Sign in Subscribe
Energy management

Hybrid CNN-LSTM Model Achieves Superior Performance in Chiller Power Consumption Forecasting

Strategic Insights

— 2 min read

Researchers at the University of Malaysia Pahang have developed a novel approach to forecasting chiller power consumption in commercial buildings, using a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model optimized by the Barnacles Mating Optimizer (BMO). This innovative method has demonstrated superior performance in optimizing building energy management systems and reducing operational costs. According to the researchers, the proposed model achieves the lowest Root Mean Square Error (RMSE) of 0.5523 and the highest R² value of 0.9435, showing statistically significant improvements over other optimization methods.

Key Takeaways:

  • The proposed CNN-LSTM-BMO model achieves superior performance in chiller power consumption forecasting, with an RMSE of 0.5523 and an R² value of 0.9435.
  • The model demonstrates robust convergence characteristics and superior generalization capability, making it suitable for real-world applications in building energy management systems.
  • The study compares the proposed CNN-LSTM-BMO model against other metaheuristic optimization algorithms, including Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Differential Evolution (DE).
  • The CNN-LSTM-BMO model shows statistically significant improvements over other optimization methods, as confirmed by paired t-tests (P < 0.05).
  • The research contributes to the advancement of accurate and efficient chiller power consumption forecasting methodologies, offering practical implications for Heating, Ventilation, and Air Conditioning (HVAC) system optimization and energy efficiency improvements in commercial buildings.
  • The study was supported by Umpsa and conducted by researchers at the University of Malaysia Pahang, including Zuriani Mustaffa and Mohd Herwan Sulaiman.

Statistics:

  • The CNN-LSTM-BMO model achieves an RMSE of 0.5523, which is lower than other optimization methods.
  • The CNN-LSTM-BMO model achieves an R² value of 0.9435, which is the highest among other optimization methods.
  • The study compares the proposed CNN-LSTM-BMO model against five other metaheuristic optimization algorithms.
  • The models were evaluated using comprehensive performance metrics, including RMSE, R², and paired t-tests.
  • The study was published in the journal Next Energy in 2025, with the article available for free at https://doi-org.sdpl.idm.oclc.org/10.1016/j.nxener.2025.100321.

Sources:

  • NewsRx. Researchers from University of Malaysia Pahang Detail New Studies and Findings in the Area of Networks (Chiller power consumption forecasting for commercial building based on hybrid convolution neural networks-long short-term memory model with ...). Journal of Engineering. July 7, 2025; p 4450.
  • Chiller power consumption forecasting for commercial building based on hybrid convolution neural networks-long short-term memory model with barnacles mating optimizer. Next Energy, 2025,8():100321. The publisher for Next Energy is Elsevier.

Read more