Traffic-Signal-Control-master
所属分类:其他
开发工具:Python
文件大小:102KB
下载次数:9
上传日期:2020-04-14 11:15:57
上 传 者:
寇大哥哥
说明: 附有论文,交通信号灯识别源程序。采DDPG深度强化学习方法。给出了LOSS损失函数图像
(With the paper, traffic signal recognition source program.)
文件列表:
.idea (0, 2018-09-27)
.idea\Traffic Signal Control.iml (455, 2018-09-27)
.idea\misc.xml (210, 2018-09-27)
.idea\modules.xml (296, 2018-09-27)
.idea\vcs.xml (180, 2018-09-27)
.idea\workspace.xml (20638, 2018-09-27)
Agent (0, 2018-09-27)
Agent\__pycache__ (0, 2018-09-27)
Agent\__pycache__\dqn_nature.cpython-35.pyc (6057, 2018-09-27)
Agent\__pycache__\dqn_nips.cpython-35.pyc (6312, 2018-09-27)
Agent\__pycache__\replay_memory.cpython-35.pyc (1482, 2018-09-27)
Agent\__pycache__\tf_utils.cpython-35.pyc (2160, 2018-09-27)
Agent\dqn_nature.py (5961, 2018-09-27)
Agent\dqn_nips.py (5195, 2018-09-27)
Agent\networks.py (10675, 2018-09-27)
Agent\pg.py (3785, 2018-09-27)
Agent\replay_memory.py (932, 2018-09-27)
Agent\tf_utils.py (1706, 2018-09-27)
dqn.inp (48738, 2018-09-27)
images (0, 2018-09-27)
images\Q_Mix Q.png (47628, 2018-09-27)
images\T_Mix Q.png (21546, 2018-09-27)
pg_control.py (1730, 2018-09-27)
train_vis.py (2461, 2018-09-27)
vis_env.py (9122, 2018-09-27)
## Intelligent Traffic Signal Control
This Project is a traffic control system based on DQN [(arxiv:1312.5602)](https://arxiv.org/abs/1312.5602) on [Vissim](http://vision-traffic.ptvgroup.com/en-us/products/ptv-vissim/). It's an original implement that intelligent traffic signal control via deep reinforment learning on partial urban traffic net. Choosing fine hyper-parameters, agent could learn to how to improve the performance of global net in a long term.
#### Dependencies
- Vissim 4.3.0
- Python 3.5
- Tensorflow 1.2.0
- other common packages like pandas numpy matplotlib pywin32
- ...
#### About Vissim
VisEnv.py wrapped the orignal api into the open.ai style. For now, speed, travel time, queued vehicles count interfaces are provided.
Use this like:
``` python
fron vis_env import *
env = VisEnv()
...
for epi in range(episodes):
env.reset()
env.test = True
for _ in range(steps):
next_state, reward, done = env.step(action)
env.write_summary(epi, dir)
```
#### Experiments
The performance of DQN is not so good among the series of reinforcement learning algorithm, but agent are still capble to act appropriately in our traffic enviroment.
![Queue](https://github.com/Linging/Traffic-Signal-Control/blob/master/images/Q_Mix%20Q.png)
![Travel Time](https://github.com/Linging/Traffic-Signal-Control/blob/master/images/T_Mix%20Q.png)
#### TODO
More reinforment learning models like dueling-DQN, DDPG, to further improve the performance of agent, and to solve the large discrete actions space problem.
Intelligent Traffic Signal Control
近期下载者:
相关文件:
收藏者: