基于LSSA-XGBOOST改进算法的高体鰤鱼类体质量预测模型

doi:10.14012/j.cnki.fjsc.2023.05.002

摘要/Abstract

摘要：

为构建利用体质量判断的精准投喂模型,需实时获取鱼群体质量状态,基于LSSA-XGBOOST算法,通过对工船养殖实测的高体鰤(Seriola dumerili)体长、体宽和体质量数据进行分析,构建以体长、体宽两项体态特征数据为输入、体质量数据为输出的高体鰤体质量预测模型。结果显示,与常规数学模型拟合相比,LSSA-XGBOOST模型拟合的相关性系数R²提高约10%;与传统BP神经网络和粒子群优化BP相比,LSSA-XGBOOST模型误差平方和R²提升约3%,这为构建基于体质量判断的高体鰤精准投喂模型提供了理论依据。

关键词: LSSA-XGBOOST, 高体鰤, 体长, 体质量, 关系

Abstract:

In order to build an accurate feeding model using mass judgment and obtain the body mass state of S.dumerili in real time,this study built a body mass prediction model based on LSSA-XGBOOST algorithm.Firstly,the data of body length,body width and body mass measured by the culture experiment ship were detected by extreme studentized deviate (ESD) method,and the abnormal points were removed.Secondly,the chaotic random number generator was used to complete the initial population optimization of the SSA algorithm to improve its searching ability.The optimized SSA algorithm was used to optimize three parameters of the optimal tree depth,the optimal learning rate and the optimal number of iterations of the XGBOOST model.Finally,the LSSA-XGBOOST model with body length and body width as the input and body mass data as the output was constructed.The experimental results showed that,compared with the conventional mathematical model fitting,the LSSA-XGBOOST model fitting correlation coefficient R² increased by about 10%.Compared with the traditional BP neural network and PSO particle swarm optimization BP,the error square and R² were improved by about 3%,and the mean absolute error (MAE),mean square error (MSE) as well as the root mean square error (RMSE) were significantly reduced.It could be seen that LSSA-XGBOOST model was more accurate for predicting the body mass of small samples of S.dumerili,and the construction of LSSA-XGBOOST model was greatly significant for users to grasp the growth state of S.dumerili and build accurate feeding model for mass judgment.

Key words: LSSA-XGBOOST, Seriola dumerili, body length, body mass, relationship

中图分类号:

S917.4

俞国燕, 左仁意, 严俊, 罗樱桐, 朱琪珩. 基于LSSA-XGBOOST改进算法的高体鰤鱼类体质量预测模型[J]. 渔业研究, 2023, 45(5): 427-437.

YU Guoyan, ZUO Renyi, YAN Jun, LUO Yingtong, ZHU Qiheng. Body mass of Seriola dumerili prediction model based on LSSA-XGBOOST improved algorithm[J]. Journal of Fisheries Research, 2023, 45(5): 427-437.

图/表 17

图1 实验船整体图

Fig.1 Overall diagram of test vessel

图2 初始随机分布

Fig.2 Initial random distribution

图3 混沌随机分布

Fig.3 Chaotic random distribution

图4 LSSA/SSA初始种群分布直方图

Fig.4 LSSA/SSA initial population distribution histogram

图5 XGBOOST树评分方式

Fig.5 XGBOOST tree scoring method

图6 SSA-XGBOOST 算法流程图

Fig.6 SSA-XGBOOST algorithm flow chart

图7 体长、体质量关系散点图(ESD剔除) 注:红色点为剔除数据。图8同此。

Fig.7 Scatter diagram of body length and body mass (ESD-excluded) Notes:The red dot represented excluded data.The same as in figure 8.

