Optimizing Operations to Boost Your Ability to Complete in Any Market Environment

to Boost your Ability to Compete in any EnvironmentOPTIMIZING OPERATIONSPerry Nordh, P.Eng

20-Jun-2016

Honeywell Confidential - © 2016 by Honeywell International Inc. All rights reserved.

• Profit Stepper

Honeywell’s Advanced Control Solutions

• Profit Controller • Profit Sensor Pro

• Profit Optimizer: Real-time, Dynamic Optimization

*Optimum

Minimum Effort Move

Past Future

Assumed Values

CV

Predicted

Unforced

Response

MV

Control Funnel

Optimal Response

Setpoint • Control Performance Monitor


Profit Optimization Suite

One Consistent technology platform- MPC and Real Time Optimization

- Flexible Modeling environment

- Unmatched operational awareness

- Lowest lifecycle cost

Seconds Hours/Days

Profit LoopExperion

Profit Controller(C300/ACE)

Profit Suite- Profit Controller

- Profit Sensor

- Profit Stepper

Profit Optimizer(DQP

Multi-Unit

Optimization)

Profit Executive(SuperDQP

Multi-Asset

Optimization)

Continuum of Control Solutions

Single Variable

Linear Control

Multiple-Variable Non –Linear

Control and Optimization

Profit CPM (Control Performance Monitor)


Profit Control and Optimization Roadmap3

Profit Optimization Suite R431

• Incremental release (simple install)

• CAB enhancements

• Engineering Studio updates

• Updated CV move block

Control Performance Monitor R600

• Next generation Diagnostic workflow

- Tune/Fix/Investigate (Expert Guidance)

• Intuition platform

- Coexistence with DynAMo

- Improved security/user management

Profit Optimization Suite R440

- APC Agnostic Multi-Unit Dynamic optimization

- Optimization engine interior point enhancements

- Operator Guidance enhancements for

optimization

- CAB enhancements

- Engineering Studio updates

- UOP Oleflex and Hydroprocessing toolkits

CPM R610

- Next generation MPC monitoring

- Optimization monitoring

- Integration with Intuition alerting (Pulse)

Profit R500

- Gate to gate Optimization

- Planning integrationExperion Profit Controller

- Profit Controller in CEE (C300/ACE)

- Part of Experion, includes design and visualization


R440 Focus – More out of the same Assets

• Optimization

- Optimize in a wider range of situations and environments

- Utilize UOP expertise

- Optimize across multiple environments

- Operator Guidance for Profit Optimizer

• User Experience

- New thin client

- Limits repository

- Models from Historical data

• Infrastructure

- Updates


SS Controller TargetsFeasible SS Controller Targets

SS

Model

Traditional Optimization Overview

• Build SS model for Optimizer

- Linear, NL, large scale

- Represents plant behavior

- Define objective function

• Calculate SS targets to

determine the end goal

• Realize the operating value by

dealing with dynamics

- Ensure feasibility

- Prevent constraint violations

- Push plant to SS

• Update the model, objective

function and dynamic

compensator

- Compensate for changing

conditions

- Repeat cycle

SS Optimizer

Economics and Ranges

SS Targets

Process

Dynamic Compensator

Limits

Process Changes

Feasibility

Feedback

LP Override Target Ramp Rates

Without Profit Optimization,

the LP Override must be engineered to provide

feasible SS controller targets and a realistic ramp

rate to the MPC controller

MPC Controller

Ramped Path


Profit Optimizer Concepts

Profit

controller

Model matrix

MVs

CVs

Base structure

Optimization Objective Function

Can Have multiple objective functions but only one executing

Profit Optimizer (DQP)

MVs

CVs



DV1 DV

x

x

x

MV1 MV2 MV1 MV2 DV1DV

x

x

x

MV1 MV2 MV1 MV2

x

x

x

x

x

x

x

x

x

x

x

x

DV1

DV

x

x

x

MV1 MV2 MV1 MV2

BM

Bridge models add dynamic feed forward dependence

Bridge Model inputs

Bridge Model output



MVs

CVs

y y y y y y yCombined

Constraints

and Aux data

• Combined Constraint Models are multi-unit steady state CVs (no feed back).

