Customer load characteristics provides very critical information for power system operation. The accuracy of load forecasting and the effective cases of system generation and network planning can be enhanced by the investigation of customer load characteristics. In this thesis, the intelligent meters have been used to collect the power consumption within each 15 minutes of study customers, which have been selected by stratified sampling method. The typical load patterns of each customer class have been derived. The load composition and the power system load profile of Fengshan District in Taipower have been solved by the typical load patterns and the power consumption of each custom class. To investigate the distribution system reliability, the service territory of Fengshan district is divided into six service areas. The forecasting of yearly peak loadings for each area over the future 20 years is performed by the time series method based on the historical load data and load characteristics. By using the forced outage rate ¡]FOR¡^ of main transformers in the substations, the loss of load expectation¡]LOLE¡^ which corresponds to yearly peak loading of each service area is solved. By this way, the capacity expansion planning of main transformers to meet the service reliability can therefore be derived. To further enhance the distribution system planning, the capacity transfer capability of main transformers and the tie line flow capacity between different areas are considered too. It is found that the expansion planning of main transformers by the proposed methodology can provide better cost effectiveness of transformer investment to satisfy the service reliability as well as the system peak loading.

All generation facilities have to report their generator un-availabilities to their respective Independent System Operators (ISOs). The un-availability of a generator is determined in terms of its probability of failure. Generators may serve the role of two kinds, base units which operates all the time and the others are peaking units which operate only for periods of time depending on load requirement. Calculation of probability of failure for peaking units using standard formulas gives pessimistic results owing to its time spent in the reserve shut down state. Therefore the normal two state representation of a generating unit is not adequate. A four state model was proposed by an IEEE committee to calculate the forced outage rate (unavailability) of such units. This thesis examines the representation of peaking units using a four-state model and performs the analytical calculations and Monte Carlo simulations to examine whether such a model does indeed represent the peaking units properly.