应用ARIMA和XGBoost分析并预测我国近海捕捞产量

武宇辰; 徐海龙

doi:10.12284/hyxb2025130

应用ARIMA和XGBoost分析并预测我国近海捕捞产量

doi: 10.12284/hyxb2025130

武宇辰^1,,
徐海龙^{1, 2, 3, ,}

1.
天津农学院水产学院，天津 300384
2.
天津市水产生态及养殖重点实验室，天津 300384
3.
渔业生态量化评估与管理实验室（天津农学院），天津 300384

基金项目: 天津市海洋生物资源养护工作项目（ML2020-HPF059）。

详细信息

作者简介:
武宇辰（1994—），女，天津市人，主要从事海洋渔业资源评估研究。E-mail：yuchenwu0107@163.com

通讯作者:
徐海龙，副教授，主要从事海洋渔业资源评估研究。E-mail：beiji80@163.com

中图分类号: S937.3
计量
- 文章访问数: 138
- HTML全文浏览量: 63
- PDF下载量: 18
- 被引次数: 0
出版历程
- 收稿日期: 2025-08-06
- 修回日期: 2025-09-18
- 网络出版日期: 2025-10-11
- 刊出日期: 2025-11-30

Application of ARIMA and XGBoost to analyze and predict China’s coastal capture production

Wu Yuchen^1
,,
Xu Hailong^{1, 2, 3
, ,}

1.
Department of Fishery Sciences, Tianjin Agricultural University, Tianjin 300384, China
2.
Tianin Key Laboratory of Aqua-Ecology and Aquaculture National Demonstration Center for Experimental Aqua-Ecology, Tianjin 300384, China
3.
Fisheries Ecological Quantitative Assessment and Management Laboratory (Tianin Agriculture University), Tianjin 300384, China

摘要

摘要: 为更全面地了解和掌握我国近海海域渔业资源开发利用现状和发展趋势，本文基于近海海域捕捞量渔获量数据，分别以1980年、1985年、1991年、1997年、2003年、2009年和2016年为时间序列起点，构建单一ARIMA及其与XGBoost组合模型，选用拟合精度最高的模型对我国2024−2028年近海捕捞渔获量进行预测。结果显示，单一ARIMA模型的平均绝对百分比误差（MAPE）和决定系数（R²）分别为0.11%～12.12%和0.4561～0.9794，而组合模型的MAPE和R²分别为0.12%～12.12%和0.7584～0.9933；其中，1980−2021年构建的ARIMA（1, 2, 1）组合模型的拟合效果最佳，其MAPE和R²分别为0.12%和0.9933，该模型预测2024−2028年我国近海海洋捕捞渔获量为9.5×10⁶ t左右，捕捞渔获量整体呈现缓慢上升趋势。研究表明，两种模型的预测精度均随着时间序列数据长度的增加，呈现出先下降后上升的趋势，最长时间序列预测精度和拟合度均最佳；组合模型的预测精度显著高于单一ARIMA模型；2024−2028年渔获量预测值显示其增幅小于0.1%。
- 不同时间序列 /
- 渔获量 /
- ARIMA模型 /
- XGBoost模型 /
- 预测
Abstract: To comprehensively understand and analyze the current status and developmental trends of capture resource utilization in China’s coastal waters, this study constructs individual ARIMA models and their hybrid models with XGBoost, using fishery production data from 1980, 1985, 1991, 1997, 2003, 2009, and 2016 as temporal reference points. The model demonstrating the highest fitting accuracy was selected to forecast the marine catch in China's coastal waters from 2024 to 2028. The results indicate that the individual ARIMA models exhibited Mean Absolute Percentage Errors (MAPE) ranging from 0.11% to 12.12% and coefficients of determination (R²) between 0.4561 and 0.9794, while the hybrid models showed MAPE values of 0.12%−12.12% and R² values of 0.7584−0.9933. Notably, the ARIMA(1, 2, 1) hybrid model constructed with 1980 capture data demonstrated optimal fitting performance, with MAPE and R² values of 0.12% and 0.9933, respectively. This model forecasts a marine catch of 9.5×10⁶ t in China’s coastal waters from 2024 to 2028, indicating a gradual upward trend. The research findings reveal that the predictive accuracy of both model types initially decreases and subsequently increases with the extension of time series data, achieving optimal prediction accuracy and fitting degree with the longest time series. The hybrid model demonstrates significantly superior predictive accuracy compared to the individual ARIMA model. The forecasted catch values for 2024−2028 indicate an increase of less than 0.1%.
- different time series /
- catch volume /
- ARIMA model /
- XGBoost model /
- prediction

