search

DantEzio / Koopman_PINN_SWMM

Combine Koopman operator and PINN to establish simulation method/solver of pipeline model, and use it as surrogate model of SWMM
GNU General Public License v3.0
0 stars 0 forks source link

Dosen't work #1

Open Nieyuxi183 opened 3 weeks ago

Nieyuxi183 commented 3 weeks ago

When i used a new swmm model example,koopman_pinn_swmm didn't work effectively.Could you tell me how to adjust the parameters of a,q,h?

Figure_2

DantEzio commented 3 weeks ago

I meet a very similar situation just like this. I am still working on this recently and don't have any solution. I will let you know when I have some progress. Btw, can you provide your swmm inp and inflow data for me to reference? Appreciate!

Nieyuxi183 commented 3 weeks ago

I think it may be related to the input step size of q_input，a,q=q_input[0:100:2,0] revised to q_input[0:50:2,0],h=q_input[0:500:2,0] revised to q_input[0:100:2,0],This is related to q_input.shape, but that's not enough.Of course, I used a small inp example. This is the modified version of a, q, h.Also, do you have any relevant documentation for this algorithm? I have doubts about the formulas involved

bbh @.***

 

------------------ 原始邮件 ------------------ 发件人: "DantEzio/Koopman_PINN_SWMM" @.>; 发送时间: 2024年8月19日(星期一) 下午3:42 @.>; @.**@.>; 主题: Re: [DantEzio/Koopman_PINN_SWMM] Dosen't work (Issue #1)

I meet a very similar situation just like this. I am still working on this recently and don't have any solution. I will let you know when I have some progress. Btw, can you provide your swmm inp and inflow data for me to reference? Appreciate!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

DantEzio commented 3 weeks ago

This is a new algorithm I just worked on, no document at all. Since this problem bothers me for almost 4 months, I am really thinking that maybe I should close this repository until this problem is fixed. Otherwises these codes are nothing but misleading. Also, I couldn't find the inp and 'modified version of a, q' in your comment and email. Could upload your data to your own GitHub? Then I can download them and use them to fix this new algorithm.

Nieyuxi183 commented 3 weeks ago

This is a new algorithm I just worked on, no document at all. Since this problem bothers me for almost 4 months, I am really thinking that maybe I should close this repository until this problem is fixed. Otherwises these codes are nothing but misleading. Also, I couldn't find the inp and 'modified version of a, q' in your comment and email. Could upload your data to your own GitHub? Then I can download them and use them to fix this new algorithm. Don't give up!! D32D0A63@B01F6877 B402C466 616A3562@8F924669 B402C466

Nieyuxi183 commented 3 weeks ago

I don't know how to upload files on github. Do you have email or wechat? I think h is more stable than a and q, and it is easier to learn about PINN. However, the q fluctuates too much, which can easily generate noise. Therefore, it is not appropriate to study the h of the node and the q of the link together.

DantEzio commented 3 weeks ago

And it makes sense, maybe we should only focus on h?