强化学习用于PID的参数整定和电流补偿控制。

Introduce

This is the implementation of the paper “Control Strategy of Speed Servo Systems Based on Deep Reinforcement Learning”.

For more paper information, please checkout the the paper Link.

Class PID

The class PID algorithms has no training process.

python run.py --choose_model class_pid --curve_type trapezoidal --height 1000 --run_type test

Plot the result:

python plot_result.py --choose_model class_pid --curve_type trapezoidal --height 1000 --run_type test

Parameters Adjustment Of PID

Search the parameter of PID Based DDPG Algorithm.

python run.py --choose_model search_pid_parameter --curve_type trapezoidal --height 1000 --run_type train

experimental operation process:

Electric Current Compensation Of RL-PID

python run.py --choose_model search_electric --curve_type trapezoidal --height 1000 --run_type train

test

python run.py --choose_model search_electric --curve_type trapezoidal --height 1000 --run_type test

Plot the result:

python plot_result.py --choose_model search_electric --curve_type trapezoidal --height 1000 --run_type train

the result show:

Dependencies

The code was tested under Ubuntu 16 and uses these packages:

• tensorflow-gpu==1.14.0
• atari-py==0.2.6
• gym==0.17.3
• numpy==1.91.3

more packages described in requirements.txt

Citing

If you find this open source release useful, please reference in your paper:

Chen P, He Z, Chen C, et al. (2018). Control strategy of speed servo systems based on deep reinforcement learning[J]. Algorithms, 2018, 11(5): 65..