HTML全文

图 1 1980−2023年我国近海海域渔获量变化特征

Fig. 1 Characteristics of changes in fish catch in China’s coastal waters from 1980 to 2023

下载: 全尺寸图片幻灯片

图 2 不同捕捞渔获量的时间序列差分后的ACF与PACF图

Fig. 2 ACF and PACF plots of time series differences in different fishing catches

下载: 全尺寸图片幻灯片

图 3 不同时间序列最优模型的模型检验图

Fig. 3 Model verification diagram for optimal models of different time series

下载: 全尺寸图片幻灯片

图 4 1980年ARIMA（1, 2, 1）与XGBoost组合模型预测未来2024−2028年我国近海海域捕捞渔获量

Fig. 4 1980 ARIMA (1, 2, 1) and XGBoost combined model predicting future fishing catches in China’s coastal waters from 2024 to 2028

下载: 全尺寸图片幻灯片

表 1 平均绝对百分比误差评价模型预测准确性水平

Tab. 1 Average absolute percentage error evaluation model prediction accuracy level

模型预测能力	平均绝对百分比误差（MAPE）
极佳	MAPE < 5%
优良	5% ≤ MAPE < 10%
合理	10% ≤ MAPE < 20%
预测不准确	MAPE ≥ 20%

下载: 导出CSV

表 2 近海渔获量1980−2023年结构性断点检验结果

Tab. 2 Results of structural break tests for coastal fish catches from 1980 to 2023

断点个数	0	1	2	3	4	5	6
BIC	335.94	231.78	208.29	190.77	173.45	159.45	159.03
断点位置		24(2003)	23(2002)	10(1989)	10(1989)	10(1989)	6(1985)
			32(2011)	22(2001)	22(2001	18(1997)	12(1991)
				32(2011)	29(2008)	24(2003)	18(1997)
					37(2016)	30(2009)	24(2003)
						37(2016)	30(2009)
							37(2016)

下载: 导出CSV

表 3 不同捕捞渔获量的时间序列ADF检验结果

Tab. 3 ADF test results for different fishing catch time series

指标	1980	1985	1991	1997	2003	2009	2016
p值	0.789	0.77	0.706	0.223	0.108	0.338	0.3818

下载: 导出CSV

表 4 不同捕捞渔获量的时间序列差分后的ADF检验结果

Tab. 4 ADF test results after differencing time series of different fishing catches

指标	1980	1985	1991	1997	2003	2009	2016
一阶差分p值	0.117	0.1778	0.119	0.01	0.01	0.0248	0.0178
二阶差分p值	0.01	0.01	0.01

下载: 导出CSV

表 5 不同时间序列ARIMA（p, d, q)模型不同系数组合结果

Tab. 5 Results of different coefficient combinations of ARIMA (p, d, q) models for different time series

时间序列起点	1980年	1985年	1991年	1997年	2001年	2009年	2016年	2016年
构建模型	ARIMA(0, 2,0)	ARIMA(0, 2, 0)	ARIMA(0, 2, 0)	ARIMA(1, 1, 0)	ARIMA(1, 1, 0)	ARIMA(1, 1, 0)	ARIMA(1, 1, 0)	ARIMA(1, 1, 1)（具有漂移项）
	ARIMA(1, 2, 0)	ARIMA(1, 2, 0)	ARIMA(1, 2, 0)	ARIMA(0, 1, 1)	ARIMA(0, 1, 1)	ARIMA(0, 1, 1)	ARIMA(0, 1, 1)	ARIMA(0, 1, 0)（具有漂移项）
	ARIMA(0, 2, 1)	ARIMA(0, 2, 1)	ARIMA(0, 2, 1)	ARIMA(0, 1, 0)	ARIMA(0, 1, 0)	ARIMA(0, 1, 0)	ARIMA(0, 1, 0)	ARIMA(1, 1, 0)（具有漂移项）
	ARIMA(1, 2, 1)	ARIMA(1, 2, 1)	ARIMA(1, 2, 1)	ARIMA(1, 1, 1)	ARIMA(1, 1, 1)	ARIMA(1, 1, 1)	ARIMA(1, 1, 1)	ARIMA(0, 1, 1)（具有漂移项）

