Transfer learning with deep neural networks for model predictive control of HVAC and natural ventilation in smart buildings

Chen, Yujiao; Tong, Zheming; Zheng, Yang; Samuelson, Holly; Norford, Leslie K.

doi:10.1016/j.jclepro.2019.119866

Cited by 172 publications

(54 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The ANN network was implemented using Keras [17], with TensorFlow [18] as the backend. According to Chen et al [19], the design of neural networks has several aspects of concern. A model should have sufficient width and depth to capture the underlying pattern of the data.…”

Section: Methodsmentioning

confidence: 99%

Increasing Neurons or Deepening Layers in Forecasting Maximum Temperature Time Series?

2020

View full text Add to dashboard Cite

Weather forecasting, especially that of extreme climatic events, has gained considerable attention among researchers due to their impacts on natural ecosystems and human life. The applicability of artificial neural networks (ANNs) in non-linear process forecasting has significantly contributed to hydro-climatology. The efficiency of neural network functions depends on the network structure and parameters. This study proposed a new approach to forecasting a one-day-ahead maximum temperature time series for South Korea to discuss the relationship between network specifications and performance by employing various scenarios for the number of parameters and hidden layers in the ANN model. Specifically, a different number of trainable parameters (i.e., the total number of weights and bias) and distinctive numbers of hidden layers were compared for system-performance effects. If the parameter sizes were too large, the root mean square error (RMSE) would be generally increased, and the model’s ability was impaired. Besides, too many hidden layers would reduce the system prediction if the number of parameters was high. The number of parameters and hidden layers affected the performance of ANN models for time series forecasting competitively. The result showed that the five-hidden layer model with 49 parameters produced the smallest RMSE at most South Korean stations.

show abstract

Section: Methodsmentioning

confidence: 99%

Increasing Neurons or Deepening Layers in Forecasting Maximum Temperature Time Series?

2020

View full text Add to dashboard Cite

show abstract

“…Rather than training the RL controller on every individual building, it would be more efficient and scalable if we could train the RL controller on a small number of buildings and then apply them to larger building stocks. Transfer learning technique has been used in MPC based building controls [28], however, no successful application of transfer learning is found in the RL-based building controls. Vázquez-Canteli and Nagy (2019) [29] reviewed the use of reinforcement learning for demand response applications.…”

Section: Previous Reviewsmentioning

confidence: 99%

Reinforcement learning for building controls: The opportunities and challenges

2020

View full text Add to dashboard Cite

Building controls are becoming more important and complicated due to the dynamic and stochastic energy demand, on-site intermittent energy supply, as well as energy storage, making it difficult for them to be optimized by conventional control techniques. Reinforcement Learning (RL), as an emerging control technique, has attracted growing research interest and demonstrated its potential to enhance building performance while addressing some limitations of other advanced control techniques, such as model predictive control. This study conducted a comprehensive review of existing studies that applied RL for building controls. It provided a detailed breakdown of the existing RL studies that use a specific variation of each major component of the Reinforcement Learning: algorithm, state, action, reward, and environment. We found RL for building controls is still in the research stage with limited applications (11%) in real buildings. Three significant barriers prevent the adoption of RL controllers in actual building controls: (1) the training process is time consuming and data demanding, (2) the control security and robustness need to be enhanced, and (3) the generalization capabilities of RL controllers need to be improved using approaches such as transfer learning. Future research may focus on developing RL controllers that could be used in real buildings, addressing current RL challenges, such as accelerating training and enhancing control robustness, as well as developing an opensource testbed and dataset for performance benchmarking of RL controllers.

show abstract

“…The genetic algorithm (GA) is an optimization methodology inspired by the “survival of the fittest” rule in the theory of evolution. It is an efficient and robust method to solve complex optimization problems [ 32 , 33 , 34 , 35 ]. In gear-related problems, it has been used to optimize the tooth profile [ 35 ], maintain the reliability of gear transmission [ 36 ], and improve the load-sharing performance of planetary gears [ 37 ].…”

Section: Optimization Schemementioning

confidence: 99%

Residual Stress Distribution Design for Gear Surfaces Based on Genetic Algorithm Optimization

et al. 2021

Self Cite

View full text Add to dashboard Cite

The rolling contact fatigue of gear surfaces in a heavy loader gearbox is investigated under various working conditions using the critical plane-based multiaxial Fatemi–Socie criterion. The mechanism for residual stress to increase the fatigue initiation life is that the compressive residual stress has a negative normal component on the critical plane. Based on this mechanism, the genetic algorithm is used to search the optimum residual stress distribution that can maximize the fatigue initiation life for a wide range of working conditions. The optimum residual stress distribution is more effective in increasing the fatigue initiation life when the friction coefficient is larger than its critical value, above which the fatigue initiation moves from the subsurface to the surface. Finally, the effect on the fatigue initiation life when the residual stress distribution deviates from the optimum distribution is analyzed. A sound physical explanation for this effect is provided. This yields a useful guideline to design the residual stress distribution.

show abstract

Transfer learning with deep neural networks for model predictive control of HVAC and natural ventilation in smart buildings

Cited by 172 publications

References 29 publications

Increasing Neurons or Deepening Layers in Forecasting Maximum Temperature Time Series?

Increasing Neurons or Deepening Layers in Forecasting Maximum Temperature Time Series?

Reinforcement learning for building controls: The opportunities and challenges

Residual Stress Distribution Design for Gear Surfaces Based on Genetic Algorithm Optimization

Contact Info

Product

Resources

About