Multivariable Predictive Control E-Book

Sandip K. Lahiri

• Herausgeber: John Wiley & Sons
• Kategorie: Fachliteratur
• Sprache: Englisch

Seitenzahl: 529

Veröffentlichungsjahr: 2017

Beschreibung

A guide to all practical aspects of building, implementing, managing, and maintaining MPC applications in industrial plants

Multivariable Predictive Control: Applications in Industry provides engineers with a thorough understanding of all practical aspects of multivariate predictive control (MPC) applications, as well as expert guidance on how to derive maximum benefit from those systems. Short on theory and long on step-by-step information, it covers everything plant process engineers and control engineers need to know about building, deploying, and managing MPC applications in their companies.

MPC has more than proven itself to be one the most important tools for optimising plant operations on an ongoing basis. Companies, worldwide, across a range of industries are successfully using MPC systems to optimise materials and utility consumption, reduce waste, minimise pollution, and maximise production. Unfortunately, due in part to the lack of practical references, plant engineers are often at a loss as to how to manage and maintain MPC systems once the applications have been installed and the consultants and vendors’ reps have left the plant. Written by a chemical engineer with two decades of experience in operations and technical services at petrochemical companies, this book fills that regrettable gap in the professional literature.

• Provides a cost-benefit analysis of typical MPC projects and reviews commercially available MPC software packages
• Details software implementation steps, as well as techniques for successfully evaluating and monitoring software performance once it has been installed
• Features case studies and real-world examples from industries, worldwide, illustrating the advantages and common pitfalls of MPC systems
• Describes MPC application failures in an array of companies, exposes the root causes of those failures, and offers proven safeguards and corrective measures for avoiding similar failures

Multivariable Predictive Control: Applications in Industry is an indispensable resource for plant process engineers and control engineers working in chemical plants, petrochemical companies, and oil refineries in which MPC systems already are operational, or where MPC implementations are being considering.

Leseprobe

Table of Contents

Cover

Title Page

Figure List

Table List

Preface

1 Introduction of Model Predictive Control

1.1 Purpose of Process Control in Chemical Process Industries (CPI)

1.2 Shortcomings of Simple Regulatory PID Control

1.3 What Is Multivariable Model Predictive Control?

1.4 Why Is a Multivariable Model Predictive Optimizing Controller Necessary?

1.5 Relevance of Multivariable Predictive Control (MPC) in Chemical Process Industry in Today’s Business Environment

1.6 Position of MPC in Control Hierarchy

1.7 Advantage of Implementing MPC

1.8 How Does MPC Extract Benefit?

1.9 Application of MPC in Oil Refinery, Petrochemical, Fertilizer, and Chemical Plants, and Related Benefits

References

2 Theoretical Base of MPC

2.1 Why MPC?

2.2 Variables Used in MPC

2.3 Features of MPC

2.4 Brief Introduction to Model Predictive Control Techniques

References

3 Historical Development of Different MPC Technology

3.1 History of MPC Technology

3.2 Points to Consider While Selecting an MPC

References

4 MPC Implementation Steps

4.1 Implementing a MPC Controller

4.2 Summary of Steps Involved in MPC Projects with Vendor

References

5 Cost–Benefit Analysis of MPC before Implementation

5.1 Purpose of Cost–Benefit Analysis of MPC before Implementation

5.2 Overview of Cost–Benefit Analysis Procedure

5.3 Detailed Benefit Estimation Procedures

5.4 Case Studies

References

6 Assessment of Regulatory Base Control Layer in Plants

6.1 Failure Mode of Control Loops and Their Remedies

6.2 Control Valve Problems

6.3 Sensor Problems

6.4 Controller Problems

6.5 Process‐Related Problems

6.6 Human Factor

6.7 Control Performance Assessment/Monitoring

6.8 Commonly Used Control System Performance KPIs

6.9 Tuning for PID Controllers

References

7 Functional Design of MPC Controllers

7.1 What Is Functional Design?

7.2 Steps in Functional Design

References

8 Preliminary Process Test and Step Test

8.1 Pre‐Stepping, or Preliminary Process Test

8.2 Step Testing

8.3 Development of Step‐Testing Methodology over the Years

Reference

9 Model Building and System Identification

9.1 Introduction to Model Building

9.2 Key Issues in Model Identifications

9.3 The Basic Steps of System Identification

9.4 Model Structures

9.5 Common Features of Commercial Identification Packages

References

10 Soft Sensors

10.1 What Is a Soft Sensor?

10.2 Why Soft Sensors Are Necessary

10.3 Types of Soft Sensors

10.4 Soft Sensors Development Methodology

10.5 Data‐Driven Methods for Soft Sensing

10.6 Open Issues and Future Steps of Soft Sensor Development

References

11 Offline Simulation

11.1 What Is Offline Simulation?

11.2 Purpose of Offline Simulation

11.3 Main Task of Offline Simulation

11.4 Understanding Different Tuning Parameters of Offline Simulations

11.5 Different Steps to Build and Activate Simulator in an Offline PC

11.6 Example of Tests Carried out in Simulator

11.7 Guidelines for Choosing Tuning Parameters

References

12 Online Deployment of MPC Application in Real Plants

12.1 What Is Online Deployment (Controller Commissioning)?

12.2 Steps for Controller Commissioning

References

13 Online Controller Tuning

13.1 What Is Online MPC Controller Tuning?

13.2 Basics of Online Tuning

13.3 Guidelines to Choose Different Tuning Parameters

References

14 Why Do Some MPC Applications Fail?

14.1 What Went Wrong?

14.2 Failure to Build Efficient MPC Application

14.3 Contributing Failure Factors of Postimplementation MPC Application

14.4 Strategies to Avoid MPC Failures

References

15 MPC Performance Monitoring

15.1 Why Performance Assessment of MPC Application Is Necessary

15.2 Types of Performance Assessment

15.3 Benefit Measurement after MPC Implementation

15.4 Parameters to Be Monitored for MPC Performance Evaluation

15.5 KPIs to Troubleshoot Poor Performance of Multivariable Controls

15.6 Exploitation of Constraints Handling and Maximization of MPC Benefit

References

16 Commercial MPC Vendors and Applications

16.1 Basic Modules and Components of Commercial MPC Software

16.2 Major Commercial MPC Software

16.3 AspenTech and DMCplus

16.4 RMPCT by Honeywell

16.5 SMOC–Shell Global Solution

16.6 Conclusion

References

Index

End User License Agreement

List of Tables

Chapter 01

Table 1.1 Typical Payback Period of MPC

Table 1.2 Typical Benefits of MPC Implementation in CPI

Table 1.3 Typical Benefits of MPC implementation in Refinery

Chapter 02

Table 2.1 Description of CV, MV, and DV in a Simple Distillation Column Shown in Figure 2.4

