基于LSTM-Kalman模型的蛋鸡产蛋率预测方法

doi:10.3969/j.issn.1004-1524.2021.09.17

浙江农业学报 ›› 2021, Vol. 33 ›› Issue (9): 1730-1739.DOI: 10.3969/j.issn.1004-1524.2021.09.17

基于LSTM-Kalman模型的蛋鸡产蛋率预测方法

吉训生¹(), 姜晓卫¹^,^*(), 夏圣奎²

1.江南大学物联网工程学院,江苏无锡 214122
2.南通天成现代农业科技有限公司,江苏南通 226600

收稿日期:2020-11-19 出版日期:2021-09-25 发布日期:2021-10-09
通讯作者: 姜晓卫
作者简介:* 姜晓卫,E-mail: 6191905023@stu.jiangnan.edu.cn
吉训生(1969—),男,江苏海安人,博士,教授,研究方向为工业信号感知与处理、物联网技术及其应用。E-mail: jixunsheng@163.com
基金资助:
江苏省重点研发计划(现代农业)(BE2018334)

Research on prediction of laying rate by hens based on LSTM-Kalman model

JI Xunsheng¹(), JIANG Xiaowei¹^,^*(), XIA Shengkui²

1. School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China
2. Nantong Tiancheng Modern Agricultural Technology Co., Ltd., Nantong 226600, China

Received:2020-11-19 Online:2021-09-25 Published:2021-10-09
Contact: JIANG Xiaowei

摘要/Abstract

摘要：

产蛋率是评价蛋鸡产蛋性能的重要指标之一,因其具有时序性和非线性等特点,且其影响变量众多、存在复杂的耦合关系,难以实现精准预测。由于传统神经网络预测过程的非记忆性难以处理时序性问题,该文章提出蛋鸡产蛋率的LSTM-Kalman预测方法,使用主成分分析提取影响蛋鸡产蛋率的关键变量,通过LSTM神经网络预测蛋鸡产蛋率,采用Kalman滤波对LSTM预测的结果进行动态调整,作为最终预测结果。数据分析结果表明:LSTM-Kalman模型预测产蛋率的平均绝对误差、均方误差和皮尔逊相关系数分别为0.312 8、0.435 3和0.975 2,明显优于传统的BP神经网络、极限学习机等预测方法;通过2栋鸡舍生产数据的交叉测试验证,模型的预测准确率达到97.14%和98.71%,表明模型具有较强的泛化能力,能够满足蛋鸡产蛋率预测的实际需要,可以为蛋鸡养殖环境数据的精准调控提供参考。

关键词: 蛋鸡产蛋率, Kalman滤波, LSTM神经网络, 预测准确率

Abstract:

Laying rate is one of the important indexes to evaluate laying performance of hens, it has the characteristics of time-varying, nonlinearity, and complex coupling. So it is difficult to predict the laying rate accurately. The traditional neural network prediction model does not have memory function and cannot be used on the time sequence prediction. So the LSTM-Kalman prediction model is proposed. Firstly, principal component analysis was used to extract the key influencing variables of laying rate of hens. Then LSTM neural network was used as a static prediction model to predict the laying rate of hens. Kalman filter was used to dynamically adjust the result of LSTM prediction to obtain the final prediction results. The data analysis showed that the model’s average absolute error, mean square error and Pearson correlation coefficient were 0.312 8, 0.435 3 and 0.975 2, respectively, which was significantly better than traditional prediction methods, including BP neural network and extreme learning machine. The mutual testing and verification, based on the production data of two hen coops, showed the prediction accuracy of the model were 97.14% and 98.71%, respectively. The model had strong generalization ability and could meet the actual needs of layer production rate prediction. This paper provided a reference for precise control of layer breeding environment data.

Key words: laying rate of hens, Kalman filter, LSTM neural network, prediction accuracy

中图分类号:

S831.4

吉训生, 姜晓卫, 夏圣奎. 基于LSTM-Kalman模型的蛋鸡产蛋率预测方法[J]. 浙江农业学报, 2021, 33(9): 1730-1739.

