Refinery LP yield backcasting: reconcile crude-assay predictions, CDU/VDU cut-point temperatures, and fractionation gap variances

Topics: refinery planning software, hydrocarbon supply chain software, crude oil trading software.

Refinery crude traders and supply planners purchase crude oil cargoes based on marginal values ("shadow prices") generated by a Linear Programming (LP) refinery planning model. However, library crude assays inside the LP software are often 2 to 5 years old, while real-time CDU operators must lower the Heavy Diesel draw cut-point temperature whenever cold-flow properties (Cloud Point / Cold Filter Plugging Point) tighten or when side-stripper tray efficiency degrades. While O024 governs the physical daily tank-to-unit mass balance, this guide (O125) defines the LP-to-actual yield backcasting engine, three-way margin variance attribution, ASTM D86 fractionation monitoring, and ERP/LP model update workflow.

Scope and governance boundary

This article specifies the refinery planning, process engineering, and ERP profitability-accounting workflow for yield backcasting:

• What this workflow governs: Reconciling planned LP product slates (LPG, Naphtha, Kerosene/Jet, ULSD Diesel, VGO, Vacuum Residue) against reconciled unit mass balances, three-factor backcast variance decomposition (Assay Drift + Cut-Point Shift + Fractionation Loss), Gross Product Worth (GPW) financial attribution, and LP library parameter updates (ERPNext Production Plan).
• What this workflow does not replace: It does not replace closed-loop Advanced Process Control (APC) multivariable controllers on the CDU tower, rigorous tray-by-tray thermodynamic simulation (Aspen HYSYS / Petro-SIM), or statutory custody-transfer crude bill-of-lading surveying.

1. Step 1 — Record intake & baseline identification: Never blame Crude Distillation Unit (CDU) operations for the entire gap between the monthly **Commercial Linear Programming (LP) Plan** and **Actual Distillate Yields**: decompose total yield variance into three orthogonal drivers—**Crude Assay Quality Drift**, **Intentional Cut-Point Temperature Shifts**, and **Tower Fractionation Efficiency Slump**.
2. Step 2 — Cross-system reference & custody linkage: Run a structured **Three-Step LP Backcast** at the close of every crude switch or 72-hour run: re-solve the LP simulator sequentially with (1) actual delivered crude blend proportions and lab TBP assays ([ASTM D2892 / D5236](https://www.astm.org/d2892-20.html)), (2) actual DCS draw cut-point temperatures, and (3) reconciled mass-balance yields ([Energy Institute HM 68](https://publishing.energyinst.org/)).
3. Step 3 — Exception quarantine & authority check: Monitor adjacent side-draw fractionation sharpness using the **[ASTM D86](https://www.astm.org/d0086-23.html) $(5\%_{\text{Heavy}} - 95\%_{\text{Light}})$ Gap/Overlap** alongside stripping-steam ratios so valuable diesel ($340^\circ\text{C–}365^\circ\text{C}$) is not slumped into lower-value Atmospheric Gas Oil (AGO) or vacuum feed.
4. Step 4 — Verified disposition & audit trail closure: Feed verified CDU fractionation factors and updated crude assay vectors back into the commercial LP library before traders bid on the next spot crude cargo.

Why comparing raw LP plan yields to actual tank dips misleads management?

When a refinery's middle-distillate (Jet + Diesel) yield comes in $2.1\text{ wt%}$ below the monthly LP forecast, a simplistic "Plan vs Actual" report triggers finger-pointing between the Commercial Crude Desk and CDU Operations because three distinct physical shifts are conflated:

1. Delivered Crude Cargo Assay Drift vs Library Assay: Export crude streams (Urals, Basrah Medium, WCS, WTI) are pipeline blends whose True Boiling Point (ASTM D2892) curve shifts month to month as upstream fields or condensate spiking ratios change. If a delivered cargo contains $0.9\text{ wt%}$ less natural $150^\circ\text{C–}365^\circ\text{C}$ distillate in the crude itself than the LP library assumed, operations could never recover those barrels even with a perfect distillation column.
2. Constraint-Driven Cut-Point Shifts: The commercial LP model may assume the CDU can cut Heavy Diesel all the way up to a $365^\circ\text{C}$ TBP end-point. In actual operation, if the crude's paraffinic wax content spikes, the CDU console operator must reduce the Heavy Diesel draw temperature to $358^\circ\text{C}$ TBP to meet the $-5^\circ\text{C}$ Cloud Point specification—intentionally leaving $0.75\text{ wt%}$ of heavy diesel molecules in the Atmospheric Gas Oil (AGO / FCC feed) draw.
3. True Tower Fractionation Degradation (Tail Overlap): If side-stripper reboiler duty drops, stripping steam falls below $0.25\text{ kg steam/kg oil}$, or pumparound internal reflux is under-circulated, light diesel molecules slump downward into the AGO/Residue draw (ASTM D86 $95%$ Diesel temperature exceeds $5%$ AGO temperature). That residual variance is a genuine unit operating loss.

