神经网络软测量应用于乙烯裂解深度控制

Application of neural network soft sensor on ethylene cracking severity control

基于神经网络建模方法,建立了乙烯装置裂解深度控制软测量模型,利用裂解深度控制器与裂解炉平均出口温度(COT)控制器串级功能开发出裂解深度控制系统,并在中国石油兰州石化公司46万t/a乙烯装置上进行了工业应用。结果表明:所建模型以双烯(乙烯和丙烯)收率作为主导变量,COT、原料流量、汽烃比(稀释蒸汽与原料的质量比)、油品密度等作为辅助变量;裂解深度控制系统运行平稳,使乙烯装置双烯收率提高了0.154个百分点,燃料气平均消耗量减少了109.7 kg/h。

Based on a neural network modeling method, the soft - measuring model of cracking se- verity control in the ethylene unit was established, the cracking severity control system was developed by cascading the cracking severity controller and the average coil out temperature （COT） controller of cracking furnace, and it was applied in a 460 kt/a ethylene unit of Lanzhou Petrochemical Company, PetroChina. The results showed that in the estab- lished model, the ethylene and propylene yields were considered as the dominant variables, and the COT, feed flow, steam to hydrocarbon（ mass ratio of diluted steam to raw materiai）, and oil density, etc, were considered as the assistant variables; the cracking severity control system had been running stably, the ethylene and propylene yields increased by 0. 154 percentage points, and the average con- sumption of fuel gas decreased by 109.7 kg/h.