data change, such as the overall trend of water temperature with seasonal variations; the seasonal component captures periodic patterns, such as the diurnal fluctuation of dissolved oxygen; and the residual contains random fluctuations and anomalous information not explained by the former two. Based on the decomposition results, normal time periods not marked as anomalous were extracted, and a blank long short-term memory network model was specifically trained on the residual sequence. This model can learn the latent patterns in the residual sequence and then be used to predict the residual values ​​corresponding to the data in the marked anomalous periods. Finally, the predicted residual values ​​were recombine with the corresponding trend and seasonal terms to repair the anomalous data and generate a reconstructed complete data sequence. The reconstructed regional meteorological data sequence and aquatic environmental parameter sequence still need to address the issue of time scale inconsistency. To this end, a dynamic time warping algorithm was used to perform time-series alignment processing on the two types of sequences. This algorithm can find the optimal matching path between the two time series, effectively solving the time axis mismatch problem caused by different sampling frequencies. At the same time, image enhancement and filtering were performed on the fish monitoring images to improve image quality and suppress noise. Finally, a high-quality asynchronous multi-source dataset with comprehensive preprocessing was output, providing reliable data support for subsequent analysis.

[0022] Furthermore, in a preferred embodiment of the present invention, the step of acquiring regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond and preprocessing them, and constructing a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing, further includes: Asynchronous multi-source dataset is obtained. Preprocessed fish monitoring images are extracted from the asynchronous multi-source dataset and imported into Gaussian mixture model for background modeling and moving target segmentation. Based on the color distribution features of each pixel in consecutive frames, pixels that conform to the background distribution are classified as background, while pixels that do not conform to the background distribution are judged as foreground moving targets. After completing the segmentation of moving targets, the YOLOv5 deep neural network architecture is used to identify and locate fish bodies in the segmented foreground targets. The target detection is performed on multi-scale feature maps through its anchor box mechanism. Multi-level features are extracted by the backbone network and combined with the feature pyramid network to enhance the detection capability, and fish body target detection information is output. Based on the fish target detection information, the DeepSORT multi-target tracking algorithm is used to construct the fish school movement trajectory. Kalman filtering is used to predict the target position in the current frame based on the trajectory state of the previous frame. By fusing the correlation metric of appearance features and motion information, the detection box and trajectory are matched to obtain the fish movement trajectory information. Based on the fish movement trajectory information, fish school behavior characteristics including average swimming speed, movement path tortuosity, spatial distribution density and group aggregation are calculated. At the same time, the principal component analysis method is used to reduce the dimensionality of multidimensional water quality parameters and extract the principal component features whose cumulative contribution rate reaches the predetermined contribution threshold as key water body environmental factors characterizing the water body state, and finally the water body environmental feature vector is obtained. The asynchronous multi-source dataset is used to extract the temporal features of meteorological elements, calculate the mutual information value between fish behavior features and key environmental factors of the water body to construct environmental behavior interaction features, and combine the water body environmental feature vector and the temporal features of meteorological elements to perform multi-dimensional vector splicing to generate a multimodal pond ecological feature vector.

[0023] It should be noted that the preprocessed video sequence is imported into a Gaussian mixture model for background modeling. This model establishes a probability model for each pixel by statistically analyzing the color value distribution of each pixel in consecutive frames. Pixels that conform to this background color distribution model are classified as static background, while pixels whose color distribution significantly deviates from the background model are identified as foreground moving targets. This achieves accurate segmentation of the fish body from the complex water background and yields a binarized moving target mask. After obtaining the moving target segmentation results, the accurate identification and localization of individual fish bodies proceeds. The YOLOv5 single-stage target detection algorithm is adopted, whose backbone network (such as CSPDarknet) extracts multi-level feature maps from the input image. Subsequently, a feature pyramid network fuses features from different levels, thereby simultaneously improving the detection capability for both large and small targets. The network's anchor box mechanism predicts on multi-scale feature maps, directly outputting the bounding box coordinates and confidence score of each detected fish body. For example, even in scenarios where fish are densely packed and overlapping, the algorithm can effectively distinguish and locate individual tuna, outputting a list of fish target detection information that includes location and confidence level.

[0024] Based on the detection results, the DeepSORT multi-target tracking algorithm is applied to construct continuous fish movement trajectories. The algorithm first uses Kalman filtering to predict the target's expected position in the current frame based on its motion state (such as position and velocity) in the previous frame. Then, multiple bounding boxes detected in the current frame are matched and associated with multiple existing trajectories. This association process is achieved by fusing two metrics: first, the matching degree of motion information, i.e., calculating the Mahalanobis distance between the detected box position and the predicted position; second, the similarity of appearance features, by extracting the appearance feature vector of each detected box and calculating its cosine similarity with existing appearance features in the trajectory database. By combining these two metrics, the identity switching problem after target occlusion is solved, ultimately generating a movement trajectory for each fish. After obtaining the movement trajectory, feature indicators that quantify the dynamic behavior of the fish school are calculated. These indicators include average swimming speed, the tortuosity of the movement path (to reflect the stress level or leisure state of the fish school), the spatial distribution density of the fish school in different areas of the pond (to identify clustered or sparse areas), and the group aggregation degree (to quantify the intensity of swarm behavior).

