Combining Machine Learning with Mixed Integer Linear Programming in Solving Complex Scheduling Problems

With the increasing digitalization of industrial production processes and the quest for maximizing the synergies through more integrated operations, there is an increasing need also to automatize the decision making. In terms of scheduling, problems are becoming larger and need to consider more aspects making both the modeling and the solution of the resulting problems cumbersome. Suitable methods to deal with these problems include, e.g., simplifying the problem as necessary to speed up the optimization (i.e., balancing the optimality and solution speed where possible), using heuristics to support faster solution, deploying simulation tools to predict the values of most complex variables, using decomposition methods to divide the problem into smaller subproblems, and a rich mixture of all of the above. This paper discusses various approaches to support optimization by using machine learning and related challenges in implementing them.