ITS-based bi-level online energy management strategy for extended-range fuel cell vehicles

The stochastic nature of real-world driving conditions poses a significant challenge to the energy efficiency of Range-Extended Fuel Cell Vehicles. Traditional rule-based strategies lack adaptability to dynamic traffic, while global optimization methods are often computationally prohibited for real-time applications. To address these issues, a bi-level online energy management strategy incorporating Intelligent Transportation Systems (ITS) is proposed. In the upper layer, a rolling horizon speed prediction model is established by integrating real-time traffic data from commercial map API with onboard GPS correction. Subsequently, a Learning Vector Quantization neural network is utilized to identify driving patterns from the predicted speed sequences. Based on the prediction, a mileage-domain Slow-declining SOC reference trajectory planning algorithm with dynamic slope correction is developed. This algorithm generates a trajectory that approximates the global optimum, which is validated against Dynamic Programming(DP) benchmarks. The lower layer employs a state machine with hysteresis control and dynamic power allocation to regulate the fuel cell system, ensuring precise tracking of the reference trajectory while maintaining high-efficiency operation. Evaluated through a “Performance Boundary” framework, the simulation results demonstrate that the proposed strategy achieves an 8.63% reduction in equivalent hydrogen consumption and increasing the average efficiency of the fuel cell system by 5.31% compared with the non-predictive CD-CS baseline. More importantly, by transforming highly nonlinear optimization into an <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mi>O</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>1</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:math> time complexity rule-based execution, the proposed framework closes a substantial performance gap toward the DP theoretical global optimum without the heavy computational burden of online iterative algorithms.