[0025] Concurrently, principal component analysis was performed on multidimensional water quality parameters (such as dissolved oxygen, water temperature, and pH) to extract principal components whose cumulative contribution rate exceeded a predetermined threshold (e.g., 85%). These principal components can retain key information from the original water quality data to the greatest extent possible with lower dimensionality, forming a feature vector characterizing the overall state of the water body. Finally, multi-source information fusion was performed, extracting temporal features of meteorological elements from asynchronous datasets and calculating the mutual information value between fish behavior characteristics and key environmental factors of the water body. This value quantifies the statistical dependence between behavioral changes and environmental fluctuations, thus constructing features reflecting the interaction between the environment and behavior. Finally, the water body environmental feature vector, the temporal features of meteorological elements, and the environmental-behavioral interaction features based on mutual information were concatenated to generate a multimodal pond ecological feature vector, serving as a digital summary of the pond ecosystem state and providing high-quality input for subsequent meteorological-water dynamic coupling modeling and prediction.

[0026] Furthermore, in a preferred embodiment of the present invention, the step of constructing a meteorological-water dynamic coupling model using a spatiotemporal graph neural network, and combining the multimodal pond ecological feature vector analysis to analyze the response of the pond water body to meteorological changes, to obtain water body situation response prediction information, specifically includes: A multimodal pond ecological feature vector is obtained. Based on the geospatial layout of the pond, the aquaculture water area is discretized into a regular grid structure. Each grid cell is defined as a node of the graph. Based on the multimodal pond ecological feature vector, node features corresponding to the spatial location are assigned. The node features consist of environmental parameters, meteorological elements and fish behavior indicators of the corresponding location. The K-nearest neighbor algorithm based on spatial distance is used to establish physical proximity edges. Historical water flow velocity and direction data are used to calculate the hydrodynamic correlation between nodes. Functional connection edges are established for nodes whose correlation exceeds the threshold, and a dynamic graph structure containing geometric spatial proximity and water movement correlation is constructed. A dynamic coupling model of meteorology and water bodies is constructed using a spatiotemporal graph neural network. The dynamic graph structure is input and stacked spatiotemporal convolutional blocks are used to extract hierarchical spatiotemporal features. Based on Chebyshev multinomial approximation spectral graph convolution, the spatial dependencies between nodes are captured by multi-order neighborhood feature aggregation, and high-order spatial feature representations are obtained by stacking multi-layer graph convolutions. The high-order spatial feature representation is input into the gated recurrent unit network for time dynamic modeling. The degree of retention of historical state information is controlled by updating the gate, and the combination of the current input and historical state is determined by the reset gate. The importance weight of each time step in the hidden state sequence is calculated, and adaptive focusing is performed on the impact period of key meteorological events. Finally, the spatiotemporal feature representation is output. The temporal features of meteorological elements are extracted from the multimodal pond ecological feature vector and converted into meteorological context vectors by an encoder. The cross-attention weights between the meteorological context and the state of all nodes are calculated using a multi-head attention mechanism and fused with the spatiotemporal hidden state of each node to obtain weighted node features. The weighted node features are input into the decoder, which consists of a temporal deconvolution layer and a fully connected layer, to map the high-dimensional features into the predicted environmental parameters of each grid node in the future time step. The predicted environmental parameters of all nodes in each time step are reconstructed into two-dimensional field data according to spatial coordinates. The Kriging spatial interpolation algorithm is used to generate a spatial distribution map of the water body situation. Based on the spatial distribution map of the water body situation, the parameter change rate, extreme values ​​and duration at each spatial location are extracted to generate water body situation response prediction information, which characterizes the spatiotemporal evolution of the pond water environment under future meteorological scenarios.

[0027] It should be noted that a dynamic coupling model of meteorology and water body is constructed to predict the response of pond water to meteorological changes. First, based on the acquired multimodal pond ecological feature vectors, the aquaculture area is discretized into a regular grid structure according to its geographic spatial layout. Each grid cell is defined as a node in the graph. The node features consist of environmental parameters (such as water temperature and dissolved oxygen), meteorological elements (such as air temperature and wind speed), and fish behavior indicators (such as swimming speed) at the corresponding location, thus transforming the pond ecosystem into a graph data structure. Next, the K-nearest neighbor algorithm is used to establish physical proximity edges between nodes to reflect geographical adjacency. At the same time, combined with historical water flow velocity and direction data, the hydrodynamic correlation between nodes is calculated. Functional connection edges are established for node pairs with correlation exceeding a preset threshold. For example, in areas with consistent water flow direction, even if the physical distance is large, a connection is established to capture the movement correlation of the water body. Finally, a dynamic graph structure that simultaneously includes geometric spatial proximity and water movement correlation is constructed.

