Industrial Practice of Model Predictive Control *

Transcription

1 Industrial Practice of Model Predictive Control * Workshop at CPC-FOCAPO, Tucson January 8th, 017 Joseph Lu * Primarily through the lens of practices in the process industries From Process Unit to Plantwide Control & Page 1

2 How are MPCs typically justified in process industries? Model predictive control applications are typically financially evaluated and justified by its benefits on a per-application basis. More often, provides only the necessary feasibility condition for process optimization, whereas the latter provides the sufficient financial justification for its inception. Typical Sources of Benefits: Increased Throughput Improved Yields Reduced Energy Consumptions Decreased Operating Costs (e.g. catalyst) Improved Product Quality Consistency Increased Operating Flexibility Improved Process Stability Improved Process Safety Improved Operator Effectiveness Better Process Information From Process Unit to Plantwide Control & Page

3 MPC Applications and Benefits of running applications across many industries Industry-wide, the number could exceed 10,000+ Generally accepted ranges for MPC and optimization benefits (per process unit) Petrochemicals Ethylene VCM Aromatics (50KBPD) Chemicals Ammonia Polyolefins Industrial Utilities Cogeneration/Power Systems Pulping Bleaching TMP (Thermo Mechanical Pulping) Oil Refining Crude Distillation (150 KBPD) Coking (40 KBPD) Hydrocracking (70 KBPD) Catalytic Cracking (50 KBPD) Reforming (50 KBPD) Alkylation (30 KBPD) Isomerization (30 KBPD) Benefits (/yr) -4% Increase in Production 3-5% Increased Capacity / 1-4% Yield Improvement 3.4M - 5.3M US$ Benefits (/yr) -4% Increased Capacity / -5% Less Energy/Ton -5% Increase in Production/Up to 30% faster grade transition Benefits ($0.01/bbl) Benefits (/yr) -5% Decrease in Operating Costs Benefits (/yr) 10-0% Reduction in Chemical Usage $1M-$M Benefits (US$/yr).7M - 7.0M.M - 4.8M 3.3M - 7.6M.4M - 5.4M 1.8M - 4.7M 1.1M -.8M.3M - 1.8M From Process Unit to Plantwide Control & Page 3

4 Typical Payback Time by Industry Typical Payback Time: Oil and Gas Refining Chemicals Petrochem MMM Pulping/Paper Industrial Power 1 to months 3 to 6 months 4 to 6 months 6 to 8 months 10 to 1 months From Process Unit to Plantwide Control & Page 4

5 Simplified Process Unit Unit-Wide Objectives Three Common Objectives: Produce more product(s) of same (or better) quality Satisfy all safety/quality/operating constraints Maximize the product value and/or minimize operating cost Products: Gas ($x1) Operating Constraints Feed ($y) Raw Gasoline ($x) Naphtha ($x3) Kerosene ($x4) Light Gas Oil($x5) Heavy Gas Oil($x6) Fuel ($z) Residue ($x7) From Process Unit to Plantwide Control & Page 5

6 Model Predictive Range Control Versatile control needs in MIMO applications Regulatory control Reference tracking Disturbance rejection / minimum variation Traditional SISO control Constraint handling Prevents another set of variables from exceeding their limits Transition control Moves the operating point of a process unit from x to y Role of individual output can change Degrees of freedom Model Predictive Range Control Output prediction: Additional degrees of freedom are used for Local disturbance rejection/absorption : e.g., y A u ~ y The set-point y r is changed into a set-range ylo y y A slack variable s is introduced in the MPC formulation min y; (independent tuning vs. control response) y y LO y HI penalty HI u* arg min A penalty control variable u u* arg min y u, s LO u ~ y r penalty A s control variable u ~ y s y Q HI u Regulatory Control Q R u R Range Control control variable From Process Unit to Plantwide Control & Page 6

7 Range MPC Design and Solution Performance Specification (via a funnel): high limit low limit current time correction horizon T s prediction horizon controlled variable manipulated variable time Range control benefits Robustness Minimum effort control (when the control solution is non-unique) Smoother control, minimum high frequency excitation Range control algorithm Two-stage analytical approach a) solve the range control problem u A u ~ * arg min y s u, s b) If soln to (a) is not unique, solve for the minimum effort control Q u R ; y LO s y HI set-point correction horizon prediction horizon controlled variable u* arg min u u R ; y LO A u ~ y y Online algorithm similar to the active-set QP: (here assume R = 0) HI current time T s manipulated variable time u* arg min A u w ; y ~ y w y ~ y u, w u* arg min u, w A A act free Q LO w u w act free Q HI When the solution is not unique, a secondary optimization can be performed to find the minimum effort solution. Let Q 1 ( A k ) act U R V k k T k u arg min U R V T * k k k act u u w From Process Unit to Plantwide Control & Page 7

