抄録
The chemical properties of BDFs were predicted by neural network analysis with the properties of the raw edible oils as the input learning characteristics. The model used in the analysis had a multi layer perception structure. 22 kinds of typical edible oils on the market were converted to BDFs using the transesterification of methanol with alkali catalyst method. To build the model of causal relation between raw materials and products, the measured properties of nine raw oils and BDFs were used as learning data. To verity the model, the properties of the other 13 BDFs were predicted from the properties of the raw oils. The kinematic viscosity, density, and cloud point of the BDFs were predicted well using suitable combinations of input characteristics of the raw oils. The accuracy of the predicted values depended on the combinations of the selected characteristics. However, the acid value of BDFs was poorly predicted.
