王 魏1, 2，邓长辉1, 2，关晓燕3，何东钢1, 2，李响1, 2
(1大连海洋大学信息工程学院，辽宁 大连 116023；
2辽宁省海洋信息技术重点实验室，辽宁 大连 116023；
3辽宁省海洋水产科学研究院，辽宁 大连 116023)

摘要：溶氧浓度软测量模型是用于预测水产养殖水质中溶氧变化情况的重要手段，对于控制鱼类生长存活至关重要。阐述了建立溶氧浓度软测量模型的目的和意义，综述了溶氧浓度软测量模型的研究现状。溶氧浓度软测量模型分为神经网络和优化算法组合和支持向量机和优化算法组合，总结和分析了这两类建模方法存在的问题，提出了采用混合软测量方法建立溶氧浓度模型的改进方案；最后对溶氧浓度软测量模型今后的研究方向进行了简要展望，指出融合多种信号并结合专家知识、机器学习等智能方法，基于机理和数据驱动的混合软测量是建立溶氧浓度模型的发展方向。

关键词：水产养殖；溶氧；软测量；神经网络

Advances in soft measure model of dissolved oxygen concentration in aquaculture– A review
WANG Wei1,2, DENG Changhui1,2 , GUAN Xiaoyan3 , HE Donggang1,2 , LI Xiang1,2
(1 College of Information Engineering, Dalian Ocean University;
2 Key Laboratory of Marine Information Technology of Liaoning Province, 116023, Dalian, China;
3 Liaoning Ocean and Fisheries Science Research Institute, 116023, Dalian, China)

Abstract: The soft measure model of dissolved oxygen concentration is an important tool which is used to predict the change of dissolved oxygen concentration in aquaculture process. It is essential for the growth and survival of fish. To promote the soft measure research of dissolved oxygen concentration and industrial applications, the purpose and significance were introduced, and the present situation of soft measure model of dissolved oxygen concentration was also reviewed. According to the related research papers, there were two methods of modeling soft measure of dissolved oxygen concentration: combination of neural networks and optimization algorithms, and combination of support vector machine and optimization algorithms. The problems of the two methods were summarized and analyzed, and the improved methods were proposed which was based on mixed modeling method. The future research direction and priorities in this field were briefly discussed. The intelligent measure strategy which combines multi-source information fusion with expert knowledge and machine learning, and mixed modelling which combines mechanism model with data driven model was proposed as the development direction of the field. 

Key words：aquaculture; dissolved oxygen; soft measure; neural networks