散式流化和聚式流化的BP神经网络识别  

Identification of Particulate and Aggregative Fluidization Using BP Neural Network

作　　者：周树孝[1] 彭黔荣[1] 周勇[1] 石炎福[1] 余华瑞[1] 

机构地区：[1]四川大学化学工程学院,四川成都610065

出　　处：《高校化学工程学报》2004年第4期459-464,共6页Journal of Chemical Engineering of Chinese Universities

基　　金：国家自然科学基金; 海外青年基金资助项目(29928005)。

摘　　要：由于流化系统的复杂性和非线性的特性,到目前为止,仍没有一个能很好地判断这两类流态化的可靠方法,而人工神经网络能够进行复杂的逻辑操作和实现对非线性系统流型的识别。由此,在文献数据(含15种颗粒和11种流体)的基础上,利用三层BP人工神经网络,提出了一种识别散式流化和聚式流化的新方法。以经过归一化处理的3个无因次准数pff()/rrr-、mfRe和mfFr(判别因子)作为神经网络的三个输入结点,以散式流化、过渡状态和聚式流化对应于神经网络的三个输出结点,由训练样本集得到隐层结点数等最佳网络参数,然后对待测样本进行了流型识别。研究结果表明,神经网络用于散式流化和聚式流化的模式识别,结果与实际较一致。新方法优于传统的识别方法,具有较好的应用前景。Due to the complexity and strong nonlinear characteristic of the fluidization system, there is still no reliable method to identify particulate and aggregative fluidization. The artificial neural network (ANN) can perform complicated logical operation and accomplish the identification of the flow regime for nonlinear system, thus, a method for identification of particulate fluidization and aggregative fluidization was proposed by means of three-layer BP artificial neural network based on literature data corresponding to 15 kinds of solid and 11 kinds of fluid. Taking three characteristic parameters (ρp-ρf/ρf, Remf and Frmf) which are disposed specifically as identified factors, as the input nodes, and taking particulate fluidization, transitional fluidization and aggregative fluidization as the output nodes, the best network parameters such as the number of hidden nodes were gained by the training-samples. Then the test-samples were identified. The results show that pattern recognition of fluidization regimes by artificial neural network agrees well with the fact. The new method provides better results than the traditional ones, and has promising applications.

关 键 词：散式流态化 聚式流态化 BP神经网络 流型识别 

分 类 号：TQ021.1[化学工程] TQ051.13