Neural network based model predictive control for a steel pickling process

https://doi.org/10.1016/j.jprocont.2008.09.003Get rights and content

Abstract

A multi-layer feedforward neural network model based predictive control scheme is developed for a multivariable nonlinear steel pickling process in this paper. In the acid baths three variables under controlled are the hydrochloric acid concentrations. The baths exhibit the normal features of an industrial system such as nonlinear dynamics and multi-effects among variables. In the modeling, multiple input, single-output recurrent neural network subsystem models are developed using input–output data sets obtaining from mathematical model simulation. The Levenberg–Marquardt algorithm is used to train the process models. In the control (MPC) algorithm, the feedforward neural network models are used to predict the state variables over a prediction horizon within the model predictive control algorithm for searching the optimal control actions via sequential quadratic programming. The proposed algorithm is tested for control of a steel pickling process in several cases in simulation such as for set point tracking, disturbance, model mismatch and presence of noise. The results for the neural network model predictive control (NNMPC) overall show better performance in the control of the system over the conventional PI controller in all cases.

Introduction

It has been known that many chemical industrial plants cause environmental problems due to the usage of chemicals in their production lines. One such industry is the steel pickling plant which is a fundamental industry in Thailand and has long existed and served the country’s steel demand. The steel pickling process utilizes concentrated chemicals in the production lines and the wastewater released from the process contains hazardous materials and usually causes major environmental problems. Therefore, production scheduling and control of this pickling process are inevitably needed to minimize the amount of hazardous material contained in the released wastewater and also to maintain the concentration of acid solution in the tanks in order to obtain the maximum reaction rate at the same time.

The steel pickling process presents many challenging control problems, including: nonlinear dynamic behavior; multivariable interactions between manipulated and controlled variables and constraints on manipulated and state variables. A number of control approaches and algorithms that are able to handle some of the above process characteristics have been presented in the academic literature in recent years. Bequette [1] gives a review of these various approaches such as the internal model approaches, differential geometric approaches, reference system synthesis techniques, including internal decoupling and generic model control (GMC), model predictive control (MPC) and also various special and ad hoc approaches. Many of these approaches are not able to handle the various process characteristics and requirements met in industrial applications and some of the approaches can only be applied for special classes of models.

MPC appears to be one of the general approaches which can handle most of the common process characteristics and industrial requirements in a satisfactory way. It also seems to be the approach which is most suitable for the development of a general and application independent software, which is essential for the development of cost-effective applications. However, the key in the successful use of MPC in solving these process problems is the existence of an accurate model. Chemical processes such as this steel pickling process have been traditionally controlled using linear system analysis even though they are inherent nonlinear process. However to obtain accurate model for the steel pickling process and predicting its interacting and nonlinear behavior is actually highly difficult.

Recently, neural networks have been successfully applied in the identification and control of nonlinear processes. Neural networks offer alternative nonlinear models for implementing MPC in industrial systems [2], [3], [4], [5]. Different ways of neural models being embedded in MPC systems were reviewed by two recent surveys [6], [7]. It is noted that while neural network modeling and control techniques are investigated for nonlinear systems, the current methods proposed and tested by simulations and some implementations to laboratory rigs are mainly for single-input single-output (SISO) systems [8], [9]. Others applications of neural networks for chemical process modeling and MPC have also been investigated for SISO systems [10], [11], [12], [13] but very few investigations into iterative multistep neural network predictions in MPC based control for MIMO chemical processes have been reported.

A multivariable neural network modeling and neural network model predictive control (NNMPC) technique are investigated in this paper for application to a steel pickling process which is commonly found in the steel industries of Thailand. The process involves removal of surface oxides (scales) and other contaminants out of metals by an immersion of the metals into an aqueous acid solution, which consists of three acid baths in series. The highly nonlinear dynamic behavior, multivariable in nature and interaction between baths cause this process to be difficult to control by conventional controllers. It is, therefore, the aim and contribution of this work to apply an iterative multistep neural network prediction model in a predictive control strategy for controlling such a nonlinear system. To demonstrate the robustness of the proposed control strategy, tests involving set point tracking with introduction of various disturbances including model mismatch and noise are performed in these studies.

Section snippets

Process description

The steel pickling process consists of two major steps: pickling and rinsing steps [14], [15]. The purpose of the pickling step is to remove surface oxides (scales) and other contaminants on the metals by an immersion of the metals into an aqueous acid solution. Metals are immersed in pickling baths, containing 5%, 10% and 15% by weight of hydrochloric acid (HCl), respectively, in order to remove the scales from the metals. The metals move counter current to the acid stream. The reaction

Neural network modeling

Neural networks have the advantages of distributed information processing and the inherent potential for parallel computation. In many cases, when sufficiently rich data are available, they can provide fairly accurate models for nonlinear controls when model equations are not known or only partial state information is available [16], [17]. Due to their parallel processing capability, nonlinearity in nature and their ability to model without a priori knowledge, neural networks can be used

Neural network model based predictive control

The neural network MPC strategy developed in this work is shown in Fig. 4. In this approach the neural network model is used to predict future outputs several steps in future over the prediction horizon (p). The output from the first prediction, C(k + 1) will be used as inputs for the next prediction in predicting C(k + 2). With this iterative procedure, we can predict the multiple output P steps in the future as shown in Fig. 5a–c. Other inputs are obtained as from the previous values. This

Simulation results

The multivariable NNMPC strategy is initially applied to control the concentration of HCl in the 5% HCl, 10% HCl and 15% HCl bath to the normal values of 1.40 (5% by weight HCl), 2.87 (10% by weight HCl) and 4.41 (15% by weight HCl) mol/l by adjusting the manipulated variables F2, F3 and F5, respectively. The simulations are divided into four cases of control studies, which are the set point tracking case, disturbance case, model mismatch case and noise case, respectively.

For the set point

Conclusions

The application of a neural network model based predictive controller to a nonlinear multivariable chemical process is investigated. Since the real chemical processes are nonlinear and multivariable interacting systems, which make them difficult to control by using conventional controllers, model based advance control techniques are then required to obtain tighter control. However, in many cases it is even impossible to obtain a suitable process model due to the complexity of the underlying

References (20)

There are more references available in the full text version of this article.

Cited by (0)

View full text