Coupled Control And Simulation Analysis Of Engine Throttle And Supercharger Based On DRL

The intensification of the energy crisis has forced innovation in various fields.The engine industry is subject to strict regulations.New engine technologies must be adopted to improve engine fuel economy.Among them,turbocharged engine technology is widely used.With the same engine displacement,the compressor can increase the intake air volume to increase the engine torque and power,and increase the engine's power by 40% under the same fuel economy.Due to the driving conditions of household cars at low /medium loads,manufacturers generally match the turbine to a low speed to ensure the torque output of the engine at low speed,and use a wastegate valve to deflate when high speed and large load are used.Bypass excessive exhaust gas and reduce the boost pressure of the compressor.Because the hysteresis characteristic of the turbine seriously affects its control,the engine cannot maintain the optimal working state in each transient state,thereby reducing the engine efficiency.This article uses the Deep Reinforcement Learning(DRL)algorithm for engine control.As both the turbocharger and the throttle are the direct factors affecting the intake air volume,this study focuses on the supercharger and the throttle.In terms of strategy,it imitates the adaptive algorithm in artificial intelligence.It tries to replace artificial calibration of engine supercharger and throttle with AI,and combines engine technology and artificial intelligence technology to achieve the purpose of controlling throttle and supercharger.First,the one-dimensional control model of the engine is established in GT-Power,and it meets the accuracy requirements of simulation through certain error verification.Then the throttle controller and turbocharger controller are removed based on the original model,and this model is used as a joint simulation the environment of the platform.The speed,transmission ratio,fuel consumption,and target vehicle speed in the engine properties are used as the state values of the algorithm.The program's return action is the action value of the throttle valve and the waste bypass valve.The criterion for judging the pros and cons of the strategy is the power following and the fuel consumption.And the lowest.The joint simulation platform is used for training,and the final strategy is loaded into the complete FTP75 operating condition for verification.The results show that the deep reinforcement learning algorithm can complete the power following of the operating condition by coupling the throttle and the turbocharger,and the fuel consumption rate reduced by 6.4%.Finally,the simulation data was imported into STAR-CCM + for three-dimensional simulation.The flow field information of the cylinder and the intake and exhaust passages were analyzed intuitively.It was proved that in the coupled control,the throttle and the supercharger can adjust their own action values through synergy,which can effectively avoid the interference of the two,so as to reduce the exhaust back pressure and reduce the fuel consumption rate while ensuring the engine torque output.This study builds a one-dimensional joint engine deep learning engine platform through GT-Power,Matlab,and Python.It conducts preliminary exploratory research on artificial intelligence algorithms in the field of throttle and supercharger coupling control,and under specific operating conditions the feasibility of the algorithm is verified.