Abstract:

To improve the sewage removal efficiency of aquaculture tanks， a funnel-type sewage collection device was designed. Taking a dual-channel circular aquaculture tank with bidirectional jet inlet as an example， a three-dimensional two-phase flow field model was established. Based on the verification of grid independence and the effectiveness of numerical methods， the velocity distribution， water flow uniformity index， water circulation energy utilization rate， and the migration law， agglomeration effect， and removal rate of sewage with different densities were studied. The results show that： The funnel can increase the axial velocity gradient of the aquaculture tank， form a central low-pressure circulation， and promote the agglomeration and discharge of sewage. With the increase of inlet mass flow rate， the mixing capacity of the middle and upper water bodies in the tank is enhanced， but the turbulence and secondary flow in the bottom water body are intensified， causing the sewage removal rate to show a "first decrease then increase" trend. Suspended solids are easily driven by water buoyancy and mainly discharged through the overflow outlet， with the removal rate increasing with the flow rate. The migration path and removal rate of medium-density sediments fluctuate significantly with the flow rate. Sediments are dominated by gravity， and excessive flow rate can cause turbulent disturbance， reducing the removal rate.

Abstract:

To address the inefficient dewatering of solid waste in seawater aquaculture effluent （characterized by high moisture， viscosity， and fine particles）， a chamber filter press was designed based on the slurry's measured physicochemical properties. A multi-chamber CFD model was established in FLUENT to simulate the filtration process. Operational parameters were optimized using a Box-Behnken response surface methodology targeting filter cake moisture content. The optimal combination was 0.85 MPa feed pressure， 0.15 medium porosity， and 20% feed concentration， giving a predicted moisture content of 46.65%. Prototype tests yielded an average cake moisture content of 51.46%， with the filtrate containing 14.3 mg/L suspended solids at pH 7.55， meeting discharge standards. The study demonstrates that the chamber filter press， optimized through integrated CFD simulation and response surface methodology， achieves effective solid-liquid separation， providing a viable solution for the reduction and resource recovery of aquaculture solid waste.

Abstract:

To investigate the wave-dissipation performance of novel hydraulic structures in fishing ports under different wave conditions and to ensure safe berthing and operation of fishing vessels， this paper innovatively designs two new types of perforated hydraulic structures： non-connected holes and connected holes. Through physical model tests， the variation patterns of the reflection coefficients of these two new perforated hydraulic structures under different wave parameters are systematically analyzed to explore their wave-dissipation effects. The experimental results show that under composite wave conditions with large wave heights and long periods， the non-connected structure exhibits lower reflection coefficients compared to the connected-hole structure， indicating better wave-dissipation performance. In contrast， the connected structure demonstrates superior wave-dissipation effects under short-period waves and low-water-level conditions. This study can serve as a basis for structural selection in engineering design according to the characteristics of the design wave spectrum， providing important references for the optimized design of novel hydraulic structures such as breakwaters and wharves in fishing ports.

Abstract:

As nearshore aquaculture space approaches saturation， offshore farming is gradually shifting from coastal areas to the deep sea. However， the marine environment in the deep sea is more severe， imposing higher demands on the structural safety performance of aquaculture facilities.The main structure of the truss-type aquaculture platform is rigid， offering excellent resistance to wind and waves as well as deformation. Research focuses on structural stability and stress characteristics， including platform attitude deviation， decay of restoring torque， capsizing risk， and mooring forces.To investigate the impact of mooring anchor chain failure on the hydrodynamic characteristics of truss-type aquaculture platforms， this study conducts a comparison of the hydrodynamic response characteristics of a truss-type aquaculture platform under two conditions： intact mooring system and single mooring chain failure based on physical model flume tests. The results indicate that mooring chain failure reduces the positioning capability of the mooring system on the aquaculture platform. Under irregular wave action， single mooring chain failure does not change the variation law of structural motion response with period，but it does change the amplitude of the motion response：the surge response of the aquaculture platform increases by a maximum of 7.45%， the pitch response increases by a maximum of 16.61%， and the heave response decreases by 3.99%. Meanwhile， as the significant period increases continuously， the influence of mooring chain failure on both surge and pitch responses gradually diminishes， while its influence on the heave response continues to increase. Under the action of irregular wave-current coupling， the changes in pitch response and mooring force on the wave-current-facing side are the most significant. After mooring chain failure， the effective value of pitch motion increases by 1.19 times， with a maximum tilt angle of 15.59°， and the effective value of mooring force on the incident wave-current side increases by 2.85 times. The research results provide a theoretical reference for the design of truss-type aquaculture platforms and their mooring systems in deep and open seas.

