Optimal Condition Prediction of Biodiesel Production using Neural Network and Simulated Annealing Techniques

Abstract

In this article, an implementation of the data driven technique has been applied for optimal condition prediction of biodiesel synthesis. It is known that price of biodiesel still has not been competed with petroleum diesel due to the efficiency synthesis limitation. Although, statistical design of experiment can be applied for optimising synthesis condition and minimising cost of investigation, it still consumes time and cost of investigation. Therefore, this article proposes a guideline for determining optimal synthesis condition using data driven technique. As there are many synthesis conditions reported in scientific journals and books, these data can be used as raw data for supervised learning. In here, 80 synthesis conditions and %yield of synthesis from the literature were assigned as input and output data, respectively. The data were trained by using neural network which was selected as supervised learning technique. The model obtained from the learning had the highest error of 2.44% when validated by eight new data synthesis conditions predictions. The learning model is then used to predict and find the synthesis conditions that are expected to produce the highest %yield with optimisation techniques. It was found that by searching for the expected synthesis conditions, the maximum %yield would not exceed the maximum value used in training machine due to the searching conditions were only within the domain of the training data. Therefore, if the search is to be opted out of the practice data, a method must be found to make the search outside the trained data. Such a problem is a challenging one for further research.