NEURAL ADAPTIVE CONTROL STRATEGY FOR HYBRID ELECTRIC VEHICLES WITH PARALLEL POWERTRAIN
MetadataShow full item record
In a hybrid electric vehicle (HEV) with parallel powertrain, the system can be controlled by splitting the required power between the electric propulsion machine and internal combustion engine (ICE) to meet specific goals related to fuel consumption, efficiency, performance, and/or emissions. This power splitting scenario, which is of great hybridization importance, is in fact the control strategy or energy management of the hybrid vehicle. Performance of the system depends on the control strategy, which needs to be robust, stable, reliable, and independent from uncertainties. This Ph.D. research is focused on model based control strategies, which are proposed for parallel hybrid powertrains, showing significant advantages in performance and fuel economy. If a model based control strategy is used to develop the hybrid power management algorithm, the accuracy of the model data needs to be high for proper control. Therefore, this type of management method is parameter sensitive. Implementing system identification features into this algorithm reduces the effect. As a result, the proposed controller algorithm learns the existing component parameters while operating. Furthermore, combining the base controller with an online tuner, which simultaneously optimizes the controller for current conditions, will improve the performance of the power management. In addition, this Ph.D. thesis presents a novel neural adaptive equivalent consumption minimization strategy (ECMS) and applies it to a hybrid representative sport utility vehicle (SUV) with parallel powertrain. The ECMS is a model based optimal control strategy and is based on the minimization of both fuel consumption and battery charge usage by introducing the equivalent coefficient between them. Proper operation of the controller depends on the accuracy of the model. It also depends on the correct selection of the equivalent coefficient. In this Ph.D. thesis, specific neural network structures are proposed for both coefficient selections by drive cycle recognition and for precise model building by system identification. This thesis also presents a novel fast solution method of ECMS algorithm for real time applications.