search

georkara / Chargym-Charging-Station

Chargym simulates the operation of an electric vehicle charging station (EVCS) considering random EV arrivals and departures within a day. This is a generalised environment for charging/discharging EVs under various disturbances (weather conditions, pricing models, stochastic arrival-departure EV times and stochastic Battery State of Charge (BOC) at arrival). This is an open source OpenAI Gym environment for the implementation of Reinforcement Learning (RL), Rule-based approaches (RB) and Intelligent Control (IC).
87 stars 22 forks source link

Question about the results obtained in the training. #5

Closed RSanchez996 closed 3 months ago

RSanchez996 commented 9 months ago

Hi, I have been testing and retraining the neural network for a few days now, and I can't get the results to improve. The best result I have obtained with the mean reward with DDPG is -24. I have tried adjusting the hyperparameters as detailed in the paper, but the behavior is still not improving. Could you give me a little guideline to try to improve the training results and get my results closer to yours? They really seem to look very good! Great job!

160582221 commented 3 months ago

Hi，I have the same problem. Can you tell me how you solved it?

RSanchez996 commented 3 months ago

Hi, unfortunately, with the Stable_Baselines3 training algorithms I have not been able to improve the rewards. I still have the same problem.

160582221 commented 3 months ago

Hi, thank you for your response. I'm experiencing slow training speed while training the DDPG algorithm. Have you encountered a similar problem?

RSanchez996 commented 3 months ago

Hello, Yes, correct, it is an algorithm that takes more computational load. If you have a GPU with CUDA, you can configure the training algorithm to use the GPU instead of the CPU and speed up the training slightly.