Energy Management Strategy Based On Deep Model Predictive Control For Plug-in Hybrid Electric Bus

In recent years,due to continuous deterioration of the urban environment,frequent haze,and increasing consumption of petroleum resources,developing new energy vehicles became an important topic at major conferences.Pure electric vehicles have not been widely used because of limited battery life.However,the plug-in hybrid vehicle has achieved a good balance between energy consumption and driving mileage,and has drawn a lot of research attention.This thesis proposes a short-term driving cycle prediction model based on deep learning and multi-source information fusion.Plug-in Hybrid Electric Bus(PHEB)is employed as the research object,and the predictive model is applied to model predictive control(MPC)based energy management strategy(EMS).Then,an optimized EMS named DRN-MPC is proposed.The proposed strategy is compared with traditional MPC strategies and the fuel economy experimental results show that DRN-MPC outperforms others by 3.32%.The main work includes:(1)Acquiring dynamic parameters of the Yutong Series-parallel hybrid bus,the parameter matching is analysed and the theoretical derivation results show that the designed parameters meet the vehicle performance requirement.The backward simulation model of the PHEB was constructed with the designed parameters under Matlab.(2)Designing two types of deep learning predictive models.One is a deep neural network(DNN)based on historical driving cycle information,called DNN,and the other is based on video information fusion and deep residual network(DRN),called DNN-res.On the China Transit Bus Driving Cycle(CTBDC)data set,the prediction results of DNN are compared with traditional prediction models such as Markov Chian Monte Carlo(MCMC)and Back propagation Neural Networks(BP).On the Common.AI public dataset,the results of the Deep Residual Network model DNN-res of fusing video information were compared with those of other models.The experimental results show that the DNN outperform traditional method by 35.1%,while the DNN-res further improves by 24.3% over DNN without using video information.(3)Applying the DNN prediction model to deep model predictive control(DMPC)framework ending with a novel EMS DNN-MPC based on historical driving cycle information.On the CTBDC,this thesis compares the fuel consumption of four energy management strategies and discusses the impact of different forecasting time horizon on the fuel economy and real-time performance of the strategies.The experimental results show that the DNN-MPC strategy is 2.82% higher than SMPC,and the 10 s prediction time horizon achieves a good balance between real-time performance of energy distribution and fuel consumption.(4)Applying the DNN-res prediction model to DMPC framework ending with a novel EMS DRN-MPC based on fusion information.On the Common.AI data set,this thesis compares the energy-saving effects of various energy management strategies.The DRN-MPC strategy is 3.32% higher than traditional strategy SMPC.