CV x x x


Profit Optimizer Concept

• Multiple applications built for good control and optimization performance

• Global objective function to fully utilize degrees of freedom

Time

Controller

Local optimizer

App 1

Controller

Local optimizerApp 2

Controller

Local optimizer App 3

Controller

Local optimizer App 4

Global SS

Overall Optimization

Horizon

Local optimizer


Profit Controller 2

Profit Optimizer

Ensures Optimal Dynamic Global Path

Profit Optimizer

Profit Controller 1

Global Economics and Ranges

SS MV Targets

Gains

MV Economics

Process 1

Profit Controller 3

Process 3Process 2

Limits

Optimal

Dynamic Path

Local Feasible SS MV targets

QP Override 1 QP Override 2 QP Override 3

Feedback

SS

Model

Profit Bridge

Gains

Profit Optimizer Link

Other Controller 2

Profit Optimizer coordinates

the economics of multiple

Profit Controllers for multi-unit

plant-wide optimization

Optimal

Dynamic Path

Optimal

Dynamic Path

Local Feasible SS MV targets Local Feasible SS MV targets


-5 -4 -3 -2 -1 0 1 2 3 4 5-5

-4

-3

-2

-1

0

1

2

3

4

5

x1

x2

Minimization of the JAE function

Start Point

Solution

1

65

43 2

7

Profit Optimizer will be enhanced to include

‘Hessian Updating’ which will:

- Allow optimizer to use

curvature-of-the-surface information

compared to simple gain updating

- Allow superior handling of unconstrained

(interior point) solutions

- Improve ability to deal with process non-linearities

- Faster convergence on the optimum solution

- Provide additional benefits of around 10%

(depending on process non-linearities)

R440 Optimization Enhancements

*Optimum


Starting Point

x2

x1

)x-(1)x-100(x f(x) 2

1

22

12

Rosenbrock's

Banana Function:

• Change Sequential QP (SQP) to Time-Sequenced QP (TSQP)

• Update the Hessian matrix and model-predicted constraints along the sequence

• Solve the time-sequenced QP successively inside the control feedback loop- If the model error or disturbance is negligible, the TSQP gets the same solution as SQP (or SS RTO)

- If the model error or disturbance is modest, the TSQP gets feedback corrections along the way

- If the model error or disturbance is/becomes large (over time), the TSQP should do better.

- We all know: Disturbances are part of life in process industries!

Our Approach: modify SQP for real-time use12

Update the objective

function, model gains,

future predictions and

constraints at a user-

specified frequency

(or condition)

Actual Optimal

P-Optimal Point


Honeywell Optimization Solutions

-5 -4 -3 -2 -1 0 1 2 3 4 5-5

-4

-3

-2

-1

0

1

2

3

4

5

x1

x2

Minimization of the JAE function

Start Point

Solution

Profit Optimizer – Dynamic Optimization

• Dynamic optimization

– no steady-state detection is needed

– 3-5 minute execution frequency

– Full formulation objective function w/ QP solver

• Cooperative optimization approach

– Shares common models with Profit Controller application(s)

– Calculates optimization speed for each controller depending on underlying process dynamics & settling time

– Patented bridge model technology: full dynamic relationships across the optimization scope.

• Gain scheduling for non-linear enhancement (via Profit Controller/Optimizer Gain Mappers)

– Direct move in the most profitable direction

– Global optimum when possible


OptimizationFuture Direction: Gate to gate


Refinery Model – An APC View

Stra

igh

t-Run

an

d/o

r Ble

nd

ing

Blending Components

Straight-run products

APC model has a greater resolution but limited scope• It captures essential unit or area operating constraints

• Optimizes unit production to possible detriment of plant


Refinery Model – A Planner’s View

The model has a high level of abstraction• It captures essential material and energy balances in holistic view of the refinery

• It leaves out many unnecessary or even obscuring details for the economic optimization

Stra

igh

t-Ru

n a

nd

/or B

len

din

g

Blending Components

Straight-run products

Feed

y

y

y

p

p

p

nn

2

1

2

1

Product


Planning – Execution Gap

Plan Execute

Planner uses aggregate yield models to predict

product makes for raw materials.

