The optimal design of large-scale energy systems can be found by posing the problem as an integrated multi-period planning and scheduling mathematical programming problem. Due to the complexity of the accompanying mathematical programming problem decomposition techniques are often required but they to are plagued with converge issues. To address these issues we have derived a set of valid inequalities to strengthen the formulation, and we have developed a machine learning framework to identify which subproblems should be solved in each iteration of the decomposition procedure. We illustrate the effectiveness of the valid inequalities and the machine learning framework through the use of a case study wherein we examine the development of an industrial scale hydrogen-based energy system. The results show that when the valid inequalities are applied a priori and the machine learning framework is employed to identify candidate subproblems the computational burden of the procedure is significantly reduced.