• Primary control scope: refinery planning software across operations and connected ERP/EAM records.
• Segregation boundary: Operational status updates never auto-close engineering, safety, or finance sign-off gates.
• Verification cadence: Reconcile open exceptions, overrides, and late adjustments at each shift or period close.

Refinery LP yield backcasting and margin attribution control matrix

Illustrative worked example: 72-hour CDU-1 middle-distillate yield backcast and $$535,500$ margin reconciliation

Illustrative refinery planning scenario with synthetic figures to demonstrate three-way LP backcast decomposition arithmetic; not a client case study.

Consider a refinery completing a 72-hour campaign on Crude Distillation Unit CDU-1 processing $300,000.0\text{ MT}$ of a 70% Basrah Medium / 30% WTI crude slate. During the run, the commercial crack spread differential between Middle Distillates (Jet + ULSD Diesel) and Atmospheric Gas Oil / VGO (FCC Feed) is $+$85.00/\text{MT}$ (every tonne of diesel slumped into gasoil loses $$85.00$ of refinery margin).

Step 1: Observe total middle-distillate yield shortfall vs LP plan

• Commercial LP Plan ($Y_{\text{plan}}$): Predicted middle-distillate ($150^\circ\text{C–}365^\circ\text{C}$ TBP cut) yield $= \mathbf{34.80\text{ wt%}}$ ($104,400.0\text{ MT}$).
• Reconciled actual yield ($Y_{\text{actual}}$): Measured middle-distillate yield $= \mathbf{32.70\text{ wt%}}$ ($98,100.0\text{ MT}$).
• Total middle-distillate variance ($\Delta Y_{\text{total}}$): $$\Delta Y_{\text{total}} = 32.70\text{ wt%} - 34.80\text{ wt%} = \mathbf{-2.10\text{ wt%}}\quad (-6,300.0\text{ MT})$$ $$\Delta\text{Margin}_{\text{total}} = -6,300.0\text{ MT} \times $85.00/\text{MT} = \mathbf{-$535,500.00}$$

Step 2: Execute the Three-Way LP Backcast Decomposition

To determine why $6,300.0\text{ MT}$ of distillate disappeared into heavier gasoil/residue cuts, the planning engineer runs two intermediate LP backcasts:

1. Backcast #1 (LP_Actual_Assay $\rightarrow Y_{\text{assay}} = 33.90\text{ wt%}$):
   • Holding the planned $150^\circ\text{C–}365^\circ\text{C}$ cut points fixed, the engineer substitutes the laboratory ASTM D2892 TBP curve of the actual feed tank composite sample. Because the delivered Basrah cargo was $0.6^\circ\text{API}$ heavier in the gasoil/residue range than the library assay, theoretical distillate yield at $365^\circ\text{C}$ drops to $33.90\text{ wt%}$ ($101,700.0\text{ MT}$): $$\Delta Y_{\text{assay}} = 33.90\text{ wt%} - 34.80\text{ wt%} = \mathbf{-0.90\text{ wt%}}\quad (-2,700.0\text{ MT}\ /\ -$229,500.00)$$
2. Backcast #2 (LP_Actual_CutPoint $\rightarrow Y_{\text{cut}} = 33.15\text{ wt%}$):
   • Using the actual lab assay, the engineer lowers the Diesel/AGO upper TBP cut point in the model from $365^\circ\text{C}$ to the actual $358^\circ\text{C}$ cut point used by CDU operators to hold diesel Cloud Point below $-6.0^\circ\text{C}$ without cold-flow additive: $$\Delta Y_{\text{cut}} = 33.15\text{ wt%} - 33.90\text{ wt%} = \mathbf{-0.75\text{ wt%}}\quad (-2,250.0\text{ MT}\ /\ -$191,250.00)$$
3. Residual Tower Fractionation Variance ($\Delta Y_{\text{frac}}$):
   • Comparing reconciled unit output ($32.70\text{ wt%}$) against Backcast #2 ($33.15\text{ wt%}$) isolates physical column fractionation inefficiency: $$\Delta Y_{\text{frac}} = 32.70\text{ wt%} - 33.15\text{ wt%} = \mathbf{-0.45\text{ wt%}}\quad (-1,350.0\text{ MT}\ /\ -$114,750.00)$$