下载: 导出CSV

表 6 不同时间序列ARIMA模型AIC与BIC结果

Tab. 6 AIC and BIC results for ARIMA models of different time series

模型	1980年		1985年		1991年		1997年		2001年		2009年		2016年
模型	AIC	BIC	AIC	BIC	AIC	BIC	AIC	BIC	AIC	BIC	AIC	BIC	AIC	BIC
ARIMA(0, 2, 0)	1189.08	1190.664	1044.659	1046.093	871.02	872.3864
ARIMA(1, 2, 0)	1189.817	1192.871	1045.746	1048.482	872.05	874.7882
ARIMA(0, 2, 1)	1189.136	1192.19	1045.188	1047.924	871.53	874.2651
ARIMA(1, 2, 1)	1187.622	1192.022	1044.125	1048.017	870.88	874.9828
ARIMA(1, 1, 0)							719.2408	721.575	543.8703	544.8272	364.6756	363.1303	154.9083	154.5178
ARIMA(0, 1, 1)							718.6045	720.9386	543.3656	544.3224	364.6206	363.0753	154.8550	154.0738
ARIMA(0, 1, 0)							717.5984	718.5946	541.802	542.4424	359.7715	359.8564	155.2634	154.4823
ARIMA(1, 1, 1)							721.4796	724.0865	546.5954	547.08	369.332	365.5574	156.2417	154.5178
ARIMA(1, 1, 1) （具有漂移项）													∞	∞
ARIMA(0, 1, 0) （具有漂移项）													159.3684	152.5872
ARIMA(1, 1, 0) （具有漂移项）													179.2876	154.1159
ARIMA(0, 1, 1) （具有漂移项）													∞	∞

下载: 导出CSV

表 7 不同时间序列最优ARIMA模型MAPE结果

Tab. 7 MAPE results of the optimal ARIMA model for different time series

时间序列起始	模型	真实值/t	2022年	2023年	MAPE	R²
时间序列起始	模型	真实值/t	9508508	9574869	MAPE	R²
1980	ARIMA(1, 2, 1)	预测值/t	9543167	9562882	0.12%	0.9795
	ARIMA(1, 2, 1)	相对误差	0.36%	−0.13%	0.12%	0.9795
	ARIMA(0, 2, 0)	预测值/t	9555031	9595785	0.35%	0.9755
	ARIMA(0, 2, 0)	相对误差	0.49%	0.22%	0.35%	0.9755
1985	ARIMA(1, 2, 1)	预测值/t	9490108	9426732	0.87%	0.9681
	ARIMA(1, 2, 1)	相对误差	−0.19%	−1.55%	0.87%	0.9681
	ARIMA（0, 2, 0）	预测值/t	9555528	9597361	0.36%	0.9623
	ARIMA（0, 2, 0）	相对误差	0.49%	0.23%	0.36%	0.9623
1991	ARIMA(1, 2, 1)	预测值/t	9449443	9316839	1.66%	0.9174
	ARIMA(1, 2, 1)	相对误差	−0.62%	−2.69%	1.66%	0.9174
	ARIMA(0, 2, 0)	预测值/t	9555835	9593291	0.35%	0.9030
	ARIMA(0, 2, 0)	相对误差	0.50%	0.19%	0.35%	0.9030
1997	ARIMA(0, 1, 0)	预测值/t	9531036	9843329	1.52%	0.8283
1997	ARIMA(0, 1, 0)	相对误差	0.24%	2.80%	1.52%	0.8283
2003	ARIMA(0, 1, 0)	预测值/t	9531036	9843329	1.52%	0.8283
2003	ARIMA(0, 1, 0)	相对误差	0.24%	2.80%	1.52%	0.8283
2009	ARIMA(0, 1, 0)	预测值/t	9531036	9843329	1.52%	0.8283
2009	ARIMA(0, 1, 0)	相对误差	0.24%	2.80%	1.52%	0.8283
2016	ARIMA(0, 1, 0) 具有漂移项	预测值/t	8760971	8007358	12.12%	0.7320
	ARIMA(0, 1, 0) 具有漂移项	相对误差	−7.86%	−16.37%	12.12%	0.7320
	ARIMA(0, 1, 0)	预测值/t	9514584	9514584	0.28%	0.4561
	ARIMA(0, 1, 0)	相对误差	0.06%	−0.63%	0.28%	0.4561