Chapter 05

Table 5.1 Typical Value of Factor β

Table 5.2 Average Value and Standard Deviation of Quality Parameters

Chapter 06

Table 6.1 Typical Performance of Control Loops in Industry

Table 6.2 Ziegler‐Nichols Tuning Parameters

Table 6.3 Recommended PID Tuning Parameters

Table 6.4 IMC Tuning Parameters

Chapter 08

Table 8.1 Difference between Normal Step Testing and PRBS Testing

Chapter 11

Table 11.1 Simulation Initial Condition File for MVs

Table 11.2 Simulation Initial Condition File for CVs

Table 11.3 Controlled Variables with Their Limits for Simulation Studies

Table 11.4 Controlled Variables with Their Limits for Simulation Studies

Chapter 14

Table 14.1 Retaining Initial MPC Benefits after 12 Months

Table 14.2 Commercial MPC Monitoring Tools

List of Illustrations

Chapter 01

Figure 1.1 Flow scheme of a simple distillation column using multivariable model predictive controller

Figure 1.2 Hierarchy of plant‐wide control framework

Figure 1.3 Expected cost vs. benefits for different levels of controls

Figure 1.4 Typical benefit of MPC

Figure 1.5 MPC stabilization effect can increase plant capacity closer to its maximum limit

Figure 1.6 Reduced variability allows operation closer to constraints by shifting set point

Figure 1.7 Operating zone limited by multiple constraints

Figure 1.8 Opportunity loss due to operator action

Figure 1.9 Advance control implementations by one of the major MPC vendors

Figure 1.10 Spread of MPC application across the whole spectrum of chemical process industries

Chapter 02

Figure 2.1 Optimum operating point vs. operator comfort zone

Figure 2.2 Different module of MPC

Figure 2.3 A general MPC calculation

Figure 2.4 Schematic of distillation column

Figure 2.5 Model of distillation column

Figure 2.6 CV prediction due to past MV change

Figure 2.7 Model reconciliation and bias update

Figure 2.8 Operating region of a distillation column with two manipulated variables and six controlled variables

Figure 2.9 Revised CV trajectory and steady state error

Figure 2.10 Develop a detail plan of MV movement to drive the steady state error to zero

Figure 2.11 Controlled variables predictions with and without control moves

Figure 2.12 Manipulated variables move plan for distillation column

Chapter 03

Figure 3.1 Brief history of development of MPC technology

Chapter 04

Figure 4.1 Different steps in MPC implementation project

Figure 4.2 Schematics of steps involved in MPC project with vendor

Chapter 05

Figure 5.1 Benefit estimation procedure

Figure 5.2 Stabilizing effect of MPC and moving of set point closer to limit

Chapter 06

Figure 6.1 Various probable reasons of failure of control loops

Figure 6.2 Valve sizing problem detection by process gain

Figure 6.3 Typical trends when valve stiction presents

Figure 6.4 Typical trends of the process having hysteresis and backlash

Chapter 07

Figure 7.1 Different steps in functional design

Chapter 08

Figure 8.1 Expectation matrix (√ definite response expected, X no response expected, ? response is doubtful)

Figure 8.2 Basic concept of step test

Chapter 09

Figure 9.1 Advantages and disadvantages of various model structures

Figure 9.2 Flowchart of identification process

Figure 9.3 System identification structure

Chapter 10

Figure 10.1 Types of soft sensors

Figure 10.2 Steps involved in developing reliable soft sensors

Figure 10.3 Artificial neural network architecture

Figure 10.4 Schematic of SVR using an e‐insensitive loss function

Chapter 11

Figure 11.1 Different tuning parameters

Figure 11.2 Hard and soft limits

Chapter 12

Figure 12.1 The schematic of interface of MPC controller and DCS

Figure 12.2 Schematic of online commissioning of the controller

Chapter 13

Figure 13.1 Effect of move suppression (or MV weight) on CV and MV trajectory

Figure 13.2 Effect of CV give up on CV trajectory and CV error

Chapter 14

Figure 14.1 Benefit loss over time

Figure 14.2 Contributing failure factors of postimplementation of MPC applications

Figure 14.3 Strategies for avoiding MPC failures

Chapter 16

Figure 16.1 Major linear MPC companies and their products

Figure 16.2 Basic structure of MPC software

Figure 16.3 Comparison of different MPC identification technology

Figure 16.4 DMCplus product package

Figure 16.5 MPC project outline: Conventional vs. adaptive approach

Figure 16.6 Optimization in adaptive control mode

Figure 16.7 RMPCT product package

Figure 16.8 History of SMOC

Figure 16.9 SMOC product package

Guide

Cover

Table of Contents

Begin Reading

Pages

iii

iv

v

xix

xx

xxi

xxiii

xxiv

xxv

xxvi

xxvii

xxviii

xxix

xxx

xxxi

xxxii

xxxiii

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

Multivariable Predictive Control

Applications in Industry

Sandip Kumar Lahiri

This edition first published 2017© 2017 John Wiley & Sons Ltd

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

The right of Sandip Kumar Lahiri to be identified as the author of this work has been asserted in accordance with law.

Registered OfficesJohn Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USAJohn Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

Editorial Office111 River Street, Hoboken, NJ 07030, USAThe Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Wiley also publishes its books in a variety of electronic formats and by print‐on‐demand. Some content that appears in standard print versions of this book may not be available in other formats.

