An adaptive neuro-fuzzy logic based systematic approach of modelling virtual driver for real world driving

Competition to be at the helm of automotive market with regards to fuel economy has led to development of new optimization techniques. Vehicle parameters although is the primary influencing factor, driving behavior is an equally important factor that influences fuel consumption considerably. In the current scenario it has become a paramount challenge for automobile OEMs to offer unparalleled fuel economy for a wide range of driving behavior. Conventional process is to conduct a series of real world trials with drivers of wide range of driver aggressiveness, collect data, calibrate, optimize and validate the same in real world. Few major setbacks in this traditional process like tediousness, accuracy, repetitiveness and most importantly development cost have led to investigating alternative solutions to overcome this challenge. To overcome these challenges, it is very important that majority of calibration activities that are done on road, should be performed in lab. The substitute solution comes in the form of a simulation model which is a package of all parameters that define a real world driver behavior for the virtual environment. One way of realizing this model is using traditional control logic for understanding transient behavior of the real world driver based on already available data and applying it for a different drive cycle. This paper intents to provide a method of understanding and training the real world driver behavior based on already available data, with adaptive neuro-fuzzy interface. The driver model will initially be validated for the legislative cycle to check the accuracy level of the model, and various real world drive cycles will be fed to the model to check the driver behavior. This paper will also compare, Adaptive Neuro Fuzzy Interface System (ANFIS) methodology with conventional way of modeling driver (PID) and hence the benefit of using a fuzzy solution.