(607e) A Novel Bayesian Probabilistic Approach for Multi-Objective Optimization of Biofuel Processes Under Uncertainty

Though fossil fuels are the most commonly used energy sources, they are not renewable and are limited in supply. In addition, the combustions of fossil fuels can generate a range of pollutants into the atmosphere including carbon dioxide and sulphur dioxide, which have baneful influence on our environment and climate.  Therefore, alternative and renewable energy sources have been gaining growing attention worldwide. As one of the promising renewable energy sources and especially transportation fuels, biodiesel can be produced from a wide variety of renewable resources such as vegetable oils, animal fats, recycled restaurant greases, switch grasses, microalgae, etc. Meanwhile, it can be either used alone or blended with petroleum-based diesel products. Since there are various kinds of staring materials and different ways to carry out the transesterification, it is highly desirable to optimize biofuel process operation in order to obtain the best product yield, operating profit, energy efficiency, environmental sustainability, waste reduction and carbon emissions. 

In this work, a novel Bayesian inference based probabilistic approach is developed for multi-objective optimization of biofuel processes under inherent system uncertainty. There are various kinds of objective functions to be optimized, such as biofuel yields, material operating costs, energy and utility costs, biofuel production wastes, and total carbon footprint. Due to the operational uncertainty and complexity, the conditional probability density distributions of different objective functions given decision variables are first estimated from historical process data. Then the multiple objective functions are incorporated into a single objective function through posterior probability based weights. After estimating the posterior probabilities of the objective functions, a Bayesian inference based differential evolution (DE) strategy with respect to the estimated probability density functions is proposed to search for the best operating conditions with the optimal overall eco-benefit in terms of the combined profits, energy efficiency, material consumption, wastes, and carbon emission. A simulated biorefinery is built to demonstrate the effectiveness of the proposed approach in handling multiple objective functions and constraints towards the optimal solution under system uncertainty. Future research will extend the proposed approach to dynamic real-time optimization of biofuel processes.