下载: 导出CSV

表 8 组合模型预测渔获量数据的准确性验证结果

Tab. 8 Validation results of the accuracy of combined model predictions of fish catch data

时间序列起始年份	组合模型	MAPE/%	R²
1980	ARIMA(1, 2, 1)	0.12%	0.9934
1980	ARIMA(0, 2, 0)	0.35%	0.9925
1985	ARIMA(1, 2, 1)	0.87%	0.9888
1985	ARIMA(0, 2, 0)	0.36%	0.9873
1991	ARIMA(0, 2, 0)	0.35%	0.9701
1991	ARIMA(1, 2, 1)	1.66%	0.9737
1997	ARIMA(0, 1, 0)	1.52%	0.9393
2016	ARIMA(0, 1, 0)	0.28%	0.7584
2016	ARIMA(0, 1, 0)具有漂移项	12.12%	0.8688
注：1997年、2003年、2009年我国近海海域全国捕捞渔获量时间序列构建模型为同一模型，因此用1997年代表此3个模型。

下载: 导出CSV

表 9 滚动单步预测与多步预测 MAPE结果

Tab. 9 Rolling single-step forecast and multi-step forecast MAPE results

真实值/t	2022年	2023年	MAPE
真实值/t	9574869	9508508	MAPE
滚动单步预测值/t	9543167	9562882	0.12%
多步直接预测/t	9435027	9786429	0.73%

下载: 导出CSV

参考文献(31)

