This paper addresses the problem of maximizing the power produced by an organic rankine cycle (ORC) waste heat recovery system on board a diesel-electric railcar. A simplified model of the system allows the formulation of an optimal control problem that can be solved via dynamic programming (DP). To increase the smoothness and the accuracy of the solution obtained offline using the implementation of DP known as level-set DP, an improved version is developed, making use of adaptive grids for discretization. The analysis of the resulting optimal trajectory for the ORC control problem provides useful insight for both control design and system design. Based on these results, the optimal control problem is reformulated allowing online implementation via dynamic real-time optimization. The proposed approach is validated on a realistic simulator, showing significant benefits in the amount of recovered energy when compared with the classical, quasi-static approach found in ORC literature. (C) 2015 Elsevier Ltd. All rights reserved.