siemens / powergym Goto Github PK

Thanks for this PowerGym provided by the author. This is very useful to me. But as a novice in programming, there are still some difficulties to understand. Can the author upload a simple supporting training algorithm to help me learn？

Thank you very much if you can reply.

Sorry, I have some problems again.
I noticed that you wrote 'the load profiles are randomly partitioned into two haves, one for training and the other for testing.'For example, there are only 24 data in powergym master \ systems \ 13bus \ loadshape \ 000 \ 611.csv. Even if there are 73 folders like this, the amount of data is not enough. So I'd like to ask how your training data is composed? What the time scale about the load data?
Thank you very much for your reply.

I have just gone through the paper "PowerGym: A Reinforcement Learning Environment for Volt-Var Control in Power Distribution Systems". I have found it insightful and thanks for sharing this repository. I have understood that the file random_agent.py performs random actions but I was also looking for agents trained by Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) and the training implementation (as mentioned in the paper) codes. Do I need to code those or other reinforcement learning by myself using the action space and observation space mentioned at env.py ? Or I have missed those training codes?

An error occurred while running random_agent.py ,as

Traceback (most recent call last):
File "E:/powergym-master/random_agent.py", line 225, in
run_random_agent(args, worker_idx=None, use_plot=args.use_plot, print_step=False)
File "E:/powergym-master/random_agent.py", line 53, in run_random_agent
env = make_env(args.env_name, worker_idx=worker_idx)
File "E:\powergym-master\powergym\env_register.py", line 287, in make_env
return Env(folder_path, base_info, dss_act)
File "E:\powergym-master\powergym\env.py", line 240, in init
self.load_profile = LoadProfile(
File "E:\powergym-master\powergym\loadprofile.py", line 24, in init
for f in os.listdir(self.loadshape_path):
FileNotFoundError: [WinError 3] 系统找不到指定的路径。: 'E:\powergym-master\systems\13Bus\loadshape'

How to solve this problem?

I was going through the paper "PowerGym: A Reinforcement Learning Environment for Volt-Var Control in Power Distribution Systems" again! The reinforcement learning allows to enable online training which is an advantage of the model I guess. But I was wondering what would be the solution if we add or remove components like bus, voltage regulator, capacitor bank to the trained power distribution system.
My Idea: Say there is an addition of load bus. We will simulate it's active power and reactive power and concatenate this data to the historical data of the remaining distribution system. Then train the model with additional load bus. We can hope that this would not effect the overall performance badly. Such simulation can be made for the controller devices also, like we consider to had that capacitor bank at the historical data and start training. While training the AI will turn the switch on or off the load bus. And finally we will learn to operate such devices.

Kindly, tell me if my idea is promising or not. It would be helpful if you please share your thoughts regarding this issue.