基于饮水行为参数的育肥猪个体饮水量量化方法  

Quantification method for the individual water intake of fattening pigs based on drinking behavior parameters

作　　者：吴锦辉 刘健 周振宇 庄晏榕 冀横溢 滕光辉 WU Jinhui;LIU Jian;ZHOU Zhenyu;ZHUANG Yanrong;JI Hengyi;TENG Guanghui(College of Water Resources and Civil Engineering,China Agricultural University,Beijing 100083,China;Key Laboratory of Facility Agriculture Engineering of Ministry of Agriculture and Rural Affairs,Beijing 100083,China)

机构地区：[1]中国农业大学水利与土木工程学院,北京100083 [2]农业农村部设施农业工程重点实验室,北京100083

出　　处：《中国农业大学学报》2024年第11期179-187,共9页Journal of China Agricultural University

基　　金：重庆市技术创新与应用发展专项重点项目(cstc2021jscxdxwtBX0006)。

摘　　要：为解决传统猪只个体饮水量预测方法精度不高和特征输入单一的问题,以包括饮水次数、饮水时长、水温、日龄及体重在内的猪只饮水行为参数为特征输入,分别构建基于线性回归、K最近邻回归、随机森林回归、支持向量回归、极限梯度提升及BP神经网络的单因素与多因素回归预测模型,并以经过标准分数预处理的育肥期猪只饮水数据为样本,对个体饮水量量化方法进行研究。结果表明:1)相比机器学习回归模型预测结果,线性回归模型能更好地揭示饮水时长与饮水量之间的线性相关性,尽管其模拟精度有限;2)相较于单一特征输入的回归模型,多特征输入的模型纳入更多影响因素,在预测饮水量上更为精准,且机器学习回归模型能更有效地捕捉饮水行为参数与饮水量之间的非线性关系;3)BP神经网络在模型适应度和模拟精度方面表现出明显优势,更适用于育肥猪个体饮水量量化。在智能化养殖环境中,根据饮水量建立的完整反馈数据可用于创建每只猪当天的评价指标数据库,为进一步分析猪只健康状况提供关键数据。To address the issues of low accuracy and limited feature inputs in traditional methods of predicting individual water intake of pigs,swine drinking behavior parameters including the number of drinking actions,drinking duration,water temperature,age,and weight were taken as input features.This study constructed both single-factor and multi-factor regression prediction models based on linear regression,K-nearest neighbors regression,random forest regression,support vector regression,extreme gradient boosting,and back propagation neural networks.The fattening pig drinking water data are preprocessed with Z-Score standardization to explore methods of quantifying individual water intake.The results show that:1)Compared to other machine learning regression model predictions,the linear regression model better reveals the linear correlation between drinking duration and water intake,although its simulation accuracy is limited.2)Compared to models with a single feature input,models with multiple feature inputs include more influencing factors and are more accurate in predicting water intake,and the machine learning regression models are more effective in capturing the nonlinear relationships between drinking behavior parameters and water intake.3)The back propagation neural network shows significant advantages in model fit and simulation accuracy,making it more suitable for quantifying individual water intake in fattening pigs.In an intelligent farming environment,complete feedback data based on water intake can be used to create a daily evaluation index database for each pig,providing key data for further analysis of pig health.

关 键 词：机器学习 育肥猪 行为参数 网格搜索 个体饮水量量化 

分 类 号：S24[农业科学—农业电气化与自动化] TP399[农业科学—农业工程]