Modern automotive engines are controlled by the electronic control unit, in which open-loop control based on sets of look-up tables and closed-loop control using proportional-integral-derivative are employed for steady-state and transient control respectively. Incorporating new technologies into engines is one way to further reduce emission pollutants and improve engine performance. However, it also makes traditional offline model based engine calibration be inefficient and tedious even for experienced engineers, because more number of control variables and parameters of look-up tables should be considered and tuned. Another way to reduce emissions is to improve transient control performance of air-fuel ratio (AFR), which strongly affects the conversion efficiency of the three-way catalytic converters and finally affects emissions. However, the engine build-in AFR controller, lacking adaptive capability, cannot guarantee long-term control performance. Online modeling based calibration, which can totally make use of the information from collected measurements, is an efficient method comparing to traditional approaches. Also, online system identification based adaptive control can overcome the problem of lacking adaptive capability. However, to accomplish the two tasks, a fast and efficient online learning algorithm for modeling is necessary. Online extreme learning machine (OELM) is a good choice for both applications. However, the base model training of original OELM highly restricts the direct use of OELM in many applications. Therefore, this thesis firstly proposes an improved version of OELM, called initialtraining-free OELM (ITF-OELM), in which the base model training in the original OELM becomes unnecessary. Thus, the online learning algorithm becomes simpler and can be easily applied to more online applications. Based on the developed ITF- iii OELM algorithm, an efficient point-by-point engine calibration using sequential design of experiment strategy is proposed to improve the calibration performance and to reduce the number of measurements. Besides, an ITF-OELM based system identification for adaptive engine AFR regulation is developed. To verify the effectiveness of the proposed approaches, simulations on virtual engine and experiments on real engine for both ITF-OELM applications are conducted. The results show that the proposed engine calibration can be carried out with significant fewer experiments and time, and the proposed AFR adaptive control strategy is feasible and effective.