基于数据驱动的一类复杂系统动态特性分析

Dynamical Performance Analysis for a Class of Complex Systems Based on Data Driven

摘要:

针对一类复杂的无法对其机理建模的离散时间系统,根据采集的两年工艺参数数据,结合复杂工艺特点,提出了基于数据驱动的系统动态特性建模方法,构建了时间序列受控回归滑动平均(CARMA)胞映射模型。在模型结构确定的基础上,采用改进的量子行为粒子群优化(IQPSO)算法对系统参数进行辨识。算法通过设计新的粒子更新式增加了粒子的多样性,避免了算法的早熟收敛；算法通过在后期将搜索到的最优值传递给神经元作为初始权值,利用神经元增强算法的局部搜索能力,实现了算法探索与开发的平衡,达到对模型参数进行快速精确辨识的目的。在转化为状态空间模型基础上,根据胞映射理论对系统进行了稳定性分析,通过对胞映射作图快速获得平衡胞,利用动态优化原理,找到所有的周期胞和吸引域,达到对系统稳定性分析的目的。利用现场工艺数据进行仿真,结果证明了所提方法的有效性。

Abstract:

A model and the stability analysis are proposed for a class of dynamical systems which is difficult in modeling due to the complex physicochemical change by data-driven method. A CARMA cell mapping model is established, whose parameters are identificated by quantum-behaved particle swarm optimization. To avoid the premature convergence of the IQPSO algorithm, the particles position are updated and accurate search capacity is enhanced by applying the neural networks that train the optimal value as the initial weighted value. Based on the model, the stability performance is analyzed by cell mapping theory and dynamic optimization principle, which is helpful to find the whole periodic cell and attractive domain. Simulation studies are included to demonstrate the effectiveness of the proposed approach.