[0028] Subsequently, a spatiotemporal graph neural network is used to construct the model. The aforementioned dynamic graph structure is input, and hierarchical feature extraction is performed through stacked spatiotemporal convolutional blocks. In the spatial dimension, a spectral graph convolution method based on Chebyshev multinomial approximation is used to aggregate multi-order neighborhood information of each node, thereby capturing the spatial dependencies between nodes. In the temporal dimension, the obtained high-order spatial feature representation is input into a gated recurrent unit network. The update gate controls the retention of historical state information, and the reset gate determines the combination of the current input and historical state, thus modeling temporal dynamics. For example, it focuses on tracking water changes before and after a rainstorm event, ultimately outputting a representation of fused spatiotemporal features. To incorporate meteorological influences, temporal features of meteorological elements (such as temperature and precipitation sequences) are extracted from the multimodal feature vector and converted into meteorological context vectors by an encoder. A multi-head attention mechanism is used to calculate the cross-attention weights between the meteorological context and the states of all nodes. For example, higher weights are assigned to nodes in high-pressure areas during typhoon weather. Then, the weighted meteorological information is fused with the spatiotemporal hidden state of each node to obtain enhanced weighted node features. Next, these features are input into a decoder consisting of a temporal deconvolution layer and a fully connected layer. The high-dimensional features are mapped to predicted environmental parameters (such as dissolved oxygen concentration) for each grid node in the future time step. The predicted values ​​are then reconstructed into two-dimensional field data according to spatial coordinates, and a smooth spatial distribution map of the water body situation is generated using a Kriging spatial interpolation algorithm. Finally, based on the spatial distribution map of the water body situation, the rate of change of parameters (such as the rate of decrease of dissolved oxygen), extreme values ​​(such as the location of the lowest water temperature), and duration (such as the duration of the hypoxic state) at each spatial location are extracted to generate water body situation response prediction information. This allows for a quantitative characterization of the spatiotemporal evolution of the pond water environment under future meteorological scenarios. For example, a report could be output indicating that the northwest area of ​​the pond may experience a sustained risk of hypoxia after a rainstorm.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of evaluating the disaster resistance of tuna in the target aquaculture pond after the current weather impact ends, screening superior disaster-resistant germplasm resources based on the evaluation results, constructing a parent selection matrix by combining genetic pedigree analysis, and finally obtaining a recommended disaster-resistant breeding program, specifically includes: After the target meteorological impact event ends, environmental monitoring data and fish behavior response data throughout the entire meteorological event are collected, and after standardization and preprocessing, a meteorological response dataset is generated. Based on the meteorological response dataset, the principal component analysis method is used to reduce the dimensionality of all environmental feature parameters to obtain the principal component scores of each environmental feature parameter. The main environmental feature parameters are then selected as environmental stress feature vectors by using the principal component scores. Subsequently, fish behavior response data were used to evaluate the environmental adaptability, stress recovery speed and growth stability of each individual during the current weather impact. Combined with the environmental stress feature vector, the TOPSIS comprehensive evaluation method was used to calculate the comprehensive evaluation value of the disaster resistance of each tuna individual. Based on the comprehensive evaluation value of each tuna individual's disaster resistance ability, the fish population is divided into three levels: strong disaster resistance, medium disaster resistance, and weak disaster resistance using a set grading threshold. Strong disaster resistance individuals are selected as candidate superior disaster resistance germplasm resources. Genome analysis was performed on the selected candidate individuals with superior disaster resistance. Biological samples were collected from the candidates and their SNP genotype data were obtained through whole-genome sequencing. Genome-wide association analysis was used to correlate disaster resistance phenotype data with genotype data. A mixed linear model was used to control for the influence of population structure and to identify SNP sites that were significantly associated with disaster resistance traits. Genotype feature vectors related to disaster resistance are constructed for each individual using the identified significant SNP loci to generate genotype feature spectra. Parental selection and optimization are carried out by combining the comprehensive evaluation value of disaster resistance ability of each tuna individual. The best linear unbiased prediction method is used to estimate the breeding value of each individual. By calculating the kinship coefficient and genetic distance between individuals, a parental selection affinity matrix is ​​established. At the same time, considering the heritability and selection intensity of disaster resistance traits, the gene flow algorithm is used to optimize the parental combination scheme and generate a parental selection matrix that maximizes the genetic gain of offspring. Based on the parental mating matrix, the candidate superior disaster-resistant germplasm resources are screened again to obtain the optimal disaster-resistant individuals and mark them, generating a recommended individual table for tuna breeding at the current time.

[0030] It should be noted that weather conditions are a significant factor affecting the profitability of tuna pond farming. Therefore, cultivating and raising tuna that are better adapted to adverse environments is crucial for improving farming efficiency. After the target weather event, environmental monitoring data (including water temperature fluctuation range, duration of extreme dissolved oxygen levels, and salinity changes) and fish behavioral response data (including stress response duration, feeding recovery speed, and changes in swimming activity) were collected throughout the event. These multi-source data were standardized and quality-controlled to form a structured weather response dataset. Based on this dataset, principal component analysis was used to reduce the dimensionality of environmental characteristic parameters. By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, principal components with a cumulative contribution rate exceeding 85% were extracted, and their corresponding characteristic parameters (such as the magnitude of sudden temperature changes and the duration of hypoxia) were used as environmental stress feature vectors. Subsequently, the fish behavioral response data were used to evaluate the performance of each individual fish during the disaster. By calculating three indicators—environmental adaptability (based on the degree of behavioral deviation from the normal range), stress recovery speed (time required to recover from a stressed state to normal behavior), and growth stability (weight change rate during a disaster)—and combining them with environmental stress feature vectors, the TOPSIS comprehensive evaluation method is used to calculate the comprehensive disaster resistance rating of each individual. This method constructs positive and negative ideal solutions to calculate the relative closeness of each evaluated object to the optimal solution, thereby obtaining a standardized evaluation value between 0 and 1. Based on the comprehensive disaster resistance rating, two grading thresholds of 0.8 and 0.5 are set to divide the fish population into three levels: those with a rating greater than 0.8 are highly resilient, those between 0.5 and 0.8 are moderately resilient, and those less than 0.5 are weakly resilient.

[0031] Furthermore, genomic analysis was performed on candidate individuals, fin tissue samples were collected, and high-density SNP genotype data were obtained using whole-genome resequencing technology. Genome-wide association analysis was conducted to correlate disaster resistance phenotype data with genotype data, and a mixed linear model was used to control for the influence of kinship, identifying SNP loci significantly associated with disaster resistance traits. Genotype feature vectors for each individual were constructed using the identified significant SNP loci, and combined with a comprehensive disaster resistance evaluation value, the optimal linear unbiased prediction method was used to estimate the individual's breeding value. By calculating the kinship coefficient and genetic distance between individuals, a parental compatibility matrix was established, and a gene flow algorithm was used to optimize parental combination schemes, generating a parental selection matrix that maximizes offspring genetic gain. Finally, based on the parental selection matrix, considering factors such as breeding value, parental compatibility, and genetic diversity, the optimal 20% of individuals were selected from the candidate superior disaster-resistant germplasm resources and marked, generating a tuna breeding recommendation individual table containing individual numbers, breeding values, and recommended mating schemes, providing a scientific basis for subsequent breeding work. Meanwhile, the growth performance and status of the corresponding individuals are monitored during the subsequent breeding process. Before the completion of the breeding cycle, they are screened and evaluated again to obtain breeding recommendations that are suitable for the local meteorological environment of the breeding area, thus obtaining tuna with better adaptability to the breeding environment.

