基于动态多尺度分析的水产养殖水质非线性组合预测模型研究

Aquaculture water quality deterioration is the primary factor induced aquaculture disease outbreaks or even large quantities of death, aquaculture water quality forecasting precision is always the need to solve the difficult problem for aquaculture. The project plan to volatile ammonia nitrogen, nitrite and key water quality parameters, such as dissolved oxygen as the research object, using Geostatistic and multifactor analysis of variance methods, studying the key parameters of the spatial and temporal variation characteristics under the condition of different water, clear water quality parameters and the mutual relationship between ecological environment factors; Using multidimensional signal processing technology, find the feature extraction method suitable for aquaculture water quality parameters, obtain reliable comprehensive water quality parameters of the characteristic matrix; Analysis of single forecasting method performance and the applicable scope,the research of data mining technology, a combined forecasting effectiveness criteria, swarm intelligent multi-objective optimization algorithm with weight determination method, explore suitable for aquaculture water quality, and a method for determining nonlinear combination weighting adaptive initial combination forecast model is set up; Combination weight coefficient is realized by using swarm intelligent multi-objective optimization algorithm and adaptive optimization model parameter adjustment, has built the learning of multi-scale analysis, high precision and dynamic nonlinear combination forecast model of aquaculture water quality, precision prediction of aquaculture water quality, early warning and control, for aquaculture water quality strategy management and prevention of aquatic disease outbreaks to provide scientific basis and method guidance.

养殖水质恶化是诱导水产疾病爆发甚至大批量死亡首要因素，水质精准预测一直是水产养殖业亟需解决的棘手难题。本项目拟以极不稳定的氨氮、亚硝酸盐和溶解氧等关键水质参数为研究对象，采用地统计学和多因素方差分析方法，研究不同条件下水质关键参数时空变异特性，明确水质参数与生态环境因子相互作用关系；利用多维信号处理技术，找到适宜于养殖水质参数特征提取方法，获得全面可靠的水质参数特征矩阵；分析单项预测方法性能及适用范围，研究数据挖掘技术、组合预测有效性准则、群集智能多目标优化算法与权重确定方法，探索适宜于养殖水质的非线性组合权重自适应确定方法，建立初步的组合预测模型；采用群集智能多目标优化算法实现组合权重系数和模型参数自适应优化调整，构建具有自学习、精度高和动态多尺度分析的养殖水质非线性组合预测模型，实现养殖水质精准预测预警与控制。为水产养殖水质精准调控管理与预防水产疾病爆发提供科学依据和方法指导。

为解决水产养殖水质精准预测难题，本课题开展了基于动态多尺度分析的水产养殖水质非线性组合预测模型研究。提出了基于系统动力学、主成分分析法和粗糙集理论等水质关键参数时空变异特性的研究方法，探索了水质参数与生态环境因子相互作用关系，确定养殖水质关键影响因子；提出了基于小波分析、经验模态分解、集合经验模态分解、深度学习等多维信号处理技术的水产养殖水质参数特征提取方法，实现养殖水质特征动态多尺度提取，构建全面描述的水质参数特征矩阵；采用人工蜂群、改进粒子群算法、改进人工蜂群等群集智能优化算法，解决水产养殖水质的非线性组合权重系数和模型参数自适应确定问题；研究了ARIMA、GAWNN、BP神经网络(BPNN)、灰色系统理论(GST)、非线性回归方法（NRM）、支持向量机（SVM）、最小二乘支持向量回归机(LSSVR)、极限学习机（ELM）、长短期记忆网络（LSTM）等单项预测方法性能和适用范围，组合预测有效性准则，提出了基于融合ARIMA和GA算法优化的WNN、基于数据融合和IPSO-LSSVM、基于WA-ABC-WLSSVR、基于人工蜂群优化机理模型、基于EMD-ELM、基于EEMD-IABC、基于EMD-IPSO-ELM氨氮软测量、基于DBN-LSSVR、基于DBN-LSTM的多变量多步等多种养殖水质非线性组合预测模型，基于粗糙集融合支持向量机的河蟹养殖水质预警模型，实现提高水产养殖水质预测精度，本课题研究为水产养殖水质非线性组合预测提供了科学方法和理论依据。依托本项目资助，共发表论文37篇，其中SCI论文18篇，EI论文9篇；申请专利26项、发明专利10项，其中以已授权发明专利3件，实用新型专利授权8件，授权计算机软件著作权7项。