On-line suboptimal control strategies for a power-assist hybrid electric vehicle

Two on-line suboptimal control strategies for a developed power-assist hybrid electric vehicle (PAHEV) were conducted in this paper. One is a one-line time-independent optimization, while the other is a time-dependent optimization combined with the Neural-Network (NN) technique for the on-line implementation. The first method discretizes all variables including control inputs, disturbances, states, and then globally searches the best solutions (minimum value) according to a preset cost function. The second method can be separated into two phases. Phase 1 is the off-line optimization for specific driving cycles using Dynamic Programming (DP) theory. Phase 2 concludes the optimal results in phase 1 for the on-line control by the NN training, where the NN inputs are four driving pattern indices and the outputs are three polynomial coefficients derived from DP results. Simulation results show cases with different types of transmissions and cost functions. Results will be also compared with the experimental data of a prototype PAHEV, which consists of a 2.2L SI engine, an 18kW BLDC (Brush-less DC) motor and a 288V nickel metal hydride battery set.