Have you ever been in a meeting where your LP planning analyst was asked to explain the logic behind a specific decision or proposal and gotten the response, “That’s what the LP model said?” 

That’s one of the most dangerous phrases known to refinery profitability. When a refinery planner says that to me, it’s a loud warning that numbers are being plugged in and solutions being accepted without any critical analysis of the input or output.

The LP Model Itself is Inaccurate

How often do you actually validate your refinery linear program model?  Do the feedstock assays for refinery raw material inputs receive proper updates?  Are submodel yield vectors and constraints maintained as equipment capabilities change?

Upgrades in hardware, catalyst, and distillation tower performance are just some examples of things that require continuous model updates. Evaluate the major short term and long term constraints in your refinery and make sure they’re included in your models. Otherwise you’re making decisions with incomplete information.

While LP model updates require diligence, many models may not even include key parameters for making the right decisions in the first place.  For example, most models for hydroprocessing are focused on feed sulfur, but in many situations the nitrogen content is just as important.  Similarly, conrad carbon content in resid can have great impacts on Coker yields but many models don’t see that impact. 

I would venture to even say that the best value spent on a refinery engineer is the money put to obtaining the highest caliber LP engineer possible.  What other tool do you see making million dollar decisions on a daily basis?

Insufficient Analysis

Just because someone has the word planning analyst in their title doesn’t mean they know how to run a model. Often times the idea of a refinery model gives people false security in the output data. If a computer came up with answer it must be right? Right?

A model can only optimize around the limits it’s been given. Over constraining or improperly constraining variables can have enormous impacts on the refinery optimization results. This emphasizes the importance of having the planning “analyst” put effort in analyzing the results.

The incentive on each constrained parameter must be scrutinized, and the case run results must be vetted. Each model is only as good as the person running it. If the operator is content with plugging in numbers and accepting the results without further evaluation that model result isn’t a very good one.

Inaccurate Inputs

The final issue to be aware of is the quality of the inputs into your model. The most universal example is pricing. Since the majority of models are optimizing around profitability, it’s pretty apparent that the pricing assumptions in your model are absolutely critical. The ability to accurately estimate prices for feedstocks and products can be the difference between an excellent decision and a very poor one.

Comparably, constraints need to also be reflected properly.  How many times have you seen a LP analyst constrain unit capacity to reflect a pseudo-limit?  I’ve seen many instances where a LP planner places a rate limit on the FCC to represent a fuel gas constraint in the refinery.  Instead of placing a spec on off-gas production or refinery fuel gas balance, this artificial limit on throughput will now limit the degrees of freedom the model can optimize around.

All refineries face challenges with ensuring the accuracy of their refinery LP models. The false sense of security that computer generated decisions provide can be dangerous. Often times, just focusing on a critical validation of the refinery model and it’s result can provide an upper hand on your competition.  The end goal of the refinery business is to make money, and no tool can deliver a better refinery optimization direction than a well tuned linear program model.