Limit of Liability/Disclaimer of WarrantyIn view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

Library of Congress Cataloging‐in‐Publication Data

Names: Lahiri, Sandip Kumar, 1970– author.Title: Multivariable predictive control : applications in industry / Sandip Kumar Lahiri, Supra International Private Ltd, Vadodara, India.Description: First edition. | Hoboken, NJ, USA : Wiley, 2017. | Includes bibliographical references and index. | Description based on print version record and CIP data provided by publisher; resource not viewed.Identifiers: LCCN 2017010553 (print) | LCCN 2017012540 (ebook) | ISBN 9781119243519 (pdf) | ISBN 9781119243595 (epub) | ISBN 9781119243601 (cloth)Subjects: LCSH: Predictive control. | Multivariate analysis.Classification: LCC TJ217.6 (ebook) | LCC TJ217.6 .L34 2017 (print) | DDC 629.8–dc23LC record available at https://lccn.loc.gov/2017010553

Cover design by WileyFront Cover Image: FotoBug11/ShutterstockBack Cover Image: Paulo Vilela/Shutterstock

To my parents, wife Jinia and two lovely children Suchetona and Srijon

Figure List

Figure 1.1

Flow scheme of a simple distillation column using multivariable model predictive controller

Figure 1.2

Hierarchy of plant‐wide control framework

Figure 1.3

Expected cost vs. benefits for different levels of controls

Figure 1.4

Typical benefit of MPC

Figure 1.5

MPC stabilization effect can increase plant capacity closer to its maximum limit

Figure 1.6

Reduced variability allows operation closer to constraints by shifting set point

Figure 1.7

Operating zone limited by multiple constraints

Figure 1.8

Opportunity loss due to operator action

Figure 1.9

Advance control implementations by one of the major MPC vendors

Figure 1.10

Spread of MPC application across the whole spectrum of chemical process industries

Figure 2.1

Optimum operating point vs. operator comfort zone

Figure 2.2

Different module of MPC

Figure 2.3

A general MPC calculation

Figure 2.4

Schematic of distillation column

Figure 2.5

Model of distillation column

Figure 2.6

CV prediction due to past MV change

Figure 2.7

Model reconciliation and bias update

Figure 2.8

Operating region of a distillation column with two manipulated variables and six controlled variables

Figure 2.9

Revised CV trajectory and steady state error

Figure 2.10

Develop a detail plan of MV movement to drive the steady state error to zero

Figure 2.11

Controlled variables predictions with and without control moves

Figure 2.12

Manipulated variables move plan for distillation column

Figure 3.1

Brief history of development of MPC technology

Figure 4.1

Different steps in MPC implementation project

Figure 4.2

Schematics of steps involved in MPC project with vendor

Figure 5.1

Benefit estimation procedure

Figure 5.2

Stabilizing effect of MPC and moving of set point closer to limit

Figure 6.1

Various probable reasons of failure of control loops

Figure 6.2

Valve sizing problem detection by process gain

Figure 6.3

Typical trends when valve stiction presents

Figure 6.4

Typical trends of the process having hysteresis and backlash

Figure 7.1

Different steps in functional design

Figure 8.1

Expectation matrix (√ definite response expected, X no response expected, ? response is doubtful)

Figure 8.2

Basic concept of step test

Figure 9.1

Advantages and disadvantages of various model structures

Figure 9.2

Flowchart of identification process

Figure 9.3

System identification structure

Figure 10.1

Types of soft sensors

Figure 10.2

Steps involved in developing reliable soft sensors

Figure 10.3

Artificial neural network architecture

Figure 10.4

Schematic of SVR using an e‐insensitive loss function

Figure 11.1

Different tuning parameters

Figure 11.2

Hard and soft limits

Figure 12.1

The schematic of interface of MPC controller and DCS

Figure 12.2

Schematic of online commissioning of the controller

Figure 13.1

Effect of move suppression (or MV weight) on CV and MV trajectory

Figure 13.2

Effect of CV give up on CV trajectory and CV error

Figure 14.1

Benefit loss over time

Figure 14.2

Contributing failure factors of postimplementation of MPC applications

Figure 14.3

Strategies for avoiding MPC failures

Figure 16.1

Major linear MPC companies and their products

Figure 16.2

Basic structure of MPC software

Figure 16.3

Comparison of different MPC identification technology

Figure 16.4

DMCplus product package

Figure 16.5

MPC project outline: Conventional vs. adaptive approach

Figure 16.6

Optimization in adaptive control mode

Figure 16.7

RMPCT product package

Figure 16.8

History of SMOC

Figure 16.9

SMOC product package

Table List

Table 1.1

Typical Payback Period of MPC

Table 1.2

Typical Benefits of MPC Implementation in CPI

Table 1.3

Typical Benefits of MPC implementation in Refinery

Table 2.1

Description of CV, MV, and DV in a Simple Distillation Column Shown in

Figure 2.4

Table 5.1

Typical Value of Factor β

Table 5.2

Average Value and Standard Deviation of Quality Parameters

Table 6.1

Typical Performance of Control Loops in Industry

Table 6.2

Ziegler‐Nichols Tuning Parameters

Table 6.3

Recommended PID Tuning Parameters

Table 6.4

IMC Tuning Parameters

Table 8.1

Difference between Normal Step Testing and PRBS Testing

Table 11.1

Simulation Initial Condition File for MVs

Table 11.2

Simulation Initial Condition File for CVs

Table 11.3

Controlled Variables with Their Limits for Simulation Studies

Table 11.4

Controlled Variables with Their Limits for Simulation Studies

Table 14.1

Retaining Initial MPC Benefits after 12 Months

Table 14.2

Commercial MPC Monitoring Tools

Preface

In chemical process industries, there is an ongoing need to reduce cost of production and increase profit margin. Due to cut‐throat competition on a global level, the major chemical industries are now competing to optimize raw material and utility consumption, to reduce waste, to reduce emission, and to minimize pollution. Multivariable model predictive control (MPC) is considered as an excellent tool to achieve those goals. The benefit of implementing MPC are many. MPC optimizes the plant operation on a continuous basis, reduces waste and utility consumption, minimizes raw material consumption, and maximizes production. Due to these benefits, all major chemical industries, petrochemical industries, and oil refineries throughout the globe are implementing MPC in their plants.

However, there are no dedicated books available to discuss the basic concepts of MPC, provide practical guidelines, and explain industrial application procedures.

