人工神经网络在低温甲醇洗系统优化中应用

Application of artificial neural net work in optimization of rectisol process

采用正交试验法确定人工神经网络(ANN)训练样本集的输入参数,利用基于严格机理模型的低温甲醇洗模拟系统(RPS)进行模拟计算,得到样本的输出期望值后,对改进的BP网络进行训练. 结果表明,ANN成功地模拟了低温甲醇洗系统,其模型可作为“黑箱”模型代替RPS的严格模型. 运用复合形法基于ANN模型对低温甲醇洗系统的重要工艺条件进行优化,可节省计算时间. 优化结果表明,装置的CO2产量提高,氨冷负荷降低；所描述的优化策略可用于解决大型实际复杂系统的操作条件优化问题. 这一结果为工厂优化操作指明了方向.

The modified back-propagation artificial neural network (ANN) was trained by orthogonal experiment to get the input parameters and the rectisol process simulator (RPS) based on exact mathematical models was used to obtain the output parameters. The results show that ANN is successful in simulating the rectisol process and can be used as 'black box' model to replace the rigorous RPS models. Complex algorithm was used to optimize the critical process conditions of rectisol system, which uses the trained ANN as a mathematical model, so that the optimization problem was greatly simplified and the computational time decreases significantly. The optimization results show that the product quantity of CO2 is increased and the load of refrigerated ammonia is reduced. These results can be used to guide the operation of ammonia plant.