Spatial Predictive Modeling of Power Outages Resulting from Distribution Equipment Failure: A Case of Thailand

This research develops a location-based predictive model for distribution equipment failure for use in preventative maintenance scheduling and planning. This study focuses on equipment-related failures because they are one of the main causes of outages in Thailand. Geographic Information Systems (GIS) data was integrated with asset data to predict the equipment failure of distribution equipment. Data on assets and outages from the Provincial Electricity Authority (PEA) was merged with GIS data from multiple sources, including elevation data, weather data, natural landmarks, and points of interest (POIs). Data was split into four regional datasets, and Random Forests (RF) feature selection and structural equation modeling was used to identify and confirm the most important features in each region. Logistic regression and RF regression were then used to estimate failures. RF regression was more effective than logistic regression at estimating equipment failure. The asset age and electrical load were significant predictors of outages. There were also geographic features that were significant predictors in each region, but which features affected outages varied by region. Thus, the study concluded that the approach developed could be used in preventative maintenance planning with some modification for regional characteristics, including geographic location and patterns of urbanization and industrialization.