冶金动态过程的自组织经验进化预测方法研究

A self-organization evolutionary predicting method has been presented to solve the conflict between extending experience used for predicting dynamical process and information over-load. A model of adaptive resonance theory, ART2, has been used for vector quantifying of spatio-temporal pattern of dynamical information. A software propotype of the model for the prediction of silicon content of hot metal tapped from blast furnace has been designed and programmed. The propotype has been tested in a range of parameters of the model in which vigilance are from 0.8 to 1.0 and filter are from 0.15 to 0.35. It has been shown that the hit ratio for prediction of silicon content of hot metal tapped from blast furnace No. 1 of Tianjin Iron Plant is less than 75% when accuracy of prediction is 0.10% of silicon content. As the number of variables in prediction model increases, the predictablity of the model decreses. In most case of online test, the model passed the Turing Test based on the prediction criteria of hit ratio on trend, hit ratio and mean squared error. The data mining has been used for acquiring second hand prediction knowledge when the model works at poor first hand prediction experience. With the help of second hand prediction knowledge, the predictability has been improved. The hit ratio is 76.8% for 88 tappings of hot metal based on data mining operation on 450 tappings data in comprison with the hit ratio of 67% which is based on data mining operation on 400 tappings data of hot metal. The results are helpful for nourishing research on prediction technique and improving process control of complex metallurgical process.

提出自组织经验进化预测方法用于解决动态过程预测中经验范围与信息超载之间的矛盾，运用自适应共振理论研究动态信息的时空模式量化方法，研究预测经验的自组织表达、贮存、进化和累积算法以及实验方法。以高炉铁水含硅量预测为例对该方法予以软件实现并进行在线检验。项目研究对丰富预测技术研究内容以及提高复杂冶金过程控制水平具有重要意义。