The main idea of writing this book is to fill this gap with the following people in mind: managers, process engineers, control engineers, operators working in the process industries, and chemical engineering students who want to pursue process control career.

MPC is normally implemented by an external MPC consultant company or experts such as AspenTech, Honeywell, and Shell. The practicing process engineers or process control engineers working in the plant normally have much less exposure or knowledge to implement MPC. The available books in market on MPC don’t cover the practical aspects to implement commercial MPC software.

The available books on MPC emphasize unnecessary theoretical details, which are normally not required by the practicing engineers, and those theories have very little relevance for commercial implementation of MPC software. This book discusses the practical aspects of MPC implementation and maintenance. The consultants or experts coming from MPC vendor companies normally implement MPC, hand over the technology to client plant, and then leave. After they leave, the responsibility goes to plant process engineers and control engineers to keep the MPC software running, derive maximum benefit from it, and sustain those benefits by proactive maintenance. So plant engineers need to have a thorough understanding about the different features of MPC software and key implementation steps. Often, due to unavailability of literature on this subject, plant engineers lack the knowledge and understanding of MPC.

The book is intended to build an overall understanding of MPC implementation and how to derive maximum benefit from MPC. It covers everything that a practicing process engineer or process control engineer needs to know to build an effective MPC application. Practical considerations of MPC implementation are emphasized over unnecessary theoretical details. The book covers a wide range of subjects of MPC applications, starting from an initial functional design stage to final implementation stage. Readers will also get enlightened as to why many MPC applications fail in industries across the globe. The root causes of this failure are discussed in detail so that readers of the book can safeguard and take preventive and corrective action beforehand to avoid MPC failure.

As this book covers a wide range of topics, the materials are organized in such a way that helps the reader to locate the relevant chapters quickly, to be able to understand them readily, and to apply them in the right context. The book is organized in the following way.

Overview of Contents

Chapter 1 gives an overview of the importance of multivariable predictive control (MPC) in chemical process industries in the context of today’s competitive business environment. The benefits of implementing MPC over normal Proportional‐Integral‐Derivative (PID)‐type regulatory control and how MPC brings this benefit in real commercial chemical plant are explained in detail here. A brief description of MPC working principle is also discussed. The purpose of process control in chemical process industries (CPIs) is to ensure safety, maintain product quality and operational constraints while trying to maximize economic benefit. Traditionally, PID controllers are used in CPIs. However, PID controllers are not efficient to handle multivariable processes with significant interactions. Multivariable model predictive optimizing controller understands these process interactions and makes multiple small moves with the help of its model predictive capability. By doing this, it slowly brings the process to the most economic operating zone while maintaining all the process parameters within their limits. MPC acts as a supervisory controller above base‐level PID control and is situated at the middle of a multilevel control hierarchy. The relevance of multivariable predictive control (MPC) in chemical process industry in today’s business environment is very high while industries are struggling to reduce operating cost, maximize profit margin, and reduce waste. MPC stabilizes the process by utilizing its model predictive capability and thus allows the operation near to constraints. MPC is applied in oil refinery, petrochemical, fertilizer, and chemical plants across the globe and they bring huge amount of profit. The chapter ends with practical examples of MPC implementations in various process industries starting from petrochemicals, petroleum refinery to fertilizer and many other chemical plants.

Chapter 2 deals with theoretical foundation of MPC. Different variables and commonly used terms in MPC are introduced in the chapter. Different features of MPC controller are explained in detail. A simple algorithm explains the reader the underlying calculation steps of MPC technology. Simplified dynamic control strategy of MPC controllers are discussed in detail to develop an understanding of how it works. One of the major features of MPC is its future prediction and constraint handling capability. The theoretical background of these two main features is explained in detail with examples.

Model predictive controllers (MPCs) have many features. They are multivariable controllers with model‐based predictive capability. They continuously optimize the process by rigorously planning and executing small movement in manipulated variables (MVs). As simple architecture, they have data collection module, control variable (CV) prediction module, steady‐state optimization module, and dynamic optimization module. The process starts with reading current value of controlled variable and MV, and using its internal process model it predicts the future value of controlled variable. In every execution, it reconciles this prediction value with actual process measurements to compensate for model inaccuracies. Also, it calculates the size of the control process (i.e., number of MV and CV available for control purpose) in every execution and sees whether any change in size is made by panel operator. Its steady‐state optimization module then calculates the optimum targets for CV and MV and feeds this information to dynamic module to plan detail MV movement to achieve those targets. Depending on various tuning parameters and MV–CV limits, dynamic module initially plans for a series of MV movements so that those targets can be achieved and process can be brought to the most economic optimum zone. The first step of MV movement is actually implemented through PID controllers and all other moves are discarded. In next execution, again all the calculations are repeated.

The chapter explains all of the aforementioned features in a simple way.

Historical developments of different MPC technology are described in detail in Chapter 3. First‐generation MPC was developed in 1970s. Over the years, MPC technology went through various modifications and additions of different features and reached currently as fifth generations MPC technology. The genesis of these developments over the years, the need, and innovations at different generations are discussed in this chapter. MPC control algorithm is developed over the years starting from 1970s. The initial IDCOM, an acronym for Identification and Command and Dynamic matrix control (DMC) algorithms represent the first generation of MPC technology (1970–1980); they had an enormous impact on industrial process control and served to define the industrial MPC paradigm. Engineers at Shell Oil continued to develop the MPC algorithm and addressed the weakness of first‐generation algorithm by injecting quadratic program (QP) in DMC algorithm. The QDMC algorithm can be regarded as representing a second generation (1980–1985) of MPC technology, comprising algorithms that provide a systematic way of implementing input and output constraints. However, after initial phase, MPC technology slowly started to get huge profit and gain wider acceptance during the 1990s. The Identification and Command, modified version (IDCOM‐M), Hierarchical constraint control (HIECON), Single Multivariable Control Architecture (SMCA), and Shell Multivariable Optimizing Controller (SMOC) algorithms represent a third generation of MPC technology (1985–1990); others include the predictive control technology (PCT) algorithm sold by Profimatics, and the RMPC algorithm sold by Honeywell. In the era of 1990–2000, increased competition and the mergers of several MPC vendors have led to significant changes in the industrial MPC landscape. Major MPC companies started acquisition and wanted to dominate the market. AspenTech and Honeywell got out as the winners of this phase and represent fourth‐generation MPC (1990–2000). Today, we are witnessing a further technology development that is not so much focused on improving the algorithms, but to improve the development steps. This represents fifth‐generation algorithm (2000–2015). The focus is put to make those steps smoother, faster, and easier, for both the developer and the client, and to do as much as possible remotely. The chapter enlightens readers on all of the aforementioned areas.