[1]	金显仕, 田洪林, 单秀娟. 我国近海渔业资源研究历程及展望[J]. 水产学报, 2023, 47(11): 119310. Jin Xianshi, Tian Honglin, Shan Xiujuan. Development and prospects of studies on inshore fisheries resources in China[J]. Journal of Fisheries of China, 2023, 47(11): 119310.
[2]	粮农组织. 2022年世界渔业和水产养殖状况: 努力实现蓝色转型[R]. 罗马: 联合国粮食及农业组织. 2022. FAO. The State of World Fisheries and Aquaculture 2022: Towards Blue Transformation[R]. Rome: FAO. 2022.
[3]	Costello C, Ovando D. Status, institutions, and prospects for global capture fisheries[J]. Annual Review of Environment and Resources, 2019, 44: 177−200. doi: 10.1146/annurev-environ-101718-033310
[4]	Froese R, Zeller D, Kleisner K, et al. What catch data can tell us about the status of global fisheries[J]. Marine Biology, 2012, 159(6): 1283−1292. doi: 10.1007/s00227-012-1909-6
[5]	韩杨. 1949年以来中国海洋渔业资源治理与政策调整[J]. 中国农村经济, 2018(9): 14−28. Han Yang. Marine fishery resources management and policy adjustment in China since 1949[J]. Chinese Rural Economy, 2018(9): 14−28.
[6]	刘子飞. 我国近海捕捞渔业管理政策困境、逻辑与取向[J]. 生态经济, 2018, 34(11): 47−53. Liu Zifei. Research on management policies of marine capture fishery resources: issues, logic and choice[J]. Ecological Economy, 2018, 34(11): 47−53.
[7]	丁依婷, 胡志远, 董帝渤. 基于深度学习和时空特征融合的海洋渔船密度预测方法[J]. 应用海洋学学报, 2024, 43(2): 350−359. Ding Yiting, Hu Zhiyuan, Dong Dibo. A density prediction method for fishing vessel based on deep learning and fusion of spatial-temporal features[J]. Journal of Applied Oceanography, 2024, 43(2): 350−359.
[8]	张忠, 陈新军, 余为. 基于灰色理论系统的西非海域捕捞渔获量预测[J]. 上海海洋大学学报, 2023, 32(4): 818−828. Zhang Zhong, Chen Xinjun, Yu Wei. Catch prediction off the coast of West Africa based on grey theory system[J]. Journal of Shanghai Ocean University, 2023, 32(4): 818−828.
[9]	黄冬梅, 高静霞. 基于SDSS的渔业资源预测决策系统[J]. 计算机工程, 2010, 36(6): 270−272,275. Huang Dongmei, Gao Jingxia. Fishery resources forecasting decision system based on SDSS[J]. Computer Engineering, 2010, 36(6): 270−272,275.
[10]	李辉华, 郭弘艺, 唐文乔, 等. ARIMA模型在预测长江靖江段沿岸鱼类渔获量时间格局中的应用[J]. 水产学报, 2008, 32(6): 899−905. Li Huihua, Guo Hongyi, Tang Wenqiao, et al. ARIMA model application to predict temporal pattern of fish catches of coastal area at Jingjiang Reach of the Yangtze River[J]. Journal of Fisheries of China, 2008, 32(6): 899−905.
[11]	董江水, 诸英富. 时间序列分析模型在江苏省河蟹总产量预测中的应用[J]. 金陵科技学院学报, 2008, 24(3): 102−105. Dong Jiangshui, Zhu Yingfu. Application of time series analysis model on total yield of Chinese mitten-handed crab in Jiangsu Province[J]. Journal of Jinling Institute of Technology, 2008, 24(3): 102−105.
[12]	程炎宏, 樊伟. 东海区海洋捕捞产量的时间序列分析研究[J]. 中国水产科学, 2001, 8(3): 31−34. Cheng Yanhong, Fan Wei. Study of time-serial analysis of marine capture yield in East China Sea region[J]. Journal of Fishery Sciences of China, 2001, 8(3): 31−34.
[13]	蔡格菁, 傅海彬, 蒋仁斌, 等. 基于ARIMA模型的渔业经济预测及其优化[J]. 计算机与现代化, 2019(4): 87−91. Cai Gejing, Fu Haibin, Jiang Renbin, et al. ARIMA based fishing economy prediction model and its optimization[J]. Computer and Modernization, 2019(4): 87−91.
[14]	宋大德, 汪金涛, 陈新军, 等. 时间序列分析模型在黄海南部小黄鱼资源量预测中的应用[J]. 海洋学报, 2020, 42(12): 26−33. Song Dade, Wang Jintao, Chen Xinjun, et al. Application of time series analysis model on stock prediction of small yellow croaker (Larimichthys polyactis) in the southern Yellow Sea[J]. Haiyang Xuebao, 2020, 42(12): 26−33.
[15]	王啸, 刘文俊, 张健. 基于ARIMA的海洋尼诺指数对中西太平洋黄鳍金枪鱼年际CPUE的影响[J]. 南方水产科学, 2023, 19(4): 10−20. Wang Xiao, Liu Wenjun, Zhang Jian. Effect of Oceanic Niño index on interannual CPUE of yellowfin tuna (Thunnus albacares) in Western and Central Pacific Ocean based on ARIMA model[J]. South China Fisheries Science, 2023, 19(4): 10−20.
