Energy management system for hybrid electric vehicles

Stringent legislations regulate exhaust and particle emissions emitted as a result of fuel combustion in Motor Vehicles that use Internal Combustion Engines (ICEs). This encourages the manufacturers to investigate alternative propelling technologies, which paves the way of developing environmentally friendly vehicles. Vehicle Performance Emission standards recently sets also Greenhouse gas emissions particularly CO2 which implies indirect limitations on fuel economy behaviour of vehicles. Hence, Hybrid Electric Vehicles (HEVs) increase their popularity to unlock these limitations by introducing electricity based driving. Development of an Energy Management System that improves the total powertrain system efficiency as well as a harmonic and effective operation of various components under their optimal conditions is a very critical but equivalently challenging part of the overall vehicle and powertrain design. On the other hand, HEV Energy Management System development, as will be depicted along this thesis, is an optimisation problem aiming to improve regularly compromising efficiency behaviours of various powertrain components. This study targets to solve this problem, not only by utilizing the most appropriate modelling and simulation techniques but also by formulating it as an optimization problem and then solving it by numerical and optimal control techniques. Plug-in Hybrid Electric Vehicles (P-HEV), become more popular with the recent development in battery technologies. On the other hand, they are considered as an interim technology towards a Zero Emission Vehicle Technology, Fully Electric Vehicles. In principle, these vehicles have an additional charging interface that allow to be charged through electrical means. This charging process is independent from any on-board energy conversion, therefore is kept usually out of the energy management problem. Consequently, depending on the power requirements and range characteristics, they have relatively larger battery capacity. This of course implies, less fuel consumption and exhaust emission over the same range. This thesis primarily defines the development of Energy Management System of HEV and P-HEV as an optimization problem, and investigates various methods in theoretical and practical basis. Firstly, the optimization problem of Energy Management for HEV and P-HEV are depicted as two distinct problems. For a HEV, along a driving mission (e.g. Drive Cycle), the initial charge and final charge equivalence of the battery poses a constraint for the optimization problem definition. Additionally, limited capacity of the battery allows only limited usage of the state of the charge. This implies that the state if the charge is forced to remain inside a limited charge band. Finally, physical behaviours of the hybrid components such as speed and torque define further system constraints for the optimization problem. For P-HEV, the situation is slightly different: due to the charging behaviour of the system over an external source, the charge balance is not mandatory. Having the knowledge of theoretical optimal behaviour of the system is substantially important since it defines a "Benchmark Solution" which is used to compare other available local optimal solutions. However, theoretical solutions providing the global optimal behaviour are not feasible for real applications mainly due to their computational burden. In this thesis, such a global optimal behaviour is obtained using Dynamic Programming (DP) for both HEV and P-HEV. DP formulates the energy optimization problem in discrete form for system states and time. Then by considering aforementioned system constraints, it produces the solution iteratively from the final time towards the start of the problem. The quasi-static vehicle and powertrain model developed for HEV and P-HEV is solved via this DP algorithm. There is a clear influence of gear shifting schedule on fuel consumption. In order to analyse this correlation, a second DP solution is obtained, where the shifting rules are modelled as control inputs. Although this approach significantly improves the fuel consumption behaviour, there is no obvious optimal shifting characteristics, which can be captured for offline usage out of this global behaviour. Computational and storage capacity requirements and the dependency of a-priori knowledge of the overall load of the overall cycle prevents the usage of DP solution to in real world conditions. Therefore, a casual alternative solution, namely Equivalent Consumption Minimization Strategy (ECMS), has been applied to the aforementioned Hybrid Vehicle topology. Both physical analogy and optimal control theory is capable to describe the ECMS. Due its instantaneous optimization characteristics, the computational load is relatively low with respect to DP solution. In principle, ECMS relies on calculation of an "equivalent fuel consumption value" based on the battery energy. This virtual consumption constitutes with the real consumed fuel, a "total equivalent fuel consumption". In fact, this value establishes a cost function, which needs to be minimized instantaneously. The equivalence factor balances the fuel equivalence of the battery energy with respect to