A Novel Adaptive PMP-Based Energy Management Strategy for Fuel Cell Vehicle Based on Short-Term Driving Condition and Historical States of Power System

Energy management strategy (EMS) based on vehicle speed prediction has been widely used in fuel cell vehicles (FCVs). Actually, not only the actual power demand but also other factors affect the optimal power allocation between fuel cell system (FCS) and battery. However, this relationship is difficult to express in formulas especially under urban conditions because the power demand fluctuates greatly under the above conditions. To address the issue, a novel EMS for FCV based on short-term power demand and FCS output power is proposed. In the offline part, the short-term SOC change rate is used to characterize short-term power allocation. Besides, the average of short-term power demand and the FCS output power are selected as input factors. The feedforward neural network is used to learn the relationship of the above three state variables based on historical driving cycles. In the online part, a long short-term memory (LSTM) network is used to predict the short-term speed based on the vehicle historical motion states, then the power demand sequence can be calculated using dynamic model. Based on the short-term power demand and FCS output power, the well-trained feedforward network predicts the short-term optimal SOC change rate, then the short-term reference SOC can be obtained, which is used to find the optimal co-state of Pontryagin's minimum principle (PMP) for each horizon. In the local level, shooting method is used to determine the co-state for each horizon. The results show that LSTM network meets the require of short-term speed prediction. In addition, the proposed EMS makes the terminal SOC close to the target value and reduces the total cost by 12.14% compared with the benchmark strategy.