Book Description
The fuel economy prediction in an automobile is a significant and complex issue. There are numerous variables involved in a vehicle’s daily usage that influence its fuel economy. This problem is even more complex for a hybrid electric vehicle (HEV), due to the presence of the supervisory controller overseeing the energy management strategy. The control strategies implemented in production vehicles involve the use of hundreds of calibration parameters in the form of Lookup Tables (LUTs). The work described in this document aims to lay the groundwork in resolving this complex issue of fuel economy prediction in an HEV using a model based optimization approach. There are two distinct aspects of the approach utilized here: 1) Calibration and Validation of the Vehicle Models, 2) Optimization of the Supervisory Controller. An Open Loop Vehicle Model is utilized for the calibration and validation aspect. Experimental data corresponding to a driving distance of ~36,000 km collected over the span of 2 years is made available. The vehicle models used for the research represent the same vehicle on which this data was obtained. The calibration, validation and optimization tasks need to consider different weather patterns across the year to aid in accurately estimating the fuel economy. The primary reason for the use of an open loop model for the calibration and validation aspect is to eliminate the effects of the vehicle controller so that an accurate representation of the 'Vehicle Plant' is available. This thesis details the methodology undertaken for validating the open loop model. A novel technique of converting a look-up table into a surface fit to calibrate the same is implemented and the results are discussed. Once validated, the model truly represents the actual vehicle behavior and the results obtained from the optimization performed on it are reliable. The optimization techniques used through the work described here and in further research, are termed as "Derivative-free Simulation based Optimization". There is an absence of a definitive analytical function to describe the control variables as a function of the objective, and there is a vehicle simulator/model in this loop, that ultimately warrants the use of such methods. Finally, implementation of this validated plant in the closed loop model is illustrated using commonly available derivative-free optimization methods.