Intelligent analysis of economic benefits in marine ranching is essential for fine-grained aquaculture management. However, existing data-driven methods rely heavily on large labeled datasets and often suffer from under fitting in small-sample scenarios. To address this problem, this paper proposes two prompt-engineering-based intelligent analysis methods: a zero-shot economic benefit analysis method, named Prompt-Engineered Sea Ranching Economic Analysis (PE-SEA), and a sample-augmentation-based economic benefit analysis method, named Prompt-Augmented Pretraining for Sea Ranching Economic Analysis (PAP-SEA).Both methods use prompt engineering to guide large language models in analyzing marine ranching-related data and predicting future economic benefits. PE-SEA performs zero-shot inference based on GPT-4 and does not rely on additional training data; PAP-SEA uses ChatGPT for sample augmentation, while model training and inference are both completed locally. PE-SEA is more suitable for performance-prioritized scenarios, while PAP-SEA is more suitable for scenarios with higher requirements for data security and privacy protection. To facilitate practical application, this paper develops a real-time intelligent analysis tool called SeaPastureGPT for the economic benefits of marine ranching. Experiments are conducted on real monitoring datasets from three different sea areas and aquaculture species. The results show that both methods outperform existing mainstream methods across six evaluation metrics (precision, recall, F1, area under the ROC curve, Matthews correlation coefficient, and accuracy). In particular, PE-SEA improves the F1 by 70.67% compared with the best-performing baseline method called KnowledgeGraph-BERT. Overall, our methods offer new insights and practical guidance for both academic research and industrial applications.
大语言模型(Large Language Model,LLM)[12]的兴起为解决样本稀缺问题提供了新的方法。LLM依托海量开源语料完成自监督预训练,具备丰富的语言知识与领域知识,能够实现小样本或零样本学习下的高质量文本生成与语义推理,有效解决了样本稀缺与人工标注成本高的问题[13-14]。同时,提示工程的发展能够引导大语言模型完成不同的下游任务。现有的研究表明,通过设计合理的提示模板,大语言模型能够在软件跟踪、实体识别、数据增强等任务中展现出优越性能[15]。
随着通用语料和模型的训练参数不断增加,ChatGPT、DeepSeek以及豆包等大语言模型受到研究者们的关注。ChatGPT是基于GPT-3.5(1 750亿参数量)或GPT-4(万亿级参数量)的自回归语言模型[12],其训练语料涵盖海量网络数据,同时融合了人类反馈的强化学习(Reinforcement Learning from Human Feedback,RLHF)算法,生成的文本更符合人类表达习惯与价值偏好。现有研究表明,ChatGPT无需额外的预训练或微调,在实体识别[17]、数据增强[14]等下游任务中均表现出卓越性能,且其具备的跨领域知识与语义推理能力,使其能够直接处理海洋牧场经济效益分析这类跨生态、养殖、经济的复合任务。因此,本文使用ChatGPT大语言模型,依托其出色领域知识,实现海洋牧场经济效益的零样本分析与样本增强。
WUD, RENZ J, HANR G, et al. Exploration and practice of offshore wind power and marine ranching integration development technology[J]. Chinese Fisheries Economics, 2023, 41(3): 78-84. DOI:10.3969/j.issn.1009-590X.2023.03.010(Ch ).
LIY H, YINJ, LINH. How does artificial intelligence and marine economy “empower each other”?[J]. China Telecommunication Trade, 2025(11): 14-17. DOI:10.3969/j.issn.1671-3060.2025.11.006(Ch ).
[5]
KÜHNB, CAYETANOA, FINCHAMJ I, et al. Machine learning applications for fisheries—At scales from genomics to ecosystems[J]. Reviews in Fisheries Science & Aquaculture, 2025, 33(2): 334-357. DOI:10.1080/23308249.2024.2423189 .
[6]
SAVILLER, FUJIWARAA, HATANAKAK, et al. AI-powered decision support system for mariculture: Real-time fish mortality prediction with random forest[J]. Aquacultural Engineering, 2026, 112: 102621. DOI:10.1016/j.aquaeng.2025.102621 .
[7]
GHAEMIS, GHOLMOHAMADIH, ANVARI-MOGHADDAMA, et al. Optimizing power consumption in aquaculture cooling systems: A Bayesian optimization and XGBoost approach under limited data[J]. Applied Sciences, 2025, 15(11): 6273. DOI:10.3390/app15116273 .
[8]
JONGJARAUNSUKR, TAPARHUDEEW, MATONDANGP. An ensemble learning technique for predicting mortality rate in red tilapia (Oreochromis Niloticus Linn.) fingerlings[J]. Journal of Fisheries and Environment, 2024, 48(1): 37-50. DOI: 10.3390/app15116273 .
[9]
WUA Q, LIK L, SONGZ Y, et al. Deep learning for sustainable aquaculture: Opportunities and challenges[J]. Sustainability, 2025, 17(11): 5084. DOI:10.3390/su17115084 .
[10]
YANGX T, ZHANGS, LIUJ T, et al. Deep learning for smart fish farming: Applications, opportunities and challenges[J]. Reviews in Aquaculture, 2021, 13(1): 66-90. DOI:10.1111/raq.12464 .
[11]
ZHANGM Q. Marine fishery resource dynamic prediction based on CNN-XGBoost fusion model[J]. Scientific Reports, 2026, 16: 3244. DOI:10.1038/s41598-025-33175-4 .
