Load modeling has been extensively studied to understand the behavior of power systems. The essential problem of load modeling is that it is very hard to precisely describe a large collection of heterogeneous physical devices. These devices not only have different characteristics but also change depending on the various conditions such as weather, time, economic conditions, etc.

To analyze the behavior, a large number of devices are first grouped into similar loads. Then, these loads are replaced with equivalent circuits and logics for calculation. Many parameters are typically needed to describe the load characteristics. One aspect of model simplification has to do with the number of parameters, as more parameters definitely have the potential to offer better accuracy. However, more parameters will also make the system and computation more complicated.

This dissertation introduces a new approach to simplify complex load models and estimate the parameters. One of emerging trends in power systems involves the use of information technology. This dissertation focuses on a method based on information geometry which combines information theory with computational differential geometry to derive global estimation results. The approach sheds a new light on difficulties commonly encountered when fitting widely used models to the measurement data.

Simulations are performed based on a conventional composite load model and the new WECC Composite Load Model. The results are then compared with the full original model and the reduced parameter model to verify the effectiveness of reduction via information geometry.