A hybrid deep learning model based on Rime optimization and multi-head attention for cooling load prediction in public buildings
Affiliation of Author(s):
建筑设备科学与工程学院
Key Words:
Cooling load prediction
Large public building
Bidirectional long short-term memory
Multi-head self-attention mechanism
Improving the rime optimization algorithm
Deep learning
Abstract:
Accurate prediction of cooling load is a prerequisite for the control of air conditioning systems and is of great
significance for energy saving in buildings. This study proposes a hybrid deep learning model (AHRIME-CNNBiLSTM-MHSA) based on Rime optimization and Multi-Head Attention for cooling load prediction in large public
buildings. Firstly, an importance analysis of input features is conducted using random forests, and key features
are selected to reduce the dimensionality of input feature parameters. Secondly, a hybrid deep learning model is
established by integrating data extraction with Convolutional Neural Network (CNN), bidirectional learning of
temporal features with Bidirectional Long Short-Term Memory (BiLSTM), and the mining of key information
with the Multi-Head Self-Attention mechanism (MHSA). Based on this, the improved Adaptive Hybrid Rime
Optimization algorithm (AHRIME) is employed to optimize the hybrid model, determining the AHRIME-CNNBiLSTM-MHSA model with the optimal combination of parameters. Finally, the performance of the proposed
cooling load prediction model is validated using actual data by establishing control groups. The results indicate
that CNN and MHSA can enhance the effectiveness of BiLSTM in processing information and improve prediction
accuracy. AHRIME can obtain the optimal parameters of the model, thereby enhancing predictive performance.
Paper Publications
