Prediction of Operational Strategy for Plant Cell Suspension Culture by Neural Network

• Main Authors: Yu Chi-Ming, 余齊明
• Other Authors: 黃世佑
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/01383783623790850133

碩士 === 國立臺灣大學 === 化學工程學研究所 === 87 === Artificial Neural networks were employed to simulate the cell growth and secondary metabolite production of plant cell culture system. Using experimental data of plant suspension cell culture, the optimal operation condition and the parameters for scaling up the bioreactor were investigated. Feed forward back propagation artificial networks was adopted for network structure. One or two hidden layers which were connected with the complexity of the input data was tried. Sigmoid type transfer function was implemented to the hidden layer while pure linear transfer function was used in output layer. Training a network was carried out with Levenberg-Marquardt method which exhibited a rapid and accurate performances. To train the network, adjustment of the numbers of artificial neurons in the hidden layer was necessary, thus avoiding the overfitting and underfitting .With inoculum density, rotational speed of agitator, cultivation period as input values and dry cell weight and L-DOPA content as output values, performance of proposed neural network was justified. Prediction of dry cell weight was satisfactory because of its simple physiological reaction. While the prediction of L-DOPA production was difficult due to its physiological complexity. Eddy length scale was employed to simulate the cultivation system. The result was not as good as predicted. It may be attributed to the shear stress characteristics of an agitated bioreactor. Extrapolation went successfully as the learning was abundant and covered almost all the characteristic data. Mixing time of bioreactor with different impellers were determined by neutralization of acid and alkali in a reactor. Gate turbine impeller exhibited the best mixing performance. The disk turbine showed longer mixing time, and the flat-blade turbine exhibited longest mixing time. There were no distinct dead zones for three impellers. The effect of liquid viscosity on mixing time was remarkable. This must be an important index for designing and scaling-up a bioreactor.