# Simulation and Optimisation of Air Conditioning Systems using Machine Learning

**Authors:** Rakshitha Godahewa, Chang Deng, Arnaud Prouzeau, Christoph Bergmeir  
**arXiv:** [2006.15296](https://arxiv.org/abs/2006.15296)  
**Date:** 2020-06-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2006.15296v1/

## Key points

- Use of machine learning approaches to optimize setpoints in unoccupied rooms
- Static thermal setpoints lead to energy wastage and increased expenses
- Proposal of a deep-learning model based on Recurrent Neural Networks (RNN)
- RNNs enable learning across short series and focus on specific operation modes of AC systems
- Evaluation of RNN model against state-of-the-art models shows significant improvement in prediction accuracy
- Analysis of RNN model usage in optimizing energy consumption in a real-world scenario using temperature data from a university lecture theatre
- Implementation of RNN model leads to savings of around 20% compared to traditional temperature controlling models without optimization techniques
- Innovative approach leveraging machine learning for optimizing air conditioning systems
- Potential for substantial energy savings and cost reduction through intelligent temperature control during unoccupied periods in buildings

## Summary

This paper titled "Simulation and Optimization of Air Conditioning Systems using Machine Learning" explores the use of machine learning approaches to optimize the setpoints used in a particular room during its unoccupied periods. In building management, static thermal setpoints are typically used to maintain the inside temperature of a building at a comfortable level regardless of occupancy. However, this strategy often leads to significant energy wastage and increased energy expenses. To address this issue, the authors propose a deep-learning model based on Recurrent Neural Networks (RNN). This model can predict the temperatures of future periods when a specific room is unoccupied. By using these predicted temperatures, optimal thermal setpoints for the room during its unoccupied period can be defined. The authors demonstrate that RNNs are particularly suitable for this learning task as they enable learning across many relatively short series. This capability is necessary to focus on specific operation modes of the air conditioning (AC) system. The prediction accuracy of their RNN model is evaluated against state-of-the-art models and shows significant improvement. Furthermore, the usage of the RNN model in optimizing energy consumption in a real-world scenario is analyzed using temperature data from a university lecture theatre. Based on simulations, it is shown that implementing the RNN model can lead to savings of around 20% compared to traditional temperature controlling models that do not utilize optimization techniques. Overall, this paper presents an innovative approach to optimizing air conditioning systems by leveraging machine learning techniques. The results highlight the potential for substantial energy savings and cost reduction through intelligent temperature control during unoccupied periods in buildings.