[0032] Figure 2 A flowchart of the second method for optimizing tuna pond aquaculture strategy based on meteorological impact analysis, provided in an embodiment of the present invention; like Figure 2 As shown, the present invention provides a second method flowchart for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, comprising: S202, obtain the preset aquaculture management and control strategy of the target aquaculture pond, extract the preset water state benchmark features of the target aquaculture pond as the initial control sequence through the preset aquaculture management and control strategy, initialize the model prediction control framework, and construct a dual-objective optimization function and set constraints with the objectives of maximizing economic benefits and minimizing aquaculture risks. S204, Obtain water body situation response prediction information, initialize the water body situation response prediction information into the model prediction control framework and generate a multi-dimensional search space, introduce a multi-objective reptile algorithm to perform population initialization in the multi-dimensional search space, generate an initial crocodile population P0 for optimization of aquaculture regulation strategy. S206, calculate the fitness value of each crocodile individual in the population based on the initial crocodile population P0 to obtain an initial target matrix F0 containing the fitness values ​​of all individuals; update the position for entering the encirclement phase based on the initial crocodile population P0 and the initial target matrix F0; S208, during the encirclement process, simulate the high-speed movement strategy of crocodiles moving quickly and the local search strategy of crocodiles crawling slowly close to the ground. The selection of the movement strategy is determined by the ratio of the current iteration number to the total iteration number of the corresponding crocodile individual. After all crocodile individuals have completed the position update, the new crocodile population P1 is output. S210, based on the new crocodile population P1, adopts coordinated predation and cooperative hunting strategies to update the position during the hunting phase. When implementing the coordinated predation strategy, individuals move towards the center point determined by the positions of multiple elite individuals in the current population. When implementing the cooperative hunting strategy, individuals move closer to the position of another randomly selected elite individual. S212, In the hunting phase, a dynamic sensitivity parameter and a random perturbation factor are introduced to avoid getting trapped in local optima and to adaptively adjust the search focus range. The dynamic sensitivity parameter adaptively decays as the number of iterations increases, and the hunting phase is completed to finally output the crocodile population P2. S214. Perform non-dominated sorting and unset screening based on the crocodile population P2. Establish Pareto dominance relationships between individuals by traversing and comparing the fitness values ​​of all individuals in population P2, and divide all individuals into non-dominated fronts of different levels. Calculate the crowding distance of individuals located in the same non-dominated front on the corresponding front, and sort them in combination with the Pareto level. S216. Based on the ranking results, individuals with the highest rank and crowding greater than a preset threshold are stored as high-quality solutions in the external archive. An elite retention strategy is adopted to maintain the external archive. Non-dominated optimal solutions generated in each iteration are stored in the external archive. When the new generation of population is generated, elite individuals in the archive are selected first. Pareto optimal solution set is output from the external archive through repeated iterations. Finally, the optimal aquaculture control strategy under the current weather influence is generated.

[0033] It should be noted that, firstly, the water state baseline characteristics of the target aquaculture pond, including parameters such as dissolved oxygen maintenance range and temperature fluctuation threshold, are extracted from the pre-set aquaculture management strategy library as the initial control sequence. When initializing the model predictive control framework, a bi-objective optimization function is constructed with the objectives of maximizing economic benefits and minimizing aquaculture risks. Economic benefits consider feed conversion rate and growth rate, while aquaculture risks include the probability of hypoxia and disease incidence. Constraints such as equipment operating power and water quality safety range are also set. After obtaining the water state response prediction information, it is input into the model predictive control framework to generate a multi-dimensional search space containing equipment control parameters and water quality state variables. A multi-objective reptile search algorithm is introduced for optimization. Compared with classic metaheuristic algorithms such as the gray wolf algorithm and the whale algorithm, the reptile algorithm has a faster convergence speed and higher optimization accuracy. An initial crocodile population P0 is generated through Latin hypercube sampling, with each individual representing a complete control strategy scheme, including decision variables such as aerator operation sequence, feeding amount, and water exchange scheme. Based on the initial population P0, the fitness value of each individual is calculated to form the initial objective matrix F0. During the encirclement phase, the movement strategy is dynamically selected based on the ratio of the current iteration count to the total number of iterations. When the ratio is less than 0.3, a high-speed movement strategy is used for global exploration, such as significantly adjusting the aerator power setting; when the ratio is greater than 0.7, a local search strategy is used for fine-tuning, such as fine-tuning the feeding interval. After all individuals have completed their position updates, a new population P1 is output. Based on population P1, the hunting phase begins, using coordinated predation and cooperative hunting strategies to update positions. The coordinated predation strategy moves individuals towards the center point determined by three elite individuals, while the cooperative hunting strategy moves individuals towards randomly selected superior individuals. A dynamic sensitivity parameter is introduced in this phase, with its value linearly decreasing from 1.0 to 0.1, while a Gaussian random perturbation with a standard deviation of 0.1 is added to effectively avoid getting trapped in local optima. After the hunting phase is completed, population P2 is output. Non-dominated sorting is performed on population P2, and individuals are divided into different levels of non-dominated fronts according to Pareto dominance. The crowding distance between individuals within the same front is calculated and sorted based on the Pareto level. The top 10% of individuals with crowding levels above average are stored in an external archive, and an elite retention strategy is employed to maintain archive diversity. After iteration, the Pareto optimal solution set of non-dominated solutions is output from the external archive, ultimately generating the optimal aquaculture control strategy under the current meteorological influence, including specific equipment control parameters and expected water quality improvement effects. Through the effective combination of multi-objective optimization and predictive control, intelligent generation and dynamic optimization of aquaculture strategies are achieved.

