DiffusionControl / results /pendulum /constrained_full.log
matCercola18's picture
Add files using upload-large-folder tool
929e74d verified
Raw History Blame Contribute Delete
1.57 kB
/home/cercola/matte_server/DiffusionControl/nonlinear/data_generation/pendulum_lmi.py:65: UserWarning: Solution may be inaccurate. Try another solver, adjusting the solver settings, or solve with verbose=True for more information.
prob.solve(solver=cp.SCS, verbose=False)
=== CONSTRAINED OC ===
Constrained optimal control on the pendulum (held-out N=40, tracking spec ISE ≤ 0.8× prior)
PRIOR (spec-unaware): median ISE = 0.0333, median ∫u² = 6.730 (ISE = 1.0× ⇒ violates the 0.8× spec)
STEERED toward min ∫u² s.t. ISE ≤ 0.8× prior (barrier λ=200.0):
meets spec: 0.95 median ISE = 0.0272 median ∫u² = 7.247 certified 1.00
energy used vs prior (the cost of the tighter spec): +8%
ENERGY REGRET vs per-plant constrained oracle: median +0.1% (n=38)
energy-vs-spec constrained-optimal frontier (steered):
ρ median ISE_max median ∫u² (meeting spec) meets certified
0.70 0.0233 4.988 0.50 0.88
0.80 0.0266 6.786 0.95 1.00
0.90 0.0299 6.924 1.00 1.00
1.00 0.0333 6.708 1.00 1.00
INTERPRETATION: a non-quadratic, hard-CONSTRAINED optimal-control problem solved through
the differentiable non-linear pendulum — minimise control energy subject to a tracking
spec. The spec-unaware prior violates the tighter spec; steering meets it at the minimum
energy (near the per-plant constrained optimum), staying certified; a tighter spec costs
more energy (the constrained-optimal frontier).