Model predictive control (MPC) based approaches are being increasingly employed for solving the Energy Management (EM) problem of Plug-In Hybrid Electric Vehicles (PHEV). Model uncertainty, un-modelled disturbances and uncertainties due to driver behavior in real-world driving necessitate approaches that extend nominal MPC. This paper proposes a MPC approach modeling the future power demand using Markov Chain (MC) and employing an Explicit MPC (EMPC) algorithm to solve the resulting finite-time horizon constrained optimal control problem. The stochastic power demand modeling approach does not require a-priori driving cycle knowledge explicitly. Further the MC transition probability matrices can be learnt and reconfigured to accommodate changing driving characteristics. EMPC based on multi-parametric programming ensures reduced computational complexity and provides guaranteed stability. The proposed method has been implemented for the EM of a PHEV powertrain with a downsized 2-cylinder combustion engine at an Engine-In-The-Loop (EiL) testbed. The results from the simulation study and EiL implementation show performance close to implicit or online MPC with knowledge of the future power request in standardized and representative real-world driving scenarios.