CAJ | 학술논문

提出一种利用回声状态网络(echo state network, ESN)建立复杂分布参数系统模型的灰箱建模方法。此建模方法可以充分利用已知机理模型的结构信息和回声状态网络的逼近能力，可更好地描述和解释出系统各变量之间的因果关系，使模型的“灰箱”化程度更高。首先，根据系统方程和先验知识将初始系统特征团引入ESN储备池中，赋予网络节点实际物理意义，并以此建立结构逼近神经网络模型；然后，通过逐步回归分析方法，结合递归最小二乘算法选择最优系统特征团，并对网络结构进行优化，建立起描述系统特性关系的灰箱模型。本文以实验室规模的管式聚合反应过程作为实验对象，建立以温度分布为输出的数学模型，结果表明所提出的灰箱建模方法行之有效。
제출일충이용회성상태망락(echo state network, ESN)건립복잡분포삼수계통모형적회상건모방법。차건모방법가이충분이용이지궤리모형적결구신식화회성상태망락적핍근능력，가경호지묘술화해석출계통각변량지간적인과관계，사모형적“회상”화정도경고。수선，근거계통방정화선험지식장초시계통특정단인입ESN저비지중，부여망락절점실제물리의의，병이차건립결구핍근신경망락모형；연후，통과축보회귀분석방법，결합체귀최소이승산법선택최우계통특정단，병대망락결구진행우화，건립기묘술계통특성관계적회상모형。본문이실험실규모적관식취합반응과정작위실험대상，건립이온도분포위수출적수학모형，결과표명소제출적회상건모방법행지유효。
An approach of grey-box modeling with Echo State Network (ESN) is developed for modeling dynamic processes with nonlinear characteristics. This method can take full advantage of the already known structural information of the mechanism model at the early stage of modeling and make better use of the approximation ability of neural networks, thus resulting in higher accuracy of grey-box modeling. By combination the prior knowledge and systematic equations into ESN state pool, structure approaching neural network (SAAN) is established based on system feature block, and it is given actual significance. Then the optimal fundamental genes were chosen through recursive least square method with stepwise regression analysis to optimize the structure of SANN, so as to get the grey-box model. Detailed process of modeling was described in modeling of tubular polymerization reaction in laboratory scale. The simulation result proves that the approach is effective.ocesses heat exchanger network synthesis by taking place.