图8 体宽、体质量关系散点图(ESD剔除)

Fig.8 Scatter diagram of body width and body mass(ESD-excluded)

图9 Gauss 函数拟合曲线

Fig.9 Gauss function fitting curve

图10 Logistic函数拟合曲线

Fig.10 Fitting curve of Logistic function

图11 幂函数拟合曲线

Fig.11 Power function fitting curve

图12 Von Bertalanffy 方程

Fig.12 Von Bertalanffy equation

图13 SSA群体迭代适应度变化曲线

Fig.13 Change curve of SSA group iterative fitness

图14 预测值与真实值对比图注:RMSE为均方根误差;MSE为均方误差;RPD为相对百分比差异。

Fig.14 Comparison between predicted value and real value Notes:RMSE is root mean square error;MSE is mean square error;RPD is the relative percentage difference.

图15 神经网络算法效果对比图

Fig.15 Comparison of neural network algorithm effects

表1 7种拟合模型拟合度R2对比

Tab.1 Comparison of fitting degree R2 of 7 fitting models

模型 Model	自变量 Independent variable	因变量 Dependent variable	相关性系数 R²
Gauss	体长	体质量	0.822 5
Gauss	体宽	体质量	0.708 5
Logistic	体长	体质量	0.772 5
Logistic	体宽	体质量	0.708 4
幂指数 Power-exponent	体长	体质量	0.771 9
幂指数 Power-exponent	体宽	体质量	0.709 6
W=aL^b	体长	体质量	0.771 0
BP	体长+体宽	体质量	0.877 5
PSO-BP	体长+体宽	体质量	0.910 5
LSSA-XGBOOST	体长+体宽	体质量	0.947 9

表2 LSSA-XGBOOST模型与神经网络算法各项误差对比

Tab.2 Comparison of errors between LSSA-XGBOOST model and neural network algorithm

模型 Model	平均绝对误差 MAE	均方误差 MSE	均方根误差 RMSE
BP	13.988	300.165 5	17.325 3
PSO-BP	13.716	243.015 2	16.412 4
LSSA-XGBOOST	9.042	203.533 0	14.266 5