8 WFO Dehydrator WFO Atmos Flash Tower WFO Stripper Ketone Stripper PW Stripper PW Atmos Flash Twr WFO Press Flash Twr WFO Final Flash Twr PW Final Flash Twr PW Press Flash Twr PW Mix Surge Slop Solvent Double Pipe Chiller Repulp Filt Feed Double Pipe Exchanger Double Pipe Exchanger Double Pipe Chiller Primary Filt Feed Double Pipe Exchanger Double Pipe Exchanger Double Pipe Chiller Double Pipe Chiller Non-Square Control Example Solvent Dewaxing Unit Steam Feed Refrig Refrig Refrig Primary Filtrate Refrig Wet Dry Refrig Dewaxed Oil Repulp Filtrate Reflux Reflux Reflux Reflux Reflux Reflux Reflux Reflux Vacuum Steam Steam WFO Mix Htr Steam Wax Steam From Process Unit to Plantwide Control & Page 8

9 Solvent Dewaxing Unit Illustration Tank 1: Mixture with solvent Tank : Cooled mixture (wax slurry) Tank 3: Chilled solvent (wax slurry) Wax + Oil Mixture 1 3 Filtering Wax Solvent Lubricant Oil Solvent Inventory Control Tank 4: Filtrate with wax removed Tank 5: Wet solvent (oil removed) Tank 6: Dried solvent for reuse From Process Unit to Plantwide Control & Page 9

10 Model Dynamics of Solvent Control From the application reported by Hall, et. al., Primary Filtrate Level Repulp Filtrate Level Dry Solvent Drum Level Moist Solvent Drum Level Slop Solvent Tank Level Slack Wax Tank Level Primary Filtrate Flow 1.4e 1 3 s 1.6e 1.6e e s 3 1 e s 3 1 5s 5s Repulp Kickback Flow 1.4e 1 3 s 1.39e 1 3 s Slop Solvent Flow.16e 1 3 s 1.3e 1 3 s Slack Wax Flow 1.08e 1 4 e s 3 1 e 1 s 5s 5 th Dilution Flow Ratio e s s s 10s From Process Unit to Plantwide Control & Page 10

11 Simulation Results 15 Controlled Variables % Manipulated Variables 5 Time[Min] % Min Filter Wash Grade Change From Process Unit to Plantwide Control & Page 11

12 Pulp and Paper Example: Thermomechanical Pulping (TMP) Refiner TMP refiner a naturally non-square 4x5 interacting system Objective: consistent quality from coordinated control (decoupling) Customer estimated benefits: ~$13.75M/Year From Process Unit to Plantwide Control & Page 1

13 DEBUTANIZER DISTILLATE STRIPPER CONDENSATE STRIPPER DEPROPANIZER PROPYLENE SPLITTER DEMETHANIZER DEETHANIZER CAUSTIC ETHYLENE SPLITTER Ethylene Plantwide Control and Fig 1. Aerial photo of an ethylene plant Ethylene Plant Overview Feed and products Ethylene plants are very flexible and can process a wide variety of feed stocks, ranging from gases (Ethane, Propane, LPG etc), naphthas (light and full range) to distillates and gas oils. Main products are polymer grade ethylene and propylene. Typical market driving forces Ethylene being a bulk commodity is sensitive to market conditions. Depending on the local economics, operations objectives are a combination of yield improvement, production maximization and energy intensity reduction. Operating flexibilities (variables for optimization) Main operating degrees of freedom include feed selection, furnace feed rates, cracking severity, dilution steam, CGC suction pressure, typical column variables (reflux, reboiler and pressure), converter temperature and H ratio and refrigeration compressor suction pressure. Complex chemical reactions and yield prediction Ethylene yield from the furnaces cannot be directly measured. Therefore we use a yield model to estimate the yields (& derivatives) for control and optimization. Technip s on-line SPYRO, the industry standard, has been integrated into s UniSim Design as an extension Unit Op. Process and Operating Characteristics Universal: No product blending Stringent quality requirements Slow dynamics from gate to gate Gradual furnace and converter coking Furnace and converter decoking Frequent furnace switching Site specific: Feed quality variations Product demand changes Sensitivity to ambient conditions Periodic switching (e.g., dryers) Advanced Control and Goals Stabilize operation Minimize product quality give away Maximize selectivity & yield Minimize converter over-hydrogenation Minimize ethylene loss to methane/ethane recycle. FURNACES Feed System Profit Controllers OIL QUENCH TCG WATER QUENCH CGC ETHYLENE REFRIG COLD BOX Fig. Process overview flow diagram PROPYLENE REFRIG ETHYLENE HYDROGEN FUEL GAS ACETYLENE RECOVERY UNIT ETHANE COCRACK MAPD CONVERTER PROPYLENE MIXED C4 PROPANE COCRACK TCG From Process Unit to Plantwide Control & Page 13