Acceptance tests and operational readiness criteria

1. ACT-O125-01 (Mass-Balance Closure Pre-Requisite Gate): The LP backcasting workflow must reject raw un-reconciled flowmeter inputs and only execute variance decomposition against closed daily/campaign mass balances (Energy Institute HM 68) where hydrocarbon loss/gain closure is verified within $\pm 0.3\text{ wt%}$.
2. ACT-O125-02 (Orthogonal 3-Way Variance Identity Check): For every cut (Naphtha, Kerosene, Diesel, VGO, Residue), the sum of $\Delta Y_{\text{assay}} + \Delta Y_{\text{cut}} + \Delta Y_{\text{frac}}$ must equal $Y_{\text{actual}} - Y_{\text{plan}}$ to $0.00\text{ wt%}$ rounding accuracy, and cross-product sum across all cuts must equal $0.00\text{ wt%}$ (mass conservation).
3. ACT-O125-03 (ASTM D86 Gap/Overlap Alert Trigger): Whenever $\Delta Y_{\text{frac}}$ on Kerosene or Diesel falls below $-0.30\text{ wt%}$, the system must automatically pull the latest LIMS ASTM D86 $95%$ light-key and $5%$ heavy-key distillation temperatures and flag side-stripper/pumparound degradation.
4. ACT-O125-04 (Closed-Loop LP Assay Vector Calibration): If $\lvert\Delta Y_{\text{assay}}\rvert > 0.50\text{ wt%}$ persists across 2 consecutive cargoes of the same crude grade, the ERP/LP governance log must generate a mandatory Assay Library Revision Task for the commercial crude evaluation team.

Next steps

1. Automate weekly 3-step LP backcast runs after mass-balance reconciliation: Stop reporting undifferentiated "plan vs actual" yield tables that mask crude quality drift as plant operating failure.
2. Use fast GC-Simulated Distillation (ASTM D7169) on feed tanks before crude switches: Feed actual feed-tank TBP curves into the short-term scheduler before the crude hits the desalter.
3. Evaluate cold-flow additive economics against cut-point depression losses: As shown in Step 2, spending $$18,000$ on Cold Flow Improver (CFI) to raise the diesel cut point from $358^\circ\text{C}$ back to $365^\circ\text{C}$ recovers $$191,250$ of distillate downgrade loss.
4. Track ASTM D86 $(5%{\text{AGO}} - 95%{\text{Diesel}})$ fractionation gaps on the console KPI dashboard: Give operators real-time visibility into diesel tail slump.

Frequently Asked Questions

How should refinery planning teams reconcile LP model predicted distillate/residue yields against actual crude-unit cut-point temperatures, lab TBP assays, and daily mass-balance backcasts?

Teams should govern refinery planning software by linking every transaction to its source evidence, timestamp, and assigned owner before permitting downstream posting. Distinct from O024 (daily tank reconciliation); O125 closes the loop between commercial crude selection (expected assay yields in the LP model) and actual CDU/VDU draw rates, identifying tower fractionation inefficiency vs crude-quality drift. Rather than relying on informal spreadsheet overrides, the workflow enforces a documented verification gate and preserves the original observation whenever a correction is recorded.

How does hydrocarbon supply chain software handle exceptions and mismatches?

When a discrepancy occurs during hydrocarbon supply chain software, the system routes the blocked record into an owned exception queue rather than silently dropping or auto-adjusting the balance. Assigned operational, quality, and finance owners must review and approve exceptions before posting. Each exception retains the original quantity, date, and counterparty reference until the designated reviewer resolves the variance with supporting evidence.

What should teams test before automating crude oil trading software?

Before scaling crude oil trading software, run a bounded pilot using historical records that include partial receipts, late updates, duplicate messages, and rejected lines. Verify that the integration preserves idempotency, maintains a clear segregation of duties between operational entry and financial approval, and produces an auditable trail for every state transition.

Summary and next actions

1. Map the current refinery planning software record boundaries, system of record, and named approval owners before changing workflow rules.
2. Separate operational status updates from engineering, safety, or financial disposition sign-off.
3. Validate exception queues, duplicate handling, and cutoff reconciliation in a bounded pilot before production rollout.