Abstract:

The hardness of the flesh， low temperature， and large body weight of super-chilled tuna make cutting and processing highly dangerous， with large processing errors and low efficiency. To address the cutting and processing of ultra-low temperature tuna with a large saw， CCD dual-vision calibration， retractable push rod fixture， and spinal line dynamic tracking technology were studied， and an integrated intelligent processing system combining detection， handling， and processing was designed. CCD （Charge coupled device） dual vision camera was used for automatic identification of fish body posture. Characteristic parameter measurements and analyses of ultra-low temperature tuna were conducted， a special retractable push rod clamp for tuna handling was designed， and clamping surface texture selection and clamping force calculation analyses were performed. Finally， the impact of tuna spinal line dynamic tracking cutting processing errors was analyzed and cutting process was simulated by employing PDPS （Process design and process simulate）. Research results show that the designed ultra-low temperature tuna intelligent processing system operates smoothly with no safety hazards. The adopted fish posture automatic identification facilitates precise clamping and handling by the robot. The fish bone path cutting accuracy is ±1.5mm， and the single detection time， handling time， and cutting time are 2 s， 5 s and 4 s， respectively. The operation efficiency of each process link is higher than that of manual cutting， meeting the processing requirements.

Abstract:

As an emerging field in China's agricultural development， offshore aquaculture relies on intelligent monitoring as a fundamental prerequisite for ensuring its economic benefits and quality， and it is also an important basis for risk identification and early warning. To further serve industrial development， this paper focuses on intelligent monitoring technologies for offshore aquaculture. Based on existing research findings， it classifies and sorts out the current development status of these technologies， dividing them into monitoring technologies for aquaculture equipment， cultured organisms， and aquaculture environments， which cover the main contents of the offshore aquaculture process.In addition to the traditional monitoring of water quality， structural health， and fish， this paper further points out that with the development and expansion of the offshore aquaculture industry， the monitoring of underwater noise and safety environments of platforms will also become increasingly important. Based on the technical characteristics at the present stage， this paper systematically summarizes the urgent problems that need to be solved currently， including in-situ measurement capabilities， equipment quality， and noise monitoring standards， as well as front-end sensing technologies and back-end data processing support technologies.This paper also looks forward to the future development trends of the technology， including digital intelligence driven， multi-modal integration and fusion， biofouling protection， and system construction. As the intelligence level of monitoring technologies continues to improve， a complete intelligent data support system for land-sea interaction will be formed， which will further promote the development and growth of the marine digital industry.

Abstract:

Current intelligent feeding boats face low efficiency and insufficient positioning accuracy when transferring between multiple ponds. Therefore， a split-type pond-transfer structural design scheme and a fusion positioning algorithm are proposed. This system uses a modular combination of a catamaran and an intelligent feeding vehicle， together with a pond-transfer ramp and docking limit devices. By optimizing the mechanical structure to design a power switching device， the rapid combination and separation of the boat body and feeding vehicle are achieved. An adaptive strong tracking Kalman filter fusion positioning algorithm based on UWB and IMU is proposed to improve positioning accuracy and robustness in complex water environments. The algorithm is simulated and compared using MATLAB， and pond-transfer ramp experiments verify the passability and operational efficiency under different slopes and load conditions. Pond-transfer ramp tests show that the feeding vehicle transfer time is under 25 seconds under various slopes and loads， much shorter than manual transfer time， and operations are smooth. Simulation results of the positioning algorithm indicate that compared to single UWB positioning， the ASTKF algorithm reduces mean square error by 98.9% and maximum error by 70.6%. The ASTKF algorithm has strong real-time response and recovery capabilities against dynamic interference. During actual boat positioning tests， the root mean square error is 5.41 cm， 27.3% lower than the AEKF algorithm. The split-type pond-transfer scheme and the proposed fusion positioning algorithm effectively solve the problems of low transfer efficiency and poor positioning accuracy in traditional feeding boats. This study enhances the level of intelligent operations and economic efficiency in complex water areas， providing reliable technical support and application demonstration for aquaculture automation.