14 DeC3 C3 Splitter Fractionator DeC1 DeC C Splitter Quench CR C3R Profit Controller Envelopes RMPCT 11 RMPCT 13 RMPCT 9 Cold Box CH Converter RMPCT 1 RMPCT 1 RMPCT 15 RMPCT RMPCT 8 RMPCT 10 RMPCT 14 MAPD Converter From Process Unit to Plantwide Control & Page 14

15 DeC3 C3 Splitter Fractionator DeC1 DeC P Quench CR C3R Plantwide Dynamic Profit Optimizer RMPCT 13 RMPCT 11 RMPCT 9 Cold Box CH Converter RMPCT 1 RMPCT 1 RMPCT RMPCT 8 RMPCT 10 RMPCT 14 MAPD Converter RMPCT 15 From Process Unit to Plantwide Control & Page 15

16 Typical Results of Ethylene Plant APC Execution Frequencies: - lers: ½ to 1 minute - Global Optimizer: 1 minute - Gain updating: 5 minutes Typical Benefits - Production Increase: $1.5-3Millions - Energy Savings: Additional Implementation Details Design Typical design comprises of individual Profit Controllers for each of the units, a Profit Optimizer on top of all the controllers, a UniSim model for inferentials and gain mapper, which is a Profit Bridge app that calls the UniSim model in real time to update the gains in both Profit Controllers and Profit Optimizer once every 5 mins. Modeling Dynamic controller models are identified, in either open loop or closed loop, using Profit Stepper. Profit Optimizer model is automatically aggregated from all Profit Controllers. Bridge Models are identified from historical data (no step testing is required) for modeling the dynamic interactions among all MPC controllers. The UniSim modeling is fairly simple as only material balance calculations are required for gain updating. Commissioning Profit Controllers are commissioned first to provide stable unit operations. Profit Bridge is then commissioned to provide gain updating for the furnace controllers. Next, Profit Optimizer is commissioned, and Bridge Model predictions are verified against data. Finally, a plantwide objective function is populated and optimization moves are validated. Continuous operation of the controllers and optimizer starts after operator training Normalized Olefin Production Rate Trend Before APC Decoking campaigns; one furnace out of service After APC Ethane Impurity Content (in ppm) in Product Ethylene Production is increased and product variability reduced... Tighter control and less quality give-away Maintenance Typically little maintenance is required to keep the ethylene APC/ running. Clients either dedicate a parttime control engineer to monitor and perform minor services as needed, or depend on s BGmax quarterlyvisit services Base Before APC After APC From Process Unit to Plantwide Control & Page 16

17 vend 7 Nontraditional Space: Discrete Manufacturing vendor1 1.1 si 1. si vendor Semiconductor Multi-factory Supply Chain.1..3 Fab1 P1 = Mats Mfg = Inv Hold Fab P1,P Fab3 P Sort1 P1 A Sort P1,P Sort3 P Asm pp 3.4 pp vendor3 vendor Asm 9 Semiconductor vendor5 6 vendor6 3.8 pp 3.9 ram Test1 B C Test D Fuse1 Fuse 7.4 pp pp 7.6 Box vend8 46 Blue = Intel Red = Mat. Sub. Green = Cap. Sub. 7.5 Box F = Prod Mfg = Transport 3.1 Asm Test3 6.3 Courtesy of Karl Kempf Intel From Process Unit to Plantwide Control & Page 17