[0034] Furthermore, in a preferred embodiment of the present invention, the adaptive optimization of the control strategy based on water monitoring data during real-time control specifically includes: The optimal aquaculture regulation strategy under the current meteorological influence is obtained. The optimal regulation command sequence for aquaculture ponds in the current time domain is generated through the model predictive control framework and sent to various execution agencies to regulate the water body of aquaculture ponds. During the regulation process, the first control command of the optimal regulation command sequence for the aquaculture pond is executed first, and the actual water body regulation response data is obtained by real-time water environment monitoring through the deployed heterogeneous sensor network. The data is then compared with the preset water body state benchmark characteristics of the target aquaculture pond to obtain the deviation sequence. The deviation sequence is compared with the preset tolerance range. If it is not within the preset tolerance range, the water body response situation prediction information at the current moment is obtained based on the actual water body regulation response data. This information is then input into the model prediction control framework for sample update and re-optimization of the aquaculture regulation strategy. The updated optimal regulation command sequence for the aquaculture pond is then obtained for the next moment's aquaculture pond regulation.

[0035] It should be noted that after obtaining the optimal aquaculture control strategy under the current meteorological influence, it is transformed into a specific sequence of equipment control commands through a model predictive control framework. This sequence contains equipment operating parameters for multiple control time domains in the future, including power adjustment commands for aerators, feeding amount and frequency settings for feeders, and valve opening control for the water circulation system. These commands are sent to the corresponding actuators in real time, initiating active regulation of the aquaculture pond water. During the execution of the control strategy, the first control command in the control sequence is executed first, while the real-time response status of the water body is continuously monitored through a heterogeneous sensor network deployed in the pond. The monitoring data includes key water quality parameters such as dissolved oxygen concentration, water temperature, and pH value. These actual monitoring data are compared with preset water body state benchmark characteristics to calculate the real-time control deviation sequence. For example, if the dissolved oxygen concentration in a certain area does not reach the target range within the expected time after the aerator is turned on, the deviation is recorded and the cause is analyzed. The calculated deviation sequence is dynamically compared with a preset tolerance range, which is set according to different water quality parameter characteristics and aquaculture requirements. When the deviation value continues to exceed the tolerance range, a new short-term water body response trend prediction is made based on the latest actual monitoring data. The system model parameters are updated using the recursive least squares method, and the corrected prediction information is fed back to the model predictive control framework. After receiving the updated prediction information, the model predictive control framework initiates a strategy re-optimization process. First, the sample database is incrementally updated. Then, the optimization problem is resolved based on the latest system state, and the equipment operating parameters in subsequent control time domains are adjusted to compensate for the previous control deviation. The updated control command sequence generated after re-optimization will be implemented in the next control time domain, thus forming a closed-loop control mechanism of "monitoring-evaluation-adjustment" to ensure that the aquaculture water body is always maintained in an ideal state. This adaptive optimization mechanism can effectively cope with the impact of uncertainties such as sudden changes in meteorological conditions and fluctuations in equipment performance on the control effect.

[0036] Furthermore, the method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis provided by this invention also includes the following steps: The heterogeneous sensor array is deployed to continuously monitor the changes in pond water and the behavior of fish underwater during the current weather period. The collected data on changes in pond water and fish underwater are processed using timestamp synchronization and spatial coordinate registration methods. The extended Kalman filter algorithm is used to perform dynamic noise cancellation and state estimation to form a multidimensional spatiotemporal monitoring dataset. Based on the multidimensional spatiotemporal monitoring dataset, the Kriging spatial interpolation algorithm is used to transform the discrete water quality monitoring data into a spatially continuous spatial distribution field of water parameters. At the same time, the kernel density estimation algorithm is used to extract the spatial distribution features of fish behavior monitoring data, generating a fish activity heat map that reflects the spatial differences in the intensity of fish activity. Spatial heterogeneity analysis and abnormal region identification are performed based on the spatial distribution map of water parameters and the thermal distribution map of fish activity. The edge detection algorithm is used to identify gradient change regions in the distribution of environmental parameters. Combined with the low-density regions in the fish behavior thermal map, the overlapping regions of environmental anomalies and behavioral anomalies are determined through spatial overlay analysis. Density-based spatial clustering algorithms are used to perform cluster analysis on overlapping areas, and spatially adjacent and similar outliers are aggregated into candidate weak control sub-regions. Multi-dimensional risk assessment and priority ranking are then performed on the identified candidate weak control sub-regions. A comprehensive assessment model was established, with four evaluation indicators preset: area size, degree of environmental anomaly, duration of anomaly, and intensity of fish escape behavior. The entropy weight method was used to determine the weight of each indicator, and the comprehensive risk score of each area was calculated. Based on the risk score, the area was divided into three risk levels: high, medium, and low, and a list of vulnerable areas was generated in order of priority. For high-priority weak areas, conduct cause analysis and equipment demand assessment, identify the dominant environmental factors causing regional anomalies through correlation analysis, assess the coverage blind spots and inadequate performance of existing control equipment in combination with regional spatial location characteristics, and determine the types and technical requirements of equipment that need to be added or optimized based on the analysis results. Based on the spatial distribution characteristics and causal analysis results of the weak areas, a greedy algorithm is used to optimize the equipment layout. The principle is to achieve the maximum coverage with the minimum number of equipment. The optimal installation location and orientation of the new equipment are calculated, and the layout optimization suggestion scheme including equipment type and installation coordinates is output.

