Supercritical carbon dioxide extraction of Ledum palustre L. essential oils optimizing with uniform design and BP neural network

To obtain the optimal essential oils extraction conditions with supercritical carbon dioxide, uniform design and the artificial neural network (back propagation, BP) were applied in extraction of ledum palustre L. Number of neuron in hide layer and some functions for learning, training and transfer were chosen through one-factor experimental design. Four factors (six levels in each factor) were considered in uniform design. Five factors (extraction time, extraction temperature, extraction pressure, particle size, kind of leaves or stems) were considered in BP network. The average error of network prediction was 0.0116. Analysis of correlation and comparison showed that the experimental values and the predicted values were not significant. The effect of each factor on extraction yield and the optimal extraction conditions were studied with this model. The highest extraction yield of essential oils from ledum palustre L. stems could reach 1.82% at 17.5℃, 375 bar for 1 h while the essential oils extraction yield of ledum palustre L. leaves was 2.65% at 15℃, 275 bar for 3 h. It may provide a new study method for the extraction of volatile oil of ledum palustre L.