Abstract:

To enhance the management of waterweed in crab farming ponds， this paper integrates and optimizes image processing methods and the path planning of waterweed clearing boat，aiming to develop a high-quality waterweed control strategy.The research employs unmanned aerial vehicle （UAV） photogrammetry to acquire multi-temporal imagery of crab ponds， and proposes a refined YOLO11n-seg network architecture. Specifically， the Neck stage is enhanced with a lightweight High-level Screening Path Aggregation Network-Dysample （HSPAN-D） module incorporating dynamic upsampling operators， the original C3k2 module is replaced with a lightweight feature extraction module C3k2_Faster_EMA， and an EfficientHead lightweight segmentation head is introduced. Based on model recognition and segmentation outputs， a raster map of the crab pond is constructed， and a target waterweed clearing area screening mechanism is designed. An improved A* algorithm is then developed through path optimization strategies to enable precise waterweed clearing path planning. Results demonstrate that the enhanced model achieves a 1.6% improvement in waterweed recognition precision and a 0.5% increase in mean Average Precision （mAP）， while reducing parameter count by 39.4%， computational overhead by 25.5%， and model size by 34.5%.The improved A* algorithm exhibits more precise control over waterweed area coverage compared with manual clearing paths. Relative to the conventional A* algorithm， it reduces total path length by 14.25 meters， decreases cleaning boat turning maneuversby 10 instances， and shortens average planning time by 1.97 seconds.The research confirms that the proposed lightweight enhancement strategy significantly reduces computational burden while improving recognition accuracy. Coupled with the improved path planning algorithm， it enables efficient and precise waterweed clearing in crab ponds. This study provides a valuable path planning reference for practical operations of waterweed clearing boat.

Abstract:

To address the issues of low mechanization levels and high labor intensity in the fishing process of Litopenaeus vannamei， this paper proposes an optimized structural design for the ground cage used in rail-type fishing systems within greenhouse ponds to enhance fishing efficiency. Electromagnetic coupling technology was employed to achieve automatic attachment and detachment between the fishing module and the rail lifting system. Furthermore， a trapezoidal structural model of the shrimp inlet was constructed using parametric geometric modeling. Building on this， and incorporating the cruising behavior characteristics of Litopenaeus vannamei， a shrimp-structure collision detection model based on a normal vector projection algorithm was established to dynamically simulate and quantitatively analyze the trap entry process. Response surface methodology （RSM） was utilized to systematically investigate the influence mechanisms of key parameters-specifically outer inlet length， sidewall inclination angle， and depth-on harvesting efficiency. The results demonstrated that the model possesses high reliability （R²>0.9）. Analysis of the structural parameters revealed that the outer inlet length and sidewall inclination angle had the most significant impact on harvesting efficiency （P<0.01）. Optimal harvesting efficiency was achieved when the ratio of the outer inlet length to depth was 2.08， and both the sidewall and inner inlet inclination angles were 75°. Field comparative trials indicated that， compared to the original design， the optimized fishing device increased harvesting efficiency by 14.5%， 12.8%， and 14.6% across different operational periods， respectively. The study concludes that the designed umbrella-shaped ground cage satisfies the requirements of facility-based greenhouse pond rail harvesting systems， offering a viable solution to the harvesting challenges in Litopenaeus vannamei aquaculture in terms of technical advancement， economic feasibility， and operational efficiency. This research provides a theoretical basis and technical reference for the structural optimization of fishing gear and the development of automated harvesting systems.

Abstract:

This study targets detection degradation in the pump-downstream channel （conveyor/dewatering chute） of light-luring purse-seine operations in the South China Sea， where fish exhibit small size， high speed， strong motion blur， and dense occlusion. A blur-tolerant， detail-preserving real-time detector， MGI-RTDETR， is presented. Built on RT-DETR， it integrates Multi-Scale Grouped Interaction features （MGI-FE）， Adaptive Dynamic Context Mixing （ADC-MI）， Adaptive Detail-Preserving Fusion （ADPF）， deployment-time Consolidated Re-parameterization （CRC）， and an Adaptive Reverse-Projection Upsampler （ARPU）， improving small-object boundary restoration and separability in crowded scenes without increasing inference-time complexity. A 396-image single-class dataset （train/val/test=7/1/2） is used under a unified 640×640 training/inference protocol. On the test set， the method attains Precision/Recall/F-Score of 88.54%/75.03%/81.23%， mAP50 of 81.00%， and mAP_50-95 of 28.29%， while maintaining deployment-friendly cost （44.5 GFLOPs， 13.39 M parameters，~50.46 FPS）， outperforming RT-DETR-R18 and multiple YOLO/SSD baselines. To illustrate downstream usability， ByteTrack is stacked without modifying the detector to realize line-crossing counting； counting precision （CP） at low/medium/high densities is 95.0%/80.8%/49.7%. To assess cross-vessel generalization， an external test set of 309 images from three additional vessels is evaluated， confirming trends consistent with the main experiment under varied viewpoints and illumination. The results indicate that multi-scale interaction and cross-scale adaptive fusion oriented to blur and fine-detail preservation provide a practical foundation for near-real-time， georeferenced catch analytics on working vessels， supporting transparency and data-driven fisheries management.

