Research On Medium-Speed Marine Diesel Engine Speed Control Strategy Based On Deep Reinforcement Learning

The development of shipbuilding industry is an indispensable part of the national equipment manufacturing industry.As the core of ship propulsion,the stability and reliability of diesel engine are very important.The basic problem of marine diesel engine is the stable control of speed.However,in the actual operation process of diesel engine,it is inevitable to encounter inappropriate control parameters and sudden loading and unloading of load,which will lead to frequent abnormal fluctuations in speed,and also lead to galloping in bad conditions.Therefore,effective control of speed instability of marine diesel engine is of great significance to improve its stability.In this paper,a self-tuning control algorithm based on deep reinforcement learning is designed for the speed control of a marine diesel engine in steady state,sudden changes in steady state load and transient conditions,taking a four-stroke medium-speed diesel engine of China’s own brand as the controlled object.Firstly,a simulation model is established based on the parameters of the controlled object and its accuracy is verified.The control algorithm adopts deep reinforcement learning theory to update the PI control parameters in real time,and collects the diesel engine operating parameters to update the state action value function online during the operation of the diesel engine to realise the online learning of the algorithm.The tacho control algorithm is then designed according to the algorithm and the characteristics of the controlled object,and the steady-state response,the sudden change in steady-state load response and the transient response of the tacho control algorithm are verified in simulation experiments.Finally,the hardware selection and software design of the rapid control prototype and hardwarein-the-loop simulation system are carried out to match the designed control strategy,and the semi-physical simulation experiments of the tacho control strategy are completed.After simulation and experimental verification,the results show that,compared with the traditional PI controller,the speed control strategy designed in this paper for marine mediumspeed engines can adjust the PI control parameters in real time when the control parameters are inappropriate in steady-state operation and smooth out the speed with an average of three modifications;make the overshoot peak drop by 46% on average when the ship is subject to sudden loading and unloading,and the system return speed increase by 38% on average.When the ship is subjected to sudden changes in control parameters during the speed change process,the speed deviation can be reduced by an average of 57.5%,allowing it to track to the desired speed as soon as possible.