In this article, Senior product marketing manager for the Simulink platform at MathWorks, Michael Carone, explains how to use Artificial Intelligence (AI) to improve the performance of HVAC systems.
According to a recent report from the McKinsey Global Institute, Artificial Intelligence (AI) could potentially deliver $13 trillion of global activity by 2030.
This amounts to an additional 1.2% of GDP annual growth, which would be significantly more than the 0.3% growth we saw from the steam engine that sparked the dawn of the industrial revolution.
Yet with all this potential, AI is still in its infancy. Companies are struggling with AI and that's because too many people are focused on just the intelligent AI algorithms.
There are three other requirements (or three other I’s, if you will), to being successful with artificial intelligence:
1. The discovery and use of insights from domain experts in applications where the AI will be used;
2. Tools to handle the implementation details across the entire design workflow, not just the AI piece;
3. Ensuring that there is an effective interaction between the AI and the surrounding environment, especially with people.
To illustrate these three requirements for building a successful AI, I’d like to use BuildingIQ as an example.
BuildingIQ is using MATLAB to build a cloud-hosted system that optimises the energy consumption of a building’s HVAC system.
Data is used from external sources, such as a temperature forecast and an electricity price forecast, to minimise the building’s energy consumption repeatedly with a 12-hour predictive horizon. This approach has reduced energy used by HVAC systems by 10 to 25 per cent.
Let’s go back to the first I, insights. Applying insights is about what we as engineers and scientists bring to the AI, not what the AI brings to us. We use our insights when selecting data, making tradeoffs, and evaluating results.
When working with AI, it’s important that you can use tools so that your insights are an integral part of the work. BuildingIQ does this by filtering out data, looking at poles and zeroes, and running nonlinear optimisations so that the data coming in and out of the AI algorithm is sound.
Designing an implementation is about designing the entire solution, not just the AI. If you’re a researcher, that means testing, data analysis, and reporting.
If you’re building a car, it’s about requirements gathering, modeling and simulation, and verification and validation. And if you’re creating a climate control system as is the case with BuildingIQ, it’s about designing an HVAC control system and projecting weather and energy prices that work with the AI.
The third requirement is ensuring that the AI interacts effectively with the surrounding environment and within complex human workflows. This can mean different things based on the industry and application. For example, in a car, the AI can avoid accidents, but needs to do it in a smooth way to ensure that riders still have a pleasant experience.
In a HVAC example, BuildingIQ offers a mobile app that provides information and flexibility so that people can adjust operating parameters to optimise their comfort levels.
Success with AI requires more than just developing intelligent algorithms. We must also apply insights from domain experts, design the AI within a complete implementation, and ensure the AI can interact well in its surrounding environment.
Consider your own projects and how you can use AI combining data science with engineering.