Factory aquaculture faces a difficult situation due to its high running costs, with one of the main contributing factors being the high energy consumption of aquaculture workshops. Accurately predicting the power load of recirculating aquaculture systems (RAS) is critical to optimizing energy use, reducing energy consumption, and promoting the sustainable development of factory aquaculture. Adequate data can improve the accuracy of the prediction model. However, there are often missing and abnormal data in actual data detection. To solve this problem, this study uses a time-series convolutional network–temporal sequence generation adversarial network (TCN-TimeGAN) to synthesize multivariate RAS data and train a long short-term memory (LSTM) network on the original and generated data to predict future electricity loads. The experimental results show that the data generated based on the improved TCN-TimeGAN provide more comprehensive coverage of the original data distribution, with a lower discriminative score (0.2419) and a lower predictive score (0.0668) than the conventional TimeGAN. Using the generated data for prediction, the R² reached 0.86, which represents a 19% improvement over the ARIMA model. Meanwhile, compared to LSTM and GRU without data augmentation, the mean absolute error (MAE) was reduced by 1.24 and 1.58, respectively. The model has good prediction performance and generalization ability, which benefits the RAS energy saving, production planning, and the long term sustainability of factory aquaculture.

This study presents an algorithm for shrimp counting using a small-scale labeling model. The aim is to enhance shrimp farming efficiency, reduce biosecurity risks, and minimize manual counting errors. Experimental evaluations were conducted using Litopenaeus vannamei samples divided into four growth stages. A comparison was made between the classical threshold segmentation method, small-scale labeling density estimation based on FamNet, and our proposed small-scale labeling density estimation based on FamNet-S models. The results showed that the proposed FamNet-S-based small-scale labeling density estimation method achieved a better level of accuracy than the classical FamNet model across different growth stages. In the first stage, it reduced the mean absolute error (MAE) by 8.7% and the mean squared error (MSE) by 9.6%. In the fourth stage, MAE and MSE further decreased by 18.9% and 21.6%, respectively. The research findings demonstrate that the small-scale labeling density estimation approach based on FamNet-S exhibits robust adaptability and accuracy across diverse growth stages, rendering it suitable for scenarios with limited annotated samples. It effectively tackles challenges associated with shrimp overlap and occlusion while mitigating interference from feed and excrement, thus enhancing the precision and efficiency of shrimp counting. This algorithm for small-scale labeling density estimation significantly reduces annotation workloads while facilitating rapid deployment, making it an ideal solution for counting and marking in practical aquaculture environments. The study provides a high-precision yet efficient methodology for shrimp counting and marking tasks, thus reducing labor intensity while improving farming management accuracy, thereby supporting intelligent control in shrimp farming.

To clarify the spreading law of river crab pellet feed in a centrifugal spreading mechanism and provide a physical basis for the path planning of automatic feeding boats, this study took 4.0 mm sinking extruded river crab feed as the research object. A systematic research method combining physical experiments and Discrete Element Method (DEM) simulation was established. Physical experiments were conducted to calibrate the intrinsic parameters (density, Poisson’s ratio, elastic modulus) and contact parameters (friction coefficients and restitution coefficients between feed and 304 stainless steel/ABS plastic, as well as between feed particles) of the pellet feed. On this basis, a DEM simulation model of a vibration blanking-dual disc centrifugal spreading mechanism was constructed using the multi-sphere aggregation method and the Hertz-Mindlin (no-slip) contact model. A Central Composite Design (CCD) response surface experiment was employed to investigate the spreading law, with boat speed (0.5–1.5 m/s) and spreading disc rotation speed (800–1000 rpm) as independent variables, and unilateral spreading width (W), track superposition uniformity (ω), and transverse coefficient of variation ($Cv$) as response indicators to characterize spreading range and particle distribution. The results showed that the spreading disc rotation speed had an extremely significant effect (p < 0.0001) on all three response indicators, while boat speed had no significant effect. The feed exhibited a characteristic double fan-shaped superposition distribution pattern. Through multi-objective optimization, the optimal operational parameters were determined as a boat speed of 1.0 m/s and a spreading disc rotation speed of 879 rpm, yielding a unilateral spreading width of 2.9 m, a track superposition uniformity of 88.31%, and a transverse coefficient of variation of 8.33%. This study establishes a quantitative method for analyzing feed spreading characteristics and clarifies the spreading range and particle distribution law, providing a reliable physical basis for full-coverage path planning of crab pond feeding boats.