search

emad-arezoomand / online-actor-critic-algorithm-to-solve-continues-time-infinite-horizon-optimal-control-problem

online actor critic algorithm to solve the continues time infinite horizon optimal control problem
7 stars 0 forks source link

online-actor-critic-neural-net-optimal-controler problem #1

Open HappyTiger1 opened 3 years ago

HappyTiger1 commented 3 years ago

When the learning rate or initial critic weights change, why the critic weights converge to different values? How to justify its optimality?

emad-arezoomand commented 3 years ago

notice the difference between w2change and actual formula in the paper. the convergence is proved in the paper via lyapunov analysis.

HappyTiger1 commented 3 years ago

Thank you for your answer! In the paper, the linear model has a determinstic critic weights value [1.4279 1.1612 +0.1366 1.4462 +0.1480 0.4317], but through the algorithm I can't get that. How to explain this?

HappyTiger1 commented 3 years ago

I find the answer, the persistence excitation should be added.

oujieww commented 3 years ago

@HappyTiger1 do you know how to add persistence excitation for this code， and could you share me your code？

emad-arezoomand commented 3 years ago

i updated the code , this algorithm does not yield the promised results.

youwantpeach commented 3 years ago

Have you seen this articel? I have a problem. The weights of the actor NN is W2, and how to choose the value of F1 in Theorem2. I think this paper is wrong.