参考文献 29

[1]	朱明, 张镇府, 黄凰, 等. 鱼类养殖智能投喂方法研究进展[J]. 农业工程学报, 2022, 38(7):38-47.
[2]	刘晓娟, 沙宗尧, 李大鹏, 等. 基于生物能量学模型的尖吻鲈精准投喂管理辅助决策系统构建[J]. 水生生物学报, 2021, 45(2):237-249.
[3]	沈炜皓, 崔海朋, 徐以军. 基于环境信息和鱼类行为的智能投喂系统研究[J]. 中国新技术新产品, 2021(22):33-35.
[4]	陈澜, 杨信廷, 孙传恒, 等. 基于自适应模糊神经网络的鱼类投喂预测方法研究[J]. 中国农业科技导报, 2020, 22(2):91-100. DOI
[5]	林在凡, 郭敬蓉, 洪晓林, 等. 基于Python的图像处理技术在鱼类尺寸测量中的应用[J]. 福建农机, 2021(2):38-41.
[6]	Mafalda F, Pedro I, Manuel B. Length-weight relationships for eight Chondrichthyes from the north-eastern Atlantic Ocean[J]. The Egyptian Journal of Aquatic Research, 2023, 49(1):87-90. DOI URL
[7]	Dinh Q M, Nguyen T H D, Nguyen-Ngoc L, et al. Temporal variation in length-weight relationship,growth and condition factor of Acentrogobius viridipunctatus in the Mekong Delta,Viet Nam[J]. Regional Studies in Marine Science, 2022, 55:102545. DOI URL
[8]	Shuman L A, Selyukov A G, Nekrasov I S, et al. Data on length-weight and length-length relationships,mean condition factor,and gonadosomatic index of Rutilus rutilus and Perca fluviatilis[J]. Data in Brief, 2022, 42:108067. DOI URL
[9]	Kaka R M, Jung’a J O, Badamana M, et al. Length-weight relationships of wild penaeid shrimps in Malindi-Ungwana Bay:implications to aquaculture development in Kenya[J]. The Egyptian Journal of Aquatic Research, 2019, 45(2):167-173. DOI URL
[10]	Sepa P, Coello D, Herrera M, et al. Length-weight relationship of four deep-sea chondrichthyans (Elasmobranchii & Holocephali) in Ecuadorian oceanic waters[J]. The Egyptian Journal of Aquatic Research, 2022, 48:397-399. DOI URL
[11]	Najmudeen T M, Zacharia P U, Seetha P K, et al. Length-weight relationships of three species of pelagic sharks from southeastern Arabian Sea[J]. Regional Studies in Marine Science, 2019, 29:100647. DOI URL
[12]	陈锋, 刘艳超, 魏聪, 等. 察隅弧唇裂腹鱼仔稚鱼体长体重关系研究[J]. 西藏科技, 2020(10):11-12.
[13]	林雅蓉, 王荣, 高尚武. 胶州湾桡足类的生物学研究——Ⅱ.小拟哲水蚤的个体重量测定及体长-体重关系[J]. 海洋科学, 1987(5):41-45.
[14]	张志伟, 胡伍生, 黄晓明. 回归拟合模型的神经网络方法[J]. 测绘科学, 2010, 35(S1):39-41.
[15]	韩超. 深远海养殖:走向“蓝海”的朝阳产业[J]. 农产品市场, 2021(12):26-28.
[16]	范斌, 古恒光, 杨永健. 卵形鲳鲹深水网箱深远海绿色健康养殖技术研究与应用[J]. 中国科技成果, 2020, 21(19):49-51.
[17]	陈昌生. 高体鰤人工繁殖、育苗及养殖技术的研究[C]// 中国海洋湖沼学会,中国动物学会鱼类学分会.中国动物学会鱼类分会2004年学术研讨会摘要汇编. 重庆: 中国海洋湖沼(动物)学会鱼类学分会, 2004:77.
[18]	陈昌生, 黄佳鸣, 叶加松, 等. 高体鰤人工育苗技术研究[J]. 水产学报, 1998, 22(1):40-44.
[19]	王雅英. 高体鰤鱼胚胎发育过程初步报告[J]. 中国水产, 2001(3):59-60,53.
[20]	廖志强. 高体鰤网箱养殖技术[J]. 中国水产, 2003(12):60-61.
[21]	周梦, 吕志刚, 邸若海, 等. 基于小样本数据的BP神经网络建模[J]. 科学技术与工程, 2022, 22(7):2754-2760.
[22]	曹晨曦, 田友琳, 张昱堃, 等. 基于统计方法的异常点检测在时间序列数据上的应用[J]. 合肥工业大学学报(自然科学版), 2018, 41(9):1284-1288.
[23]	陈伟, 吴布托, 裴喜平. 风电机组异常数据预处理的分类多模型算法[J]. 电力系统及其自动化学报, 2018, 30(4):137-143.
[24]	黄真理, 常剑波. 鱼类体长与体重关系中的分形特征[J]. 水生生物学报, 1999, 23(4):330-336.
[25]	薛建凯. 一种新型的群智能优化技术的研究与应用[D]. 上海: 东华大学, 2020.
[26]	吕鑫, 慕晓冬, 张钧, 等. 混沌麻雀搜索优化算法[J]. 北京航空航天大学学报, 2021, 47(8):1712-1720.
[27]	Chen T Q, Guestrin C. XGBoost:a scalable tree boosting system[EB/OL].[2023-03-09]. http://dl.acm.org/doi/10.1145/2939672.2939785.
[28]	唐亦舜, 徐庆, 刘振鸿, 等. 基于优化非线性自回归神经网络模型的水质预测[J]. 东华大学学报(自然科学版), 2022, 48(3):93-100.
[29]	邓森, 李希建, 徐畀泽. 基于PSO-BP神经网络的煤体瓦斯渗透率预测[J]. 矿业工程研究, 2022, 37(4):35-41.