文献类型：期刊文献

中文题名：船载投料系统饲料颗粒流落点预测

英文题名：Method and Experiment for Predicting Feed Particle Landing Points in Ship-Mounted Feeding System

作者：俞国燕[1,2];王涛[1,2];郭国全[1,2];刘皞春[1,2,3]

机构：[1]广东海洋大学机械工程学院,广东湛江524088;[2]广东省海洋装备及制造工程技术研究中心,广东湛江524088;[3]广东省南海海洋牧场智能装备重点实验室,广东湛江524088

年份：2024

卷号：44

期号：1

起止页码：142

中文期刊名：广东海洋大学学报

外文期刊名：Journal of Guangdong Ocean University

收录：北大核心2023、CSTPCD、、CSCD_E2023_2024、北大核心、CSCD

基金：湛江市现代海洋渔业装备重点实验室项目(2021A05023);广东省研究生教育创新计划(2023JGXM_075)。

语种：中文

中文关键词：网箱养殖;船载式投料系统;落点预测模型;混合网络模型;BP神经网络

外文关键词：cage aquaculture;ship-mounted feeding system;landing point prediction model;hybrid network;BP neural network

中文摘要：【目的】为解决网箱养殖中使用船载投料系统的饲料颗粒流落点控制问题,提出一种用于实时分割饲料颗粒流轨迹并精确预测其落点的方法(MLBP)。【方法】考虑到输料管管内参数及饲料颗粒流出口参数获取难度较大,本研究采用高速相机获取饲料颗粒流轨迹图像,并利用提出的混合网络模型分割饲料颗粒流轨迹,以获取轨迹关键信息;为准确预测饲料颗粒流落点,利用BP神经网络的优势,将轨迹信息及投料口高度作为其输入,实现饲料颗粒流落点的预测。【结果】与相关研究方法对比,结合混合网络模型与BP神经网络的MLBP方法的系统单次运行时间降低95%,同时落点预测准确度达到96%,落点的平均误差范围与平均误差百分比也分别降低32.0%和30.5%。【结论】本研究提出的MLBP方法预测精度及实时性均能满足网箱投饵作业需求,可为相关研究提供参考。

外文摘要：【Objective】In order to address the issue of controlling the dispersion of feed particle landing points in cage aquaculture employing ship-mounted feeding systems,a method termed Real time Feed Particle Trajectory Segmentation and Precise Landing Prediction(MLBP)is proposed.【Method】To address challenges in obtaining parameters inside the feeding tube and at the outlet of feed particle flow,this study employed a high-speed camera to capture images of feed particle trajectories.A proposed hybrid network model was utilized to segment the trajectories and extract key information.For accurate prediction of feed particle landing points,the advantages of a BP neural network were leveraged,with trajectory information and feeding port height serving as inputs to achieve precise predictions.【Result】The MLBP technique,which combined a BP neural network and hybrid network model, reduced single-run execution time by 95% compared to pertinent research techniques. Concurrently, 96% of landing points were predicted correctly, with the average error range and percentage down to 32.0% and 30.5%, respectively.【Conclusion】The MLBP method’s real-time performance and prediction accuracy are sufficient for cage-feeding operations and offer important information for relevant research.