Intelligent Buildings

In this elective, two assignments were completed. The first assignment was related data analysis, several python packages were used to obtain data visualisation for observational studies. The second assignment was related to MPC and ML. The outcomes of the 2nd assignment was an algorithms that delivers highest accuracy of prediction. 

Intelligent Buildings  

Intelligent buildings equipped with sensors and meters are becoming common, large amounts of data are being collected by these devices.  Smart residential and commercial buildings utilize multiple sensors to collect or generate huge amounts of data every day, especially in HVAC systems. These datasets provide operation information and insights to not only from building managers, but also building occupants. 

In the most recent technology trends, with the integration of Machine Learning (Katić et al, 2018), Internet of Things and Cloud Technology (Wen & Mishra, 2018), engineers and designers are able to analyse data from multiple resources in order to obtain a predictive model, which aims to improve building performance in terms of energy efficiency and thermal comfort.

Data Analysis

Python has evolved to be a great platform for data analysis. There is a whole ‘ecosystem’ or stack of packages which together provide a comprehensive toolkit for most kinds of data analysis.

In the first assignment, Jupyter Notebook was the chosen platform to process all the datasets. Several packages were used to perform different tasks, including data analysis, visualisation and statistical calculations.

ML in MPC system 

The training exercise was executed in MATLAB by using Classification Learner App, it is an app in the Statistics and Machine Learning Toolbox that allows training models to classify data by using supervised machine learning. Several algorithms are included in this app but mainly two categories of algorithms (SVM and Ensemble) were applied in the research. In order to achieve a predictive model that scores the highest accuracy, parameter fine-tuning was iteratively performed.

- This figure shows the procedures of performing a machine learning training and obtain a predictive model.
- The practise and results please see assignment 2.

Challenges we face

In order to implement the Model Predictive Control strategies in the development of smart building control systems, we yet need to solve the following problems: 

1. Preparing reliable data for model training

Created with Sketch.

-  Missing building information & HVAC specifications 
-  Unreliable & non- calibrated sensors

2. Creating thermal model for predictive control 

Created with Sketch.

-  Finding optimal complexity, avoiding over parametrisation & non-representativeness 
-  Comparing model prediction to measurements - avoiding over / under - fitting  
-  Scalability & adaptability - Creating replicable models to tramline the process
-  Control granularity - limited availability of sensors & actuation points

3. Implementing MPC

Created with Sketch.

-  Dealing with uncertainties - measurements, model, weather conditions and building occupants 
-  Multiple conflicting optimisation objectives 


1. Wen, J. T., & Mishra, S. (2018). Intelligent Building Control Systems: A Survey of Modern Building Control and Sensing Strategies (Advances in Industrial Control) 2018 Edition. ISBN-13, 978-3319684611.
2. Kim, J. (2018). Advancing comfort technology and analytics to personalize thermal experience in the built environment. Retrieved November 27, 2019, from website:
3. Schein, J., & Bushby, S. T. (2006). A hierarchical rule-based fault detection and diagnostic method for HVAC systems. Hvac&r Research, 12(1), 111-125.
4. Katić, K., Li, R., Verhaart, J., & Zeiler, W. (2018). Neural network based predictive control of personalized heating systems. Energy and Buildings, 174, 199-213.