Abstract:

Uneven illumination in underwater environments often leads to blurred object boundaries and severe imaging degradation， posing significant challenges for instance segmentation in aquaculture and ecological monitoring scenarios. These issues are particularly pronounced in images of aquatic organisms with complex textures， such as largemouth bass. To address the problems of weak detail representation， ambiguous object contours， and strong noise interference in underwater imagery， this paper proposes an improved underwater instance segmentation method based on the UniInst framework. The proposed approach introduces a Squeeze-and-Excitation （SE） module to enhance channel-wise responses of multi-scale features by modeling global semantic dependencies， thereby improving the discriminative capability of feature representations. In addition， a Spatial Response Modulation （SRM） module is incorporated to further suppress noisy regions and strengthen the preservation of local structural information through spatial response regulation on top of channel semantic calibration. Furthermore， an edge supervision branch is designed， in which a hybrid edge loss combining Binary Cross-Entropy （BCE） loss and the Dice coefficient is employed to guide the model to more effectively exploit boundary structural information during mask prediction. Experimental results demonstrate that the proposed method consistently outperforms the baseline and other mainstream models on instance segmentation tasks， achieving an average precision （AP） of 84.882% and an AP₇₅ of 97.882%. Notably， the proposed method exhibits superior performance in fine structural recovery and high-IoU threshold segmentation accuracy. These results indicate that the proposed approach can effectively improve instance segmentation performance in complex underwater environments， showing strong robustness and promising application potential.

Abstract:

Assessing hunger levels is a critical step in achieving precise feeding in aquaculture. Existing feeding methods often rely on manual experience， lack a basis for determining feeding amounts， and lead to overfeeding. Current common classifications of fish feeding behavior mostly depend on the intensity of water splashing during feeding to qualitatively analyze feeding intensity. However， these classification criteria are highly subjective and essentially still rely on manual experience. In aquaculture， approximately 15%-30% of feed is wasted due to overfeeding， with overfeeding contributing 12.3% of this waste.Therefore， a quantitative detection method based on the fusion of depth video and infrared video is proposed to improve the accurate determination of bass feeding intensity. The processed frame images from depth and infrared videos are input into the DAIF-MOG2 （Depth and Infrared Fusion-Mixture of Gaussians 2） optimized model for feature extraction. The extracted results are then fused and evaluated for hunger levels， ultimately providing a quantitative hunger score. For the DAIF-MOG2 optimized model， improvements were made based on the MOG2 algorithm， transforming single-stage， single-modal learning into a multi-stage， multi-modal learning approach. This enhances learning stability and compensates for the limitations of a single modality. The introduction of shape feature constraints and physical space validation constraints improves overall detection performance and robustness in complex environments. Experimental results demonstrate that the proposed multimodal fusion model for quantifying bass feeding intensity achieves an accuracy rate of 94.2%. Compared to the original MOG2 model， its overall performance improves by 51.6%. It enables rapid assessment of fish hunger levels， effectively utilizes multimodal information， ensures timely and accurate feeding of bass in practical aquaculture scenarios， and reduces feed wastage.

