Adaptive Neuro-Fuzzy Inference System (ANFIS)

The technique which is explained in the previous section was used and verified experimentally and theoretically in various kinds of applications. But this technique is not suitable for biological counterparts (Jain and Martin, 1998). The merging of these techniques into a single multidisciplinary field was carried out by Lotfi Zadeh which led to the development of soft computing techniques.

Many further techniques were added, along with the technique of soft computing, which paved way for the development of Extended Soft Computing. The utilization of any of these techniques comes under the class of soft computing.

For the purpose of tuning the fuzzy logic (FL) controller, the neural network artificial neural network (ANN) was utilized. Lee and Lee (1974) developed a model

Fuzzy process

FIGURE 3.2 Fuzzy process.

based on neurons in which multi-inputs and multi-outputs can be included. The widely used normal method will rely on a binary system for obtaining output. Many researchers were motivated to carry out their research in the area of neural networks because many issues related to this network were mentioned in IEEE Communications Magazine. To solve issues in neural networks, fuzzy-based neural networks were developed (Plevyak, 1992). Following that, Jang developed the technique of Adaptive Neural Fuzzy Inference Systems (ANFIS) (Jang, 1993). In the Department of Electrical and Computer Engineering of the Institution Superior Tecnico, Lisbon, Portugal in 1998, Dente and Costa Branco developed an electro- hydraulic system based on the Neuro-Fuzzy approach.

The similar kind of operation seen in neural networks is found in the neuro- adaptive learning method. The modeling protocol related to fuzzy will be rendered through neuro-adaptive learning techniques. This process is performed to obtain the information related to the dataset. The toolbox function named ANFIS, with the help of knowing the dataset, either inputs or outputs the construction of a Fuzzy Inference System (FIS). Parameters such as a membership function in this fuzzy system will be altered utilizing the least squares method or back propagation algorithm. The alteration made will support the fuzzy system to acquire the information regarding the dataset used for modeling. The shaping of the membership function can be done through the Neuro-Fuzzy Design application. Instead of giving the data manually the data which are to be used as input can be trained using the above-mentioned application. The working principle for ANFIS is given in Figure 3.3. The objectives of ANFIS are mentioned as follows:

From Neural Networks (NN) and Fuzzy Systems (FS) the integration of effective features:

(1) Integrating feature from FS: The information that is obtained previously is defined in a set of constraints in order to decrease the search process carried out through optimization.

Basic flow diagram of ANFIS computation

FIGURE 3.3 Basic flow diagram of ANFIS computation.

(2) Integrating feature from NN: To achieve the adjustment of parameters in FC automatically, the back propagation algorithm is utilized along with the structured network.

In certain manufacturing the ANFIS can be applied to:

  • (1) Models (to describe earlier data and to guess future behavior).
  • (2) Controllers (automated FC tuning).

Basic flow' diagram of computations in ANFIS is shown in Figure 3.3.

Fuzzy Logic Toolbox

The above-mentioned Fuzzy Logic Toolbox™ is referred to as an application. Through this application a Simulink® block, application and function can be rendered. With the help of the Simulink block the process of simulation, designing, and analyzing can be performed related to fuzzy logic. The steps required for constructing fuzzy inference systems will be explained through this application. This application is considered the origin of many techniques such as adaptive neuro-fuzzy learning and fuzzy clustering. By means of a simple logic rule the modeling of complex systems can be done using this toolbox and further, the implementation of the rules can be done in a fuzzy inference system. This toolbox can be utilized as standalone fuzzy inference engine.

Key Features of the Fuzzy Logic Toolbox

  • • The inference systems can be constructed as well as the analysis of outcomes.
  • • Membership functions which are necessary for generating fuzzy inference systems can be developed.
  • • The logic gates such as NOT, OR, and AND logic can be defined in user- defined rules.
  • • Mamdani which is found to be the standard for FIS as well as Sugeno-type is considered as standard for ANFIS fuzzy inference systems.
  • • The alteration in membership function can be done automatically utilizing fuzzy-clustering and neuro-adaptive learning techniques.
  • • It has the capability to create standalone executable fuzzy inference engines or embeddable C code.
< Prev   CONTENTS   Source   Next >