18 Semiconductor Discrete manufacturing Sorting Assembly wafer sort test TPT Distribution C1 Simple Supply Chain Sequence M1 T1 I10 C M T I0 C3 Data Model DataPDF ModelPDF M3 Fab/Sort Die/Package Inventory T3 I30 C4 Assembly/Test M4 Semi-Finished Goods Inv. Finish/Pack Components Warehouse T4 Shipping Demand (Orders) 0 0 D3 D D1 t From Process Unit to Plantwide Control & Page 18

19 ler Design One controller per product 14 controlled variables 5 fab/sort End-of-Line Inventories 7 Assembly Die Inventories (ADI) 7 die prep capacities (Work-in-Progress) 56 Tape and Reel Die Inventories (TRDI) 48 ADI + Die-Prep + TRDI inventories 90 manipulated variables (Green) 35 ships from sorts 48 ships from a hub TRDI to assembly test manufacturing (ATM) sites 7 starts from ADI 93 feedforward variables (Blue) 5 sort supply w/forecasts 48 ATM TRDI demands w/forecasts 40 TRDI cross-ships w/forecasts Current system design consists of 34 such controllers (products) per computer node One dynamic optimizer covers all 34 MPC controllers. FAB FAB FAB FAB SORT 01 SORT 0 SORT 03 SORT 04 ADI DIE PREP TRDI 01 ADI 01 DIE PREP TRDI TRDI TRDI DEMAND ADI DIE PREP TRDI 0 DEMAND ADI DIE PREP TRDI 03 DEMAND ADI DIE PREP TRDI 04 DEMAND ADI DIE PREP TRDI 05 DEMAND A model predictive control application for inventory and production management. From Process Unit to Plantwide Control & Page 19

20 K Die K Die K Die Chip Inventories at an Unnamed Site Die xxx_06 Inventory 3000 High Limit 500 APC used 000 Inventory Low Limit 0 5/1/06 5/8/06 6/4/06 6/11/06 6/18/06 6/5/06 7//06 7/9/06 Die xxx_7 Inventory 3000 Date High Limit Inventory 1500 APC used 1000 Low Limit /1/06 5/8/06 6/4/06 6/11/06 6/18/06 6/5/06 7//06 7/9/06 Die xxx_47 Inventory 1500 Date 150 High Limit 1000 APC used 750 Inventory Low Limit 5/1/06 5/8/06 6/4/06 6/11/06 6/18/06 6/5/06 7//06 7/9/06 From Process Unit to Plantwide Control & Page 0 Date

21 Refinery-wide (Production) Control/ Supply Side (upstream supply chain) Average daily production What Feed to buy? Production Plan (monthly) Generate a month-plan based on an average model What products to sell? Customer Demand (downstream supply chain) Average daily delivery Production Schedule (weekly) Feed Received What Feed to use? What products to deliver? Product to be Deliveried Translate the plan into production activities to satisfy the plant logistics A set of production schedules (e.g., feed blending schedule) Master MPC Control APC Blending/shipment schedules Control Human Coordination Control Control Slave MPC 1 Slave MPC Slave MPC n month ahead days ahead hours ahead now hours ahead days ahead months ahead Raw Materials Execution Products From Process Unit to Plantwide Control & Page 1

22 (Multiscale) MPC Cascade Structure Master MPC two-way connection two-way connection Slave MPC 1 Slave MPC Slave MPC n 1-to-n MPC cascade One Master MPC and multiple Slave MPCs The Master MPC employs a dynamic model which can be, at user s discretion, a coarser-scale model e.g., similar to the scale-level of a planning model The Slave MPCs employs a dynamic model which can be a finer-scale one If Slave MPCs are deployed at every process unit, then the Master MPC can solve the plantwide production control problem dynamically and in closed loop Currently, a steady-state version of this problem is controlled by planning solutions in open-loop From Process Unit to Plantwide Control & Page

23 Refinery-wide Production Control : JIT manufacturing Blending Components Straight-run products Straight-Run and/or Blending The intermediate tanks here are for illustration and may vary from plant to plant. From Process Unit to Plantwide Control & Page 3

