Development of the simple building electric power prediction model with local weather forecast based on clustering and silhouette algorithm

This paper presents the development of the building electric power prediction model with local weather forecast information. Annual electric power usage data of the testbed is analyzed to develop a building electricity prediction model. K-means clustering algorithm is selected as a data mining technique. Silhouette index is applied to validate clustering results. Cluster analysis of total high voltage electric power usage of the testbed is performed. Results show that time parameters such as the season and the day type are important factors to classify the total electric power usage pattern. Further analysis results of the low voltage electric power usage are presented. Correlation analysis of each low voltage usage presents the local weather condition has a huge effect on energy facility power usages. Electric power usages of other systems such as lighting and electronics are affected more by a day type than a weather condition. The building electricity prediction model is developed based on the data mining analysis results. Simplified structure and fast calculation speed of the model help it to be applied in various fields of researches.