Abstract:

To address the challenges of online acoustic monitoring of behavior and accurate biomass assessment of large yellow croaker in complex environments such as net cages， this study investigated the applicability of a horizontal split-beam scientific echosounder system in an actual aquaculture setting. A scientific echosounder （Kongsberg EK80， 200 kHz） was employed to conduct standard sphere positioning accuracy experiments in a tank. Subsequently， the system was deployed at the Zhoushan Liuheng Marine Ranch to conduct continuous 24-hour horizontal monitoring of large yellow croaker in a net cage. Data regarding attitude angles （based on the net cage coordinate system）， swimming speeds， and horizontal-aspect in situ target strength （TS） were analyzed. The results indicated that the yaw angles of fish near the center of the cage exhibited high dispersion， whereas those near the cage edges were more concentrated， demonstrating a pattern of reciprocal swimming along the cage perimeter. The pitch angles followed a normal distribution， with a mean of 0.48° and a standard deviation of 13.38° （P>0.05）. The average swimming speed was highest during feeding periods （0.321 m/s）， followed by daytime （0.242 m/s）， and lowest at night （0.123 m/s）. The target strength relative to the horizontal incident beam showed periodic variations with the yaw angle； a cosine model provided the best fit （R²=0.50）， described by the equation TS = -43.41-6.99cos（2θ）. This study provides a scientific basis for the effective monitoring of behavioral indicators， facilitating the digital and intelligent monitoring of cage aquaculture and the precise acoustic assessment of stock biomass.

Abstract:

In aquaculture， target detection serves as a fundamental basis for monitoring the behavior and evaluating the growth status of cultured organisms. However， the complexity of the underwater environment often results in degraded image quality， and the tendency of organisms to aggregate further complicates detection， leading to relatively low accuracy in target identification. To address these issues， this paper proposes an improved aquaculture target detection algorithm based on YOLO11n. A lightweight network StarNet is introduced into the backbone network to reduce the number of model parameters and computation load. Mixed Aggregation Network （MANet） is adopted in the neck network to perform multi-scale fusion on aquaculture targets and mitigate the detection deviation caused by blurred images. A Separated and Enhancement Attention Module （SEAM） is incorporated into the detection head to improve the model's detection accuracy in scenarios with biological aggregation and complex backgrounds. The Wise-MPDIoU （Wise Modified Penalized Distance Intersection over Union） loss function is adopted to replace the original one， so as to enhance the robustness of aquaculture organism detection. Experimental results demonstrate that， on the Underwater Target Detection and Classification 2020 （UTDAC2020） and Brackish datasets， the parameter count of the improved YOLO11n model was reduced by 16%. Meanwhile， the precision was improved by 1.1% and 0.2% respectively， the recall was increased by 2.8% and 0.4% respectively， and the mean average precision was elevated by 2.5% and 0.5% respectively. This model achieves high detection accuracy while ensuring lightweight design， and has been successfully deployed on hardware devices equipped with entry-level consumer-grade graphics cards， thus providing a reliable solution for aquaculture target detection tasks.

Abstract:

As a crucial means for implementing intelligent fishery supervision， electronic monitoring （EM） relies heavily on image recognition technology. A key challenge lies in deploying high-performance yet lightweight models in edge computing scenarios. This study introduces the Vision Mamba （ViM） model from the field of deep learning. By utilizing the selective state space model （SSM） to construct a bidirectional encoder， the model achieves global modeling of long-range dependencies in images while maintaining linear computational complexity. Based on the Nature Conservancy （TNC） Fisheries Monitoring dataset， a systematic performance comparison was conducted against mainstream models such as ResNet， EfficientNet， and DeiT. The results demonstrate that the ViM model exhibits outstanding performance in both efficiency and accuracy. Among lightweight models， ViM-Tiny achieved a 1.12% increase in accuracy and a 2.19% improvement in F1 score while reducing parameters by 44.28% compared to the ResNet-18 baseline. Among medium-sized models， ViM-Small， with 44.65% fewer parameters than the ResNet-101 baseline， achieved comparable accuracy （0.960 3） and F1 score （0.964 5）. The study indicates that the ViM model maintains strong fishery species classification capability while significantly reducing model complexity， achieving an excellent balance between lightweight design and high accuracy. This research provides a novel technical pathway for constructing efficient and intelligent fishery supervision systems.