Detailed unit model is generally not required

and can lead to deleterious effects for solver

Planning model required to make complex

decisions for which raw materials to purchase

and which product to produce

Current plan may be infeasible for current

operating conditions

Mismatched operation can lead to imbalances

in plant inventories

Today’s yields may not match the planner’s

yield model resulting in different operation

Assumed planning constraints for unit

performance may be too aggressive or too

conservative compared to actual performance


Key Challenge:

Plantwide economics

Local economics

Optimal feasible

Control & Optimization

Schedule & Optimization

Production Planning

Business Planning

Planning (Months)

Scheduling (Day/Weeks)

Integration Scheme? How?

App -1 App -2 App -3 App -n

Real Time Dynamic Optimization - DQP (hr)

How to get the solution layers to stay consistent and

reach the global optimum jointly?

Manage intermediate and final:

• Inventory (volumes)

• Properties (quality)

• Timeline (just in time)


Implementation of the plan today

MPC 1-1

MPC 1-2

MPC1-3

MPC1-n

Optimizer -1

MPC2 -1

MPCw -1

MPCm-1

MPCm-p

Optimizer -m


Implementation of the plan tomorrow

Profit Executive

MPC 1-1

MPC 1-2

MPC1-3

MPC1-n

Optimizer -1

MPC2 -1

MPCw -1

MPCm-1

MPCm-p

Optimizer -m

Returned values can include

the actual versus predicted

yields and proxy limits

Manage intermediate and final:

• Inventory (volumes)

• Properties (quality)

• Timeline (just in time)

Manage:

• Unit operation

• Limits

• Constraints


Profit Executive Components

Planning Execution

OptimizerWhat-If steady state analysis using current

operating conditions

ControllerDynamic operating plan moves provided to

simulator

OptimizerGenerates optimal steady state operating

plan based on actual costs

ControllerMoves plant toward optimal while accounting

for immediate constraints

PredictorComputes yield/property model biases using

operating data

APC ExtensionsBounds optimization problem based on APC

constraints


1 2 3 4 5 6 7 8 9 10 11 12 130

500

1000

1500

2000Monthly Diesel Production in a More Profitable Product Mix

KB

arr

els

/Month

1 2 3 4 5 6 7 8 9 10 11 12 130

500

1000

1500

2000Historical (Acutal) Monthly Diesel Production

Month in the Study Year

KB

arr

els

/Month

Product A

Product B

Product C

Product D

Product E

Product F

Quality Giveaway

A more profitable product mix could have been produced in the study year

The actual product mix was produced and sold in the study year

Over-Qualification: $65M/year

~$3/Barrel of Product

The purpose for estimating the quality giveaway is to show:

how less optimally the components were produced at a potentially higher cost.

the potential room for reducing the component quality while still meeting the

same demand.

Actu

al

Optim

al


Plantwide Optimization - conclusions

• The planning model is fleshed out with dynamics and used in closed-loop control

• This Profit Executive solution provides better:

- Scalability – Suitable for different plant size, small or large

- Operability – Decentralized control is retained while providing centralized, Plantwide optimization

- Real-time responsiveness – The master runs like a regular MPC controller

• Benefits of better closed-loop Plantwide control

- Refineries can give away a significant amount of product quality due to inadequate closed loop control – ~$65M/year (~$3/barrel)

- The Profit Executive solution can capture ~$22M/year (~$1/barrel) without changing the product orders.

Honeywell Confidential - © 2016 by Honeywell International Inc. All rights reserved. © 2015 by Honeywell International Inc. All rights reserved.

cc

www.honeywell.com/ps

Optimization Sustainability

Optimizing Operations to Boost Your Ability to Complete in Any Market Environment

Technology