[12]
LIUS T, DUZ Z, WANGG X, et al. From traditional machine learning models to multimodal large models: A review of aquaculture[J]. Reviews in Aquaculture, 2026, 18(1): e70111. DOI:10.1111/raq.70111 .
ZHONGX, CHENL, LIUW X, et al. Research on temporal-spatial semantic-driven progressive multiview action debiasing[J]. Computer Engineering, 2025, 51(1): 1-10. DOI:10.19678/j.issn.1000-3428.0069307 (Ch ).
[15]
BROWNT B, MANNB, RYDERN, et al. Language models are few-shot learners[DB/OL].[2025-12-12]. DOI: 10.18653/v1/2021.emnlp-main.734 .
[16]
MINB N, ROSSH, SULEME, et al. Recent advances in natural language processing via large pre-trained language models: A survey[J]. ACM Computing Surveys, 2023, 56(2): 1-40. DOI:10.1145/3605943 .
[17]
DAIH X, LIUZ L, LIAOW X, et al. AugGPT: Leveraging ChatGPT for text data augmentation[EB/OL]. 2023: arXiv: 2302.13007.
[18]
ZHAOZ H, WANGS, GUJ C, et al. Chatcad+: Toward a universal and reliable interactive CAD using LLMs [J]. IEEE Transactions on Medical Imaging, 2024, 43(11):3755-3766. DOI: 10.1109/TMI.2024.33983500 .
[19]
LIUY H, OTT M, GOYALN, et al. RoBERTa: A robustly optimized BERT pretraining approach[EB/OL]. 2019: arXiv: 1907.11692.
[20]
REZAYIS, DAIH X, LIUZ L, et al. ClinicalRadioBERT: Knowledge-infused few shot learning for clinical notes named entity recognition[C]//Machine Learning in Medical Imaging. Cham: Springer, 2022: 269-278. DOI:10.1007/978-3-031-21014-3_28 .
[21]
LEEU, JUNGH, JEONY, et al. Few-shot is enough: Exploring ChatGPT prompt engineering method for automatic question generation in English education[J]. Education and Information Technologies, 2024, 29(9): 11483-11515. DOI:10.1007/s10639-023-12249-8 .
[22]
EKINS. Prompt engineering for ChatGPT: A quick guide to techniques, tips, and best practices[J]. ASEE Journal of Engineering Technology, 2026, 43(1): 52-68. DOI:10.18260/jet.43.1.5 .
[23]
HANX, ZHAOW L, DINGN, et al. PTR: Prompt tuning with rules for text classification[J]. AI Open, 2022, 3: 182-192. DOI:10.1016/j.aiopen.2022.11.003 .
[24]
HAOD Q, XUB Z, LENGJ, et al. Elucidating the drivers of aquaculture eutrophication: A knowledge graph framework powered by domain-specific BERT[J]. Sustainability, 2025, 17(19): 8907. DOI:10.3390/su17198907 .
XUH. Research on marine ranch water quality prediction model and monitoring system based on deep learning[D]. Jinan: Qilu University of Technology, 2024(Ch). DOI: 10.3390/w15152760 .
[27]
KIMY H, CHOS, LEEP S. Wave height classification via deep learning using monoscopic ocean videos[J]. Ocean Engineering, 2023, 288: 116002. DOI:10.1016/j.oceaneng.2023.116002 .
[28]
SCHULHOFFS, ILIEM, BALEPURN, et al. The prompt report: A systematic survey of prompt engineering techniques[EB/OL]. 2024: arXiv: 2406.06608. DOI: 10.1201/9781032692319-3 .
[29]
HUE, SHENY L, WALLISP, et al. LoRA: Low-Rank adaptation of Large Language Models[DB/OL].[2025-12-12].
CHEW X, DOUZ C, FENGY S, et al. Towards a comprehensive understanding of the impact of large language models on natural language processing: Challenges, opportunities and future directions[J]. Science in China (Information Sciences), 2023, 53(9): 1645-1687. DOI:10.1360/SSI-2023-0113 (Ch ).
[32]
VALIZADEHASLANIT, SHIY W, WANGJ, et al. Two-stage fine-tuning with ChatGPT data augmentation for learning class-imbalanced data[J]. Neurocomputing, 2024, 592: 127801. DOI:10.1016/j.neucom.2024.127801 .
[33]
YINX Y GOUDRIAAN J, LANTINGAE A, et al. A flexible sigmoid function of determinate growth[J]. Annals of Botany, 2003, 91(3): 361-371. DOI:10.1093/aob/mcg029 .
YANGL, SHIB F. Credit risk evaluation model and empirical research based on focal loss modified cross-entropy loss function[J]. Chinese Journal of Management Science, 2022, 30(5): 65-75. DOI:10.16381/j.cnki.issn1003-207x.2020.2188 (Ch ).
[36]
AUGERT, SAROYANE. Overview of the OpenAI APIs[M]//Generative AI for Web Development. Berkeley, CA: Apress, 2024:87-116.
[37]
JAINS M. Hugging face[M]//Introduction to Transformers for NLP. Berkeley, CA: Apress, 2022:51-67.
YAOS Y, YANGS T, LIY M. Research on performance measure indicators of machine learning classification based on confusion matrix and corresponding Python practice[J]. Hans Journal of Data Mining, 2022, 12(4): 351-367. DOI:10.12677/HJDM.2022.124033 (Ch ).