Control And Simulation Analysis Of Variable Geometry Turbocharger Based On DRL

Turbo has a problem of "turbo lag".Variable geometry turbocharger(VGT)is a good way to solve the problem.Meanwhile,the boost control of a VGT-equipped diesel engine is difficult mainly due to its strong coupling with Exhaust Gas Recirculation(EGR)system and large lag resulting from time delay and hysteresis between the input and output dynamics of engine's gas exchange system.Traditionally,VGT is controlled by PID,but PID control will have problems such as overshoot and poor followability.So the author applied deep reinforcement learning(DRL)algorithms which excels at solving a wide variety of Atari and board games to the control of VGT.And DRL is an area of machine learning that combines the deep learning approach and reinforcement learning(RL).However,there seem few studies that analyze the latest DRL algorithms on real-world powertrain control problems.In this context,one of the latest model-free DRL algorithm i.e.Deep Deterministic Policy Gradient(DDPG)algorithm is built in this paper to develop and finally form the strategy to track the target boost pressure under transient driving cycles.The specific research content is as follows:A simplified model of a one-dimensional supercharged engine is established in GT-Power software.In order to speed up the simulation calculation,the mean value engine model used this time,which main modules include intercooler module,mean value cylinder module,VGT module,EGR Module and accelerator pedal module.at the same time,a three-dimensional simplified engine model was established for transient analysis,and the mesh division was completed.At the same time,a simplified three-dimensional engine model was established for transient analysis,and the mesh division was completed.Python compatible with Tensorflow as the algorithm design language was applied in this study.Meanwhile,in order to apply the DRL algorithm built in Python to the diesel engine environment,it is proposed to use Matlab/Simulink as the program interface,so that the two-way transmission among Python,Matlab/Simulink and GT-Suite can be realized.So the experimental simulation platform is completed.The engine speed,the actual boost pressure,the target boost pressure and the current vane position are chosen to group a four-dimensional state space.And the vane position controlled by membrane vacuum actuator is selected as the control action in this DDPG algorithm.After the neural network training is completed,the control results of the PID strategy and the DDPG strategy are compared and analyzed.The results show that under the common driving cycle(FTP-72),the integral absolute error(IAE)values with the proposed algorithm in the first 80%(defined as the training segment)and the last 20%(defined as the inspection segment)are reduced by 10.28% and 17.90% respectively for the control performance and generality index,compared with a fine-tuned PID.Meanwhile,using a fine-tuned PID controller as a benchmark,the result shows that the control performance based on the proposed DDPG algorithm can realize a good transient control performance from scratch by autonomously learning the interaction with the environment,without relying on model supervision or complete environment models.In addition,the proposed strategy is able to adapt to the changing environment and hardware aging over time by adaptively tuning the algorithm in a self-learning manner on-line,making it attractive to real plant control problems whose system consistency may not be strictly guaranteed and whose environment may change over time.