Categorize the Power Grid Projects with SOM Method

Unsupervised learning is to find the potential features without any supervised information, which is much more extensive and robust in the real applications. In this paper, we proposed a self organized map(SOM) based method to categorize the power grid areas based on their features. Further, we analyze the areas in certain categories, which have the similar features that show the voltage transformers are overwhelming. Our SOM based method can significantly find the right categories, in comparison to the KMeans, the DBSCAN, and the Birch. Since the SOM based method is computing cost, we employed a static graphic to build the algorithm and then implement and run it with tensorflow. We evaluate our method with the Calinski-Harabaz and the Silhouette-Coefficient. The experimental studies demonstrate that our method has a better accuracy.