Refinery Planning, Yield Accounting & Blend Optimization · Manufacturing & Industrial
Should you build or buy Refinery Planning & LP Optimization (Crude/Blend Linear Programming)?
Refinery planning and LP optimization software uses linear (and nonlinear) programming to determine the optimal crude slate, process unit operating modes, and product blending economics for a specific refinery. Planning engineers use it to run weekly or monthly scenarios that model how different crude purchases and unit configurations translate into margin.
The build-vs-buy decision for Refinery Planning & LP Optimization turns on how deeply proprietary your refinery's physical model needs to be and how close AI-assisted LP tooling has come to matching decades of validated, plant-specific crude assay libraries; the specifics of your unit set and trading strategy decide it.
Build it, buy it, or bridge?
When building makes sense
Building a refinery LP from scratch makes sense when your plant's unit configuration is genuinely unusual enough that no commercial model maps to it cleanly, and when you have the engineering talent to maintain what you build. The LP solver itself is not the barrier — open-source and commercial solvers are widely available. What makes this hard is the validated crude assay library, the unit yield vectors calibrated against actual plant performance, and the recursion handling that keeps the model stable when crudes and operating modes shift. Oil majors have historically done this, and some maintain internal teams that shadow commercial tools. If your planning department already employs engineers who understand refinery physics at that depth, and your crude slate is non-standard enough that PIMS or GRTMPS configurations would always be an approximation, the build argument gets real. The annual model health-check cadence that vendors provide — where they recalibrate yield vectors against recent plant data — is the piece most often underestimated in a build business case.
When buying makes sense
Buying makes sense for most refiners because what you're purchasing is not just a solver but a body of validated refinery physics that took decades to build. Aspen PIMS, Haverly GRTMPS, and Honeywell RPMS all carry mature crude assay libraries and unit yield models tested against real plant behavior across many configurations. For a refinery that runs monthly crude selection exercises against multi-million-dollar margin decisions, being 2% off on a yield vector costs more than a decade of license fees. The vendor also provides annual health checks that recalibrate the model against recent operating data — a discipline that most internal teams don't sustain. The software is deeply embedded in planning workflows and well-utilized, which means the switching cost is high, but so is the cost of misallocating crude on a flawed self-built model. If your unit set is conventional and your team lacks refinery-modeling specialists, buying is the lower-risk path.
The desk read
Buying earns its keep when you consider what's actually inside tools like Aspen PIMS or Haverly GRTMPS: not a generic LP solver, but decades of accumulated refinery physics encoded in crude assay libraries, unit yield vectors, and recursion-handling logic validated against real plant behavior. Crude selection and product slate optimization drive multi-million-dollar margin decisions every month, and the LP model informing them needs to reflect that specific plant's unit set, tanks, and blending economics accurately.
The build case gets more serious as AI-era tools make LP formulation and scenario modeling more accessible. The gap today isn't the solver; it's the validated model library and the annual health-check cadence that vendors run to keep the refinery physics current. A self-built optimizer that gets unit yields wrong by 2% can cost more in misallocated crude than a decade of license fees. Teams evaluating AVEVA Refinery Planning or Honeywell RPMS alongside the Aspen and Haverly options will find that question, model fidelity versus build cost, is the real axis of the decision.
Frequently asked
What is Refinery Planning & LP Optimization software?
Refinery planning and LP optimization software uses linear (and nonlinear) programming to determine the optimal crude slate, process unit operating modes, and product blending economics for a specific refinery. Planning engineers use it to run weekly or monthly scenarios that model how different crude purchases and unit configurations translate into margin.
When does building Refinery Planning & LP Optimization make sense?
Building makes sense when your refinery's unit configuration is unusual enough that commercial models can't map to it cleanly, and you have a engineering team capable of developing and maintaining validated crude assay libraries and unit yield vectors. Most refiners underestimate the ongoing model health-check burden that vendors handle in a buy scenario.
When does buying Refinery Planning & LP Optimization make sense?
Buying makes sense for most commercial refiners because tools like Aspen PIMS and Haverly GRTMPS include decades of validated refinery physics — crude assay libraries, unit yield models, recursion handling — that would take years and significant engineering resources to replicate. The margin risk of a miscalibrated self-built model outweighs license costs when crude selection drives multi-million-dollar decisions monthly.
What are the main Refinery Planning & LP Optimization vendors?
Representative vendors include AspenTech Aspen PIMS / PIMS-AO, Honeywell RPMS (Refinery and Petrochemical Modeling System), Princeps Refinery Operations Planning, Haverly Systems GRTMPS. B4 Pro scores the full set.
What is the difference between LP and NLP in refinery planning?
LP (linear programming) handles most crude selection and product slate optimization problems efficiently at scale. NLP (nonlinear programming) becomes necessary when blend property relationships — like gasoline octane blending — are non-additive and require nonlinear equations to model accurately. Most modern refinery planning tools support both, switching to NLP for blend-pool interactions.