WNN优化模型在模拟超临界二氧化碳中固体溶解度的应用

Applization of optimized wavelets neural networks in simulating the solubility of solids in supercritical carbon dioxide

为了更好地模拟超临界二氧化碳(SCCO2)中固体的溶解度,提出了一种余弦常数r为0.1和采用部分动量法的Morlet自适应小波神经网络优化模型,并以温度T、压力P、溶质的摩尔体积V、熔点温度Tm、熔化热ΔHfus、色散溶解度参数δ1和极性及氢键溶解度参数δ为输入参数,对SCCO2中两种脂肪酸的溶解度进行了模拟。经过980次迭代得到了该模型对学习2样本的最小模拟误差为1.34%,对预测样本的模拟误差为8.89%,都小于活度系数模型和其它两种常用小波神经网络模型,此结果表明该模型是一种SCCO2中固体溶解度的较好模拟模型。

To simulate the solubility of solids in supercritical carbon dioxide （SC CO2） better, a optimized self-adapted Morlet wavelets neural networks （WNN）model with the constant of cosine r=0.1 and the partial momentum method was first introduced to simulate the solubility of 2 fatty acids in SC CO2 with the pressure P, the temperature T and the mol volume V, melting point Tm, fusing heat ΔH^fus, dispersion solubility parameter δ1, the polar and hydrogen-bonding solubility parameter δ2 of the solid using as the input parameters of the model. The least average absolute relative deviations （AARD） of the optimized model after 980 iterating times for the learning sots was 1.34%, and the AARD for the predicting sets was 8.89%, which were less than those of other models such as activity efficient model, the Morlet and Mexican hat WNN used usually. The results showed that the optimized model can simulate the solubility of solids in SC CO2 well.