Abstract

Hybrid electric vehicle (HEV) is one of the main types that will be
developed in the next twenty years because of its low fuel consumption, low
emissions and comparatively easiness for being merchandized. However, as it
has at least two sets of propulsion systems, the hardware structure and the
energy management system is complex. The optimal match of various components
and the optimal combination of a large number of control parameters are
therefore required. The thesis proposes to solve such kinds of HEV
propulsion system optimization problems using the evolutionary computation
techniques. Single objective optimization and multi-objective optimization
will be both explored based on the HEV propulsion system simulation software
ADVISOR by treating the typical performance parameters of the components and
the major control strategy parameters as optimization variables together.

Genetic algorithm (GA) is employed for the single objective optimization.
This algorithm has conquered the deficiencies of the traditional
optimization methods that require calculating the derivative of the
objective function and that are easily trapped in the local optimal areas.
But the shortcoming of GA is its poor search efficiency. The thesis suggests
integrating the gradient-based algorithm SQP into GA, and thus combining the
global search ability of GA and fast convergence ability of SQP, which
results in two hybrid algorithms GA-SQP-I and GA-SQP-II. The result of a HEV
study case demonstrates that these algorithms improve the fuel economy of
the original HEV dramatically. In this part, another global optimization
method, DIRECT is also tested. It is a deterministic method, not as GA,
therefore possessing relatively stronger robustness. Though the precision of
the optimum it finds usually is not high, it has fast convergence rate and
is also an appropriate choice for many engineering problems.

The topic of the latter half of the thesis is the multi-objective
optimization. A multi-objective evolutionary algorithm (MOEA), NSGA-II is
employed to optimize the fuel consumption and different emissions (CO，NOx，
CO, etc) of a parallel HEV simultaneously. The analysis of the output
Pareto-optimal solutions shows that the MOEA not only improves the
performances of the original vehicle described above but also provides the
users with a large flexibility for the design and control of the HEV,
therefore enabling the users choose different components and control
strategies according to different requirements.

How to accelerate the convergence rate of the MOEAs has been explored based
on the idea of combining MOEAs with gradient-based algorithms through
epsilon-Constraint method. A hybrid algorithm NSGA-SQP is then developed.
Nevertheless, the numeric simulation study shows that for some test problems
the hybrid algorithm converges very fast, while for some problems especially
for those which have numerous local optimal traps it does not behave better
than the original NSGA-II, which calls for further studies.