ANFIS based Distillation Column Control

This paper presents a control strategy that combines the predictive controller and neuro-fuzzy controller type of ANFIS. An Adaptive Network based Fuzzy Interference System architecture extended to cope with multivariable systems has been used. The neurofuzzy controller and predictive controller are works parallel. This controller adjusts the output of the predictive controller, in order to enhance the predicted inputs. The performance of the control strategy is studied on the control of Distillation Column problem. The results confirmed the control quality improvement with MPC and multi-loop PID controller.

The controller always aims to achieve the process variable to the given set point value. This is the main task of the properly designed controller. The controller should also provide some flexibility in case of change in set point and disturbances. Today there are many methods for designing intelligent controllers, such as predictive controller, fuzzy control, neural networks and expert systems. Various combinations of these controllers give a number of design possibilities. Artificial Neural Networks (ANNs) and Fuzzy Logic (FL) have been increasingly in use in many engineering fields since their introduction as mathematical aids by McCulloch and Pitts, 1943, and Zadeh, 1965, respectively. Being branches of Artificial Intelligence (AI), both emulate the human way of using past experiences, adapting itself accordingly and generalizing. While the former have the capability of learning by means of parallel connected units, called neurons, which process inputs in accordance with their adaptable weights usually in a recursive manner for approximation; the latter can handle imperfect
information through linguistic variables, which are arguments of their corresponding membership functions.

Although the fundamentals of ANNs and FL go back as early as 1940s and 1960s, respectively, significant advancements in applications took place around 1980s. After the introduction of back-propagation algorithm for training multi-layer networks by Rumelhart and McClelland, 1986, ANNs has found many
applications in numerous inter-disciplinary areas . On the other hand, FL made a great advance in the mid 1970s with some successful results of laboratory experiments by Mamdani and Assilian. In 1985, Takagi and Sugeno  contributed FL with a new rule-based modeling technique.Obtaining a mathematical model for a system can be rather complex and time consuming as it often requires some assumptions such as defining an operating point and doing linearization about that point and ignoring some system parameters, etc. This fact has recently led the researchers to exploit the neural and fuzzy techniques in modelling complex systems utilizing solely the input-output data sets. Although fuzzy logic allows one to model a system using human knowledge and experience with IF- THEN rules, it is not always adequate on its own. This is also true for ANNs, which only deal with numbers rather than linguistic expressions. 

This deficiency can be overcome by combining the superior features of the two methods. To achieve the most accurate set point, an appropriate extensions and improvements in the intelligent control is needed. The predictive control is added with neuro – fuzzy controller to get this task. The prediction is a future value of model. Using this states and suitable optimization criterion, it is possible to get more accurate values and the controlling becomes more effective.
This paper uses a control strategy that enhances a fuzzy controller with a self-learning capability for achieving prescribed control objectives. In this sense, an extended Adaptive-Networkbased Fuzzy Inference System (ANFIS) architecture is employed , so that a fuzzy inference system is built for achieving a desired input/output mapping. The learning method used allows the tuning of parameters both of the membership functions and the consequents in a Sugeno–type inference system.

No comments:

Post a Comment

leave your opinion