24 JIT manufacturing viewed as a control problem Scope: the distillate production pool in an oil refinery It is simpler but still captures most of the important issues Crude Feed 1 Crude Feed Crude Unit 1 Crude Unit Vac Unit 1 Light Distillate Hydrotreater Heavy Distillate Hydrotreater Imports Component Tank 1 Component Tank Moving Avg daily delivery: 5 KBarrels 8 KBarrels 10 KBarrels 7 KBarrels Component Tank 3 10 KBarrels Crude Unit 3 Vac Unit Hydrocracker Blending Crude Feed 3 Material Balance & Inventory Control From Process Unit to Plantwide Control & Page 4

25 Just-in-Time Manufacturing Quality Control (1) Quality Control - Sulfur Legend: Control handles Sweet 0.1% Sulfur Medium 0.3% Sulfur Crude Unit 1 Crude Unit Vac Unit 1 50~300 ppm S 100~400 ppm S Light Distillate Hydrotreater 4~0 ppm S Heavy 4~0 ppm S Distillate Hydrotreater Imports Component Tank 1 Component Tank 1 ppm S 10 ppm S 10 ppm S 10 ppm S Component Tank 3 10 ppm S Sour 1.% Sulfur Crude Unit 3 Vac Unit 100~600 ppm S Hydrocracker 1~ ppm S Blending From Process Unit to Plantwide Control & Page 5

26 Just-in-Time Manufacturing Quality Control () Quality Control - Cloud Point Legend: Control handles Sweet Medium Crude Unit 1 Crude Unit Jet Cut Stove Cut Jet Cut Stove Cut Vac Unit 1 Jet Cloud -30 to -40 ⁰C Stove Cloud -5 to -10 ⁰C Light Distillate Hydrotreater Heavy Distillate Hydrotreater = Feed Cloud -5 to -45 ⁰C Imports Component Tank 1 Component Tank Component Tank 3 Cloud Spec (winter) -36 ⁰C -44 ⁰C -40 ⁰C -30 ⁰C -4 ⁰C Sour Crude Unit 3 Jet Cut Stove Cut Vac Unit Gasoils Hydrocracker -40 to -55 ⁰C Blending Few if any (closed-loop) control solutions are available for this class of problems! From Process Unit to Plantwide Control & Page 6

27 1) Production amount ) Product quality Controlled Variables (CVs) Control Model Matrix for the Master MPC 1) Unit feed rates ) Control factors that affect the product quality Manipulated Variables (MVs) 1) Demands ) Others DVs CVs MVs Crude1 Feed Crude Feed Crude3 Feed Hydrocracker Conv Cr Unit Cuts Hydrotreater Sulfur Product-i delivery CompTank.Level CompTank.Sulfur CompTank.Cloud CompTank3.Level CompTank3.Sulfur CompTank3.Cloud More CVs Each (i, j) entry contains a dynamic model ( ) : k bs 1 g i, j s k s( s 1) s s Often simple dynamics such as e or e would suffice as Notice that there is no setpoint in this control problem just constraint control From Process Unit to Plantwide Control & Page 7

28 Master MPC for JIT Manufacturing in closed loop? CVs MVs Crude1 Feed Crude Feed Crude3 Feed Hydrocr acker Conv Cr Unit Cuts Hydrot reater Sulfur Produc t-i deliver y Master MPC Tank.Level Tank.Sulfur Tank.Cloud Tank3.Level Tank3.Sulfur Tank3.Cloud - More CVs Slave MPC 1 Slave MPC Slave MPC n Yes, if the solution for JIT manufacturing from the Master MPC is implementable by the Slave MPCs. No, if the solution from the Master MPC can cause one or more Slave MPCs to wind up (i.e. sustained CV constraint violations). The Real Challenge: How to get the Master to understand the Slave s constraints well enough so that the Master s solution will respect them CV CV 7.3 CV CV4 MV1 MV MV3 MV4 MV5 MV6 MV7 MV8 Feed From Process Unit to Plantwide Control & Page 8

29 Slave CVs Slave1 CVs Master CVs Master-Slave Models: a side-by-side view Below is an illustration of the model structure for combined units: Master MV/DV index Different Tasks Master MPC Conjoint MVs Plantwide Production Control Inventory control Quality control Cost and profit optimization a.b Slave MV index Slave MPC 1 Free MVs Slave MPC Unit Process Control Safe/smooth operation Unit efficiency Intermediate Quality Local The relationship is a bit convoluted, so let s first look at the simplest case: 1 conjoint variable per Slave MPC From Process Unit to Plantwide Control & Page 9

