燃料电池混合动力汽车深度强化学习能量管理优化.pdfVIP

燃燃燃料料料电电电池池池混混混合合合动动动力力力汽汽汽车车车深深深度度度强强强化化化学学学习习习能能能量量量管管管理理理优优优化化化

摘要:针对配备有锂电池与超级电容的燃料电池混合动力汽车,为降低车辆总体运行成本,延长能量源寿命,本

文提出一种基于深度强化学习的能量管理策略.首先,依据超级电容高功率密度特性,建立基于模糊自适应滤波的

功率分层结构,并依据燃料电池与锂电池的经验退化模型,建立能量源退化的成本函数,采用等效消耗最小策略平

衡氢耗成本与能量源退化成本,以最小化总体运行成本为目标来优化能量源功率分配;然后,引入优先经验回放与

软更新以提高深度强化学习的离线训练效率;最后,在多种工况下进行仿真,结果表明,与未考虑退化的策略相比,

本文所提出策略在全球统一轻型车辆测试循环下可使氢耗量降低11.8%,并可有效减缓燃料电池与锂电池的退化

速率,降低燃料电池混合动力汽车的总体运行成本.

关键词:燃料电池混合动力汽车;深度强化学习;能量源退化;等效消耗最小;数据驱动;能量管理策略

Energymanagementoptimizationoffuelcellhybridelectricvehicle

basedondeepreinforcementlearning

Abstract:Forthefuelcellhybridelectricvehicleequippedwithlithiumbatteryandultracapacitor,toreducetheoverall

operationcostandprolongthelifespanofenergysources,anenergymanagementstrategybasedondeepreinforcement

learningisproposedinthispaper.Firstly,accordingtothehighpowerdensitycharacteristicsofultracapacitor,apower

hierarchicalstructurebasedonafuzzyadaptivefilterisestablished,andbasedontheempiricaldegradationmodelof

fuelcellandlithiumbattery,thecostfunctionofenergysourcedegradationisestablished.Theequivalentconsumption

minimumstrategyisusedtobalancethehydrogenconsumptioncostandenergysourcedegradationcost,andthepower

allocationofenergysourcesisoptimizedtominimizetheoveralloperationcost.Then,prioritizedexperiencereplayandsoft

updateareintroducedtoimprovetheoff-linetrainingefficiencyofdeepreinforcementlearning.Finally,thesimulationis

carriedoutundervariousdrivingcycles.Theresultsshowthatcomparedwiththestrategywithoutconsideringdegradation,

theproposedstrategyreduceshydrogenconsumptionby11.8%undertheworldlightvehicletestcycle,andcaneffectively

slowdownthedegradationrateoffuelcellandlithi