重油加氢裂化过程稳态模型的建立

Building of static model for residual hydrocracking processes

为保证石油化工系统安全生产，采用偏最小二乘法（ＰＬＳ）与人工神经网络（ＡＮＮ）相结合的方法，建立了重油加氢裂化过程的稳态模型，解决了体系复杂、影响因素多、现场数据的噪声等对模型的运算速度和精度不易满足要求的困难。对ＰＬＳ的主矢量所张成的子空间进行了数学分析，并对该方法的降维去噪能力进行了讨论。仿真实例表明，ＰＬＳ－ＡＮＮ法与ＡＮＮ法相比，在训练与预报精度都有所改善的条件下，不仅具有良好的降维效果，且有更强的除噪能力；与线性ＰＬＳ法相比具有更好的非线性映射能力。

In order to ensure the safety production of the petrochemical system, the partial least square and artificial neural networks (PLS ANN) method was suggested for building steady model in residual hydrocracking systems. This method can solve the problem of low computation speed and percision caused by data noise and complexity of the systems. A mathematical analysis for PLS main component was finished, and the ability of reducing dimension and noise filtration of this method was discussed. The result of a case study shows that PLS ANN method not only has the advanced ability on reducing data dimension, but also presents the ability on noise filtration comparing with ANN method and non linear project ability comparing with linear PLS.