30 The Concept of Proxy Limit An Illustration for 1 Conjoint Variable Case Steps in the hydrocracker example: 1) The HC feed is configured as MV3 in the Master. ) Its current value is 33 3) The Master makes an inquiry call to the Slave to maximize the feed, subject to the Slave constraints 4) The call returns the following: a) the maximum feed = 38 b) 4 active HC constraints are shown below, which are CV1, CV3, MV4 & MV5 (others not shown) 5) The master uses the maximum feed limit of 38 as a proxy limit for all Slave constraints. Refinery MV3 = 33 Hydrocracking Blending Components Straight-run products Straight-Run and/or Blending Opt Value: Opt Value 100 CV CV 7.3 CV CV4 Feed MV MV3 MV4 MV5 MV6 MV7 MV8 MV Feed g g g g yd yd yd yd ( s) ( s) ( s) ( s) 1 3 n Products From Process Unit to Plantwide Control & Page 30

31 Approximate the feasible region defined by the Slave constraints for (or more) conjoint variables Master CMV (k+1) MV1 Low Lmt MV1 High MV High Master MVs Current Operating Point (0,0) k k+1 MV Low Master CMV (k) Conjoint MVs Free MVs 8 inquiry calls: Call # c t d t b LO t min c x A x c d t x conj x Where, conj A conj f 0 x A I G free b HI Slave MPC (i) From Process Unit to Plantwide Control & Page 31

32 Proxy Limits Customer Example To what extent is the Master MPC allowed to change the Slave unit s operations? Hydrocracker Conversion - Actual vs Maximum From Process Unit to Plantwide Control & Page 3

33 Proxy Limits Customer Example () Heavy Distillate HTR Feed Rate - Actual vs Maximum From Process Unit to Plantwide Control & Page 33

34 Higher cloud point means that more low-cost (heavier) components are used to produce the same products Master s Control and Results: Component Quality Changes in Tank #3 Tank3.Cloud Point in Deg C Tank3.Flash Point in Deg C Lower flash point means that more low-cost (lighter) components are used to produce the same products Tank3.90pct Point in Deg C Tank3.Qualities (After - Master MPC) Historical Operation Data (Before) Tank3.Sulfur in PPM Sulfur is controlled at its high limit which means much lower catalyst/utility costs Higher 90pct point means that more low-cost (heavier) components are used to produce the same products Component production costs are significantly reduced! From Process Unit to Plantwide Control & Page 34

35 Feed Cost Saving from Optimizing the Volumetric Gains Net Result: Crude Reduction 4 3 Incremental Changes in Crude Unit Feed Rate Crude Unit 1 deltafeed Crude Unit deltafeed Crude Unit 3 deltafeed Combined Crude deltafeed 1 Average crude saving is ~ 0.5% for meeting the same product demand. Huge cost saving! From Process Unit to Plantwide Control & Page 35

36 Future Work A 3-level MPC Cascade for a larger part of the supply chain? For example: a combination of multiple refineries and fuel/crude depots Level-3 Master two-way connection two-way connection Master MPC Master MPC Master MPC two-way connection two-way connection two-way connection two-way connection two-way connection two-way connection Slave MPC 1 Slave MPC Slave MPC n Slave MPC 1 Slave MPC Slave MPC n Slave MPC 1 Slave MPC Slave MPC n From Process Unit to Plantwide Control & Page 36

37 Closing Remarks MPC is the control solution of choice in the process industries optimization is often applied in tandem with MPC Close to 10 thousand MPC controllers have been implemented in different industries. Oil refining, petrochemical, pulp & paper, coal-to-fuel, alumina, LNG, shale gas just to name a few >100 plantwide optimization solutions have been implemented A majority of them are for ethylene plants Estimated benefits delivered: $10 billions in total ($5 billions by since 1996) Solution Operability and User Preference Monolithic solutions are not desirable Decentralized control flexible to turn parts of control ON or OFF when addressing process upsets, maintenance issues, etc. Gate-to-gate, plantwide optimization greater benefits and no need to use any transfer prices Multiscale MPC cascade solution is being developed, which Performs gate-to-gate, plantwide control and optimization It can capture a significantly greater amount of benefits (~$1/barrel) Leverages the existing planning model (structure) Streamlines the current workflow from planning, scheduling to APC From Process Unit to Plantwide Control & Page 37

38 Questions? From Process Unit to Plantwide Control & Page 38