Researchers and practitioners have proposed a variety 
of solutions to reduce electricity consumption and 
curtail peak demand. This research focuses on load 
control by improving the operations in existing building 
HVAC (Heating, Ventilating and Air-Conditioning) systems 
and by aggregating individual loads based on optimization 
studies. Emphasis is placed on electricity rates and 
climate data in California, where electricity costs have 
been of particular concern. The optimization problem in 
this research is multi-objective in the sense that we aim 
to reduce building load while maintaining an acceptable 
level of comfort. The first part of this research focuses 
on optimizing controls in a single building. A simple 
three-zone VAV system model is built in EnergyPlus (E+). 
The cost function structures and the potential difficulties 
associated with simulation-based optimization are discussed. 
Discontinuity and nonlinearity are of major concern. Two 
optimization algorithms are tested and applied to a variety 
of problems: Direct Search (DS) and Genetic Algorithms (GA). 
An E+ simulation based GA optimization environment is 
developed in Matlab. DS is found to be efficient with small 
problems in this research, while GA works in almost any 
situation with the price of intensive computation. A few 
operations guidelines are proposed. The second part of this 
research presents three ways of optimizing load control in 
an aggregation pool: Enumeration, multi-GA and model-based 
nonlinear optimization. Enumeration relies on expert rules 
to find a small set of feasible solutions through automated 
E+ simulations and search exhaustively for the optimal 
solution. Multi-GA solves the aggregation problem in the 
Matlab GA environment with sequential E+ simulations as the 
function evaluator. Because simulation-based optimization is 
very computationally intensive in handling multiple buildings, 
the model-based approach develops for each aggregation 
participant a time series model and several regression models 
to predict individual load profiles under load control. It 
then applies an interior-point-method-based commercial solver 
LOQO to these simplified building models. This system is fast 
and easy to scale up. Certain precision is lost due to 
modeling simplifications, but the results are still satisfactory 
for practice purposes. Overall, load aggregation offers load 
diversification opportunities among participants and improves 
the aggregated load profile. Load shedding later individual 
load profiles in a way that enhances the aggregation performance.