[16]	李辉华, 郭弘艺, 唐文乔, 等. 长江下游靖江段沿岸贝氏䱗渔获量的时间格局及ARIMA模型预测[J]. 中国水产科学, 2009, 16(3): 357−364. Li Huihua, Guo Hongyi, Tang Wenqiao, et al. Temporal pattern of Hemiculter bleekeri catches at Jingjiang Section of the Yangtze River and catch forecast by using ARIMA model[J]. Journal of Fishery Sciences of China, 2009, 16(3): 357−364.
[17]	Soykan C U, Eguchi T, Kohin S, et al. Prediction of fishing effort distributions using boosted regression trees[J]. Ecological Applications, 2014, 24(1): 71−83. doi: 10.1890/12-0826.1
[18]	韩秋影, 黄小平, 施平. 生态补偿在海洋生态资源管理中的应用[J]. 生态学杂志, 2007, 26(1): 126−130. Han Qiuying, Huang Xiaoping, Shi Ping. Ecological compensation and its application in marine ecological resources management[J]. Chinese Journal of Ecology, 2007, 26(1): 126−130.
[19]	黄硕琳, 唐议. 渔业管理理论与中国实践的回顾与展望[J]. 水产学报, 2019, 43(1): 211−231. Huang Shuolin, Tang Yi. Review and prospect of theories of fisheries management and China’s practice[J]. Journal of Fisheries of China, 2019, 43(1): 211−231.
[20]	Bai Jushan, Perron P. Estimating and testing linear models with multiple structural changes[J]. Econometrica, 1998, 66(1): 47−78. doi: 10.2307/2998540
[21]	Vaihola S, Kininmonth S. Ecosystem management policy implications based on Tonga main tuna species catch data 2002-2018[J]. Diversity, 2023, 15(10): 1042. doi: 10.3390/d15101042
[22]	徐海龙, 韩颖, 谷德贤, 等. 渔业资源自然死亡估算方法研究进展[J]. 水产科技情报, 2019, 46(3): 160−164,171. Xu Hailong, Han Ying, Gu Dexian, et al. Methods of estimated natural mortality in fishery resources assessment[J]. Fisheries Science & Technology Information, 2019, 46(3): 160−164,171.
[23]	King K B S, Giacomini H C, Wehrly K, et al. Using historical catch data to evaluate predicted changes in fish relative abundance in response to a warming climate[J]. Ecography, 2023, 2023(8): e06798. doi: 10.1111/ecog.06798
[24]	Aldrin M, Aanes F L, Tvete I F, et al. Caveats with estimating natural mortality rates in stock assessment models using age aggregated catch data and abundance indices[J]. Fisheries Research, 2021, 243: 106071. doi: 10.1016/j.fishres.2021.106071
[25]	李琪, 刘淑德, 王琨, 等. 渔获量时间序列长度对基于CMSY方法的资源评估结果的影响[J]. 海洋学报, 2023, 45(3): 27−39. Li Qi, Liu Shude, Wang Kun, et al. Effects of lengths of catch time series on stock assessment using CMSY method[J]. Haiyang Xuebao, 2023, 45(3): 27−39.
[26]	Miyagawa M, Ichinokawa M, Yoda M, et al. Commentary: stock status assessments for 12 exploited fishery species in the Tsushima Warm Current region, Southwest Japan and East China, using the CMSY and BSM methods[J]. Frontiers in Marine Science, 2021, 8: 703039. doi: 10.3389/fmars.2021.703039
[27]	Ju Peilong, Tian Yongjun, Chen Mingru, et al. Evaluating stock status of 16 commercial fish species in the coastal and offshore waters of Taiwan using the CMSY and BSM methods[J]. Frontiers in Marine Science, 2020, 7: 618. doi: 10.3389/fmars.2020.00618
[28]	Bouch P, Minto C, Reid D G. Comparative performance of data-poor CMSY and data-moderate SPiCT stock assessment methods when applied to data-rich, real-world stocks[J]. ICES Journal of Marine Science, 2021, 78(1): 264−276. doi: 10.1093/icesjms/fsaa220
[29]	Zhai Lu, Liang Cui, Pauly D. Assessments of 16 exploited fish stocks in Chinese waters using the CMSY and BSM methods[J]. Frontiers in Marine Science, 2020, 7: 483993. doi: 10.3389/fmars.2020.483993
[30]	周淑婷, 邹晓荣, 邹邦郁. 基于改进的XGBoost模型的大西洋大眼金枪鱼渔场预报[J]. 大连海洋大学学报, 2025, 40(3): 481−488. Zhou Shuting, Zou Xiaorong, Zou Bangyu. Prediction of bigeye tuna (Thunnus obesus) fishing ground in the Atlantic Ocean based on the modified XGBoost model[J]. Journal of Dalian Ocean University, 2025, 40(3): 481−488.
[31]	袁红春, 高子玥, 张天蛟. 基于改进的XGBoost模型预测南太平洋长鳍金枪鱼资源丰度[J]. 海洋湖沼通报, 2022, 44(2): 112−120. Yuan Hongchun, Gao Ziyue, Zhang Tianjiao. Prediction of albacore tuna abundance insouth Pacific based on improved XGBoost model[J]. Transactions of Oceanology and Limnology, 2022, 44(2): 112−120.