Your browser does not support JavaScript!

Oleksandr Zdir

BMS Engineer


Category

Other

Present and future of Machine Learning in Building Management Systems

5 Minutes

Send to your friends

Share

  1. What is Machine Learning?
  2. Areas where ML proved to be useful in Building Management Systems.
  3. Future of Machine Learning in BMS.


What is Machine Learning?

 

The general term ‘Machine Learning’ denotes a variety of mathematical, statistical and computational methods for developing algorithms that can solve a problem not in a direct way, but based on searching for patterns in a variety of input data. The description may be confusing if you see it for the first time so let’s start with an example. Imagine you are showing your 3-year old son a dog on the street. You want him to be able to distinguish dogs from other animals. You definitely can not show him all the possible dog breeds, but after a few encounters, he will probably be able to identify the dog with high accuracy even if he will see a completely new breed. How is this possible? Our brain is a complex machine that can connect different parts of gathered information, create patterns and calculate the conclusion. You showed your son a few examples, and now he is good to go. Perfect. Can we do the same with machines? Yes, and that's what we call ‘Machine Learning’. We are creating an algorithm that analyses different data points and expects it to indicate exceptions from the pattern. The more data a computer has to analyse the better chances of successful prediction. And the key is that the decision is calculated not according to a clear formula, but according to the established dependence of the results on a specific set of characteristics and their values. For example, if the figure is round-shaped and has no corners or edges it is probably a circle.

 

Picture 1. Example of data identification.

 

Therefore, Machine Learning is used for diagnostics, forecasting, recognition and decision-making in various applied areas: from medicine to banking. For example, your bank may block your account if the transactions are way out from your usual behaviour. Normally your expenses are 65% of your income and all the previous transactions were made in the US, and all of a sudden all your funds are being transferred to a random PayPal account registered in Mexico. It was not a bank employee who decided to block your account, but a previously programmed Machine Learning algorithm detected sufficient deviation from your 'usual behaviour and pre-blocked your account with possible acceptance by a human employee. That's Machine Learning in action.

 

 

Machine Learning in Building Management systems.

 

The major aim of the Building Management System (BMS) is to reduce energy usage in a facility by optimising heating, cooling, ventilation and lighting. Controls Engineers and facility operators are doing their best to keep the system running optimally. But the problems with humans is that their ability to analyse large amounts of data is limited. Nobody would stand at the entrance of the office and count every entering employee and visitor to calculate the precise occupancy in the building and set the Air Handle Unit performance accordingly. But Machine Learning (ML) algorithms can take into consideration thousands of data points: Outside Air Temperature; Humidity; Weather Forecast; Occupancy; Actual Temperatures in the offices; Number of people who entered the building and so on. With that data, ML  can make better predictions and detect anomalies which would be probably overlooked by maintenance stuff.

Imagine how useful ML can be in an unexpected change in tenants behaviour. For example, the CEO of the company has a birthday and invites the whole company for a coffee in the conference room. No building operator knows about this. But the algorithm detected that the level of CO2 decreased on the whole floor but increased in the conference room. The comfort parameters have been adjusted, more fresh air is supplied, everyone feels comfortable and some amount of energy is being saved on uninhabited areas of the office.

The ML algorithms can be especially useful in malfunction detection. Their huge advantage is that they are not limited to the detection of specific, predetermined issues. Instead, they can scan all building data and metrics, and report any anomaly it encounters. Sensor failure, increase in energy usage, leakage of liquids and much more.

 

Picture 2. Example of malfunction detection.

 

 

Future of Machine Learning in BMS.

 

In the world of growing concerns about climate change governments are trying to push-forward all kinds of technologies that can reduce carbon emissions. Those which have proven effectiveness would inevitably be implemented in the nearest future in developed countries. Among them, Machine Learning. Building controls become more complex year after year and the amount of data is becoming impossible to analyse without computing power.
We already can see successful examples of implementing Machine Learning on a large scale. DeepMind, owned by Google has successfully used machine learning algorithms to reduce the company’s energy consumption in data centres by nearly 40%. Such examples are showing us that, we are getting closer in an attempt to connect Cloud Computing with its advantages with the data we get from Building Operation. It's just the beginning. When it comes to Building Management Systems, it’s more common that, Engineers are making decisions based on Machine Learning hints. The real revolution will occur when Machine Learning would be on that level of safety and precision, that it would have the authority to make decisions by itself.

Although the technology is promising, as always, we have some challenges in its way.

  • Large amounts of data need to be identified, recognized, stored and analysed.
  • Interoperability. We still have multiple vendors of equipment, different standards for tagging data, different types of communication among the BMS systems. Before we can implement Machine Learning on a large scale we need to figure out how to use the different types of data that have been deployed in buildings.
  • Reliability. Before this technology can be widely accepted we need to see more successful statistics, clear regulations and a proven track record. There is a risk of misuse of this technology due to a lack of wide acceptance and therefore experience.
  • Safety. There is always additional risk involved where decisions could be made without supervision by a responsible professional. The wrong set of data or a breach in cybersecurity may cause damage to the Building Management System operation.


We will definitely see more successful examples of the implementation of ML in the nearest future. With significant investments from the market leaders, we may expect resolving challenges we face rather sooner than later. We believe ML can be one of the key drivers of the optimization of Building Operations in the next decade.

Other articles by this author

Other / Smart Buildings

Introduction to communication protocols in Building Management Systems

read the article