http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Deep learning-based heat optimization techniques for forecasting indoor temperature changes
Latika Jindal,Nitika Vats Doohan,Swati Vaidya,Hemlata Patel,Arpit Deo 대한공간정보학회 2024 Spatial Information Research Vol.32 No.1
The traditional climate compensator technique requires extensive modifcations and signifcantly relies on manual knowledge when it comes to winter heating in buildings. This paper suggests a deep-learning-based optimization procedure for heating systems to improve the initial control strategy. To begin, we recommend using a deep MTDN (Multiple Time Diference Network) to deduce the thermodynamic laws governing the dynamics of indoor temperature changes and make future predictions about the temperature in a given space. The network is and follows physical rules; next, M is used as a simulator, and the evaluation index representing the human thermal reaction issued as the appropriate reward item. Next, a strategy optimizer based on the SAC (Soft Actor-Critic) learning thought algorithm is used to train the simulator to realize a stable and excellent heating control strategy; fnally, experiments are designed using real data from a Tianjin heat exchange station to evaluate the simulator’s ability to predict future outcomes. It is confrmed that the simulator not only has high prediction accuracy but also adheres to the laws of physics and that the strategy optimizer learned approach could guarantee more stable and comfortable indoor temperature in multiple periods of random sampling, as compared to the original strategy.