| Optimization problems are frequently solved in chemical engineering to aid in the decision-making processes. Unfortunately, there are many cases where the problems cannot be solved to optimality. Since there are situations where the optimization problems cannot be solved, it would be helpful to have a tool that is capable of accurately predicting the optimization results.; For this project, machine learning techniques have been used to examine the results of solving numerous representative problem cases. Through this examination, it is often possible to extract trends that describe the optimal solution in terms of the problem's input information. The goal of this research was to apply machine learning techniques to improve chemical engineering problem-solving techniques. To reach this goal, three objectives needed to be met: (1) apply machine learning techniques to find trends in the solutions to chemical engineering problems; (2) determine whether or not the trends obtained from the training instances can be used to effectively classify future instances; and (3) use the machine learning results to generate problem solutions.; The primary contributions of this thesis are: (1) the analysis and development of a mathematical model for a novel scheduling problem; (2) the application of machine learning techniques—including decision-tree learning, nearest-neighbor classification, Winnow, and a variety of “wrapper” methods—to three chemical engineering problems that had not previously been examined in this manner; (3) the development of rules for the waste-volume reduction problem that could be applied without solving the original mathematical programming model; and (4) the development of a novel heuristic problem solving method for state-task network scheduling problems. |
