| Title |
Automated Machine Learning Pipeline System Based on Beam Search for Electric Power Industry |
| Authors |
장광선(Gwangseon Jang) ; 황명하(Myeong-Ha Hwang) |
| DOI |
https://doi.org/10.5370/KIEE.2021.70.12.1914 |
| Keywords |
Automated machine learning; AutoML; Beam search; Electric power industry |
| Abstract |
As Artificial Intelligence (AI) shows excellent performances, electric power industry also applies AI to various fields. Even though a lot of data and infrastructures for data analysis are prepared, AI experts in the industry are insufficient. Automated machine learning can be a solution in the industry to apply the excellence of AI to many fields despite the shortage of professionals. Recently, several automated machine learning services have shown good performances on general purpose. However, due to the specificity of data, they does not perform well in the electric power industry. Therefore, we develop an automated machine learning pipeline system based on beam search especially for electric power industry. The proposed system is applied to three real-word problems, one for each of regression, classification, and text classification. The significance of our work is first to apply automated machine learning to electric power industry with high performance. The performances of the models to predict hourly peak power demand and to detect illegal use of electricity for Bitcoin mining are improved by 3.42% and 3.23% respectively compared to the existing models. Moreover, our work shows 95.23% accuracy in classification of the questions’ indents of chatbot, which gives the possibility to replace the existing model. |