Intermittent “No model solution” after restarting Julia when using time-domain reduction in a myopic multistage model

[huangzhenyue61-tech]

I am running a myopic multistage GenX model with time-domain reduction (TDR). I have encountered an intermittent issue: runs based on the same original input data and the same GenX, and solver settings do not always produce consistent results.

For example, the model solved successfully in two runs. However, after I closed Julia, reopened it, and ran the model again, Stage 1 returned No model solution.

Before each run, I deleted all previously generated TDR output files, so the time-domain reduction procedure was rerun from scratch every time.

GenX then continued to Stage 2 and raised a MathOptInterface.ResultIndexBoundsError when attempting to access variable values from the Stage 1 model, which had no available primal solution.

I would like to understand whether this behavior could be caused by:

- randomness or non-determinism in the TDR procedure;
- numerical instability in the solver;
- differences in the representative periods generated across runs;
- differences in model or random-number-generator state across Julia sessions;
- or an issue in the multistage workflow when the previous stage has no primal solution.

Environment

• GenX version: v0.4.6
• Julia version: 1.9.4
• Solver: Gurobi
• Operating system: Windows 11
• Multistage mode: Myopic
• Time-domain reduction: Enabled
• Number of stages: 4

Gurobi Setting:

Feasib_Tol: 1.0e-05
Optimal_Tol: 1e-5
TimeLimit: 110000
Pre_Solve: 1
Method: 2
#Gurobi-specific solver settings
MIPGap: 1e-3
BarConvTol: 1.0e-06NumericFocus: 0.
Crossover: 0
PreDual: 0
AggFill: -1
OutputFlag: 0

Resource summary:

Resource type	Number of resources
Thermal	878
VRE	200
Hydro	31
Storage	54
Total	1163

Thank you for your time. Any guidance on reproducing, diagnosing, or resolving this behavior would be greatly appreciated.

[mohammadrezwankhan]

[huangzhenyue61-tech]

Thank you for explaining the GenX model functionality and the multistage workflow. However, what I would most like to understand is the root cause of the “No model solution” message in Stage 1.

Does this indicate that my input data or model settings have created conflicting constraints or made the model infeasible? Could it instead be caused by numerical issues in the solver, differences in the TDR results, or some other problem?

How should I further diagnose and troubleshoot this issue? For example, how can I inspect the model’s termination status, compute conflicting constraints or an IIS, and identify the specific input data or constraints responsible for the solve failure?

[huangzhenyue61-tech]

Thank you. I captured the actual solver status, and it shows:
termination_status(EP) = MathOptInterface.NUMERICAL_ERROR
primal_status(EP) = MathOptInterface.NO_SOLUTION
dual_status(EP) = MathOptInterface.NO_SOLUTION
raw_status(EP) = "Optimization was terminated due to unrecoverable numerical difficulties."

Does this mean that the optimization failed because of numerical issues? Could the problem be caused by my model inputs? How should I diagnose and troubleshoot it further?