neqsim-optimization-and-doe

NeqSim Process Optimization & DoE Skill

NeqSim ships ~30 production-grade optimization classes. Do not reinvent optimizers in Python — use what exists, then wrap it. This skill is the discoverability layer that maps engineering problems to the right class.

Decision Tree — Which Optimizer

Need to OPTIMIZE something on a flowsheet?
│
├── ML / agentic closed loop (string addresses, never-throw JSON, replayable trajectory)?
│   → auto.newOptimizer()  (AgenticProcessOptimizer — bounded Nelder-Mead on top of evaluate())
│
├── Maximize THROUGHPUT for given P_in / P_out, with equipment constraints?
│   → ProcessOptimizationEngine.findMaximumThroughput(...)
│
├── Single decision variable, custom objective, monotonic feasibility?
│   → ProductionOptimizer + OptimizationConfig(SearchMode.BINARY_FEASIBILITY)
│
├── Single var, NON-monotonic / unimodal score?
│   → ProductionOptimizer + SearchMode.GOLDEN_SECTION_SCORE
│
├── 2–10 vars, no gradients, smooth-ish?
│   → ProductionOptimizer + SearchMode.NELDER_MEAD_SCORE
│
├── Non-convex, multiple local optima, global search?
│   → ProductionOptimizer + SearchMode.PARTICLE_SWARM_SCORE
│
├── Constrained NLP (equality + inequality + bounds), gradients OK to FD?
│   → SQPoptimizer (BFGS + active-set QP + L1 merit)
│
├── 2–4 competing objectives → Pareto front?
│   → MultiObjectiveOptimizer.optimizeWeightedSum(...) or .optimizeEpsilonConstraint(...)
│   then front.findKneePoint()
│
├── DoE / parameter sweep / scenario screening?
│   → BatchStudy.builder(base).vary(...).vary(...).addObjective(...).parallelism(8).build().run()
│
├── Uncertainty quantification, P10/P50/P90, tornado?
│   → MonteCarloSimulator (triangular sampling + tornado)
│
├── Calibrate parameters to plant/lab data?
│   → BatchParameterEstimator (LM-based) — see `neqsim-model-calibration-...`
│
├── PVT / EOS regression to experiments?
│   → LevenbergMarquardt + appropriate *Function (CME, CVD, density, etc.)
│
├── External optimizer (SciPy, NLopt, Pyomo, GA, BoTorch)?
│   → ProcessSimulationEvaluator — black-box evaluate(double[] x) bridge
│
├── Compressor specifically?
│   → CompressorOptimizationHelper
│
├── Lift curves / VFP tables for reservoir simulation?
│   → PipeBeggsAndBrills + surface-rate recombination + EclipseVFPExporter — recipe in
│     `neqsim-production-optimization` § "Lift curves / VFPPROD". LiftCurveGenerator /
│     FlowRateOptimizer sweep in kg/hr and are diagnostic only; the exporter needs Sm3/d.
│
└── Auto-size equipment + apply constraints + optimize, fluent API?
    → DesignOptimizer.forProcess(p).autoSizeEquipment(1.2)
        .applyDefaultConstraints().setObjective(...).optimize()