稀土萃取分离过程的自组织特征学习及参数优化研究

The rare earth extraction separation process is a complex dynamic system, which has many characteristics, such as, complicated mechanism of chemical reaction, multiple variables, mixed variables types, strong nonlinear and coupling between multi variables, obvious change of dynamic characteristics with in working conditions and operation conditions, difficult to online detect and measure the operational indices continuously, and the accurate process model is unable to build. Along with the application of new instruments, sensor technology, and network technology in the industrial process, a lot of production process real-time data has been obtained. And, the production line operation experts have accumulated preferable experience knowledge in operation optimization, fault diagnosis, safety operation and maintenance system. In order to promote our rare earth industry, realize the comprehensive automation level of production management and operation optimization, improve the market competitiveness of enterprises, this project will focusing on the operation optimization problems of rare earth extraction separation process, and studying the real-time intelligent operation optimization method. Through the self-organizing learning of system dynamic characteristic, taking the market demand, energy saving and other conditions as constraint, to establish optimal operational scheme. Based on data-driven and experience knowledge to design optimization algorithm to realize the real-time collaborative optimization of process control and operations. Thereby, obtain the real-time optimal control and safety operation and maintenance.

稀土萃取分离过程机理复杂、变量多、类型混杂、变量之间强非线性强耦合、动态特性随工况及操作条件变化明显，其运行指标难以在线连续测量，无法准确建立过程模型 。新型仪表、传感技术、网络技术的应用在生产过程获得了大量的实时数据，生产线的操作专家也积累了丰富的系统运行优化、故障诊断、安全运行维护等经验知识。本课题拟从提升我国稀土行业综合自动化水平、实现企业生产管理和运行控制优化、提高企业市场竞争力的实际需求出发，针对稀土萃取分离过程亟待解决的操作、运行控制开展实时智能运行优化方法研究，通过对系统动态特征的自组织学习，建立以市场需求、节能降耗等为约束条件的参数优化运行方案 ，基于数据驱动及知识设计实时协同优化算法。

本项目通过对稀土萃取分离过程的分析，针对该过程机理复杂、变量多、类型混杂、变量之间强非线性强耦合、动态特性随工况及操作条件变化明显等特点研究基于人工智能技术的复杂系统（自组织）特征建模方法，提出了基于连接自组织发育的跨越-侧抑制神经网络设计方法、基于AQPSO的RBF神经网络自组织学习、自适应权重粒子群优化LS-SVM的自组织特征学习方法等；同时，基于生产过程实时数据、生产线的操作专家经验知识，针对稀土萃取分离过程中组分含量的预测及控制，研究了基于改进即时学习算法的镨/钕元素组分含量预测方法、HSV空间下基于图像检索的稀土元素组分含量快速检测方法；设计了一种稀土溶液图像采集装置及方法、一种稀土萃取过程多组分含量预测方法及系统、基于时序特征的稀土元素组分含量动态监测方法及系统、一种基于虚拟样本的铈镨和钕组分含量的预测方法及系统、一种稀土串级萃取最优工艺计算系统、一种稀土元素组分含量动态监测系统。