自适应循环发动机性能智能在线寻优算法研究

Online Intelligent Optimization Algorithm for Adaptive Cycle Engine Performance

关键词:  航空发动机  自适应循环  控制规律  性能寻优控制  强化学习  深度确定性策略

Key words:  Aero engine  Adaptive cycle  Control law  Performance seeking control  Reinforcement learning  Deep geterministic policy gradient

摘要:

为解决自适应循环发动机性能在线自主优化问题，提出一种适用于三流道自适应循环发动机的智能自主优化控制方法，通过深度确定性策略梯度算法（DDPG）在线优化压比计划，实现控制规律自主寻优。在基准发动机特性图上给出了不同外涵面积下的等推力特性曲线，在此基础上给出了基准的最低油耗控制规律曲线。当发动机性能退化或存在个体差异等偏离时，基准控制规律不再能使发动机性能最优，DDPG算法利用前期存储的数据对压比指令修正量进行训练学习，自主调整发动机控制规律。整机数值仿真结果表明，调整后亚声速巡航单位耗油率降低7.63%，与最低油耗点比较偏差0.03%。超声速巡航单位耗油率降低5.04%，与最低油耗点比较偏差0.01%。性能寻优算法可以在发动机性能偏离情况下实现发动机控制规律自主调节，达到最低油耗。

Abstract:

In order to solve the on-line performance optimization problem of three-stream adaptive cycle engine， an intelligent performance optimization control method was developed. This method used deep deterministic policy gradient （DDPG） algorithm to adjust engine pressure ratio control law online. In this paper， the equal thrust characteristic curves under different external culvert areas were given on the baseline engine characteristic chart. The minimum specific fuel consumption （SFC） control law curve was marked on the above chart. When performance degradation or characteristic deviation of the engine occurred， the baseline control law would no longer be optimal. Then the DDPG algorithm used the data stored in the previous period to train pressure ratio instruction and adjust the engine control law autonomously. The simulation results showed that， for the subsonic cruise， the SFC is decreased by 7.63%，the error between the adjusted SFC and the lowest SFC point is 0.03%. For the supersonic cruise， the SFC is decreased by 5.04% and the error between the adjusted SFC and the lowest SFC point is 0.01%. The intelligent performance optimization control method can optimize the engine control law online and decrease the SFC when the deviation of engine characteristic occurs.