fuel chemical energy. Obtaining the value of such factor is possible through a calibration process. This equivalence factor actually is different for the two main hybrid modes. Same solution is obtained by optimal control techniques as well. When the energy minimization problem is formulated via classical optimal control methods, the result of a differential equation formed by the co-state of the Hamiltonian function needs to be solved, according to Pontryagin Minimum Principle. On the other hand, a small assumption enables the solution of this differential equation. If we assume a constant behaviour of the battery charge vs. battery terminal voltage, then the Hamiltonian function becomes to an equivalent function previously formulated by ECMS. This assumption is reasonable especially for the limited charge behaviour of HEVs where the SOC dependency of the terminal voltage is inferior. Local optimal solution of ECMS and global optimal solution of DP have very similar behaviours. The main drawback of ECMS is a good equivalence factor calibration for a complete charge sustaining over a drive cycle. Therefore, an additional factor for a closed-loop control of battery state-of-charge is necessary. On the other hand, this additional term disturbs the original optimal behaviour of ECMS. The instantaneous local optimization ability of the ECMS makes it a candidate for real-time application. However, there is an additional problem disturbance effect of road topology for the charge sustaining of HEVs. Especially, on negative-slopes, if the battery charge is relatively high, only part of free breaking energy can be recovered. Similarly, on the positive slopes due to early depletion of battery charge, the higher load of the road is only covered by engine, which results higher fuel consumption. This problem can be solved by anticipation of road topology and then applying pre-charge or pre-discharge strategies. In this sense, two different charge control strategies are developed for HEVs. First method is based on the addition of a new term that correspond to the "equivalent fuel consumption energy" of the upcoming road section with positive or negative slope in classical ECMS formulation. Hence, thanks to this predictive behaviour, an accurate pre-charge and pre-discharge is done for the upcoming road section to exploit the free recuperation energy. This additional term is actually, fuel equivalent of the electrical energy consumed or recovered of the upcoming road. This value is a function of the upcoming slope value, the expected speed over the hill, the electrical energy required by the non-predictive strategy until the slope. The last one differs depending on the actual hybrid mode. Therefore, problem is analysed for two different hybrid modes, power split and parallel charging. Second method uses some of the prediction values of the first method and implement in Model Predictive Control (MPC) formulation. For that, ECMS cost function is linearized for the control inputs-power split values. Then a first order dynamics represent the relation between battery power and battery energy. All system and control constraints are considered in the MPC formulation. Here, with an assumption of taking all control values for each step-time along the control horizon as unknown constant parameters, the complete problem are solved like a parameter optimization problem. Due to the linear nature of obtained equations, this problem can be solved instantaneously via Linear Programming. All developed controllers and energy management strategies are actually tested based on two different simulation-modelling methodologies. First one is the quasi-static model of vehicle and powertrain and is based on backwards modelling methodologies. It relies on an assumption of perfect speed tracking and on calculation of road load, corresponding to this speed tracking behaviour. Second modelling methodology considers the drive cycle as a desired input speed to the driver model. Then the driver model acts as a speed tracking control that generates corresponding control inputs for the powertrain components through acceleration and brake pedals. Second modelling methodology considers all type of vehicle and powertrain dynamics and transient behaviour as well as the detailed lateral dynamics of the vehicle. In the course of this thesis, for the first modelling methodology, MATLAB/Simulink and MATLAB Script environments are used. For the second modelling platform, AVL CRUISE and IPG CarMaker tools and models are co-simulated. At the very last part of the thesis, two Driver Assistant System, Adaptive Cruise Control (ACC) and Traffic Light Assistant System are developed using Model Predictive Control approach. ACC works as a speed and position regulator to maintain a safe distance between the vehicles ahead. TLA is a similar semi-autonomous speed controller developed regulates the vehicle speed based on the information obtained from Traffic Lights to assess an on-board green-wave. Traffic Light Assailant System is additionally implemented on a real-time platform for a driver in the loop System. This integration furthermore includes the interaction of the controller functionality with the programmable Human Machine Interface to inform and interact with the driver.