In order to identify drivers' intentions, and improve the fuel economy of plug-in 4WD hybrid electric vehicles (HEVs) with their power performances guaranteed, a method for calculating the torque identification coefficient was put forward. An energy management control strategy was designed based on the engine optimal control, and the judging condition of every working mode as well as its torque distribution method was introduced. In order to avoid the inherent defects of a single optimization algorithm that the calculation time was long and it was easy to end up with a local optimal solution, design of experiment (DOE) was conducted by the method of optimal Latin-hypercube design. An approximate model was designed using the RBF (radial basis function) neural network and then optimized using the multi-island genetic algorithm. In addition, an off-line simulation was conducted to verify the optimized control strategy. The results showed that the optimized control strategy could reduce the fuel consumption per 100 km of the plug-in hybrid electric vehicle by 16.4%, without degrading the power performances. After optimization, the control strategy is validated by a hardware-in-the-loop test on dSPACE and the experimental results show that the control strategy can realize the basic energy management. What's more, with torque identification, the average velocity error is reduced by 39.9% and the fuel consumption is reduced by 8.5%.