Implementing MPC in chemical plants is itself a project and involves lot of steps. Chapter 4 gives an overview about the various stages of MPC implementation starting from an assessment of existing regulatory control, functional design of MPC, model building and final MPC implementation stages. It starts with preliminary cost–benefit analysis to evaluate approximate payback period. Assessment of base control loop and strengthening it is a basic requirement to build a solid foundation upon which MPC works. In functional design step, a list of controlled and MVs are identified. Plant step test is carried out to collect dynamic data of CV for a step change in MV. These step test data are utilized to build models in model building stage. Potential soft sensors are made where online analyzers are either not available or very costly. The suitability of developed model for control purpose is checked in off‐line simulation mode. After that, controller is commissioned in actual plant and online tuning is done to achieve the desired controller action. As a last step, performance monitoring and benefit assessment of installed MPC controller is done. An essential part of each step is to train the plant operators and engineer regarding different features of MPC and how to operate the installed MPC application. The chapter also explains the steps involved in MPC projects with vendor.

Normally, the implementation of MPC involves cost that includes MPC software, hardware cost, and MPC vendor cost. Client plants who want to implement MPC always want to know about the benefit or payback period of MPC implementations before they decide to go for MPC implementation. Chapter 5 describes cost–benefit analysis procedures before MPC implementation.

Preliminary cost–benefit analysis is usually carried out before starting MPC project. The purpose is to estimate the actual benefit after MPC implementation. A scouting study of process analysis and economic opportunity analysis is done to know the potential areas where MPC can bring profit. By its model‐based predictive capability MPC stabilizes the process and reduces variability of key process parameters. This reduction of variability enables operators to shift the set point closer to the constraints. Operation closer to constraints translated into more profit. By statistical analysis, this increase of profit due to MPC implementation is calculated. Finally, a scientific cost–benefit analysis is done to evaluate the payback period. The results of the cost–benefit analysis help the plant management to take economic decision to implement MPC in plant. An example with practical case study is also given to explain the cost–benefit analysis procedure.

Chapter 6 explains the procedure to assess the health of regulatory base control layer of plant. MPC cannot work efficiently if base control layer or regulatory control layer is weak. Hence, strengthening base control layer is an important prerequisite to build the good foundations of MPC. Over the years, process industries technical community realizes the importance of monitoring the base control loop performance. The benefits gained from detecting the weakly performed control loop and subsequently improving their performance are huge. Assessment of regulatory base control layer in plant starts with understanding different common failure mode of valves, sensor, controller, and so on. Control valves may malfunction due to hysteresis, stickiness, and improper valve sizing. Sensors exhibit different problems such as noisy indication, improper calibration, and overfiltration, to name a few. Controllers commonly have tuning problems. Sometimes, process also has problems such as variable gains and too much interaction. Due to a large number of control loops present in any moderate‐sized process industries, manual evaluation of each control loop performance is not feasible. Online systematic performance monitoring of control loops through various key performance indices (KPIs) and matrices is the need of the hour. This gives rise to a new technology/software called control performance monitoring/assessment (CPM/CPA). Performance KPIs are generated and monitored online, and they are grouped as follows: traditional KPIs, statistical‐based metrics, business/operational metrics, and advanced indices. The chapter ends with giving a short exposure of controller tuning for PID controllers.

Functional design is the most important step in MPC project. Functional design is the proper planning and design of MPC controller to achieve operational and economic objective of the plant. There is no standard procedure to be followed to do a functional design. It depends on expertise and experience of MPC vendor or control engineer, plant operating people, and plant process engineering people.

Chapter 7 explains in detail about various aspects and practical considerations of functional designs of MPC controller in actual commercial plants. This step starts with understanding of process opportunity and process constraints. Process controls objective, controller scope, and identification of CV–MV–DV list is done in this step. Exploring the potential optimization opportunity is a key job in functional design stage. Identification of any scope to implement the inferential calculations or soft quality estimators is also done in this stage. Conceptualization of economic objective of controller and form of linear program (LP) and quadratic program (QP) objective function is finalized in this step.

Functional design of MPC controller started with the identification of controlled and MV and subsequent planning for MPC model structures. Practical considerations to identify process and equipment constraints are also discussed to help the reader formulate a robust, safe, and reliable MPC model. Good step test data is of paramount importance in MPC model building and its overall functioning. How to ideally perform step test in actual shop floor of the plant and do’s and don’ts of step test are discussed in detail. The chapter also briefly explains the requirement of soft sensor building.

Chapter 8 deals with preliminary process step and step test. Step test is considered as one of the major steps in MPC project. In step testing, step change in MVs is given and the impact of it on CVs with time is collected through step test data. These data are used to build process model. Both open‐loop and closed‐loop test are practiced in industry, and both methods have their own advantages and limitations. As the MPC models are data‐driven empirical models and those data are generated in step test, it is very important to carry out this test with all precautions. The quality of developed model will be as good or as bad as step test data. Hence, it is important to know all do’s and don’ts of step testing method. To reduce the unnecessary problems in step test, a preliminary process test or pre‐stepping is done before step test. The purpose is to identify all the possible bad actors of step test and rectify them beforehand. The chapter explains various do’s and don’ts in step test.

Chapter 9 describes in detail about the various model building procedures available in commercial software. Process models are dynamic MV–CV relationship generated from step test data. In model building step, a suitable model structure with proper order is first identified. Later on, model coefficients are evaluated from step test data by statistical fitting operation. Various data cleaning methods and outliers detection are discussed. Basic steps of process identifications start with execution of step test and collection of data, pre‐processing and cleaning of data, selection of model structure and order, and determination of model parameters. There are a lot of predefined dynamic model structures available in the library of commercial identification software. Knowing those structures and their key strength and weakness and finally identifying a suitable structure to accurately model the step data is the key of system identification step.

