KCC 2021
Current Result Document :
| 한글제목(Korean Title) |
Forecasting Building Energy Consumption with Bayesian Optimized Model Architecture on Federated Data |
| 영문제목(English Title) |
Forecasting Building Energy Consumption with Bayesian Optimized Model Architecture on Federated Data |
| 저자(Author) |
Ye Lin Tun
Chu Myaet Thwal
Choong Seon Hong
|
| 원문수록처(Citation) |
VOL 48 NO. 01 PP. 0645 ~ 0647 (2021. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Forecasting energy demand of industrial and residential buildings can be an integral part of every power of plant and energy grid. Energy usage information collected by smart building management systems can be useful for training a deep neural network for energy consumption forecasting. However, manually crafting an architecture for a neural network model requires a lot of effort from experts. Moreover, fine-grained energy consumption data such as day-to-day or hourly usage is locally stored on client buildings. Due to communication restraints and privacy concerns, centralizing clients’ data for model training can be challenging. To address these challenges, we propose a Bayesian optimization based neural architecture search system for energy demand forecasting in a federated learning setting. Bayesian optimization is an iterative optimization method that can be applied to evaluate expensive black-box functions such as hyperparameter tuning. This can be integrated together with federated learning which is a distributed model training strategy that does not require centralized data for training. In this way, we construct an energy consumption forecaster model with Bayesian optimization based automatic neural architecture search system that requires no transmission of training data |
| 키워드(Keyword) |
Federated learning
energy demand prediction
recurrent neural network
Bayesian Optimization
|
| 파일첨부 |
PDF 다운로드
|