Abstract:

To accurately characterize the effects of dissolved oxygen （DO） concentration on koi behaviors， two novel quantitative indicators， Planar Motion Direction Entropy （PMDE） and Group Aggregation Index （GAI）， based on object tracking are proposed. First， a modified H-SORT object tracker was constructed by introducing the HOG contour feature into the original SORT algorithm. Then， the matching mechanism was optimized via a weighted fusion strategy to enhance the stability of real-time koi tracking. Finally， the behavioral states of koi under different DO concentrations were compared and analyzed based on the two quantitative indicators （PMDE and GAI）. The results showed that the H-SORT algorithm increased the 10-second tracking retention rate of koi by 27.8%， which effectively improved the recognition accuracy in cases of fish school occlusion. The Spearman correlation coefficient between PMDE and DO concentration reached -0.966 9， an increase of 0.240 7 compared with the average swimming speed （Avg_Speed）， indicating that PMDE can effectively capture the behavioral characteristics of koi under different DO concentrations and reflect the impacts of dynamic DO changes and fluctuations on the swimming behaviors of koi. GAI exhibited a strong negative correlation with DO concentration with a coefficient of -0.924 0， an increase of 0.202 8 compared with the average distance （AD）， and the data fluctuation of GAI was within ±0.05. The improved tracking algorithm and the proposed quantitative indicators in this study can effectively quantify the effects of DO concentration on fish school behaviors， which provides a theoretical and technical basis for water quality monitoring and early warning based on fish school behaviors in aquaculture.

Abstract:

With the rapid advancement of computational fluid dynamics （CFD） technology， the Reynolds-Averaged Navier-Stokes （RANS） model has gained widespread application in scientific research and engineering as a vital tool for turbulence simulation. For the large-scale， complex flow field data generated by RANS models， visualization techniques have become a key method for understanding and analyzing flow field characteristics due to their intuitive and visual nature. This paper systematically reviews the current state of information visualization in CFD flow field visualization， focusing on the analysis of various practical visualization methods. Traditional visualization techniques encompass scalar and vector field visualization， volumetric rendering， multiscale representation， as well as interactive and immersive visualization. Modern web technologies like WebGL offer new avenues for online visualization and cross-platform presentation of flow field data. Through analyzing and summarizing these visualization methods， this paper identifies current challenges in CFD flow visualization， including large-scale data processing， multivariable coupling， and vortex feature recognition. Discussions address these issues through CFD-driven machine learning， edge computing， and real-time interaction approaches. The paper concludes with a prospective outlook on CFD flow visualization development， providing theoretical insights for future visualization research.

Abstract:

In order to explore the application and development of artificial intelligence （AI） in the fisheries sector， this study systematically reviews the innovative application scenarios， challenges， and future directions of AI in key segments of the field. Existing research indicates that the core AI technologies applied in fisheries primarily include computer vision， machine learning， and deep learning. Their applications cover critical areas such as breeding， aquaculture， disease prevention and control， processing and quality safety， resource monitoring. Key research focuses include： utilizing AI technologies for fisheries breeding； achieving intelligent feeding in the aquaculture process， real-time analysis of water quality factor data， and assessment of the aquaculture environment； rapid diagnosis of common aquatic diseases， automatic identification of abnormalities in farmed organisms， and issuing warnings； precise automated sorting and grading of aquatic products， guiding robots in accurate slaughtering and segmentation； and real-time dynamic monitoring of fishery resources. In response to current limitations in AI applications such as scarcity of high-quality data， insufficient model generalization capability， and a shortage of interdisciplinary talent， the study suggests that future efforts should focus on the following tasks： （1） Integrating AI with technologies such as gene editing and genomic selection to promote the breeding of new varieties with disease resistance， high yield， and superior quality； （2） Combining AI with Internet of Things technologies to develop practical multi-factor aquaculture prediction methods and models， enabling intelligent and precise feeding； （3） Experimenting with establishing various BP neural network models or adopting large vision models for disease prediction； （4） Integrating AI， robotics， Internet of Things， and other technologies to build an intelligent aquatic product processing system， and utilizing blockchain technology to achieve full-process traceability； （5） Emphasizing the establishment of standard databases， incorporating distinct morphological characteristics of species into learning models to improve the accuracy of identification. With the continuous advancement of AI technology， its application scenarios in the fisheries sector will continue to expand. This not only aligns with China’s requirements for resource-saving and environmentally sustainable development in fisheries， but also helps promote the digitalization， precision， and intelligent development of China’s fisheries industry.