JI Xunsheng, JIANG Xiaowei, XIA Shengkui. Research on prediction of laying rate by hens based on LSTM-Kalman model[J]. Acta Agriculturae Zhejiangensis, 2021, 33(9): 1730-1739.

图/表 14

参考文献 25

[1]	郑可锋, 祝利莉, 胡为群, 等. 数字农业技术研究进展[J]. 浙江农业学报, 2005, 17(3):170-176.
	ZHENG K F, ZHU L L, HU W Q, et al. Introduction on technology for digital agriculture[J]. Acta Agriculturae Zhejiangensis, 2005, 17(3):170-176.(in Chinese with English abstract)
[2]	朱钰, 郑屹然, 尹默. 统计学意义下的多重共线性检验方法[J]. 统计与决策, 2020, 36(7):34-36.
	ZHU Y, ZHENG Y R, YIN M. Multicollinearity test under statistical significance[J]. Statistics & Decision, 2020, 36(7):34-36.(in Chinese with English abstract)
[3]	潘迪夫, 刘辉, 李燕飞. 基于时间序列分析和卡尔曼滤波算法的风电场风速预测优化模型[J]. 电网技术, 2008, 32(7):82-86.
	PAN D F, LIU H, LI Y F. A wind speed forecasting optimization model for wind farms based on time series analysis and Kalman filter algorithm[J]. Power System Technology, 2008, 32(7):82-86.(in Chinese with English abstract)
[4]	刘先旺. 基于改进BP神经网络的鸡舍环境与产蛋性能关系模型研究[D]. 合肥: 安徽农业大学, 2018.
	LIU X W. Research on the model of the relationship between the environmental factors and the laying performance of layer house based on improved BP neural network[D]. Hefei: Anhui Agricultural University, 2018. (in Chinese with English abstract)
[5]	李飞, 蒋敏兰. 基于极限学习机的蛋鸡产蛋性能预测[J]. 中国家禽, 2019, 41(2):62-64.
	LI F, JIANG M L. Prediction of laying performance of laying hens based on extreme learning machine[J]. China Poultry, 2019, 41(2):62-64.(in Chinese)
[6]	HONG W C, DONG Y C, CHEN L Y, et al. Seasonal support vector regression with chaotic genetic algorithm in electric load forecasting[C]. Sixth International Conference on Genetic and Evolutionary Computing, IEEE, 2012:124-127.
[7]	LV Y, DUAN Y J, KANG W W, et al. Traffic flow prediction with big data: a deep learning approach[J]. IEEE Transactions on Intelligent Transportation Systems, 2015, 16(2):865-873.
[8]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8):1735-1780. DOI URL
[9]	WU T, LIU C C, HE C. Prediction of egional temperature change trend based on LSTM algorithm[J]. IEEE Information Technology Networking Electronic and Automation Control Conference, 2020, 34(4):62-66.
[10]	HUANG H H, WANG T Y, LIU J, et al. Predicting urban rail traffic passenger flow based on LSTM[J]. IEEE Information Technology Networking Electronic and Automation Control Conference, 2019, 24(3):616-620.
[11]	陈卓, 孙龙祥. 基于深度学习LSTM网络的短期电力负荷预测方法[J]. 电子技术, 2018, 47(1):39-41.
	CHEN Z, SUN L X. Short-term electrical load forecasting based on deep learning LSTM networks[J]. Electronic Technology, 2018, 47(1):39-41.(in Chinese with English abstract)
[12]	王鑫, 吴际, 刘超, 等. 基于LSTM循环神经网络的故障时间序列预测[J]. 北京航空航天大学学报, 2018, 44(4):772-784.
	WANG X, WU J, LIU C, et al. Exploring LSTM based recurrent neural network for failure time series prediction[J]. Journal of Beijing University of Aeronautics and Astronautics, 2018, 44(4):772-784.(in Chinese with English abstract)
