Physically based electric load estimation using fuzzy-neuro systems

The proper representation and modeling of components in a power system are critical to obtaining accurate simulation results. The complexity and dynamic nature of load makes modeling it very difficult and it remains to be so even after many attempts over the years. The work in this dissertation presents three approaches to estimating the composition of electric load at the distribution level. The model chosen for load representation is physically based, and hence is interpretable in terms of real world load components.;The uncertainty involved with modeling load response of an aggregation of different load types with a wide range of parameters necessitates that a relaxation be enforced in estimating composition. The first approach developed initially by Dr Abraham Ellis, and which was further extended, was to formulate the uncertainty into the composition coefficients of the different load types. The composition coefficients were obtained using a fuzzy-LP solution. The fuzzy LP approach was extended to Gaussian fuzzy sets. While this approach allows for the development of crisp simulation models for the load types, the uncertainty, which really lies in the internal load parameters and hence the load response, is not captured. The second approach is to use statistical sampling of internal load parameters and to obtain a mean power response for each of the different load classes. The probabilities density of the parameter ranges is used to perform the random sampling of parameter values. The composition coefficients are modeled as fuzzy numbers. The third approach is to model uncertainty into the load response of a particular load type by training an Adaptive-Neuro-Fuzzy-Inference-System (ANFIS) rule base to input output patterns of different load types. The fuzzy rule bases that are developed are used to formulate the fuzzy-LP problem, where now the uncertainty lies with the load component response and in the composition coefficients.;The performance of all these approaches is presented along with the detailed formulation. Particular emphasis is laid on the formulation of ANFIS and its application to estimating load composition. The application of fuzzy numbers in representing uncertainty has been shown to be very valuable and the fuzzy LP approach a robust estimation technique.