[0037] It should be noted that in traditional aquaculture models, managers often rely on limited manual observation points and experience-based judgment to implement environmental control. This approach struggles to fully grasp the spatiotemporal heterogeneity of the aquatic environment within large-scale aquaculture ponds, especially during periods of meteorological disasters. This often leads to mismatches between control measures and actual conditions, and inconsistencies between equipment layout and demand. Consequently, localized environmental deterioration may occur without timely intervention, or insufficient control capacity in certain areas may prolong the control period during meteorological impacts, directly affecting the pond's ecological stability and response rate. Therefore, this innovative approach uses spatial correlation analysis between aquatic environmental parameters and fish behavior to identify problem areas from a holistic ecosystem perspective, overcoming the limitations of traditional methods that focus solely on single water quality indicators. Furthermore, by introducing a multi-indicator risk assessment model and equipment layout optimization algorithms, management measures shift from experience-driven to data-driven, significantly improving the scientific rigor and foresight of aquaculture environmental management. By employing heterogeneous sensor arrays and advanced data processing methods, this solution can promptly detect local anomalies that are difficult to detect using traditional methods. Secondly, based on spatial analysis and clustering identification methods, it can accurately locate weak areas in management, providing clear targets for targeted improvements. Thirdly, through multi-dimensional risk assessment and prioritization, management resources can be allocated to the most critical areas, improving management efficiency. Finally, equipment layout optimization based on a greedy algorithm ensures maximum coverage with minimal investment, solving practical problems while controlling costs. Overall, this solution effectively improves the precision of aquaculture environment management, providing reliable technical support for responding to the impact of meteorological disasters, maintaining aquatic environment stability, and ensuring aquaculture safety.

[0038] Figure 3 A system 3 for optimizing tuna pond aquaculture strategies based on meteorological impact analysis is provided in one embodiment of the present invention. The system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When the processor 302 executes the program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, it performs the following steps: Regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond are acquired and preprocessed. After preprocessing, a multimodal pond ecological feature vector of the target aquaculture pond is constructed. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. The preset aquaculture management strategy for the target aquaculture pond is obtained, and the aquaculture control strategy under the current meteorological influence is formulated in combination with the water body situation response prediction information. The control strategy is adaptively optimized based on the water body monitoring data during the real-time control period. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis, and finally a disaster-resistant breeding recommendation program is obtained.

[0039] In another aspect, the present invention provides a computer-readable storage medium comprising a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When the program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis is executed by a processor, it implements the steps of the method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis as described in any of the preceding claims.

[0040] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0041] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0042] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0043] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0045] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, characterized in that, include: Regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond are acquired and preprocessed. After preprocessing, a multimodal pond ecological feature vector of the target aquaculture pond is constructed. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. The preset aquaculture management strategy for the target aquaculture pond is obtained, and the aquaculture control strategy under the current meteorological influence is formulated in combination with the water body situation response prediction information. The control strategy is adaptively optimized based on the water body monitoring data during the real-time control period. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis, and finally a disaster-resistant breeding recommendation program is obtained.

2. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The process involves acquiring regional meteorological data and pond aquaculture monitoring data for the target aquaculture pond and preprocessing them. After preprocessing, a multimodal pond ecological feature vector for the target aquaculture pond is constructed, specifically including: Regional meteorological data, including real-time observations and short-term forecasts, is obtained through meteorological stations deployed in aquaculture areas and external forecast interfaces. Pond aquaculture monitoring data, including water environment parameters and fish monitoring images, is collected through heterogeneous sensor arrays deployed in ponds, forming an asynchronous multi-source data stream and undergoing preprocessing. For regional meteorological data and water environment parameters, the local outlier detection algorithm is used to identify outliers. The relative density of each data point in its local neighborhood is calculated according to a preset time window, and the density is compared with that of neighboring data points to obtain the local outlier score. When the score exceeds the preset score threshold, it is marked as an outlier data point. The time series parameter sequence after marking outlier data points is obtained and the time series parameter sequence is decomposed into trend term, seasonal term and residual term. The trend term captures the long-term change direction of the data, the seasonal term extracts the periodic change pattern, and the residual term contains random fluctuations and outlier information, generating sequence decomposition information. After completing the sequence decomposition, the normal time period data segment is extracted based on the sequence decomposition information. The blank long short-term memory network model is called and trained through the residual sequence of the normal time period data segment. After training is completed, the residual terms corresponding to the normal data before and after the abnormal time period are input to predict the residual value of the abnormal time period. Finally, the predicted residual value is recombined with the corresponding trend term and seasonal term. The reconstructed regional meteorological data sequence and aquatic environmental parameter sequence are time-aligned using a dynamic time warping algorithm. Simultaneously, the fish monitoring images are enhanced and filtered, and finally, a preprocessed asynchronous multi-source dataset is output.

3. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The process of acquiring and preprocessing regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond, and constructing a multimodal pond ecological feature vector of the target aquaculture pond after preprocessing, further includes: Asynchronous multi-source dataset is obtained. Preprocessed fish monitoring images are extracted from the asynchronous multi-source dataset and imported into Gaussian mixture model for background modeling and moving target segmentation. Based on the color distribution features of each pixel in consecutive frames, pixels that conform to the background distribution are classified as background, while pixels that do not conform to the background distribution are judged as foreground moving targets. After completing the segmentation of moving targets, the YOLOv5 deep neural network architecture is used to identify and locate fish bodies in the segmented foreground targets. The target detection is performed on multi-scale feature maps through its anchor box mechanism. Multi-level features are extracted by the backbone network and combined with the feature pyramid network to enhance the detection capability, and fish body target detection information is output. Based on the fish target detection information, the DeepSORT multi-target tracking algorithm is used to construct the fish school movement trajectory. Kalman filtering is used to predict the target position in the current frame based on the trajectory state of the previous frame. By fusing the correlation metric of appearance features and motion information, the detection box and trajectory are matched to obtain the fish movement trajectory information. Based on the fish movement trajectory information, fish school behavior characteristics including average swimming speed, movement path tortuosity, spatial distribution density and group aggregation are calculated. At the same time, the principal component analysis method is used to reduce the dimensionality of multidimensional water quality parameters and extract the principal component features whose cumulative contribution rate reaches the predetermined contribution threshold as key water body environmental factors characterizing the water body state, and finally the water body environmental feature vector is obtained. The asynchronous multi-source dataset is used to extract the temporal features of meteorological elements, calculate the mutual information value between fish behavior features and key environmental factors of the water body to construct environmental behavior interaction features, and combine the water body environmental feature vector and the temporal features of meteorological elements to perform multi-dimensional vector splicing to generate a multimodal pond ecological feature vector.

4. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The method employs a spatiotemporal graph neural network to construct a dynamic coupling model of meteorology and water bodies. Combined with the analysis of the multimodal pond ecological feature vectors, it analyzes the response of pond water bodies to meteorological changes, obtaining water body situation response prediction information. Specifically, this includes: A multimodal pond ecological feature vector is obtained. Based on the geospatial layout of the pond, the aquaculture water area is discretized into a regular grid structure. Each grid cell is defined as a node of the graph. Based on the multimodal pond ecological feature vector, node features corresponding to the spatial location are assigned. The node features consist of environmental parameters, meteorological elements and fish behavior indicators of the corresponding location. The K-nearest neighbor algorithm based on spatial distance is used to establish physical proximity edges. Historical water flow velocity and direction data are used to calculate the hydrodynamic correlation between nodes. Functional connection edges are established for nodes whose correlation exceeds the threshold, and a dynamic graph structure containing geometric spatial proximity and water movement correlation is constructed. A dynamic coupling model of meteorology and water bodies is constructed using a spatiotemporal graph neural network. The dynamic graph structure is input and stacked spatiotemporal convolutional blocks are used to extract hierarchical spatiotemporal features. Based on Chebyshev multinomial approximation spectral graph convolution, the spatial dependencies between nodes are captured by multi-order neighborhood feature aggregation, and high-order spatial feature representations are obtained by stacking multi-layer graph convolutions. The high-order spatial feature representation is input into the gated recurrent unit network for time dynamic modeling. The degree of retention of historical state information is controlled by updating the gate, and the combination of the current input and historical state is determined by the reset gate. The importance weight of each time step in the hidden state sequence is calculated, and adaptive focusing is performed on the impact period of key meteorological events. Finally, the spatiotemporal feature representation is output. The temporal features of meteorological elements are extracted from the multimodal pond ecological feature vector and converted into meteorological context vectors by an encoder. The cross-attention weights between the meteorological context and the state of all nodes are calculated using a multi-head attention mechanism and fused with the spatiotemporal hidden state of each node to obtain weighted node features. The weighted node features are input into the decoder, which consists of a temporal deconvolution layer and a fully connected layer, to map the high-dimensional features into the predicted environmental parameters of each grid node in the future time step. The predicted environmental parameters of all nodes in each time step are reconstructed into two-dimensional field data according to spatial coordinates. The Kriging spatial interpolation algorithm is used to generate a spatial distribution map of the water body situation. Based on the spatial distribution map of the water body situation, the parameter change rate, extreme values ​​and duration at each spatial location are extracted to generate water body situation response prediction information, which characterizes the spatiotemporal evolution of the pond water environment under future meteorological scenarios.

5. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The acquisition of the preset aquaculture management strategy for the target aquaculture pond, combined with the water body situation response prediction information, to formulate an aquaculture control strategy under the current meteorological influence, specifically includes: Obtain the preset aquaculture management and control strategy for the target aquaculture pond, extract the preset water state benchmark features of the target aquaculture pond as the initial control sequence through the preset aquaculture management and control strategy, initialize the model prediction control framework, and construct a dual-objective optimization function and set constraints with the objectives of maximizing economic benefits and minimizing aquaculture risks. The water body situation response prediction information is obtained, and the water body situation response prediction information is initialized in the model prediction control framework and a multi-dimensional search space is generated. A multi-objective reptile algorithm is introduced to perform population initialization in the multi-dimensional search space to generate an initial crocodile population P0 for optimization of aquaculture regulation strategy. The fitness value of each crocodile individual in the population is calculated based on the initial crocodile population P0 to obtain an initial target matrix F0 containing the fitness values ​​of all individuals; the position is updated during the encirclement phase based on the initial crocodile population P0 and the initial target matrix F0. During the encirclement process, a high-speed movement strategy simulating the rapid movement of crocodiles and a local search strategy simulating the slow crawling of crocodiles close to the ground are used. The choice of movement strategy is determined by the ratio of the current iteration number to the total number of iterations for the corresponding crocodile individual. After all crocodile individuals have completed the position update, a new crocodile population P1 is output. Based on the new crocodile population P1, the location update during the hunting phase is carried out using coordinated predation and cooperative hunting strategies. When the coordinated predation strategy is executed, individuals move towards the center point determined by the positions of multiple elite individuals in the current population. When the cooperative hunting strategy is executed, individuals move towards the position of another randomly selected elite individual. In the hunting phase, dynamic sensitivity parameters and random perturbation factors are introduced to avoid getting trapped in local optima and to adaptively adjust the search focus range. The dynamic sensitivity parameters adaptively decay as the number of iterations increases, and the hunting phase is completed to finally output the crocodile population P2. Based on the crocodile population P2, non-dominated sorting and unset screening are performed. Pareto dominance relationships between individuals are established by traversing and comparing the fitness values ​​of all individuals in population P2. All individuals are divided into non-dominated fronts of different levels. For individuals located in the same non-dominated front, the crowding distance on the corresponding front is calculated and sorted in combination with the Pareto level. Based on the ranking results, individuals with the highest rank and crowding greater than a preset threshold are stored as high-quality solutions in the external archive. An elite retention strategy is adopted to maintain the external archive. Non-dominated optimal solutions generated in each iteration are stored in the external archive. When the new generation of population is generated, elite individuals in the archive are selected first. Through repeated iterations, the Pareto optimal solution set is output from the external archive, and finally the optimal aquaculture control strategy under the current weather influence is generated.

6. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The adaptive optimization of the control strategy based on water monitoring data during real-time control specifically includes: The optimal aquaculture regulation strategy under the current meteorological influence is obtained. The optimal regulation command sequence for aquaculture ponds in the current time domain is generated through the model predictive control framework and sent to various execution agencies to regulate the water body of aquaculture ponds. During the regulation process, the first control command of the optimal regulation command sequence for the aquaculture pond is executed first, and the actual water body regulation response data is obtained by real-time water environment monitoring through the deployed heterogeneous sensor network. The data is then compared with the preset water body state benchmark characteristics of the target aquaculture pond to obtain the deviation sequence. The deviation sequence is compared with the preset tolerance range. If it is not within the preset tolerance range, the water body response situation prediction information at the current moment is obtained based on the actual water body regulation response data. This information is then input into the model prediction control framework for sample update and re-optimization of the aquaculture regulation strategy. The updated optimal regulation command sequence for the aquaculture pond is then obtained for the next moment's aquaculture pond regulation.

7. The method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis according to claim 1, characterized in that, The process involves evaluating the disaster resistance of tuna in the target aquaculture ponds after the current weather impact ends. Based on the evaluation results, superior disaster-resistant germplasm resources are screened, and a parent selection matrix is ​​constructed by combining genetic pedigree analysis. Finally, a recommended disaster-resistant breeding program is obtained, which specifically includes: After the target meteorological impact event ends, environmental monitoring data and fish behavior response data throughout the entire meteorological event are collected, and after standardization and preprocessing, a meteorological response dataset is generated. Based on the meteorological response dataset, the principal component analysis method is used to reduce the dimensionality of all environmental feature parameters to obtain the principal component scores of each environmental feature parameter. The main environmental feature parameters are then selected as environmental stress feature vectors by using the principal component scores. Subsequently, fish behavior response data were used to evaluate the environmental adaptability, stress recovery speed and growth stability of each individual during the current weather impact. Combined with the environmental stress feature vector, the TOPSIS comprehensive evaluation method was used to calculate the comprehensive evaluation value of the disaster resistance of each tuna individual. Based on the comprehensive evaluation value of each tuna individual's disaster resistance ability, the fish population is divided into three levels: strong disaster resistance, medium disaster resistance, and weak disaster resistance using a set grading threshold. Strong disaster resistance individuals are selected as candidate superior disaster resistance germplasm resources. Genome analysis was performed on the selected candidate individuals with superior disaster resistance. Biological samples were collected from the candidates and their SNP genotype data were obtained through whole-genome sequencing. Genome-wide association analysis was used to correlate disaster resistance phenotype data with genotype data. A mixed linear model was used to control for the influence of population structure and to identify SNP sites that were significantly associated with disaster resistance traits. Genotype feature vectors related to disaster resistance are constructed for each individual using the identified significant SNP loci to generate genotype feature spectra. Parental selection and optimization are carried out by combining the comprehensive evaluation value of disaster resistance ability of each tuna individual. The best linear unbiased prediction method is used to estimate the breeding value of each individual. By calculating the kinship coefficient and genetic distance between individuals, a parental selection affinity matrix is ​​established. At the same time, considering the heritability and selection intensity of disaster resistance traits, the gene flow algorithm is used to optimize the parental combination scheme and generate a parental selection matrix that maximizes the genetic gain of offspring. Based on the parental mating matrix, the candidate superior disaster-resistant germplasm resources are screened again to obtain the optimal disaster-resistant individuals and mark them, generating a recommended individual table for tuna breeding at the current time.

8. A tuna pond aquaculture strategy optimization system based on meteorological impact analysis, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When the processor executes the program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis, it performs the following steps: Regional meteorological data and pond aquaculture monitoring data of the target aquaculture pond are acquired and preprocessed. After preprocessing, a multimodal pond ecological feature vector of the target aquaculture pond is constructed. A dynamic coupling model of meteorology and water body is constructed using a spatiotemporal graph neural network. The response of pond water body to meteorological changes is analyzed by combining the multimodal pond ecological feature vectors to obtain water body situation response prediction information. The preset aquaculture management strategy for the target aquaculture pond is obtained, and the aquaculture control strategy under the current meteorological influence is formulated in combination with the water body situation response prediction information. The control strategy is adaptively optimized based on the water body monitoring data during the real-time control period. After the current weather impact ends, the disaster resistance of tuna in the target aquaculture ponds is evaluated. Based on the evaluation results, superior disaster-resistant germplasm resources are screened and a parent selection matrix is ​​constructed by combining genetic pedigree analysis, and finally a disaster-resistant breeding recommendation program is obtained.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a program for optimizing tuna pond aquaculture strategies based on meteorological impact analysis. When the program is executed by a processor, it implements the steps of the method for optimizing tuna pond aquaculture strategies based on meteorological impact analysis as described in any one of claims 1 to 7.

Images