[13]	于家斌, 尚方方, 王小艺, 等. 基于遗传算法改进的一阶滞后滤波和长短期记忆网络的蓝藻水华预测方法[J]. 计算机应用, 2018, 38(7):2119-2123.
	YU J B, SHANG F F, WANG X Y, et al. Cyanobacterial bloom forecast method based on genetic algorithm-first order lag filter and long short-term memory network[J]. Journal of Computer Applications, 2018, 38(7):2119-2123.(in Chinese with English abstract)
[14]	KALMAN R E, BUCY R S. New results in linear filtering and prediction theory[J]. Journal of Basic Engineering, 1961, 83(1):95-108. DOI URL
[15]	BUCY R S, RENNE K D. Digital synjournal of nonlinear filter[J]. IEEE Transactions on Automatic Control, 2000, 45(3):287-289.
[16]	SUNAHARA Y. An approximate method of state estimation for nonlinear dynamical systems[J]. Journal of Basic Engineering, 1970, 92(2):385-393. DOI URL
[17]	JULIER S, UHLMANN J, DURRANT-WHYTE H F. A new method for the nonlinear transformation of means and covariances in filters and estimators[J]. IEEE Transactions on Automatic Control, 2000, 45(3):477-482. DOI URL
[18]	SONG X J, HUANG J J, SONG D W. Air quality prediction based on LSTM-Kalman model[C]// 2019 IEEE 8th Joint International Information Technology and Artificial Intelligence Conference (ITAIC). May 24-26, 2019, Chongqing. IEEE, 2019: 695-699.
[19]	王钰, 赵晓艳, 杨杏丽, 等. 基于K折交叉验证Beta分布的AUC度量的置信区间[J]. 系统科学与数学, 2020, 40(9):1564-1577.
	WANG Y, ZHAO X Y, YANG X L, et al. Confidence interval of AUC measure based on K-fold cross-validated beta distribution[J]. Journal of Systems Science and Mathematical Sciences, 2020, 40(9):1564-1577.(in Chinese with English abstract)
[20]	HIROSE Y, YAMASHITA K, HIJIYA S. Back-propagation algorithm which varies the number of hidden units[J]. Neural Networks, 1991, 4(1):61-66. DOI URL
[21]	陈啸, 王红英, 孔丹丹, 等. 基于粒子群参数优化和BP神经网络的颗粒饲料质量预测模型[J]. 农业工程学报, 2016, 32(14):306-314.
	CHEN X, WANG H Y, KONG D D, et al. Quality prediction model of pellet feed basing on BP network using PSO parameters optimization method[J]. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(14):306-314.(in Chinese with English abstract)
[22]	王伟, 张玉, 吴应淼, 等. BP神经网络模型在香菇中SO₂含量分析中的应用[J]. 浙江农业学报, 2011, 23(5):1012-1016.
	WANG W, ZHANG Y, WU Y M, et al. Application of sulfur dioxide content detection in mushrooms based on the back-propagation neural networks[J]. Acta Agriculturae Zhejiangensis, 2011, 23(5):1012-1016.(in Chinese with English abstract)
[23]	HUANG G B, ZHU Q Y, SIEW C K. Extreme learning machine: theory and applications[J]. Neurocomputing, 2006, 70(1/2/3):489-501. DOI URL
[24]	HUANG G B, ZHU Q Y, SIEW C K. Extreme learning machine: a new learning scheme of feedforward neural networks[J]. IEEE International Joint Conference on Neural Networks, 2004, 2:985-990.
[25]	HUANG G B, ZHOU H M, DING X J, et al. Extreme learning machine for regression and multiclass classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2012, 42(2):513-529.

