"人工神经网络"方法用于超临界流体萃取模拟

在15－30MPa和303－323K条件下，用超临界CO2流体萃取沙棘籽油，结果表明，最高沙棘油收率（30MPa,308K)可达到90％以上，对过程进行动力学模拟，建立了超临界萃取过程的人工神经网络（ANN）模型，以MATLAB软件为平台，编制了SFE－ANN模拟程序系统，采用3层BP网络结构，以压力，温度、萃取时间为输入，以萃取出油量为输出对网络进行训练，由此得到的网络可以对萃取速率和单位时间床高方向的萃取出油量进行准确的模拟和预测，与实验结果比较证明，训练样本集误差为0．2％，测试样本集误差为0．5％，模拟误差小于6％。

Simulations for Supercritical Fluid Extraction by Use of Artificial Neural Networks

The extraction of oil from Hippophae rhamnoides L. seeds using supercritical carbon dioxide was investigated at 15~30MPa and 303~323K. The results showed that the highest yield of oil was above 90% at 30MPa and 308K. In order to make an exact simulation for the extraction kinetics, an artificial neural network (ANN) technology was applied to the simulation of SCF extraction process of vegetable oil. Based on MATLAB software, the SFE-ANN simulation program system was developed for a fixed bed extractor, in which a 3-layer BP network structure was applied. With the operation factors (such as pressure, temperature and extraction time) as input variables of the network, and the oil yield of extraction as the output, the network was trained as optimal. The ANN network developed in this work can make good simulation and prediction for the extraction rate and oil yield along the bed height. Compared with the experimental results, the deviations for training sample and for testing sample are 0.2% and 0.5%, respectively. The simulation deviations can be controlled as low as less than 6%.