3种不同方法对肉牛胴体性状预测能力的比较研究

摘要

本研究为了寻求一种对肉牛胴体性状预测准确性较高的方法,运用DPS数据处理系统和SAS软件比较偏最小二乘回归、GM(1,N)灰色系统和BP神经网络3种常用的预测模型对肉牛胴体性状的预测能力.选择肉牛7个宰前生长性状(体高、体长、胸围、腹围、管围、宰前活体质量、平均日增体质量),对2个重要的胴体性状(胴体质量和净肉质量)进行预测.结果表明:偏最小二乘回归方法在肉牛胴体性状预测方面准确性最高;GM(1,N)灰色系统和BP神经网络预测准确度偏低.本研究还将3种预测结果相结合,取其均值,大大提高了预测的准确性.这一研究将为肉牛生产实践提供一定的科学参考.%To search for a method to predict accurately carcass traits in bovine, in this study, DPS and SAS software were used to compare the methods of partial least squares regression, GM(1, N) gray system and BP neural network, in order to observe their accuracy in predicting carcass traits in bovine. Seven preslaughter growth traits including body height, body length, chest circumference, abdominal circumference, cannon bone circumference, live weight and average daily gain were used to predict the carcass weight and meat weight. The results showed that the partial least squares regression gave the highest accuracy, while the average relative errors of GM(1,N) gray system and BP neural network were lower. In this study, the three predicted results were combined and their mean value were calculated as the predictive values, which would greatlyimprove the accuracy of prediction. The results would provide some scientific references to beefproduction.