Theoretical background of various available models and their implication in MPC is explained in detail in the chapter. One of the major requirements for robustness of MPC model is to validate the developed data‐driven MPC model from practical process knowledge so that the model captures the underlying physics of the process. This important aspect is discussed in detail to give the reader a flavor regarding efficient and robust model building. Practical considerations to prioritize MVs to control particular CV in multivariable environment are discussed in detail so that user can maximize the economic benefit after MPC implementation.

An inferential or soft sensor is a mathematical relation that calculates or predicts a controlled property using other available process data. When it is very difficult or costly to measure an important parameter online, such as distillation tower top product impurity, soft sensors are used to predict that inferential property from other easy measurable parameters such as top temperature and pressure. Sometimes, soft sensors are used as backup of an existing analyzer to reduce or eliminate dead time, both from the process and the analyzer cycle.

Chapter 10 is dedicated for soft sensors available in various process industries. What are soft sensors and how to make them is the main idea of the chapter. Various commonly used algorithms to build fast principle‐based and black‐box‐based soft sensors models are discussed in detail. Why some soft sensors fail in industry and precautions needed to make successful robust soft sensors are described in the chapter.

Usually, four types of soft sensors are used in industry, namely, first principle–based soft sensor, data‐driven soft sensors, gray model–based soft sensors, and hybrid model–based soft sensors. There are many methods to develop industrial soft sensors and usually they include the following steps: data collection and data inspection, data preprocessing and data conditioning, selection of relevant input–output variables, aligning data, model selection, training and validation, analyzing dynamics, and finally deployment and maintenance.

Due to the difficulties in developing first principle–based soft sensors, data‐driven soft sensors are very popular in industry. Major data‐driven methods for soft sensing which dominates the industry and discussed in this chapter are principle component analysis, partial least squares, artificial neural networks, neuro‐fuzzy systems, and support vector machines.

After development of process model, it is important to know how the developed controller will perform in online mode before its actual deployment in real plant. Off‐line simulation refers to running the controller in a separate off‐line PC to see the MV–CV dynamic responses of the process. One major task of off‐line simulation is to set the different tuning parameters of the controller. The purpose is to perform off‐line tuning and other corrections as much as possible so that the application runs effortlessly in actual plant at real time.

Chapter 11 is dedicated to off‐line simulation of MPC model—an important prerequisite step for the MPC online implementation. How to set up off‐line simulation in MPC software and how to derive maximum benefit from them is the main focus of the chapter. Constraint handling capability of developed MPC model can be assessed in off‐line simulation. How to learn and modify the MPC model structure and tuning parameters from off‐line simulation response is explained in detail in the chapter.

Off‐line tuning involves setting proper priority for CV and MVs, CV give up priority in case of infeasible solution, setting up optimizer speed and different coefficient of LP and QP objective function. Before starting off‐line simulations, it is important to understand the concept of different tuning parameters available in MPC software package. It is also important to understand how MPC works in a dynamic environment and how different tuning parameters can impact its performance. These are explained in detail in the chapter. Usually, there are three major categories of tuning parameters, namely, tuning parameters for CVs, tuning parameters for MVs, and tuning parameters for optimizer. Various simulation tests can be planned, configured, and run in off‐line simulator to assess the different features and functionality of the developed controller in off‐line mode. Changes in different tuning parameters are done on trial‐and‐error basis until a satisfactory dynamic performance of MPC controller is achieved.

Online deployment of MPC application in real plant means connecting the MPC controller online with the plant Distributed control system (DCS) and allowing it to take control of the plant.

Chapter 12 is dedicated to the most important steps in MPC implementation—online deployment in real plant. Various stages of real‐time deployment of MPC software and precautions to be taken in open‐ and closed‐loop deployment are described in detail. Unless these precautions are taken, MPC software may lead to bumpy control of processes and shutdown of the plant in worst case. Different vendors have different methodology to commission the controller. However, the basic steps remain the same and are as follows: setting up the controller configuration and final review of the model, building the controller, load operator station on PC near the panel operator, taking MPC controller in line with prediction mode, putting the MPC controller in closed loop with one CV at a time, observation of MPC controller performance, putting optimizer in line and observation of optimizer performance, evaluation of overall controller performance, and online tuning and troubleshooting. Monitoring of MPC model performance after deployment and understanding the weakness of the developed model (if any) is key to make robust MPC application. Readers can gain insights of these features in the chapter. It is important to understand the purpose and details of implementations of the aforementioned step to avoid any malfunctioning of controller during commissioning phase. Care should be taken such that any mistake during commissioning does not lead to plant shutdown or plant upset. After the controller commissioning, proper documentation, and training of operators, engineers on online platform was usually done. Later on, some adjustments in different MV–CV limits and controller tuning parameters are done periodically to sustain the benefit of MPC controller.

Chapter 13 is dedicated to an important aspect—MPC controller online tuning.

Online controller tuning means changing of various tuning parameters of controllers online so that an optimal and expected performance is achieved by MPC controller in actual plant environment. If performance of MPC controller is not at par, it is recommended to investigate the root cause and troubleshoot the problem rather than jumping to tune the controller. How to systematically investigate and troubleshoot the problems of MPC controller is discussed in the chapter. As MPC is a multivariable controller; any change of one tuning parameters will affect many CVs and other MVs movement. It is important to understand the impacts of various tuning parameters on dynamic performance of controller during online tuning. There is always balance and compromise in online tuning. How to get that delicate balance is key to controller tuning. With proper knowledge, tuning parameters are modified by trial and error in online controller until an acceptable optimal dynamic performance is achieved.

Unlike PID controller tuning, MPC controller tuning involves many parameters and requires deep knowledge of MPC functioning and its overall impact on process. After reading the chapter, the reader can understand various free parameters available for tuning and how to tune MPC controller to make it robust and efficient.

The chapter describes various limits and constraints applied on MV and CV in commercial MPC packages. The idea behind putting operator limit, steady‐state limit, engineering limit, and so on is discussed. How these limits impact the overall MPC performance is explained. Practical considerations to set these limits to gain maximum benefit are also discussed.

