dZheji ativ nve may rtio pe s, th ear ctu an ed with traditional PID control and PI?PD control in terms of set-point tracking and ed in th plays a 1,2]. U tempe . If the ion. On een solved, there are formance for chemical ties. In view of these,

Chemometrics and Intelligent Laboratory Systems 142 (2015) 245?254

Contents lists available at ScienceDirect

Chemometrics and Intellig l seTo achieve improved control performance, a modified form of PID controller, i.e., PI?PD controller, was proposed by Majhi and Atherton [15]. In this new control strategy, a PD controller is first placed in the model predictive control (MPC) can be a suitable choice and has been intensively studied [21?28]. However, though MPC has low requirements for process model accuracy and its control performance is goodfurnace and so on. What's more, the dynamics of this temperature loop is complex with large time delay, which poses challenges for traditional PID control [6?12]. The uncertainties in this process also add to the difficulty for PID control to achieve satisfactory performance [13,14]. ter tuning is indeed not very easy.

Even if the parameter tuning of PI?PD has b still issues of improving closed-loop control per processes with large time delays and uncertainperature should be controlled as accurate as possible [3?5].

However, due to the fact that the coke tower in the whole coke process is operated through batch mode, there are many factors that affect the outlet temperature such as the circulating oil feed flow rate, gas oil temperature fluctuations, and the air volume flowing into the parameters of PI?PD controller, it is too complex and depends on the experiences. Bettou and Charef [20] raised a PI??PD controller by using particle swarm optimization (PSO). However, it must calculate five parameters in order to get the parameters of PI?PD controller. The above facts illustrate that although PI?PD controller is promising, the parame-inner feedback loop to suppress overshoot, a placed in the outer loop in order to guara ? Corresponding author. Tel./fax: +86 571 87952233.

E-mail address: jmzhang@csc.zju.edu.cn (J. Zhang). http://dx.doi.org/10.1016/j.chemolab.2015.02.013 0169-7439/? 2015 Elsevier B.V. All rights reserved.the contrary, if the outlet plete reaction and affect ese facts, the outlet temto the ratio between the proportional coefficient in the outer loop and the differential coefficient in the feedback loop, to obtain the last parameter of PI?PD. Although this method provides a solution of adjusting thetemperature is too low, it can result in incom the quality of the final products. In view of th1. Introduction

Delay coking technology iswidely us delay coking processes, coke furnace heating of a variety of raw materials [ safe and optimal operation, the outlet regarded as a very important indicator high, it can lead to ingredients dissociat? 2015 Elsevier B.V. All rights reserved. e oil refining industry. In n important role in the nder the requirement of rature of coke furnace is outlet temperature is too response [16]. It has been shown that PI?PD controller can provide improved control performance [17]. There have already been some results.

Tan came up with a graphical method to calculate the parameters of

PI?PD controller [18], however, this method relies on the graphics and cannot correct the parameters of PI?PD controller automatically.What's more, we must first obtain the parameters of PID controller [19] and then consider regulating the additional parameter ?1, which is equalCoke furnace disturbance rejection.PID control

Outlet temperature control control performance comparDynamic matrix control optimization base outlet temperature in a coke furnace

Haisheng Li a,b, Hongbo Zhou a,b, Jianming Zhang a,? a State Key Laboratory of Industrial Control Technology, Institute of Cyber-Systems & Control, b Zhejiang Supcon Co. Ltd., Hangzhou 310053, PR China a b s t r a c ta r t i c l e i n f o

Article history:

Received 22 July 2014

Received in revised form 1 December 2014

Accepted 10 February 2015

Available online 18 February 2015

Keywords:

Dynamic matrix control

PI?PD control

Proportional?integral?deriv its simple structure and co uncertainties, PID control proportional-integral?propo is now used in a limited sco

PID. In view of the above fact (DMC) optimization, which b at the same time, simple stru on the outlet temperature of j ourna l homepage: www.end a PI controller is then ntee an overall systemnew PIPD type control for ang University, Hangzhou 310027, PR China e (PID) control is widely applied to various kinds of industrial processes with nient implementation. However, for processes with complex behavior and not always satisfy the higher requirements. Compared with PID control, nal-derivative (PI?PD) control can provide improved control performance but because its controller parameter tuning is a little bit more inconvenient than is paper proposed a new PI?PD control design based on dynamicmatrix control s the advantages of improved control performance under DMC optimization and re of PI?PD control for implementation. The proposed PI?PD controller is tested industrial coke furnace, where results show that the controller shows improved ent Laboratory Systems v ie r .com/ locate /chemolab[29?35], its implementation is not as simple as PI?PD controller.

In this study, inspired by the advantage of dynamic matrix control (DMC), a new PI?PD control method optimized by DMC is proposed.

Through combining DMC algorithm with PI?PD control, the proposed method inherits the excellent performance of DMC algorithm and the advantage of traditional PI?PD control. The results show that improved closed-loop performance is obtained compared with traditional PID control and PI?PD control. The outlet temperature control of coking furnace is considered as a case study. In addition, the new PI?PD controller canworkwell on a second order plus dead time (FOPDT) system.

The proposedmethods belong to the industrial process control category under applied chemometrics, where the control method proposed is based on the idea that a controller may be effective for process with uncertainties, which are very common in industries. The uncertainties in this paper are considered asmodel/plantmismatch that is statistically generated. Then the controller is applied to these statistical process uncertain models. 2. Traditional PID and PI?PD controllers tuning