Abstract:

To improve the prediction accuracy and robustness of ammonia nitrogen concentration in recirculating aquaculture systems， this study introduces a memory-augmented compensation LSTM model （MA-LSTM-C）. The model uses bidirectional LSTM and Conv1D parallel structure to extract long and short sequence features at the same time， and embedding a memory matrix to fuse historical information and current input through multi-head attention mechanism. An ammonia nitrogen concentration interval attention mechanism with a learnable mask was introduced to strengthen the feature expression in the key concentration interval of ammonia nitrogen. Finally， through the dynamic error compensation layer， the prediction results were adaptively corrected by combining the instantaneous error， the average error and the trend error. To accommodate varying sampling frequencies from different sensors， PCHIP interpolation based on causal constraints is utilized for interpolated sampling of existing samples. Experimental results show that compared with the standard LSTM model， the MSE， RMSE， MAE and MAPE of MA-LSTM-C are decreased by 76.19%， 51.09%， 58.47% and 55.42%， respectively， and the R² is increased by 12.35%. Ablation experiments further reveal that incorporating physical mechanisms enhances each evaluation metric by approximately an average of 40%. This model demonstrates superior capability in capturing sudden fluctuations in ammonia nitrogen concentration and offers an effective approach for predicting ammonia nitrogen levels in recirculating aquaculture systems.

Abstract:

To investigate the characteristics of CO₂， CH₄， and N₂O fluxes and their driving factors in typical grass carp （Ctenopharyngodon idella） ponds， in-situ monitoring of gas fluxes was conducted using the static chamber-gas chromatography method in ponds containing 1-year-old （GC1）， 2-year-old （GC2）， and 3-year-old （GC3） grass carp. The emission intensity and its relationship with environmental factors were analyzed. Results showed that the order of greenhouse gas fluxes was GC3 > GC2 > GC1. The mean fluxes of CO₂， CH₄， and N₂O in GC3 ponds were 42.30 mg/（m²·h）， 13.31 mg/（m²·h）， and 567.25 μg/（m²·h）， respectively. All ponds acted as sources of CH₄. Redundancy analysis indicated that gas fluxes in GC1 were primarily influenced by NO₃^--N and sediment temperature， water temperature was the dominant factor in GC2， and sediment temperature and dissolved oxygen controlled the fluxes in GC3. Analysis of direct emission factors revealed that the CO₂， CH₄， and N₂O emission factors in GC3 ponds were 132.98， 41.91， and 1.77 g/kg， respectively， all significantly higher than those in GC1 and GC2 （P<0.05）. The global warming potential per unit area followed the order GC3 （53.48 t CO₂-eq/hm²）>GC2 （31.84 t CO₂-eq/hm²）>GC1 （6.54 t CO₂-eq/hm²）， with significant differences （P<0.05）. The GHG emission intensity per unit fish yield was 0.26， 1.05， and 1.95 t CO₂-eq/t for GC1， GC2， and GC3， respectively， showing significant differences （P<0.05）. This study demonstrates that culturing size significantly affects the GHG emission intensity in grass carp ponds， with elevated temperatures and larger fish size exacerbating carbon emissions. It is suggested that regulating the aquaculture structure and key environmental factors can promote the achievement of carbon neutrality in fisheries.

Abstract:

To address the challenges of single energy structure， high carbon emissions， and insufficient power supply stability in remote and deep-sea aquaculture areas， this study proposes a microgrid optimal scheduling method based on an Improved Multi-Objective Mantis Search Algorithm （IMOMSA）， aiming to provide a green and efficient energy supply solution for aquaculture. First， a multi-objective optimization scheduling model incorporating wind-solar hybrid complementary energy is established by comprehensively considering both economic efficiency and environmental sustainability. Second， to overcome the limitations of the original MOMSA algorithm， such as insufficient solution diversity and premature convergence， Bernoulli chaotic mapping is employed to enhance the diversity of the initial population， an artificial bee colony search strategy is introduced to strengthen local exploitation， and a lens imaging-based and opposition-based learning mechanism is integrated to improve global search performance. Finally， simulation and validation are conducted using a real-world aquaculture base in Chongming District， Shanghai.The results show that， compared to the original MOMSA， the IMOMSA reduces the GD and Spacing metrics by 78.45% and 10.71%， respectively， while increasing the HV metric by 100.00%. Meanwhile， the proposed method reduces economic costs and pollutant emissions by 1.76% and 22.62%， respectively. The study demonstrates that the proposed IMOMSA effectively enhances the comprehensive performance of microgrid scheduling solutions， achieving a synergistic optimization of economic and environmental objectives. This work provides theoretical and technical support for the efficient operation of multi-energy complementary aquaculture microgrids and holds significant practical implications for promoting the green transformation， upgrading， and sustainable development of the aquaculture industry.