Flow Optimization Failure and convergence for similar boundary conditions

[Katy S]

TLDR: I have boundary conditions 1cfs apart and one flow fails to optimize while the other converges. Why is this?

Model contains bridges, Lateral Structures (flow is optimized), 1D steady flow. Tests I’ve run explained in detail below.

_____________________________________________________________________________

I have curious problem pertaining to optimizing flow.

I’m running a 1D steady flow model over a stretch of river that contains bridges and lateral structures. It runs perfectly fine for all boundary conditions (I’m using known WS for downstream), with “Flow Optimization” unchecked for all structures, but the goal is to allow for overflow at the LS and when I include flow optimization, for some solutions I receive the message, “FLOW OPTIMIZATION FAILED TO CONVERGE.”

Now the curious part is that I can use two very similar upstream boundaries and have one fail and the other work fine. For example, an upstream condition of 6057 cfs fails to optimize, while an upstream condition of 6058 cfs works fine. Water surface profiles look very similar everywhere except over a particular stretch of the river before one of the bridges, where the WS from the converged profile is ~1ft higher than the WS from the failed conditions (see figure below, WS1 = failed run, WS2 = converged run). Modeling the condition 6059 cfs provides water surface profiles similar to 6058 (converges), while the condition 6056 cfs looks similar to 6057 cfs and does not converge.

I’ve read all the questions on this forum related to LS, weirs, flow optimization, blogs, any post I could find, and worked through the log outputs for conditions that worked well and didn’t (which I will admit I haven’t found great documentation for how to interpret). When I investigate the logfile, it seems that the condition that is failing to converge is oscillating back and forth between two estimates of Q for the LS for the remaining 30 iterations.

In the end, I’d like all of my modeled conditions to converge — I only chose these particular conditions because I was hoping to narrow down the issue with two conditions so close to each other. I am modeling a large range of conditions to create a set of lookup tables of WS profiles, rather than for example, a single deterministic condition like the 100-yr flow, so I can’t really be perfectly optimizing every run via changes to coefficients, etc.

Does anyone have any suggestions about why this might be happening and how to fix it? I’ve tried a few tricks — increasing the number of iterations, creating less strict tolerances, altering the weir coefficient, running with initialized flow estimates, and haven’t seemed to figure out the problem. Ideally I’d like to be fairly confident about the outputted water surface for any condition I end up running.

The Hec geometries/model was created and validated by someone else. The model was originally developed for unsteady runs and in v5.0.3.

Thanks!

[cameron]

The model starts trying to optimize with the initial split flow that you specify. If it doesn’t optimize with the first iteration, it tries different values. If you turn on the log file for the lateral structures, you will be able to see which values it is picking and what it is trying to converge to.

The reason why it works for one and not the next is how it is picking the next guess.

Based on your profile, is looks like the issue is around station 18000-19000 where the lateral structure is just barley overtops. The weir coefficient would probably be pretty low for this type of scenario to get it to be stable.

The things to fix it are as you have done, adjust weir coefficients, tolerances, or adjust the initial split flow values. You could also adjust the weir elevation slightly to see if it helps.

You may have to end up having different geometries where the only difference is the weir coefficient.

[Anonymous]

In the flow editor, you will find under options the split flow optimization. It should also be understood that the values can be retrieved for the last iteration the software tried to converge upon. By selecting the cell/row you can extract the last value. Sometimes this helps to find the sweet spot, by finding the number at which it failed and either going up or down depending on the full output

[Teresa]

Hello everyone, I have a similar question. I’m quite new to hec ras so, maybe, this is a silly question. I have a lateral structure with a tailwater connection type: cross sections of a river. I’ve noticed that the flow optimization fails to converge when the water surface elevation inside the final reach is higher than the critical depth over the lateral structure. How can I solve this problem?

[Author]

Posts