鸡舍 Hen coop	初始数量 Initial quantity	结束数量 Ending quantity	死淘率 Death rate/%	平均温度 Average temperature/℃	平均湿度 Average humidity/%	只均采食量 Average food/g	只均饮水量 Average water/mL
5号No.5	58 570	55 298	5.75	23.34~29.85	62.69~92.12	79.69~120.85	126.48~365.61
6号No.6	59 320	57 059	3.89	23.02~29.66	67.82~90.99	77.27~118.59	110.92~490.66

鸡舍 Hen coop	初始数量 Initial quantity	结束数量 Ending quantity	死淘率 Death rate/%	平均温度 Average temperature/℃	平均湿度 Average humidity/%	只均采食量 Average food/g	只均饮水量 Average water/mL
5号No.5	58 570	55 298	5.75	23.34~29.85	62.69~92.12	79.69~120.85	126.48~365.61
6号No.6	59 320	57 059	3.89	23.02~29.66	67.82~90.99	77.27~118.59	110.92~490.66

成分 Component	方差贡献率 Variance contribution rate	累计方差贡献率 Cumulative variance contribution rate
1	65.855 9	65.855 9
2	14.086 9	79.942 8
3 4 5	11.154 9 6.432 8 2.469 4	91.097 8 97.530 6 100

成分 Component	方差贡献率 Variance contribution rate	累计方差贡献率 Cumulative variance contribution rate
1	65.855 9	65.855 9
2	14.086 9	79.942 8
3 4 5	11.154 9 6.432 8 2.469 4	91.097 8 97.530 6 100

成分 Component	主成分1 Principal component 1	主成分2 Principal component 2	主成分3 Principal component 3
只均采食量Average feed	0.118 698	-0.237 459	0.487 089
温差Temperature difference	0.166 156	0.522 511	0.167 772
平均湿度Average humidity	0.302 565	-0.071 124	0.143 339
平均温度Average temperature	0.238 405	-0.119 832	-0.111 712
只均饮水量Average water	-0.174 172	0.049 097	0.090 088

基于LSTM-Kalman模型的蛋鸡产蛋率预测方法

Research on prediction of laying rate by hens based on LSTM-Kalman model

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 25

相关文章 0

编辑推荐

Metrics

本文评价

试验 Test	均方误差 MSE	平均绝对误差 MAE	皮尔逊相关系数 r	准确率 Accuracy%
1	0.561 8	0.583 7	0.522 4	60.47
2	0.364 3	0.464 3	0.807 9	75.38
3	0.119 4	0.283 5	0.953 4	94.21
4	0.168 8	0.329 9	0.905 6	89.17
5	0.233 7	0.396 1	0.898 7	85.39

模型 Model	均方误差MSE	平均绝对误差MAE	皮尔逊相关系数r	准确率Accuracy/%
BP神经网络	1.284 8	0.860 8	0.776 7	85.47
极限学习机	1.093 4	0.802 2	0.792 6	87.34
LSTM	0.827 1	0.698 9	0.893 4	92.68
LSTM-Kalman	0.312 8	0.435 3	0.975 2	98.71

模型Model	均方误差MSE	平均绝对误差MAE	皮尔逊相关系数r	准确率Accuracy/%
BP神经网络	1.557 1	1.054 3	0.701 6	81.54
极限学习机	1.135 9	0.875 4	0.802 7	88.16
LSTM	0.795 4	0.702 9	0.900 7	91.76
LSTM-Kalman	0.235 7	0.457 2	0.974 3	97.14

试验 Test	节点数 Nodes number	学习率 Learning rate	批尺寸 Batch- size	时间步长 Time step/d
1	50	0.010	10	1
2	50	0.010	10	2
3	50	0.010	10	3
4	50	0.010	50	3
5	50	0.001	10	3

试验 Test	节点数 Nodes number	学习率 Learning rate	批尺寸 Batch- size	时间步长 Time step/d
1	50	0.010	10	1
2	50	0.010	10	2
3	50	0.010	10	3
4	50	0.010	50	3
5	50	0.001	10	3