딥러닝 기반의 가정용 가스 수요 예측 모델 개발

초록

Measurement of the residential gas usage by human had some problems such as time consuming process, non-realtime because it requires humans to visit every home. In recent years, the smart gas meter, which combines traditional gas meter with internet of things technology, is having attention as a solution that covers the restriction of the passive methods. However, the malicious usage, or packet loss by any physical communication error, is emerging as the critical problems. To solve the problem, a gas demand prediction can be used as one of the solutions because the predicted value could be used as the reference to determine whether there are the errors of the measured values. Based on the fact, we proposed the residential gas demand prediction model based on long short term memory that is one of deep learning algorithms. To demonstrate and evaluate the proposed model, we collected the real residential gas usage data for 100 households from July to September, and realized that there are usage patterns depending on the date, time of day and week. With the facts, we trained the proposed model to predict the gas demand after 48 hours from the input. As the result of that, we achieved 96.08% accuracy.

키워드