Design of fuzzy neural networks using evolutionary algorithms

• Main Author: Li, Cheng Han.
• Other Authors: Er Meng Joo
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Control engineering
• Online Access: http://hdl.handle.net/10356/40884
• Institution: Nanyang Technological University

Research has been conducted on how to develop machine intelligence. Artificial Neural Networks (ANN) has been proven to mimic human brains, and has demonstrated great potential in many industrial applications. Fuzzy logic systems borrow the idea of human linguistic information processing. Evolutionary Algorithms (EA) learns from another natural phenomenon, the biological evolution. All three techniques have strengths and weaknesses. The idea of hybrid systems gives birth to the Fuzzy Neural Networks (FNN), which is meant to combine the strength of both ANN and fuzzy logic systems. EA has also been experimented to work with ANN as well as FNN in some recent works. This project studies the architecture and learning algorithm design of the FNN using EA as a development tool to assist design. The project adopts the FNN architecture based on ellipsoidal basis functions and proposes a new learning algorithm. The algorithmic parameters of this algorithm are optimized by EA. Simulation studies on benchmark function approximation and identification problems have been carried out. The simulation results are compared with previous works such as the Dynamic Fuzzy Neural Networks (DFNN), the Generalized Dynamic Fuzzy Neural Networks (G-DFNN) and the Fast and Accurate Online Self-organizing Scheme for Parsimonious Fuzzy Neural Networks (FAOS-PFNN). A comparative study demonstrates the efficiency of the proposed FNN design and the potential of EA as a powerful development tool.