convergence-failure-mitigation

Name: convergence-failure-mitigation
Author: Power-Agent

56stars12forks

How to install convergence-failure-mitigation?

power-agent/powerskills/convergence-failure-mitigation

$npx -y skills add Power-Agent/PowerSkills --skill convergence-failure-mitigation

Installs into the current project.

Summary

Senior power-engineer playbook for power-flow convergence failures. Use whenever a base case or post-change case diverges, oscillates, or fails to solve in PSS/E, PSLF, PowerWorld, pandapower, PyPSA, surge, or OpenDSS. Triggers on "power flow diverged", "case won't solve", "did n

Files · 1

View on GitHub

SKILL.md

1	# Convergence failure mitigation
2
3	Start from the failed solve and its mismatch report. The goal is to learn why it failed, not just to force a solution.
4
5	## Preferred action order
6	1. Rule out data problems first: isolated or de-energized buses, zero or absurd branch impedances, missing or misplaced swing bus, transformer tap and phase-angle data, load/generation totals that cannot balance.
7	2. Check islanding and slack coverage: every energized island needs a slack with feasible MW. Use the tool's island or topology report before touching solver settings.
8	3. Simplify the solve, then restore: try flat start vs warm start, lock taps, switched shunts, and phase shifters, relax generator Q-limits — then re-enable controls stepwise once a solution exists.
9	4. Seed from an easier point: a DC solve, a previously solved neighbor case, or a reduced-stress version of the case, then scale load or transfer toward the target in steps.
10	5. If failure persists only near the target stress level, treat it as a feasibility boundary, not a numerical bug: reduce transfer or load, add reactive support, and flag possible voltage-collapse proximity.
11
12	## Working rules
13	- The largest mismatch bus tells you where to look — start there, not with global solver settings.
14	- Change one solver setting at a time and record what finally solved.
15	- A case solved with controls locked or Q-limits relaxed is not finished; restore realism before reporting results from it.
16	- Distinguish "the math failed" from "the operating point does not exist" — the corrective actions are completely different.
17
18	## Deliver
19	- The failure category: data error, control interaction, numerical fragility, or physical infeasibility.
20	- The specific change that made the case solve.
21	- Any realism caveats that remain (locked controls, relaxed limits) and what to re-check.

People who liked this also viewed

6 picks

Type

convergence-failure-mitigation

56stars12forks

View on GitHub

How to install convergence-failure-mitigation?

power-agent/powerskills/convergence-failure-mitigation

$npx -y skills add Power-Agent/PowerSkills --skill convergence-failure-mitigation

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Convergence failure mitigation
2
3	Start from the failed solve and its mismatch report. The goal is to learn why it failed, not just to force a solution.
4
5	## Preferred action order
6	1. Rule out data problems first: isolated or de-energized buses, zero or absurd branch impedances, missing or misplaced swing bus, transformer tap and phase-angle data, load/generation totals that cannot balance.
7	2. Check islanding and slack coverage: every energized island needs a slack with feasible MW. Use the tool's island or topology report before touching solver settings.
8	3. Simplify the solve, then restore: try flat start vs warm start, lock taps, switched shunts, and phase shifters, relax generator Q-limits — then re-enable controls stepwise once a solution exists.
9	4. Seed from an easier point: a DC solve, a previously solved neighbor case, or a reduced-stress version of the case, then scale load or transfer toward the target in steps.
10	5. If failure persists only near the target stress level, treat it as a feasibility boundary, not a numerical bug: reduce transfer or load, add reactive support, and flag possible voltage-collapse proximity.
11
12	## Working rules
13	- The largest mismatch bus tells you where to look — start there, not with global solver settings.
14	- Change one solver setting at a time and record what finally solved.
15	- A case solved with controls locked or Q-limits relaxed is not finished; restore realism before reporting results from it.
16	- Distinguish "the math failed" from "the operating point does not exist" — the corrective actions are completely different.
17
18	## Deliver
19	- The failure category: data error, control interaction, numerical fragility, or physical infeasibility.
20	- The specific change that made the case solve.
21	- Any realism caveats that remain (locked controls, relaxed limits) and what to re-check.

People who liked this also viewed

6 picks

Type