Chapter 14 enlightened the reader why some MPC application fails in industries. There are several instances all over the world that MPC brings huge profit just for 1–2 years of its implementation and then profit starts decreasing. In extreme case, some oil refineries reported that MPC application does not generate any additional benefits even after 4–5 years of implementation over the simple regularity control. Different root causes of MPC failure in industry are explained to give the reader an idea about what can go wrong. User can gain insights about how to safeguard MPC performance deterioration in the long run. The chapter ends with the various unsuccessful and failure case study of MPC application in various industries across the globe.

There are two modes of failure, namely, failure to build efficient MPC application when it was first build and gradual deterioration of MPC performance post implementation. Reasons such as capability of technology to capture benefit, expertise of implementation team, and reliability of Advance process control (APC) project methodology are responsible for the failure after it was first build. Contributing failure factors of post implementation of MPC application are attributed to the following: lack of performance monitoring of MPC application, unresolved basic control problems, poor tuning and degraded model quality, and significant process modifications and enhancement. Not only technical factor but also nontechnical failure factors are responsible for MPC gradual performance degradation. Lack of properly trained personnel, lack of standards and guidelines to MPC support personnel, lack of organizational collaboration and alignment, and poor management of control system are some of the nontechnical failure factors that need proper attention. There are three solutions, namely, technical solutions, management solutions, and outsourcing solutions to deal with MPC performance deterioration. Development of online performance monitoring of APC applications, improvement of base control layer, training of MPC engineer and console operators, development of Corporate MPC standards and guidelines, central engineering support organization for MPC, and outsourcing the solutions to MPC vendors are the major strategies to sustain MPC benefits over the years. The chapter describes all these in detail.

Chapter 15 describes the final steps of MPC implementation—its actual performance assessment after deployment in real plant. Reader can gain insight about the controller performance and the optimizer performance and how to quantify them in real monetary terms. What to monitor to assess the performance in long term is also discussed in the chapter.

Performance assessment after MPC implementation will give a true picture of how much profits are achieved by a particular application as compared with initial study before implementation. A periodical performance review will also provide an idea of how much of initial benefits are preserved over time and how much money is lost by not getting the full potential performance. This will help to justify periodical maintenance or overhaul of MPC application. Performance of model predictive control application (MPCA) can be evaluated using the following four categories: control performance (whether it is able to control all its key parameters within their desired range or not), optimization performance (whether it is able to run the plant at its limit or constraints to maximize economic benefit), economic performance (how much MPCA increase profit before and after implementation in money terms), and nontangible performance (how much operator time it saves to monitor DCS). Usually, different KPIs are created and monitored periodically for each of the aforementioned cases to determine whether performance is deteriorating over time. It is important to understand the definitions and underlying calculations of these KPIs along with their implications to safeguard the MPC performance deterioration over time. Some of the major KPIs are service factor, KPIs for financial performance, KPI for standard deviation of key process variable, KPI for constraint activity, KPI for constraint violation, KPI for inferential model monitoring, model quality, limit change frequencies for CV/MVs, active MV limit, KPIs for long‐term performance monitoring of MPC, and so on. Once performance deterioration is detected by these higher level KPIs, then some low‐level detail KPIs are dig down to know the actual problems and troubleshoot them. KPIs to troubleshoot poor performance of multivariable controls include KPIs for poor performance of the controller itself, KPIs to troubleshoot cycling, KPI for oscillation detection, KPIs for regulatory control issues, KPIs for measuring operator actions, and KPIs for measuring process changes and disturbances. How to create these KPIs and what is their significance and implications are discussed in detail in the chapter.

Chapter 16 gives an idea about the various available commercial MPC vendors and their applications. A comparative study of various MPC software available in market such as Aspen, Honeywell, and SMOC has been made. They all have different implementation strategies and different unique features and relative strengths and weakness. All these are discussed in detail in the chapter. Readers can get a flavor of commercial MPC applications in chemical process industries across the globe. MPC is a matured but constantly evolving technology. Although there is no breakthrough development in core MPC algorithm in the past five years, the commercial MPC vendors comes up with more software packages, which helps to implement and monitor MPC technology in shop floor. These MPC vendors are now offered a full range of software package that comprises some basic modules such as data collection module, MPC online controller, operator/engineer station, system identification module, PC‐based off‐line simulation package, control performance monitoring and diagnostics software, and soft sensor module (also called quality estimator module). What these different modules intended to do and various common features of these modules in commercial MPC software are discussed in detail in the chapter. The chapter also describes development history and features of three major MPC players, namely, Aspen Tech DMC‐plus, Shell Global Solutions SMOC, and Honeywell’s RMPCT. The discussion of the commercial MPC vendor and their software revolves around the following: a brief history of the development of each MPC technology, product offerings of each vendor with some of their uncommon features, and distinctive feature of their respective technology with current advancement.

The main feature of the book that differentiates it from other MPC books on the markets is its practical content, which helps readers to understand all steps of MPC implementation in actual commercial plants. The book describes in detail initial cost–benefit analysis of MPC project, MPC software implementation steps, practical considerations to implement MPC application, the steps to take after implementation, monitoring of MPC software, and evaluating its post‐performance.

Key features of the book are summarized as follows:

Readers can develop a thorough understanding of steps for building a commercial MPC application in a real plant. All the practical considerations to build and deploy an MPC model in commercial running plants are the essence of the book.

Chapter 5

describes cost–benefit analysis procedures before MPC implementation.

The stages of commercial MPC implementation, starting from an assessment of existing regulatory control, functional design of MPC, model building, and final MPC implementation stages are described in detail.

The various aspects and practical considerations of functional designs of MPC controller in actual commercial plants are discussed.

Soft sensors are discussed in detail in

Chapter 10

. Commonly used algorithms to build first principle‐based and black‐box‐based soft sensors models are explained.

How to learn and modify the MPC model structure and tuning parameters from off‐line simulation response is explained in detail.

Chapter 12

is dedicated to the most important steps in MPC implementation— online deployment in real plants. Various stages of real‐time deployment of MPC software and precautions to be taken in open‐ and closed‐loop deployment are described in detail.

Monitoring of MPC model performance after deployment and understanding the weakness of the developed model (if any) is key to make robust MPC application. Readers can gain insights of these features in

Chapter 15

.

