ELECTRIC MOTOR CONTROLLER MODEL SIMULATION USING NEURO FUZZY: COMPARISON OF GRADIENTLESS OPTIMIZATION METHOD AND PARTICLE SWARM OPTIMIZATION (PSO)

• Main Author: ZAINAKMAL SOLIHIN, MUHAMMAD
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/60597
• Institution: Institut Teknologi Bandung

The most important part of an electric vehicle is the electric motor as the main drivetrain. For an electric motor to work properly, a controller is required to control the electric motor so that the power delivered to the motor at a certain point in time is the exact amount as needed. With the time, machine learning based controller is getting more and more used, and one of the best machine learning based model for a controller is neuro fuzzy system. From that, in this research one of the neuro fuzzy models will be used, namely Artificial Neuro Fuzzy Inference System (ANFIS). The standard optimization method for a neuro fuzzy model is using gradient descent, but this method has quiet a few problems for neuro fuzzy, one of which is stuck at local optima. Hence, two optimization methods without gradient are tested in this research, namely gradientless descent (GLD) and particle swarm optimization (PSO). Using data recorded from an electric motor, an ANFIS model with 3 outputs, 2 inputs configuration is trained. From the results of the study it was found that an ANFIS model with triangular membership function and (3, 3, 3) membership configuration with PSO as the optimization method delivers the best outcome. Using triangular membership function and (3, 3, 3) as membership configuration, gives the model enough complexity while still avoiding overfit. Moreover, using PSO as the optimization algorithm leads to a higher model accuracy compared to GLD optimization despite using way less time and computational cost.