The book enlightens the reader as to why some MPC applications fail in industries. Different root causes of MPC failure in industry is explained to give the reader an idea about what can go wrong. Users can gain insights about how to safeguard MPC performance deterioration in the long run.

Chapter 16

closes out the book with a discussion of commercial MPC software applications with their distinctive features.

It is my sincere hope that readers will find the methods and techniques discussed in the book useful for understanding, functional design of MPC application, online and off‐line tuning and post monitoring of MPC. This will help the readers to build and implement effective MPC application in industry and get maximum benefit from it.

Clearly, it was not a small effort to write the book, but the absence of such MPC book in market and its requirement in large number of process industries spurred me to writing. I would like to thank Mr. Farid Khan of Reliance Industry Ltd for teaching me the basics of MPC and exposing me to the practical field of MPC.

Finally, I am truly grateful to my family: my wife Jinia and my children Suchetona and Srijon for their understanding, support, and generosity of spirit in tolerating my absence during the writing of this book.

May 2017

Sandip Kumar Lahiri

1Introduction of Model Predictive Control

1.1 Purpose of Process Control in Chemical Process Industries (CPI)

Any industrial process, especially oil refineries and chemical plants, must satisfy several requirements imposed by its design and by technical, economic, and social conditions in the presence of ever‐changing external influences (disturbances). Among such requirements, the most important ones are as follows:

Safety:

This is the most important requirement for the well‐being of the people in and around the plant and for its continued contribution to economic development. Thus, the operating pressures, temperatures, concentrations of chemicals, and so on should always be within allowable limits.

Product quality/quantity:

A plant should produce the desired quantity and quality of the final products.

Environmental regulations:

Various international and state laws may limit the range of specifications of the effluents from a plant (e.g., for ecological reasons).

Operational constraints:

The various types of equipment used in a chemical plant have constraints (limits) inherent to their operation. Such constraints should be satisfied throughout the operation of the plant (e.g., tanks should not overflow or go dry).

Economics:

The operation of the plant should be as economical as possible in its utilization of raw materials, energy, and human labor.

Reliability:

The operation of the plant should be as reliable as possible to ensure that the plant is always available to make products.

These requirements dictate the need for continuous monitoring and control of the operation of a process plant to ensure that operational objectives are met. This is accomplished through an arrangement of instrumentation and control equipment (measuring devices, valves, controllers, computers) and human intervention (plant designers, plant operators), which together constitute the control system.

There are three general classes of requirement that a control system is called on to satisfy:

Suppression of disturbances

Ensuring the stability of the process

Optimizing the performance of the process

Traditionally, PID controllers are used in CPI to perform these tasks. PID regulatory controllers efficiently ensure stability of the process and suppression of disturbance. However, due to the multivariable nature of the process and complex interactions between process parameters, PID controllers cannot make a coordinated control move to optimize the process performance. Here lies the need of model predictive control.

1.2 Shortcomings of Simple Regulatory PID Control

PID control forms the backbone of control systems and is found in most CPI. PID control has acted very efficiently as a base‐layer control for many decades. But with increased global competitiveness, process industries have been forced to reduce production costs in order to maximize profit. They must continuously operate in the most efficient and economical matter possible.

Most modern chemical processes are multivariable (i.e., multiple inputs influence same output) and exhibit strong interaction among the variables. Let us consider an operation of a boiler whose main function is to produce and deliver steam to downstream units or steam header at a specified temperature and pressure. The boiler has a drum with inlet water flow and is heated by fuel gas to produce steam. Now consider a situation where demand of the steam in downstream units increases, and it starts drawing more steam from the header. As a result, water level in the boiler will drop, vapor space above the water will expand, and consequently pressure and temperature will drop. Note that the water level in boilers is not independent and can affect the steam pressure and temperature. As a corrective action, if inlet water flow increases to control level, this will drop the boiler temperature. It will call for more heating and more evaporation, which will again lead to level drop. This demonstrates that there are very strong multivariable interactions among steam pressure, temperature, boiler level, and inlet water flow. Everything affects everything.

Now consider a conventional basic regulatory control scheme in a boiler where multiple single‐input, single‐output PID controllers are used for controlling the plant (multiloop control). Say, boiler level is controlled by inlet water flow, temperature is controlled by fuel gas flow, and boiler pressure is controlled by outlet steam flow. One basic shortcoming of PID loop controls is that they act as a single‐input, single‐output (SISO) controller in an island mode. For example, level controller will see and maintain only level with no idea what is happening with pressure and temperature. The same is true for the temperature controller, which will adjust the fuel gas base on temperature feedback and will not care for level. Now consider the previous situation, where the level starts dropping due to more drawing of steam from header. Level controller will increase inlet water flow, which will reduce the temperature. Temperature controller will increase fuel gas flow, which will again lead to level drop. Again, the level controller allows more inlet water flow to maintain level. There is a lack of coordination among the controllers, and they all act as unconnected islands. Neighboring PID loops can cooperate with each other or end up opposing or disturbing each other. This is due to loop interactions and is a serious limitation of PID regulatory controller. It is very important to understand the multivariable interactions in the chemical process plant and then try to develop model predictive control.

MPC usually stands for model predictive control. Model predictive control is used in multivariable processes where multivariable interactions among the process parameters are significant. However, MPC also stands for multivariable predictive control. MPC is used for both model predictive control and multivariable predictive control throughout this book. In industry, sometimes it is also called advanced process control or APC. Reader should appreciate that MPC stands for all of them in this book and fundamentally they refer to same model predictive control in a multi variable process environment.

Unlike the PID controller, MPC is a multi‐input, multi‐output (MIMO) controller. MPC receives all the inputs (e.g., temperature, pressure, level, fuel gas flow, inlet eater flow) and uses a predictive model to predict the output. As the name suggests, the heart of MPC is the predictive model. It then calculate the fuel gas flow, inlet water flow and other factors so that the controlled variables (temperature, pressure, level) are maintained at their set point or within their specified limit. The internal predictive model will account for all the multivariable interactions among the process parameters and adjust the manipulated variable accordingly. This is where MPC is more advantageous than multiloop PID controllers.